file
stringlengths
6
242
name
stringlengths
1
11
kernel
stringclasses
4 values
input
stringlengths
531
14.6k
output
stringlengths
176
17.2k
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmph1gg3k15
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[2600]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<11>; .reg .f32 %f<371>; .reg .b32 %r<90>; .reg .b64 %rd<43>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[43264]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[51200]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd17, [default_function_kernel0_param_0]; ld.param.u64 %rd18, [default_function_kernel0_param_1]; ld.param.u64 %rd19, [default_function_kernel0_param_2]; ld.param.u64 %rd20, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd21, %rd17; cvta.to.global.u64 %rd1, %rd19; add.u64 %rd40, %SPL, 0; mov.u64 %rd23, 0; mov.f32 %f14, 0f00000000; for.size.2 i in range(0, 6, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(0+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd40+(48+i*520)], %rd23; for.size.1 i in range(0, 2, 1): st.local.v2.f32 [%rd40+(2128+i*-2072)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.v2.f32 [%rd40+(576+i*520)], {%f14, %f14}; for.size.2 i in range(0, 4, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(64+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd40+(96+i*520)], {%f14, %f14}; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd40+(1656+i*520)], %rd23; for.size.2 i in range(0, 6, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(104+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd40+(152+i*520)], %rd23; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd40+(1192+i*520)], {%f14, %f14}; for.size.2 i in range(0, 5, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(160+i*8+j*520)], {%f14, %f14}; mov.u32 %r31, 0; for.size.1 i in range(0, 3, 1): st.local.u32 [%rd40+(720+i*520)], %r31; st.local.v2.f32 [%rd40+200], {%f14, %f14}; for.size.1 i in range(0, 3, 1): st.local.u32 [%rd40+(724+i*520)], %r31; st.local.u64 [%rd40+2280], %rd23; for.size.2 i in range(0, 12, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(208+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd40+(304+i*520)], %rd23; for.size.1 i in range(0, 2, 1): st.local.v2.f32 [%rd40+(2384+i*-2072)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.v2.f32 [%rd40+(832+i*520)], {%f14, %f14}; for.size.2 i in range(0, 4, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(320+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd40+(352+i*520)], {%f14, %f14}; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd40+(1912+i*520)], %rd23; for.size.2 i in range(0, 6, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(360+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd40+(408+i*520)], %rd23; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd40+(1448+i*520)], {%f14, %f14}; for.size.2 i in range(0, 5, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(416+i*8+j*520)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd40+(976+i*520)], %r31; st.local.v2.f32 [%rd40+456], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd40+(980+i*520)], %r31; for.size.2 i in range(0, 7, 1): for.size.1 j in range(0, 5, 1): st.local.v2.f32 [%rd40+(464+i*8+j*520)], {%f14, %f14}; cvta.to.global.u64 %rd3, %rd18; cvta.to.global.u64 %rd4, %rd20; mov.u32 %r1, %ctaid.x; mul.lo.s32 %r2, %r1, 102400; mov.u32 %r3, %tid.x; mul.hi.s32 %r32, %r3, 1321528399; shr.u32 %r33, %r32, 31; shr.s32 %r34, %r32, 2; add.s32 %r5, %r34, %r33; mul.lo.s32 %r35, %r5, 13; sub.s32 %r36, %r3, %r35; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(4+i*2), %r(36+i*-31), (13+i*627); add.s64 %rd5, %rd21, 1664; shl.b32 %r37, %r3, 2; mov.u32 %r38, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r39, %r38, %r37; add.s32 %r7, %r39, 1664; mov.u32 %r83, %r31; $L__BB0_1: bar.sync 0; mad.lo.s32 %r41, %r83, 10816, %r3; mul.wide.s32 %rd24, %r41, 4; add.s64 %rd39, %rd5, %rd24; mov.u32 %r84, %r7; mov.u32 %r85, %r31; $L__BB0_2: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(15+i*1), [%rd39+(-1664+i*208)]; st.shared.f32 [%r84+(-1664+i*208)], %f(15+i*1); add.s64 %rd39, %rd39, 3328; for.size.1 i in range(0, 2, 1): add.s32 %r(84+i*1), %r(84+i*1), (3328+i*-3312); setp.ne.s32 %p1, %r85, 208; @%p1 bra $L__BB0_2; shl.b32 %r43, %r83, 6; add.s32 %r14, %r43, %r2; mov.u32 %r86, 0; bra.uni $L__BB0_4; $L__BB0_13: add.s32 %r86, %r86, 4; $L__BB0_4: mad.lo.s32 %r16, %r86, 52, %r3; setp.gt.s32 %p2, %r16, 12799; shl.b32 %r44, %r16, 2; mov.u32 %r45, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r17, %r45, %r44; @%p2 bra $L__BB0_6; shl.b32 %r46, %r16, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(47+i*1), %r(46+i*-30), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(49+i*1), %r(14+i*35), %r(48+i*-1); mul.wide.s32 %rd25, %r50, 4; add.s64 %rd26, %rd3, %rd25; ld.global.nc.f32 %f31, [%rd26]; st.shared.f32 [%r17], %f31; $L__BB0_6: add.s32 %r18, %r16, 52; setp.gt.s32 %p3, %r18, 12799; @%p3 bra $L__BB0_8; shl.b32 %r51, %r18, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(52+i*1), %r(51+i*-33), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(54+i*1), %r(14+i*40), %r(53+i*-1); mul.wide.s32 %rd27, %r55, 4; add.s64 %rd28, %rd3, %rd27; ld.global.nc.f32 %f32, [%rd28]; st.shared.f32 [%r17+208], %f32; $L__BB0_8: add.s32 %r19, %r16, 104; setp.gt.s32 %p4, %r19, 12799; @%p4 bra $L__BB0_10; shl.b32 %r56, %r19, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(57+i*1), %r(56+i*-37), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(59+i*1), %r(14+i*45), %r(58+i*-1); mul.wide.s32 %rd29, %r60, 4; add.s64 %rd30, %rd3, %rd29; ld.global.nc.f32 %f33, [%rd30]; st.shared.f32 [%r17+416], %f33; $L__BB0_10: add.s32 %r20, %r86, 3; setp.eq.s32 %p5, %r20, 247; @%p5 bra $L__BB0_14; mad.lo.s32 %r21, %r20, 52, %r3; setp.gt.s32 %p6, %r21, 12799; @%p6 bra $L__BB0_13; shl.b32 %r61, %r21, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(62+i*1), %r(61+i*-40), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(64+i*1), %r(14+i*50), %r(63+i*-1); mul.wide.s32 %rd31, %r65, 4; add.s64 %rd32, %rd3, %rd31; ld.global.nc.f32 %f34, [%rd32]; st.shared.f32 [%r17+624], %f34; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r66, 0; mov.u32 %r87, %r66; $L__BB0_15: mad.lo.s32 %r68, %r87, 169, %r4; add.s32 %r24, %r87, %r6; shl.b32 %r69, %r68, 2; add.s32 %r71, %r38, %r69; for.size.1 i in range(0, 13, 1): ld.shared.f32 %f(1+i*1), [%r71+(0+i*4)]; mov.u32 %r88, %r66; $L__BB0_16: for.size.2 i in range(0, 2, 1): shl.b32 %r(72+i*2), %r(88+i*-15), (6+i*-4); add.s32 %r(73+i*3), %r(24+i*21), %r(72+i*2); for.size.1 i in range(0, 4, 1): ld.shared.f32 %f(35+i*1), [%r76+(10240+i*10240)]; mul.lo.s32 %r77, %r88, 13; mul.wide.s32 %rd33, %r77, 4; add.s64 %rd34, %rd40, %rd33; ld.shared.f32 %f39, [%r76]; for.size.8 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): ld.local.f32 %f(40+i*10+j*2), [%rd34+(0+i*4+j*520)]; fma.rn.f32 %f(41+i*10+j*2), %f(1+j*1), %f(39+j*-4), %f(40+i*10+j*2); st.local.f32 [%rd34+(0+i*4+j*520)], %f(41+i*10+j*2); for.size.3 j in range(0, 3, 1): ld.local.f32 %f(44+i*10+j*2), [%rd34+(1040+i*4+j*520)]; fma.rn.f32 %f(45+i*10+j*2), %f(1+j*1), %f(36+j*1), %f(44+i*10+j*2); st.local.f32 [%rd34+(1040+i*4+j*520)], %f(45+i*10+j*2); add.s32 %r88, %r88, 1; setp.ne.s32 %p7, %r88, 10; @%p7 bra $L__BB0_16; add.s32 %r87, %r87, 1; setp.lt.u32 %p8, %r87, 64; @%p8 bra $L__BB0_15; add.s32 %r83, %r83, 1; setp.lt.u32 %p9, %r83, 8; @%p9 bra $L__BB0_1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(79+i*1), %r(1+i*4), (33800+i*-32110), %r(4+i*75); mul.wide.s32 %rd35, %r80, 4; add.s64 %rd36, %rd1, %rd35; add.s64 %rd42, %rd36, 54104; mul.lo.s32 %r81, %r5, 10; mad.lo.s32 %r82, %r1, 200, %r81; mul.wide.s32 %rd37, %r82, 4; add.s64 %rd38, %rd4, %rd37; add.s64 %rd41, %rd38, 320; mov.u32 %r89, 0; $L__BB0_20: for.size.1 i in range(0, 4, 1): ld.global.nc.f32 %f(170+i*1), [%rd41+(-160+i*160)]; ld.local.f32 %f174, [%rd40]; ld.global.nc.f32 %f175, [%rd41+-320]; add.f32 %f176, %f174, %f175; mov.f32 %f177, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(178+i*3), %f(176+i*4), %f177; st.global.f32 [%rd42+(-54104+i*27040)], %f(178+i*3); ld.local.f32 %f(179+i*3), [%rd40+(520+i*520)]; add.f32 %f(180+i*3), %f(179+i*3), %f(170+i*1); for.size.4 i in range(0, 2, 1): max.f32 %f(184+i*3), %f(183+i*3), %f177; st.global.f32 [%rd42+(-24+i*27040)], %f(184+i*3); ld.local.f32 %f(185+i*3), [%rd40+(1560+i*520)]; add.f32 %f(186+i*3), %f(185+i*3), %f(172+i*1); for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(190+i*15+j*3), %f(189+i*15+j*3), %f177; st.global.f32 [%rd42+(54056+i*4+j*-108156)], %f(190+i*15+j*3); ld.local.f32 %f(191+i*15+j*3), [%rd40+(4+i*4+j*520)]; add.f32 %f(192+i*15+j*3), %f(191+i*15+j*3), %f(175+j*-5); for.size.4 j in range(0, 3, 1): max.f32 %f(196+i*15+j*3), %f(195+i*15+j*3), %f177; st.global.f32 [%rd42+(-27060+i*4+j*27040)], %f(196+i*15+j*3); ld.local.f32 %f(197+i*15+j*3), [%rd40+(1044+i*4+j*520)]; add.f32 %f(198+i*15+j*3), %f(197+i*15+j*3), %f(171+j*1); max.f32 %f370, %f369, %f177; st.global.f32 [%rd42+54104], %f370; for.size.1 i in range(0, 2, 1): add.s64 %rd(42+i*-1), %rd(42+i*-1), (676+i*-672); add.s64 %rd40, %rd40, 52; add.s32 %r89, %r89, 1; setp.ne.s32 %p10, %r89, 10; @%p10 bra $L__BB0_20; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[650]; __shared__ float pad_temp_shared[10816]; __shared__ float placeholder_shared[12800]; for (int ff_inner_init = 0; ff_inner_init < 10; ++ff_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 13; ++xx_inner_init) { int index0 = ff_inner_init * 13 + xx_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 130] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 390] = 0.000000e+00f; compute[index0 + 520] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 208; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 52 + (int)threadIdx.x; int index2 = rc_outer_outer * 10816 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 52 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 247; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x < 12800) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 102400 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 10; ++ff_inner) { for (int xx_inner = 0; xx_inner < 13; ++xx_inner) { int index5 = ff_inner * 13 + xx_inner; int index6 = rc_outer_inner * 169 + (int)threadIdx.x % 13 * 13 + xx_inner; int index7 = (int)threadIdx.x / 13 * 640 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 130] = compute[index5 + 130] + (pad_temp_shared[index6] * placeholder_shared[index7 + 2560]); compute[index5 + 260] = compute[index5 + 260] + (pad_temp_shared[index6] * placeholder_shared[index7 + 5120]); compute[index5 + 390] = compute[index5 + 390] + (pad_temp_shared[index6] * placeholder_shared[index7 + 7680]); compute[index5 + 520] = compute[index5 + 520] + (pad_temp_shared[index6] * placeholder_shared[index7 + 10240]); } } } } for (int ax1_inner = 0; ax1_inner < 10; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x * 33800 + (int)threadIdx.x / 13 * 1690 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + ax3_inner; int index9 = ax1_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x * 200 + (int)threadIdx.x / 13 * 10 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 6760] = max(compute[index9 + 130] + placeholder2[index10 + 40], 0.000000e+00f); T_relu[index8 + 13520] = max(compute[index9 + 260] + placeholder2[index10 + 80], 0.000000e+00f); T_relu[index8 + 20280] = max(compute[index9 + 390] + placeholder2[index10 + 120], 0.000000e+00f); T_relu[index8 + 27040] = max(compute[index9 + 520] + placeholder2[index10 + 160], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpmw3w4xby
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<24>; .reg .f32 %f<92>; .reg .b32 %r<185>; .reg .b64 %rd<59>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd6; cvta.to.global.u64 %rd3, %rd7; mov.u32 %r1, %tid.x; shl.b32 %r65, %r1, 1; mov.u32 %r66, %ctaid.x; mul.lo.s32 %r2, %r66, 13; mul.hi.s32 %r67, %r1, 1321528399; shr.u32 %r68, %r67, 31; shr.s32 %r69, %r67, 2; add.s32 %r4, %r69, %r68; mul.lo.s32 %r70, %r4, 13; sub.s32 %r3, %r1, %r70; mul.hi.s32 %r71, %r65, 1321528399; shr.u32 %r72, %r71, 31; shr.s32 %r73, %r71, 2; add.s32 %r74, %r73, %r72; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(5+i*70), %r74, (169+i*-156); sub.s32 %r6, %r65, %r75; shl.b32 %r76, %r1, 3; mov.u32 %r77, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r7, %r77, %r76; or.b32 %r78, %r65, 1; mul.hi.s32 %r79, %r78, 1321528399; shr.u32 %r80, %r79, 31; shr.s32 %r81, %r79, 2; add.s32 %r82, %r81, %r80; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(8+i*75), %r82, (169+i*-156); sub.s32 %r9, %r78, %r83; and.b32 %r10, %r76, -512; shl.b32 %r84, %r1, 2; mov.u32 %r85, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r12, %r85, %r84; for.size.4 i in range(0, 19, 1): add.s32 %r(86+i*2), %r1, (3250+i*3250); shl.b32 %r(87+i*2), %r(86+i*2), 3; and.b32 %r(13+i*2), %r(87+i*2), -512; and.b32 %r(14+i*2), %r(86+i*2), 63; for.size.3 i in range(0, 2, 1): shl.b32 %r(124+i*2), %r(4+i*-1), (8+i*-6); add.s32 %r(125+i*2), %r(85+i*-8), %r(124+i*2); add.s32 %r(51+i*1), %r(125+i*2), (128000+i*-127896); mov.f32 %f88, 0f00000000; mov.u32 %r181, 0; for.size.1 i in range(0, 3, 1): mov.f32 %f(89+i*1), %f88; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 415; @%p1 bra $L__BB0_3; mad.lo.s32 %r128, %r181, 10816, %r2; for.size.1 i in range(0, 2, 1): add.s32 %r(129+i*1), %r(128+i*1), %r(6+i*-1); mul.wide.s32 %rd9, %r130, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f17, [%rd10]; st.shared.f32 [%r7], %f17; for.size.1 i in range(0, 2, 1): add.s32 %r(131+i*1), %r(128+i*3), %r(9+i*-1); mul.wide.s32 %rd11, %r132, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f18, [%rd12]; st.shared.f32 [%r7+4], %f18; $L__BB0_3: setp.gt.s32 %p2, %r1, 63999; shl.b32 %r54, %r181, 6; @%p2 bra $L__BB0_5; and.b32 %r180, %r1, 63; for.size.1 i in range(0, 2, 1): add.s32 %r(133+i*1), %r(180+i*-47), %r(54+i*-44); mul.wide.s32 %rd13, %r134, 4; add.s64 %rd14, %rd2, %rd13; ld.global.nc.f32 %f19, [%rd14]; st.shared.f32 [%r12], %f19; $L__BB0_5: setp.gt.s32 %p3, %r1, 60749; @%p3 bra $L__BB0_7; for.size.1 i in range(0, 2, 1): add.s32 %r(135+i*1), %r(14+i*121), %r(54+i*-41); mul.wide.s32 %rd15, %r136, 4; add.s64 %rd16, %rd2, %rd15; ld.global.nc.f32 %f20, [%rd16]; st.shared.f32 [%r12+13000], %f20; $L__BB0_7: setp.gt.s32 %p4, %r1, 57499; @%p4 bra $L__BB0_9; for.size.1 i in range(0, 2, 1): add.s32 %r(137+i*1), %r(16+i*121), %r(54+i*-39); mul.wide.s32 %rd17, %r138, 4; add.s64 %rd18, %rd2, %rd17; ld.global.nc.f32 %f21, [%rd18]; st.shared.f32 [%r12+26000], %f21; $L__BB0_9: setp.gt.s32 %p5, %r1, 54249; @%p5 bra $L__BB0_11; for.size.1 i in range(0, 2, 1): add.s32 %r(139+i*1), %r(18+i*121), %r(54+i*-37); mul.wide.s32 %rd19, %r140, 4; add.s64 %rd20, %rd2, %rd19; ld.global.nc.f32 %f22, [%rd20]; st.shared.f32 [%r12+39000], %f22; $L__BB0_11: setp.gt.s32 %p6, %r1, 50999; @%p6 bra $L__BB0_13; for.size.1 i in range(0, 2, 1): add.s32 %r(141+i*1), %r(20+i*121), %r(54+i*-35); mul.wide.s32 %rd21, %r142, 4; add.s64 %rd22, %rd2, %rd21; ld.global.nc.f32 %f23, [%rd22]; st.shared.f32 [%r12+52000], %f23; $L__BB0_13: setp.gt.s32 %p7, %r1, 47749; @%p7 bra $L__BB0_15; for.size.1 i in range(0, 2, 1): add.s32 %r(143+i*1), %r(22+i*121), %r(54+i*-33); mul.wide.s32 %rd23, %r144, 4; add.s64 %rd24, %rd2, %rd23; ld.global.nc.f32 %f24, [%rd24]; st.shared.f32 [%r12+65000], %f24; $L__BB0_15: setp.gt.s32 %p8, %r1, 44499; @%p8 bra $L__BB0_17; for.size.1 i in range(0, 2, 1): add.s32 %r(145+i*1), %r(24+i*121), %r(54+i*-31); mul.wide.s32 %rd25, %r146, 4; add.s64 %rd26, %rd2, %rd25; ld.global.nc.f32 %f25, [%rd26]; st.shared.f32 [%r12+78000], %f25; $L__BB0_17: setp.gt.s32 %p9, %r1, 41249; @%p9 bra $L__BB0_19; for.size.1 i in range(0, 2, 1): add.s32 %r(147+i*1), %r(26+i*121), %r(54+i*-29); mul.wide.s32 %rd27, %r148, 4; add.s64 %rd28, %rd2, %rd27; ld.global.nc.f32 %f26, [%rd28]; st.shared.f32 [%r12+91000], %f26; $L__BB0_19: setp.gt.s32 %p10, %r1, 37999; @%p10 bra $L__BB0_21; for.size.1 i in range(0, 2, 1): add.s32 %r(149+i*1), %r(28+i*121), %r(54+i*-27); mul.wide.s32 %rd29, %r150, 4; add.s64 %rd30, %rd2, %rd29; ld.global.nc.f32 %f27, [%rd30]; st.shared.f32 [%r12+104000], %f27; $L__BB0_21: setp.gt.s32 %p11, %r1, 34749; @%p11 bra $L__BB0_23; for.size.1 i in range(0, 2, 1): add.s32 %r(151+i*1), %r(30+i*121), %r(54+i*-25); mul.wide.s32 %rd31, %r152, 4; add.s64 %rd32, %rd2, %rd31; ld.global.nc.f32 %f28, [%rd32]; st.shared.f32 [%r12+117000], %f28; $L__BB0_23: setp.gt.s32 %p12, %r1, 31499; @%p12 bra $L__BB0_25; for.size.1 i in range(0, 2, 1): add.s32 %r(153+i*1), %r(32+i*121), %r(54+i*-23); mul.wide.s32 %rd33, %r154, 4; add.s64 %rd34, %rd2, %rd33; ld.global.nc.f32 %f29, [%rd34]; st.shared.f32 [%r12+130000], %f29; $L__BB0_25: setp.gt.s32 %p13, %r1, 28249; @%p13 bra $L__BB0_27; for.size.1 i in range(0, 2, 1): add.s32 %r(155+i*1), %r(34+i*121), %r(54+i*-21); mul.wide.s32 %rd35, %r156, 4; add.s64 %rd36, %rd2, %rd35; ld.global.nc.f32 %f30, [%rd36]; st.shared.f32 [%r12+143000], %f30; $L__BB0_27: setp.gt.s32 %p14, %r1, 24999; @%p14 bra $L__BB0_29; for.size.1 i in range(0, 2, 1): add.s32 %r(157+i*1), %r(36+i*121), %r(54+i*-19); mul.wide.s32 %rd37, %r158, 4; add.s64 %rd38, %rd2, %rd37; ld.global.nc.f32 %f31, [%rd38]; st.shared.f32 [%r12+156000], %f31; $L__BB0_29: setp.gt.s32 %p15, %r1, 21749; @%p15 bra $L__BB0_31; for.size.1 i in range(0, 2, 1): add.s32 %r(159+i*1), %r(38+i*121), %r(54+i*-17); mul.wide.s32 %rd39, %r160, 4; add.s64 %rd40, %rd2, %rd39; ld.global.nc.f32 %f32, [%rd40]; st.shared.f32 [%r12+169000], %f32; $L__BB0_31: setp.gt.s32 %p16, %r1, 18499; @%p16 bra $L__BB0_33; for.size.1 i in range(0, 2, 1): add.s32 %r(161+i*1), %r(40+i*121), %r(54+i*-15); mul.wide.s32 %rd41, %r162, 4; add.s64 %rd42, %rd2, %rd41; ld.global.nc.f32 %f33, [%rd42]; st.shared.f32 [%r12+182000], %f33; $L__BB0_33: setp.gt.s32 %p17, %r1, 15249; @%p17 bra $L__BB0_35; for.size.1 i in range(0, 2, 1): add.s32 %r(163+i*1), %r(42+i*121), %r(54+i*-13); mul.wide.s32 %rd43, %r164, 4; add.s64 %rd44, %rd2, %rd43; ld.global.nc.f32 %f34, [%rd44]; st.shared.f32 [%r12+195000], %f34; $L__BB0_35: setp.gt.s32 %p18, %r1, 11999; @%p18 bra $L__BB0_37; for.size.1 i in range(0, 2, 1): add.s32 %r(165+i*1), %r(44+i*121), %r(54+i*-11); mul.wide.s32 %rd45, %r166, 4; add.s64 %rd46, %rd2, %rd45; ld.global.nc.f32 %f35, [%rd46]; st.shared.f32 [%r12+208000], %f35; $L__BB0_37: setp.gt.s32 %p19, %r1, 8749; @%p19 bra $L__BB0_39; for.size.1 i in range(0, 2, 1): add.s32 %r(167+i*1), %r(46+i*121), %r(54+i*-9); mul.wide.s32 %rd47, %r168, 4; add.s64 %rd48, %rd2, %rd47; ld.global.nc.f32 %f36, [%rd48]; st.shared.f32 [%r12+221000], %f36; $L__BB0_39: setp.gt.s32 %p20, %r1, 5499; @%p20 bra $L__BB0_41; for.size.1 i in range(0, 2, 1): add.s32 %r(169+i*1), %r(48+i*121), %r(54+i*-7); mul.wide.s32 %rd49, %r170, 4; add.s64 %rd50, %rd2, %rd49; ld.global.nc.f32 %f37, [%rd50]; st.shared.f32 [%r12+234000], %f37; $L__BB0_41: setp.gt.s32 %p21, %r1, 2249; @%p21 bra $L__BB0_43; for.size.1 i in range(0, 2, 1): add.s32 %r(171+i*1), %r(50+i*121), %r(54+i*-5); mul.wide.s32 %rd51, %r172, 4; add.s64 %rd52, %rd2, %rd51; ld.global.nc.f32 %f38, [%rd52]; st.shared.f32 [%r12+247000], %f38; $L__BB0_43: bar.sync 0; mov.u32 %r184, 128000; mov.u32 %r182, %r52; mov.u32 %r183, %r51; $L__BB0_44: for.size.9 i in range(0, 2, 1): ld.shared.f32 %f(39+i*9), [%r183+(-128000+i*4)]; ld.shared.f32 %f(40+i*9), [%r182+(-104+i*52)]; fma.rn.f32 %f(41+i*9), %f(40+i*9), %f(39+i*9), %f(88+i*-47); ld.shared.f32 %f(42+i*9), [%r183+(-64000+i*4)]; fma.rn.f32 %f(43+i*9), %f(40+i*9), %f(42+i*9), %f(89+i*-46); ld.shared.f32 %f(44+i*9), [%r183+(0+i*4)]; fma.rn.f32 %f(45+i*9), %f(40+i*9), %f(44+i*9), %f(90+i*-45); ld.shared.f32 %f(46+i*9), [%r183+(64000+i*4)]; fma.rn.f32 %f(47+i*9), %f(40+i*9), %f(46+i*9), %f(91+i*-44); for.size.9 i in range(0, 2, 1): ld.shared.f32 %f(57+i*9), [%r183+(-127992+i*4)]; ld.shared.f32 %f(58+i*9), [%r182+(0+i*52)]; fma.rn.f32 %f(59+i*29), %f(58+i*9), %f(57+i*9), %f(50+i*9); ld.shared.f32 %f(60+i*8), [%r183+(-63992+i*4)]; fma.rn.f32 %f(61+i*28), %f(58+i*9), %f(60+i*8), %f(52+i*9); ld.shared.f32 %f(62+i*7), [%r183+(8+i*4)]; fma.rn.f32 %f(63+i*27), %f(58+i*9), %f(62+i*7), %f(54+i*9); ld.shared.f32 %f(64+i*6), [%r183+(64008+i*4)]; fma.rn.f32 %f(65+i*26), %f(58+i*9), %f(64+i*6), %f(56+i*9); for.size.1 i in range(0, 2, 1): add.s32 %r(183+i*-1), %r(183+i*-1), (16+i*192); add.s32 %r184, %r184, 16; setp.ne.s32 %p22, %r184, 128256; @%p22 bra $L__BB0_44; add.s32 %r181, %r181, 1; setp.lt.u32 %p23, %r181, 8; @%p23 bra $L__BB0_1; ld.param.u64 %rd58, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd57, %rd58; mul.hi.s32 %r179, %r1, 1321528399; shr.u32 %r178, %r179, 31; shr.s32 %r177, %r179, 2; add.s32 %r176, %r177, %r178; mul.wide.s32 %rd53, %r176, 4; add.s64 %rd54, %rd57, %rd53; ld.global.nc.f32 %f71, [%rd54]; add.f32 %f72, %f88, %f71; mov.f32 %f73, 0f00000000; max.f32 %f74, %f72, %f73; add.s32 %r174, %r3, %r2; mad.lo.s32 %r175, %r176, 169, %r174; mul.wide.s32 %rd55, %r175, 4; add.s64 %rd56, %rd3, %rd55; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd56+(0+i*169000)], %f(74+i*3); ld.global.nc.f32 %f(75+i*3), [%rd54+(1000+i*1000)]; add.f32 %f(76+i*3), %f(89+i*1), %f(75+i*3); max.f32 %f(77+i*3), %f(76+i*3), %f73; st.global.f32 [%rd56+507000], %f83; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_inner_s = 0; ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 2; ++ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) { if ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 832) { if ((int)threadIdx.x < 416) { int index0 = (int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s; int index1 = rc_outer_outer * 10816 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) / 13 * 169 + (int)blockIdx.x * 13 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) % 13; pad_temp_shared[index0] = placeholder[index1]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 20; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3250 + (int)threadIdx.x < 64000) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3250 + (int)threadIdx.x; int index3 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3250 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3250 + (int)threadIdx.x & 63); placeholder_shared[index2] = placeholder1[index3]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { int index4 = rc_outer_inner * 13 + (int)threadIdx.x % 13; int index5 = (int)threadIdx.x / 13 * 64 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index5 + 16000]); compute[2] = compute[2] + (pad_temp_shared[index4] * placeholder_shared[index5 + 32000]); compute[3] = compute[3] + (pad_temp_shared[index4] * placeholder_shared[index5 + 48000]); } } int index6 = (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13; T_relu[index6] = max(compute[0] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 42250] = max(compute[1] + placeholder2[index7 + 250], 0.000000e+00f); T_relu[index6 + 84500] = max(compute[2] + placeholder2[index7 + 500], 0.000000e+00f); T_relu[index6 + 126750] = max(compute[3] + placeholder2[index7 + 750], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpkki9s2qm
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[1300]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<2040>; .reg .b32 %r<99>; .reg .b64 %rd<42>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.f32 %f1390, 0f00000000; mov.u32 %r95, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd1+(0+i*4+j*100)], %r95; mov.u32 %r17, %tid.x; mul.hi.s32 %r18, %r17, 1321528399; shr.u32 %r19, %r18, 31; shr.s32 %r20, %r18, 2; add.s32 %r2, %r20, %r19; mul.lo.s32 %r21, %r2, 13; sub.s32 %r22, %r17, %r21; mul.lo.s32 %r1, %r22, 13; cvta.to.global.u64 %rd16, %rd11; for.size.1 i in range(0, 324, 1): mov.f32 %f(1391+i*1), %f1390; $L__BB0_1: setp.gt.s32 %p1, %r17, 103; bar.sync 0; mad.lo.s32 %r24, %r95, 2704, %r17; mul.wide.s32 %rd17, %r24, 4; add.s64 %rd2, %rd16, %rd17; ld.global.nc.f32 %f1301, [%rd2]; shl.b32 %r25, %r17, 2; mov.u32 %r26, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r27, %r26, %r25; for.size.2 i in range(0, 4, 1): st.shared.f32 [%r27+(0+i*2080)], %f(1301+i*1); ld.global.nc.f32 %f(1302+i*1), [%rd2+(2080+i*2080)]; st.shared.f32 [%r27+8320], %f1305; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f1306, [%rd2+10400]; st.shared.f32 [%r27+10400], %f1306; $L__BB0_3: mov.u32 %r96, 0; bra.uni $L__BB0_4; $L__BB0_13: add.s32 %r96, %r96, 4; $L__BB0_4: mad.lo.s32 %r5, %r96, 520, %r17; setp.gt.s32 %p2, %r5, 15999; @%p2 bra $L__BB0_6; shl.b32 %r36, %r5, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(37+i*1), %r(36+i*-19), (-512+i*527); bfi.b32 %r39, %r95, %r38, 4, 28; add.s32 %r40, %r39, %r37; cvta.to.global.u64 %rd18, %rd12; mul.wide.s32 %rd19, %r40, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f1307, [%rd20]; shl.b32 %r41, %r5, 2; mov.u32 %r42, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r43, %r42, %r41; st.shared.f32 [%r43], %f1307; $L__BB0_6: add.s32 %r44, %r5, 520; setp.gt.s32 %p3, %r44, 15999; @%p3 bra $L__BB0_8; shl.b32 %r47, %r5, 5; add.s32 %r48, %r47, 16640; for.size.1 i in range(0, 2, 1): and.b32 %r(49+i*1), %r(48+i*-31), (-512+i*527); xor.b32 %r51, %r50, 8; shl.b32 %r52, %r95, 4; or.b32 %r53, %r51, %r52; add.s32 %r54, %r53, %r49; cvta.to.global.u64 %rd21, %rd12; mul.wide.s32 %rd22, %r54, 4; add.s64 %rd23, %rd21, %rd22; ld.global.nc.f32 %f1308, [%rd23]; shl.b32 %r55, %r5, 2; mov.u32 %r56, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r57, %r56, %r55; st.shared.f32 [%r57+2080], %f1308; $L__BB0_8: add.s32 %r6, %r5, 1040; setp.gt.s32 %p4, %r6, 15999; @%p4 bra $L__BB0_10; shl.b32 %r58, %r6, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(59+i*2), %r(58+i*-41), (-512+i*527); bfi.b32 %r62, %r95, %r61, 4, 28; add.s32 %r63, %r62, %r59; cvta.to.global.u64 %rd24, %rd12; mul.wide.s32 %rd25, %r63, 4; add.s64 %rd26, %rd24, %rd25; ld.global.nc.f32 %f1309, [%rd26]; shl.b32 %r65, %r5, 2; mov.u32 %r66, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r67, %r66, %r65; st.shared.f32 [%r67+4160], %f1309; $L__BB0_10: add.s32 %r7, %r96, 3; setp.eq.s32 %p5, %r7, 31; @%p5 bra $L__BB0_14; mad.lo.s32 %r8, %r7, 520, %r17; setp.gt.s32 %p6, %r8, 15999; @%p6 bra $L__BB0_13; shl.b32 %r69, %r8, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(70+i*2), %r(69+i*-52), (-512+i*527); xor.b32 %r73, %r72, 8; shl.b32 %r74, %r95, 4; or.b32 %r75, %r73, %r74; add.s32 %r76, %r75, %r70; cvta.to.global.u64 %rd27, %rd12; mul.wide.s32 %rd28, %r76, 4; add.s64 %rd29, %rd27, %rd28; ld.global.nc.f32 %f1310, [%rd29]; shl.b32 %r78, %r5, 2; mov.u32 %r79, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r80, %r79, %r78; st.shared.f32 [%r80+6240], %f1310; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r97, 0; $L__BB0_15: mul.lo.s32 %r94, %r2, 400; mov.u32 %r93, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r82, %r97, 169, %r1; shl.b32 %r83, %r82, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(85+i*1), %r(93+i*4), %r(83+i*11); shl.b32 %r87, %r86, 2; mov.u32 %r88, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r89, %r88, %r87; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1311+i*1), [%r(89+i*-4)+0]; for.size.2 i in range(0, 12, 1): fma.rn.f32 %f(1714+i*-1), %f(1312+i*1), %f1311, %f(1714+i*-1); ld.shared.f32 %f(1313+i*1), [%r85+(4+i*4)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(1702+i*-13), %f1324, %f(1311+i*14), %f(1702+i*-13); ld.shared.f32 %f(1325+i*1), [%r89+(64+i*64)]; fma.rn.f32 %f(1701+i*-13), %f1312, %f(1325+i*1), %f(1701+i*-13); fma.rn.f32 %f(1700+i*-13), %f1313, %f(1325+i*1), %f(1700+i*-13); fma.rn.f32 %f(1699+i*-13), %f1314, %f(1325+i*1), %f(1699+i*-13); fma.rn.f32 %f(1698+i*-13), %f1315, %f(1325+i*1), %f(1698+i*-13); fma.rn.f32 %f(1697+i*-13), %f1316, %f(1325+i*1), %f(1697+i*-13); fma.rn.f32 %f(1696+i*-13), %f1317, %f(1325+i*1), %f(1696+i*-13); fma.rn.f32 %f(1695+i*-13), %f1318, %f(1325+i*1), %f(1695+i*-13); fma.rn.f32 %f(1694+i*-13), %f1319, %f(1325+i*1), %f(1694+i*-13); fma.rn.f32 %f(1693+i*-13), %f1320, %f(1325+i*1), %f(1693+i*-13); fma.rn.f32 %f(1692+i*-13), %f1321, %f(1325+i*1), %f(1692+i*-13); fma.rn.f32 %f(1691+i*-13), %f1322, %f(1325+i*1), %f(1691+i*-13); fma.rn.f32 %f(1690+i*-13), %f1323, %f(1325+i*1), %f(1690+i*-13); for.size.14 i in range(0, 22, 1): fma.rn.f32 %f(1676+i*-13), %f1324, %f(1326+i*1), %f(1676+i*-13); ld.shared.f32 %f(1327+i*1), [%r89+(192+i*64)]; fma.rn.f32 %f(1675+i*-13), %f1312, %f(1327+i*1), %f(1675+i*-13); fma.rn.f32 %f(1674+i*-13), %f1313, %f(1327+i*1), %f(1674+i*-13); fma.rn.f32 %f(1673+i*-13), %f1314, %f(1327+i*1), %f(1673+i*-13); fma.rn.f32 %f(1672+i*-13), %f1315, %f(1327+i*1), %f(1672+i*-13); fma.rn.f32 %f(1671+i*-13), %f1316, %f(1327+i*1), %f(1671+i*-13); fma.rn.f32 %f(1670+i*-13), %f1317, %f(1327+i*1), %f(1670+i*-13); fma.rn.f32 %f(1669+i*-13), %f1318, %f(1327+i*1), %f(1669+i*-13); fma.rn.f32 %f(1668+i*-13), %f1319, %f(1327+i*1), %f(1668+i*-13); fma.rn.f32 %f(1667+i*-13), %f1320, %f(1327+i*1), %f(1667+i*-13); fma.rn.f32 %f(1666+i*-13), %f1321, %f(1327+i*1), %f(1666+i*-13); fma.rn.f32 %f(1665+i*-13), %f1322, %f(1327+i*1), %f(1665+i*-13); fma.rn.f32 %f(1664+i*-13), %f1323, %f(1327+i*1), %f(1664+i*-13); fma.rn.f32 %f1390, %f1324, %f1348, %f1390; add.s32 %r97, %r97, 1; setp.ne.s32 %p7, %r97, 16; @%p7 bra $L__BB0_15; add.s32 %r95, %r95, 1; setp.lt.u32 %p8, %r95, 32; @%p8 bra $L__BB0_1; ld.param.u64 %rd38, [default_function_kernel0_param_2]; ld.param.u64 %rd37, [default_function_kernel0_param_3]; add.u64 %rd40, %SPL, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.f32 [%rd40+(0+i*4+j*100)], %f(1714+i*-13+j*-1); mul.lo.s32 %r91, %r2, 25; cvta.to.global.u64 %rd30, %rd37; mul.wide.s32 %rd31, %r91, 4; add.s64 %rd41, %rd30, %rd31; mad.lo.s32 %r92, %r2, 4225, %r1; cvta.to.global.u64 %rd32, %rd38; mul.wide.s32 %rd33, %r92, 4; add.s64 %rd34, %rd32, %rd33; add.s64 %rd39, %rd34, 24; mov.u32 %r98, 0; $L__BB0_18: ld.global.nc.f32 %f1349, [%rd41]; ld.local.f32 %f1350, [%rd40]; add.f32 %f1351, %f1350, %f1349; mov.f32 %f1352, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(1353+i*3), %f(1351+i*4), %f1352; st.global.f32 [%rd39+(-24+i*4)], %f(1353+i*3); ld.local.f32 %f(1354+i*3), [%rd40+(100+i*100)]; add.f32 %f(1355+i*3), %f(1354+i*3), %f1349; for.size.4 i in range(0, 10, 1): max.f32 %f(1359+i*3), %f(1358+i*3), %f1352; st.global.f32 [%rd39+(-16+i*4)], %f(1359+i*3); ld.local.f32 %f(1360+i*3), [%rd40+(300+i*100)]; add.f32 %f(1361+i*3), %f(1360+i*3), %f1349; max.f32 %f1389, %f1388, %f1352; st.global.f32 [%rd39+24], %f1389; for.size.1 i in range(0, 2, 1): add.s64 %rd(41+i*-1), %rd(41+i*-1), 4; add.s64 %rd39, %rd39, 676; add.s32 %r98, %r98, 1; setp.ne.s32 %p9, %r98, 25; @%p9 bra $L__BB0_18; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[325]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; for (int ff_inner_init = 0; ff_inner_init < 25; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 25] = 0.000000e+00f; compute[index0 + 50] = 0.000000e+00f; compute[index0 + 75] = 0.000000e+00f; compute[index0 + 100] = 0.000000e+00f; compute[index0 + 125] = 0.000000e+00f; compute[index0 + 150] = 0.000000e+00f; compute[index0 + 175] = 0.000000e+00f; compute[index0 + 200] = 0.000000e+00f; compute[index0 + 225] = 0.000000e+00f; compute[index0 + 250] = 0.000000e+00f; compute[index0 + 275] = 0.000000e+00f; compute[index0 + 300] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 520] = placeholder[index2 + 520]; pad_temp_shared[index1 + 1040] = placeholder[index2 + 1040]; pad_temp_shared[index1 + 1560] = placeholder[index2 + 1560]; pad_temp_shared[index1 + 2080] = placeholder[index2 + 2080]; if ((int)threadIdx.x < 104) { pad_temp_shared[index1 + 2600] = placeholder[index2 + 2600]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 31; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x < 16000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 25; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 169 + (int)threadIdx.x % 13 * 13; int index7 = (int)threadIdx.x / 13 * 400 + ff_inner * 16 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 25] = compute[index5 + 25] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 50] = compute[index5 + 50] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 75] = compute[index5 + 75] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 100] = compute[index5 + 100] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 125] = compute[index5 + 125] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 150] = compute[index5 + 150] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 175] = compute[index5 + 175] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 200] = compute[index5 + 200] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); compute[index5 + 225] = compute[index5 + 225] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 250] = compute[index5 + 250] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7]); compute[index5 + 275] = compute[index5 + 275] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7]); compute[index5 + 300] = compute[index5 + 300] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7]); } } } for (int ax1_inner = 0; ax1_inner < 25; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 4225 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 25 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 1] = max(compute[index9 + 25] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 2] = max(compute[index9 + 50] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 3] = max(compute[index9 + 75] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 4] = max(compute[index9 + 100] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 5] = max(compute[index9 + 125] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 6] = max(compute[index9 + 150] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 7] = max(compute[index9 + 175] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 8] = max(compute[index9 + 200] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 9] = max(compute[index9 + 225] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 10] = max(compute[index9 + 250] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 11] = max(compute[index9 + 275] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 12] = max(compute[index9 + 300] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpexfepp8f
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<157>; .reg .b32 %r<31>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[416]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[32000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd5; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; mov.u32 %r2, %ctaid.x; mad.lo.s32 %r3, %r1, 13, %r2; shl.b32 %r15, %r1, 2; mov.u32 %r16, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r16, %r15; shr.u32 %r17, %r1, 3; and.b32 %r18, %r1, 7; bfi.b32 %r5, %r17, %r18, 9, 23; mov.u32 %r19, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r19, %r15; shl.b32 %r20, %r1, 5; add.s32 %r7, %r19, %r20; mov.u32 %r28, 0; mov.f32 %f144, 0f00000000; for.size.1 i in range(0, 12, 1): mov.f32 %f(145+i*1), %f144; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 103; @%p1 bra $L__BB0_3; mad.lo.s32 %r21, %r28, 1352, %r3; mul.wide.s32 %rd9, %r21, 4; add.s64 %rd10, %rd3, %rd9; ld.global.nc.f32 %f53, [%rd10]; st.shared.f32 [%r4], %f53; $L__BB0_3: shl.b32 %r24, %r28, 3; add.s32 %r25, %r5, %r24; mul.wide.s32 %rd11, %r25, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f54, [%rd12]; mov.u32 %r30, 4; for.size.2 i in range(0, 7, 1): st.shared.f32 [%r6+(0+i*4000)], %f(54+i*1); ld.global.nc.f32 %f(55+i*1), [%rd12+(256000+i*256000)]; st.shared.f32 [%r6+28000], %f61; bar.sync 0; mov.u32 %r29, %r16; $L__BB0_4: ld.shared.f32 %f62, [%r29]; add.s32 %r26, %r7, %r30; ld.shared.f32 %f63, [%r26+-4]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(64+i*2), %f(62+i*3), %f63, %f(156+i*-1); ld.shared.f32 %f(65+i*2), [%r29+(4+i*4)]; for.size.2 i in range(0, 11, 1): fma.rn.f32 %f(68+i*2), %f(67+i*2), %f63, %f(154+i*-1); ld.shared.f32 %f(69+i*2), [%r29+(12+i*4)]; ld.shared.f32 %f90, [%r26]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(156+i*-1), %f(89+i*2), %f90, %f(64+i*2); ld.shared.f32 %f(91+i*1), [%r29+(56+i*4)]; for.size.2 i in range(0, 10, 1): fma.rn.f32 %f(154+i*-1), %f(92+i*1), %f90, %f(68+i*2); ld.shared.f32 %f(93+i*1), [%r29+(64+i*4)]; fma.rn.f32 %f144, %f102, %f90, %f88; for.size.1 i in range(0, 2, 1): add.s32 %r(29+i*1), %r(29+i*1), (104+i*-96); setp.ne.s32 %p2, %r30, 36; @%p2 bra $L__BB0_4; add.s32 %r28, %r28, 1; setp.lt.u32 %p3, %r28, 64; @%p3 bra $L__BB0_1; mul.wide.s32 %rd13, %r1, 4; add.s64 %rd14, %rd4, %rd13; mad.lo.s32 %r27, %r1, 169, %r2; ld.global.nc.f32 %f103, [%rd14]; add.f32 %f104, %f156, %f103; mov.f32 %f105, 0f00000000; max.f32 %f106, %f104, %f105; mul.wide.s32 %rd15, %r27, 4; add.s64 %rd16, %rd2, %rd15; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd16+(0+i*52)], %f(106+i*2); add.f32 %f(107+i*2), %f(155+i*-1), %f103; max.f32 %f(108+i*2), %f(107+i*2), %f105; st.global.f32 [%rd16+624], %f130; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[104]; __shared__ float placeholder_shared[8000]; for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = yy_inner_init; compute[index0] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 104) { int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 1352 + (int)threadIdx.x * 13 + (int)blockIdx.x; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 8; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1000 + (int)threadIdx.x; int index4 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 64000 + ((int)threadIdx.x >> 3) * 512 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = yy_inner; int index6 = rc_outer_inner * 13 + yy_inner; int index7 = (int)threadIdx.x * 8 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index1 = (int)threadIdx.x; int index8 = (int)threadIdx.x * 169 + ax2_inner * 13 + (int)blockIdx.x; int index9 = ax2_inner; T_relu[index8] = max(compute[index9] + placeholder2[index1], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpksza1ku9
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<94>; .reg .b32 %r<64>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1280]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r24, %r1, 2; mov.u32 %r25, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r25, %r24; mov.u32 %r3, %ctaid.x; mul.lo.s32 %r26, %r3, 10240; shl.b32 %r27, %r1, 7; and.b32 %r28, %r27, -512; add.s32 %r29, %r28, %r26; and.b32 %r30, %r24, 12; or.b32 %r4, %r29, %r30; shl.b32 %r31, %r1, 4; mov.u32 %r32, _ZZ24default_function_kernel0E18placeholder_shared; for.size.4 i in range(0, 2, 1): add.s32 %r(5+i*1), %r(32+i*3), %r(31+i*-3); or.b32 %r(33+i*3), %r24, (1+i*1); and.b32 %r(34+i*3), %r(33+i*3), (13+i*1); or.b32 %r(35+i*3), %r(34+i*3), %r26; add.s32 %r7, %r38, %r28; or.b32 %r39, %r24, 3; and.b32 %r40, %r39, 15; or.b32 %r41, %r40, %r26; add.s32 %r8, %r41, %r28; mul.hi.s32 %r42, %r1, 1626496491; shr.u32 %r43, %r42, 31; shr.s32 %r44, %r42, 6; add.s32 %r10, %r44, %r43; mul.lo.s32 %r45, %r10, 169; sub.s32 %r9, %r1, %r45; mad.lo.s32 %r46, %r10, 320, %r32; add.s32 %r11, %r46, 136; shl.b32 %r47, %r9, 2; add.s32 %r48, %r25, %r47; add.s32 %r12, %r48, 1352; mov.u32 %r60, 0; mov.f32 %f89, 0f00000000; for.size.1 i in range(0, 4, 1): mov.f32 %f(90+i*1), %f89; $L__BB0_1: bar.sync 0; mad.lo.s32 %r49, %r60, 2704, %r1; mul.wide.s32 %rd9, %r49, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(21+i*1), [%rd10+(0+i*2704)]; st.shared.f32 [%r2+(0+i*2704)], %f(21+i*1); setp.gt.s32 %p1, %r1, 79; @%p1 bra $L__BB0_3; shl.b32 %r50, %r60, 4; for.size.5 i in range(0, 2, 1): add.s32 %r(51+i*1), %r(4+i*2), %r50; mul.wide.s32 %rd(11+i*2), %r(51+i*1), 4; add.s64 %rd(12+i*2), %rd1, %rd(11+i*2); ld.global.nc.f32 %f(25+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r5+(0+i*4)], %f(25+i*1); for.size.5 i in range(0, 2, 1): add.s32 %r(53+i*1), %r(7+i*1), %r50; mul.wide.s32 %rd(15+i*2), %r(53+i*1), 4; add.s64 %rd(16+i*2), %rd1, %rd(15+i*2); ld.global.nc.f32 %f(27+i*1), [%rd(16+i*2)+0]; st.shared.f32 [%r5+(8+i*4)], %f(27+i*1); $L__BB0_3: bar.sync 0; mov.u32 %r63, 136; mov.u32 %r61, %r12; mov.u32 %r62, %r11; $L__BB0_4: for.size.11 i in range(0, 2, 1): ld.shared.f32 %f(29+i*11), [%r62+(-136+i*4)]; ld.shared.f32 %f(30+i*11), [%r61+(-1352+i*676)]; fma.rn.f32 %f(31+i*11), %f(30+i*11), %f(29+i*11), %f(93+i*-62); ld.shared.f32 %f(32+i*11), [%r62+(-72+i*4)]; fma.rn.f32 %f(33+i*11), %f(30+i*11), %f(32+i*11), %f(92+i*-59); ld.shared.f32 %f(34+i*11), [%r62+(-8+i*4)]; fma.rn.f32 %f(35+i*11), %f(30+i*11), %f(34+i*11), %f(91+i*-56); ld.shared.f32 %f(36+i*11), [%r62+(56+i*4)]; fma.rn.f32 %f(37+i*11), %f(30+i*11), %f(36+i*11), %f(90+i*-53); ld.shared.f32 %f(38+i*11), [%r62+(120+i*4)]; fma.rn.f32 %f(39+i*11), %f(30+i*11), %f(38+i*11), %f(89+i*-50); for.size.11 i in range(0, 2, 1): ld.shared.f32 %f(51+i*11), [%r62+(-128+i*4)]; ld.shared.f32 %f(52+i*11), [%r61+(0+i*676)]; fma.rn.f32 %f(53+i*40), %f(52+i*11), %f(51+i*11), %f(42+i*11); ld.shared.f32 %f(54+i*10), [%r62+(-64+i*4)]; fma.rn.f32 %f(55+i*37), %f(52+i*11), %f(54+i*10), %f(44+i*11); ld.shared.f32 %f(56+i*9), [%r62+(0+i*4)]; fma.rn.f32 %f(57+i*34), %f(52+i*11), %f(56+i*9), %f(46+i*11); ld.shared.f32 %f(58+i*8), [%r62+(64+i*4)]; fma.rn.f32 %f(59+i*31), %f(52+i*11), %f(58+i*8), %f(48+i*11); ld.shared.f32 %f(60+i*7), [%r62+(128+i*4)]; fma.rn.f32 %f(61+i*28), %f(52+i*11), %f(60+i*7), %f(50+i*11); for.size.1 i in range(0, 2, 1): add.s32 %r(62+i*-1), %r(62+i*-1), (16+i*2688); add.s32 %r63, %r63, 16; setp.ne.s32 %p2, %r63, 200; @%p2 bra $L__BB0_4; add.s32 %r60, %r60, 1; setp.lt.u32 %p3, %r60, 32; @%p3 bra $L__BB0_1; mul.lo.s32 %r56, %r3, 20; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(57+i*1), %r(10+i*-7), (5+i*3375), %r(56+i*-47); mad.lo.s32 %r59, %r10, 845, %r58; mul.wide.s32 %rd19, %r57, 4; add.s64 %rd20, %rd4, %rd19; ld.global.nc.f32 %f68, [%rd20]; add.f32 %f69, %f93, %f68; mov.f32 %f70, 0f00000000; max.f32 %f71, %f69, %f70; mul.wide.s32 %rd21, %r59, 4; add.s64 %rd22, %rd3, %rd21; for.size.4 i in range(0, 4, 1): st.global.f32 [%rd22+(0+i*676)], %f(71+i*3); ld.global.nc.f32 %f(72+i*3), [%rd20+(4+i*4)]; add.f32 %f(73+i*3), %f(92+i*-1), %f(72+i*3); max.f32 %f(74+i*3), %f(73+i*3), %f70; st.global.f32 [%rd22+2704], %f83; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[5]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[320]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 676] = placeholder[index1 + 676]; pad_temp_shared[index0 + 1352] = placeholder[index1 + 1352]; pad_temp_shared[index0 + 2028] = placeholder[index1 + 2028]; if ((int)threadIdx.x < 80) { int index2 = (int)blockIdx.x * 10240 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4; placeholder_shared[index0 * 4] = placeholder1[index2]; } if ((int)threadIdx.x < 80) { int index3 = (int)threadIdx.x * 4 + 1; int index4 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 1 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 1 & 15); placeholder_shared[index3] = placeholder1[index4]; } if ((int)threadIdx.x < 80) { int index5 = (int)threadIdx.x * 4 + 2; int index6 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 2 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 2 & 15); placeholder_shared[index5] = placeholder1[index6]; } if ((int)threadIdx.x < 80) { int index7 = (int)threadIdx.x * 4 + 3; int index8 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 3 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 3 & 15); placeholder_shared[index7] = placeholder1[index8]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { int index9 = rc_outer_inner * 169 + (int)threadIdx.x % 169; int index10 = (int)threadIdx.x / 169 * 80 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index9] * placeholder_shared[index10]; compute[1] = compute[1] + (pad_temp_shared[index9] * placeholder_shared[index10 + 16]); compute[2] = compute[2] + (pad_temp_shared[index9] * placeholder_shared[index10 + 32]); compute[3] = compute[3] + (pad_temp_shared[index9] * placeholder_shared[index10 + 48]); compute[4] = compute[4] + (pad_temp_shared[index9] * placeholder_shared[index10 + 64]); } } for (int ax1_inner = 0; ax1_inner < 5; ++ax1_inner) { int index11 = (int)blockIdx.x * 3380 + (int)threadIdx.x / 169 * 845 + ax1_inner * 169 + (int)threadIdx.x % 169; int index12 = ax1_inner; int index13 = (int)blockIdx.x * 20 + (int)threadIdx.x / 169 * 5 + ax1_inner; T_relu[index11] = max(compute[index12] + placeholder2[index13], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpuxm6r4un
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<193>; .reg .b32 %r<34>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048000]; ld.param.u64 %rd2, [default_function_kernel0_param_0]; ld.param.u64 %rd3, [default_function_kernel0_param_1]; ld.param.u64 %rd4, [default_function_kernel0_param_2]; ld.param.u64 %rd5, [default_function_kernel0_param_3]; mov.u32 %r1, %tid.x; setp.gt.s32 %p1, %r1, 5407; @%p1 bra $L__BB0_2; shl.b32 %r13, %r1, 4; cvta.to.global.u64 %rd6, %rd2; mul.wide.s32 %rd7, %r13, 4; add.s64 %rd8, %rd6, %rd7; ld.global.nc.f32 %f21, [%rd8]; shl.b32 %r14, %r1, 6; mov.u32 %r15, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r16, %r15, %r14; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r16+(0+i*4)], %f(21+i*1); ld.global.nc.f32 %f(22+i*1), [%rd8+(4+i*4)]; st.shared.f32 [%r16+60], %f36; $L__BB0_2: cvta.to.global.u64 %rd9, %rd3; mul.wide.s32 %rd10, %r1, 4; add.s64 %rd1, %rd9, %rd10; ld.global.nc.f32 %f37, [%rd1]; shl.b32 %r17, %r1, 2; mov.u32 %r18, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r2, %r18, %r17; for.size.2 i in range(0, 29, 1): st.shared.f32 [%r2+(0+i*67600)], %f(37+i*1); ld.global.nc.f32 %f(38+i*1), [%rd1+(67600+i*67600)]; st.shared.f32 [%r2+1960400], %f66; setp.gt.s32 %p2, %r1, 4999; @%p2 bra $L__BB0_4; ld.global.nc.f32 %f67, [%rd1+2028000]; st.shared.f32 [%r2+2028000], %f67; $L__BB0_4: bar.sync 0; mul.hi.s32 %r20, %r1, 1626496491; shr.u32 %r21, %r20, 31; shr.s32 %r22, %r20, 6; add.s32 %r3, %r22, %r21; mad.lo.s32 %r24, %r3, 10240, %r18; add.s32 %r33, %r24, 1032204; mul.lo.s32 %r25, %r3, 169; sub.s32 %r5, %r1, %r25; shl.b32 %r26, %r5, 2; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r28, %r27, %r26; add.s32 %r31, %r28, 1352; mov.f32 %f183, 0f00000000; mov.u32 %r32, 1024000; for.size.1 i in range(0, 9, 1): mov.f32 %f(184+i*1), %f183; $L__BB0_5: for.size.21 i in range(0, 2, 1): ld.shared.f32 %f(78+i*21), [%r33+(-1032204+i*4)]; ld.shared.f32 %f(79+i*21), [%r31+(-1352+i*676)]; fma.rn.f32 %f(80+i*21), %f(79+i*21), %f(78+i*21), %f(192+i*-112); ld.shared.f32 %f(81+i*21), [%r33+(-8204+i*4)]; fma.rn.f32 %f(82+i*21), %f(79+i*21), %f(81+i*21), %f(191+i*-109); ld.shared.f32 %f(83+i*21), [%r33+(-1030156+i*4)]; fma.rn.f32 %f(84+i*21), %f(79+i*21), %f(83+i*21), %f(190+i*-106); ld.shared.f32 %f(85+i*21), [%r33+(-6156+i*4)]; fma.rn.f32 %f(86+i*21), %f(79+i*21), %f(85+i*21), %f(189+i*-103); ld.shared.f32 %f(87+i*21), [%r33+(-1028108+i*4)]; fma.rn.f32 %f(88+i*21), %f(79+i*21), %f(87+i*21), %f(188+i*-100); ld.shared.f32 %f(89+i*21), [%r33+(-4108+i*4)]; fma.rn.f32 %f(90+i*21), %f(79+i*21), %f(89+i*21), %f(187+i*-97); ld.shared.f32 %f(91+i*21), [%r33+(-1026060+i*4)]; fma.rn.f32 %f(92+i*21), %f(79+i*21), %f(91+i*21), %f(186+i*-94); ld.shared.f32 %f(93+i*21), [%r33+(-2060+i*4)]; fma.rn.f32 %f(94+i*21), %f(79+i*21), %f(93+i*21), %f(185+i*-91); ld.shared.f32 %f(95+i*21), [%r33+(-1024012+i*4)]; fma.rn.f32 %f(96+i*21), %f(79+i*21), %f(95+i*21), %f(184+i*-88); ld.shared.f32 %f(97+i*21), [%r33+(-12+i*4)]; fma.rn.f32 %f(98+i*21), %f(79+i*21), %f(97+i*21), %f(183+i*-85); for.size.21 i in range(0, 2, 1): ld.shared.f32 %f(120+i*21), [%r33+(-1032196+i*4)]; ld.shared.f32 %f(121+i*21), [%r31+(0+i*676)]; fma.rn.f32 %f(122+i*70), %f(121+i*21), %f(120+i*21), %f(101+i*21); ld.shared.f32 %f(123+i*20), [%r33+(-8196+i*4)]; fma.rn.f32 %f(124+i*67), %f(121+i*21), %f(123+i*20), %f(103+i*21); ld.shared.f32 %f(125+i*19), [%r33+(-1030148+i*4)]; fma.rn.f32 %f(126+i*64), %f(121+i*21), %f(125+i*19), %f(105+i*21); ld.shared.f32 %f(127+i*18), [%r33+(-6148+i*4)]; fma.rn.f32 %f(128+i*61), %f(121+i*21), %f(127+i*18), %f(107+i*21); ld.shared.f32 %f(129+i*17), [%r33+(-1028100+i*4)]; fma.rn.f32 %f(130+i*58), %f(121+i*21), %f(129+i*17), %f(109+i*21); ld.shared.f32 %f(131+i*16), [%r33+(-4100+i*4)]; fma.rn.f32 %f(132+i*55), %f(121+i*21), %f(131+i*16), %f(111+i*21); ld.shared.f32 %f(133+i*15), [%r33+(-1026052+i*4)]; fma.rn.f32 %f(134+i*52), %f(121+i*21), %f(133+i*15), %f(113+i*21); ld.shared.f32 %f(135+i*14), [%r33+(-2052+i*4)]; fma.rn.f32 %f(136+i*49), %f(121+i*21), %f(135+i*14), %f(115+i*21); ld.shared.f32 %f(137+i*13), [%r33+(-1024004+i*4)]; fma.rn.f32 %f(138+i*46), %f(121+i*21), %f(137+i*13), %f(117+i*21); ld.shared.f32 %f(139+i*12), [%r33+(-4+i*4)]; fma.rn.f32 %f(140+i*43), %f(121+i*21), %f(139+i*12), %f(119+i*21); for.size.1 i in range(0, 2, 1): add.s32 %r(33+i*-2), %r(33+i*-2), (16+i*2688); add.s32 %r32, %r32, 16; setp.ne.s32 %p3, %r32, 1026048; @%p3 bra $L__BB0_5; mul.lo.s32 %r29, %r3, 5; mad.lo.s32 %r30, %r3, 845, %r5; cvta.to.global.u64 %rd11, %rd5; mul.wide.s32 %rd12, %r29, 4; add.s64 %rd13, %rd11, %rd12; ld.global.nc.f32 %f152, [%rd13]; add.f32 %f153, %f192, %f152; mov.f32 %f154, 0f00000000; max.f32 %f155, %f153, %f154; cvta.to.global.u64 %rd14, %rd4; mul.wide.s32 %rd15, %r30, 4; add.s64 %rd16, %rd14, %rd15; for.size.5 i in range(0, 4, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd16+(0+i*676+j*338000)], %f(155+i*6+j*3); ld.global.nc.f32 %f(156+i*6+j*3), [%rd13+(2000+i*4+j*-1996)]; add.f32 %f(157+i*6+j*3), %f(191+i*-2+j*-1), %f(156+i*6+j*3); max.f32 %f(158+i*6+j*3), %f(157+i*6+j*3), %f154; st.global.f32 [%rd16+2704], %f179; ld.global.nc.f32 %f180, [%rd13+2016]; add.f32 %f181, %f183, %f180; max.f32 %f182, %f181, %f154; st.global.f32 [%rd16+340704], %f182; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[10]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[512000]; compute[0] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[9] = 0.000000e+00f; if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0] = placeholder[index0]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 1] = placeholder[index0 + 1]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 2] = placeholder[index0 + 2]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 3] = placeholder[index0 + 3]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 4] = placeholder[index0 + 4]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 5] = placeholder[index0 + 5]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 6] = placeholder[index0 + 6]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 7] = placeholder[index0 + 7]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 8] = placeholder[index0 + 8]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 9] = placeholder[index0 + 9]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 10] = placeholder[index0 + 10]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 11] = placeholder[index0 + 11]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 12] = placeholder[index0 + 12]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 13] = placeholder[index0 + 13]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 14] = placeholder[index0 + 14]; } if ((int)threadIdx.x < 5408) { int index0 = (int)threadIdx.x * 16; pad_temp_shared[index0 + 15] = placeholder[index0 + 15]; } int index1 = (int)threadIdx.x; placeholder_shared[index1] = placeholder1[index1]; placeholder_shared[index1 + 16900] = placeholder1[index1 + 16900]; placeholder_shared[index1 + 33800] = placeholder1[index1 + 33800]; placeholder_shared[index1 + 50700] = placeholder1[index1 + 50700]; placeholder_shared[index1 + 67600] = placeholder1[index1 + 67600]; placeholder_shared[index1 + 84500] = placeholder1[index1 + 84500]; placeholder_shared[index1 + 101400] = placeholder1[index1 + 101400]; placeholder_shared[index1 + 118300] = placeholder1[index1 + 118300]; placeholder_shared[index1 + 135200] = placeholder1[index1 + 135200]; placeholder_shared[index1 + 152100] = placeholder1[index1 + 152100]; placeholder_shared[index1 + 169000] = placeholder1[index1 + 169000]; placeholder_shared[index1 + 185900] = placeholder1[index1 + 185900]; placeholder_shared[index1 + 202800] = placeholder1[index1 + 202800]; placeholder_shared[index1 + 219700] = placeholder1[index1 + 219700]; placeholder_shared[index1 + 236600] = placeholder1[index1 + 236600]; placeholder_shared[index1 + 253500] = placeholder1[index1 + 253500]; placeholder_shared[index1 + 270400] = placeholder1[index1 + 270400]; placeholder_shared[index1 + 287300] = placeholder1[index1 + 287300]; placeholder_shared[index1 + 304200] = placeholder1[index1 + 304200]; placeholder_shared[index1 + 321100] = placeholder1[index1 + 321100]; placeholder_shared[index1 + 338000] = placeholder1[index1 + 338000]; placeholder_shared[index1 + 354900] = placeholder1[index1 + 354900]; placeholder_shared[index1 + 371800] = placeholder1[index1 + 371800]; placeholder_shared[index1 + 388700] = placeholder1[index1 + 388700]; placeholder_shared[index1 + 405600] = placeholder1[index1 + 405600]; placeholder_shared[index1 + 422500] = placeholder1[index1 + 422500]; placeholder_shared[index1 + 439400] = placeholder1[index1 + 439400]; placeholder_shared[index1 + 456300] = placeholder1[index1 + 456300]; placeholder_shared[index1 + 473200] = placeholder1[index1 + 473200]; placeholder_shared[index1 + 490100] = placeholder1[index1 + 490100]; if ((int)threadIdx.x < 5000) { placeholder_shared[index1 + 507000] = placeholder1[index1 + 507000]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index2 = rc_outer_inner * 676 + (int)threadIdx.x % 169; int index3 = (int)threadIdx.x / 169 * 2560 + rc_outer_inner * 4; compute[0] = compute[0] + pad_temp_shared[index2] * placeholder_shared[index3]; compute[5] = compute[5] + (pad_temp_shared[index2] * placeholder_shared[index3 + 256000]); compute[1] = compute[1] + (pad_temp_shared[index2] * placeholder_shared[index3 + 512]); compute[6] = compute[6] + (pad_temp_shared[index2] * placeholder_shared[index3 + 256512]); compute[2] = compute[2] + (pad_temp_shared[index2] * placeholder_shared[index3 + 1024]); compute[7] = compute[7] + (pad_temp_shared[index2] * placeholder_shared[index3 + 257024]); compute[3] = compute[3] + (pad_temp_shared[index2] * placeholder_shared[index3 + 1536]); compute[8] = compute[8] + (pad_temp_shared[index2] * placeholder_shared[index3 + 257536]); compute[4] = compute[4] + (pad_temp_shared[index2] * placeholder_shared[index3 + 2048]); compute[9] = compute[9] + (pad_temp_shared[index2] * placeholder_shared[index3 + 258048]); compute[0] = compute[0] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 1]); compute[5] = compute[5] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 256001]); compute[1] = compute[1] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 513]); compute[6] = compute[6] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 256513]); compute[2] = compute[2] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 1025]); compute[7] = compute[7] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 257025]); compute[3] = compute[3] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 1537]); compute[8] = compute[8] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 257537]); compute[4] = compute[4] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 2049]); compute[9] = compute[9] + (pad_temp_shared[index2 + 169] * placeholder_shared[index3 + 258049]); compute[0] = compute[0] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 2]); compute[5] = compute[5] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 256002]); compute[1] = compute[1] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 514]); compute[6] = compute[6] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 256514]); compute[2] = compute[2] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 1026]); compute[7] = compute[7] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 257026]); compute[3] = compute[3] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 1538]); compute[8] = compute[8] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 257538]); compute[4] = compute[4] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 2050]); compute[9] = compute[9] + (pad_temp_shared[index2 + 338] * placeholder_shared[index3 + 258050]); compute[0] = compute[0] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 3]); compute[5] = compute[5] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 256003]); compute[1] = compute[1] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 515]); compute[6] = compute[6] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 256515]); compute[2] = compute[2] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 1027]); compute[7] = compute[7] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 257027]); compute[3] = compute[3] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 1539]); compute[8] = compute[8] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 257539]); compute[4] = compute[4] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 2051]); compute[9] = compute[9] + (pad_temp_shared[index2 + 507] * placeholder_shared[index3 + 258051]); } for (int ax1_inner = 0; ax1_inner < 5; ++ax1_inner) { int index4 = (int)threadIdx.x / 169 * 845 + ax1_inner * 169 + (int)threadIdx.x % 169; int index5 = ax1_inner; int index6 = (int)threadIdx.x / 169 * 5 + ax1_inner; T_relu[index4] = max(compute[index5] + placeholder2[index6], 0.000000e+00f); T_relu[index4 + 84500] = max(compute[index5 + 5] + placeholder2[index6 + 500], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmptpjx4caw
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[33800]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<25>; .reg .f32 %f<1372>; .reg .b32 %r<161>; .reg .b64 %rd<47>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[512000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.u32 %r34, 0; mov.u32 %r150, %r34; $L__BB0_1: mul.lo.s32 %r35, %r150, 13; mul.wide.s32 %rd13, %r35, 4; add.s64 %rd14, %rd1, %rd13; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd14+(0+i*4+j*2600)], %r34; add.s32 %r150, %r150, 1; setp.ne.s32 %p1, %r150, 50; @%p1 bra $L__BB0_1; mov.u32 %r151, 0; cvta.to.global.u64 %rd15, %rd8; cvta.to.global.u64 %rd18, %rd9; $L__BB0_3: mov.u32 %r39, %tid.x; shl.b32 %r40, %r39, 1; mov.u32 %r155, 40; shl.b32 %r41, %r39, 3; mov.u32 %r42, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r43, %r42, %r41; add.s32 %r152, %r43, 160; bar.sync 0; mad.lo.s32 %r44, %r151, 43264, %r40; mul.wide.s32 %rd16, %r44, 4; add.s64 %rd17, %rd15, %rd16; add.s64 %rd46, %rd17, 160; for.size.1 i in range(0, 2, 1): add.s32 %r(153+i*1), %r(40+i*-1), (60+i*-30); $L__BB0_4: add.s32 %r45, %r153, -60; setp.gt.s32 %p2, %r45, 43263; @%p2 bra $L__BB0_7; add.s32 %r46, %r154, -30; setp.gt.s32 %p3, %r46, 21631; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f508, [%rd46+-160]; st.shared.f32 [%r152+-160], %f508; $L__BB0_7: setp.gt.s32 %p4, %r45, 43262; @%p4 bra $L__BB0_10; add.s32 %r48, %r154, -30; setp.gt.s32 %p5, %r48, 21631; @%p5 bra $L__BB0_10; ld.global.nc.f32 %f509, [%rd46+-156]; st.shared.f32 [%r152+-156], %f509; $L__BB0_10: add.s32 %r11, %r153, -40; setp.gt.s32 %p6, %r11, 43263; @%p6 bra $L__BB0_13; add.s32 %r49, %r154, -20; setp.gt.s32 %p7, %r49, 21631; @%p7 bra $L__BB0_13; ld.global.nc.f32 %f510, [%rd46+-80]; st.shared.f32 [%r152+-80], %f510; $L__BB0_13: setp.gt.s32 %p8, %r11, 43262; @%p8 bra $L__BB0_16; add.s32 %r50, %r154, -20; setp.gt.s32 %p9, %r50, 21631; @%p9 bra $L__BB0_16; ld.global.nc.f32 %f511, [%rd46+-76]; st.shared.f32 [%r152+-76], %f511; $L__BB0_16: add.s32 %r12, %r40, %r155; setp.gt.s32 %p10, %r12, 43263; @%p10 bra $L__BB0_19; add.s32 %r53, %r154, -10; setp.gt.s32 %p11, %r53, 21631; @%p11 bra $L__BB0_19; ld.global.nc.f32 %f512, [%rd46]; st.shared.f32 [%r152], %f512; $L__BB0_19: setp.gt.s32 %p12, %r12, 43262; @%p12 bra $L__BB0_22; add.s32 %r54, %r154, -10; setp.gt.s32 %p13, %r54, 21631; @%p13 bra $L__BB0_22; ld.global.nc.f32 %f513, [%rd46+4]; st.shared.f32 [%r152+4], %f513; $L__BB0_22: setp.gt.s32 %p14, %r153, 43263; @%p14 bra $L__BB0_25; setp.gt.s32 %p15, %r154, 21631; @%p15 bra $L__BB0_25; ld.global.nc.f32 %f514, [%rd46+80]; st.shared.f32 [%r152+80], %f514; $L__BB0_25: setp.gt.s32 %p16, %r153, 43262; @%p16 bra $L__BB0_28; setp.gt.s32 %p17, %r154, 21631; @%p17 bra $L__BB0_28; ld.global.nc.f32 %f515, [%rd46+84]; st.shared.f32 [%r152+84], %f515; $L__BB0_28: for.size.1 i in range(0, 2, 1): add.s32 %r(155+i*-1), %r(155+i*-1), (80+i*-40); add.s32 %r153, %r153, 80; add.s64 %rd46, %rd46, 320; add.s32 %r152, %r152, 320; setp.ne.s32 %p18, %r155, 43320; @%p18 bra $L__BB0_4; shl.b32 %r56, %r151, 8; mov.u32 %r57, %ctaid.x; mad.lo.s32 %r17, %r57, 256000, %r56; mov.u32 %r156, 0; $L__BB0_30: mov.u32 %r144, %tid.x; mad.lo.s32 %r59, %r156, 10, %r144; shl.b32 %r60, %r59, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(61+i*1), %r(60+i*-1), (-512+i*767); for.size.1 i in range(0, 2, 1): add.s32 %r(63+i*1), %r(17+i*46), %r(62+i*-1); mul.wide.s32 %rd19, %r64, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f516, [%rd20]; shl.b32 %r65, %r59, 2; mov.u32 %r66, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r67, %r66, %r65; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r67+(0+i*40)], %f(516+i*1); add.s32 %r(68+i*6), %r59, (10+i*10); shl.b32 %r(69+i*6), %r(68+i*6), 1; and.b32 %r(70+i*6), %r(69+i*6), -512; and.b32 %r(71+i*6), %r(68+i*6), 255; add.s32 %r(72+i*6), %r17, %r(71+i*6); add.s32 %r(73+i*6), %r(72+i*6), %r(70+i*6); mul.wide.s32 %rd(21+i*2), %r(73+i*6), 4; add.s64 %rd(22+i*2), %rd18, %rd(21+i*2); ld.global.nc.f32 %f(517+i*1), [%rd(22+i*2)+0]; st.shared.f32 [%r67+280], %f523; add.s32 %r156, %r156, 8; setp.ne.s32 %p19, %r156, 12800; @%p19 bra $L__BB0_30; bar.sync 0; mov.u32 %r157, 0; $L__BB0_32: mov.u32 %r158, 0; $L__BB0_33: mov.u32 %r148, %tid.x; mul.lo.s32 %r116, %r158, 13; mul.wide.s32 %rd35, %r116, 4; add.s64 %rd5, %rd1, %rd35; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 13, 1): ld.local.f32 %f(1371+i*-13+j*-1), [%rd5+(0+i*4+j*2600)]; mov.u32 %r159, 0; $L__BB0_34: mov.u32 %r140, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r139, _ZZ24default_function_kernel0E18placeholder_shared; shl.b32 %r138, %r157, 3; mov.u32 %r137, %tid.x; shl.b32 %r136, %r158, 8; mad.lo.s32 %r135, %r137, 12800, %r138; add.s32 %r134, %r135, %r136; mul.lo.s32 %r133, %r157, 1352; mad.lo.s32 %r117, %r159, 169, %r133; for.size.2 i in range(0, 2, 1): add.s32 %r(118+i*3), %r(134+i*5), %r(159+i*-40); shl.b32 %r(119+i*3), %r(118+i*-1), 2; add.s32 %r124, %r140, %r122; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(524+i*1), [%r(124+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1371+i*-1), %f(524+i*2), %f525, %f(1371+i*-1); ld.shared.f32 %f(526+i*1), [%r124+(52+i*52)]; for.size.2 i in range(0, 10, 1): fma.rn.f32 %f(1369+i*-1), %f(527+i*1), %f525, %f(1369+i*-1); ld.shared.f32 %f(528+i*1), [%r124+(156+i*52)]; for.size.3 i in range(0, 12, 1): for.size.2 j in range(0, 13, 1): fma.rn.f32 %f(1359+i*-13+j*-1), %f(537+i*13+j*1), %f525, %f(1359+i*-13+j*-1); ld.shared.f32 %f(538+i*13+j*1), [%r124+(4+i*4+j*52)]; fma.rn.f32 %f1203, %f693, %f525, %f1203; add.s32 %r159, %r159, 1; setp.ne.s32 %p20, %r159, 8; @%p20 bra $L__BB0_34; mul.lo.s32 %r141, %r158, 13; mul.wide.s32 %rd43, %r141, 4; add.s64 %rd42, %rd1, %rd43; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 13, 1): st.local.f32 [%rd42+(0+i*4+j*2600)], %f(1371+i*-13+j*-1); add.s32 %r158, %r158, 1; setp.lt.u32 %p21, %r158, 50; @%p21 bra $L__BB0_33; add.s32 %r157, %r157, 1; setp.lt.u32 %p22, %r157, 32; @%p22 bra $L__BB0_32; add.s32 %r151, %r151, 1; setp.lt.u32 %p23, %r151, 2; @%p23 bra $L__BB0_3; ld.param.u64 %rd45, [default_function_kernel0_param_3]; ld.param.u64 %rd44, [default_function_kernel0_param_2]; mov.u32 %r143, %ctaid.x; mov.u32 %r142, %tid.x; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(128+i*1), %r142, (50+i*8400); mad.lo.s32 %r(30+i*1), %r143, (500+i*84000), %r(128+i*1); cvta.to.global.u64 %rd6, %rd44; cvta.to.global.u64 %rd7, %rd45; mov.u32 %r160, 0; $L__BB0_39: add.s32 %r130, %r30, %r160; mul.wide.s32 %rd36, %r130, 4; add.s64 %rd37, %rd7, %rd36; mad.lo.s32 %r131, %r160, 169, %r31; mul.lo.s32 %r132, %r160, 13; mul.wide.s32 %rd38, %r132, 4; add.s64 %rd39, %rd1, %rd38; ld.local.f32 %f694, [%rd39]; ld.global.nc.f32 %f695, [%rd37]; add.f32 %f696, %f694, %f695; mov.f32 %f697, 0f00000000; max.f32 %f698, %f696, %f697; mul.wide.s32 %rd40, %r131, 4; add.s64 %rd41, %rd6, %rd40; for.size.4 i in range(0, 12, 1): st.global.f32 [%rd41+(0+i*52)], %f(698+i*3); ld.local.f32 %f(699+i*3), [%rd39+(2600+i*2600)]; add.f32 %f(700+i*3), %f(699+i*3), %f695; max.f32 %f(701+i*3), %f(700+i*3), %f697; for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd41+(624+i*4+j*-620)], %f(734+i*39+j*3); ld.local.f32 %f(735+i*39+j*3), [%rd39+(4+i*4+j*2600)]; add.f32 %f(736+i*39+j*3), %f(735+i*39+j*3), %f695; max.f32 %f(737+i*39+j*3), %f(736+i*39+j*3), %f697; for.size.4 j in range(0, 11, 1): st.global.f32 [%rd41+(56+i*4+j*52)], %f(740+i*39+j*3); ld.local.f32 %f(741+i*39+j*3), [%rd39+(5204+i*4+j*2600)]; add.f32 %f(742+i*39+j*3), %f(741+i*39+j*3), %f695; max.f32 %f(743+i*39+j*3), %f(742+i*39+j*3), %f697; st.global.f32 [%rd41+672], %f1202; add.s32 %r160, %r160, 1; setp.ne.s32 %p24, %r160, 50; @%p24 bra $L__BB0_39; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8450]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[128000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 50; ++ff_outer_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 13; ++xx_inner_init) { int index0 = ff_outer_inner_init * 13 + xx_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 650] = 0.000000e+00f; compute[index0 + 1300] = 0.000000e+00f; compute[index0 + 1950] = 0.000000e+00f; compute[index0 + 2600] = 0.000000e+00f; compute[index0 + 3250] = 0.000000e+00f; compute[index0 + 3900] = 0.000000e+00f; compute[index0 + 4550] = 0.000000e+00f; compute[index0 + 5200] = 0.000000e+00f; compute[index0 + 5850] = 0.000000e+00f; compute[index0 + 6500] = 0.000000e+00f; compute[index0 + 7150] = 0.000000e+00f; compute[index0 + 7800] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 2164; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2 < 43264) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 10 + (int)threadIdx.x < 21632) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2; int index2 = rc_outer_outer * 43264 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2; pad_temp_shared[index1] = placeholder[index2]; } } if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2 < 43263) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 10 + (int)threadIdx.x < 21632) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2; int index2 = rc_outer_outer * 43264 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x * 2; pad_temp_shared[index1 + 1] = placeholder[index2 + 1]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 12800; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 10 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 256000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 10 + (int)threadIdx.x >> 8) * 512 + rc_outer_outer * 256 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 10 + (int)threadIdx.x & 255); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 50; ++ff_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int xx_inner = 0; xx_inner < 13; ++xx_inner) { int index5 = ff_outer_inner * 13 + xx_inner; int index6 = rc_outer_inner * 1352 + rc_inner * 169 + xx_inner; int index7 = (int)threadIdx.x * 12800 + ff_outer_inner * 256 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 650] = compute[index5 + 650] + (pad_temp_shared[index6 + 13] * placeholder_shared[index7]); compute[index5 + 1300] = compute[index5 + 1300] + (pad_temp_shared[index6 + 26] * placeholder_shared[index7]); compute[index5 + 1950] = compute[index5 + 1950] + (pad_temp_shared[index6 + 39] * placeholder_shared[index7]); compute[index5 + 2600] = compute[index5 + 2600] + (pad_temp_shared[index6 + 52] * placeholder_shared[index7]); compute[index5 + 3250] = compute[index5 + 3250] + (pad_temp_shared[index6 + 65] * placeholder_shared[index7]); compute[index5 + 3900] = compute[index5 + 3900] + (pad_temp_shared[index6 + 78] * placeholder_shared[index7]); compute[index5 + 4550] = compute[index5 + 4550] + (pad_temp_shared[index6 + 91] * placeholder_shared[index7]); compute[index5 + 5200] = compute[index5 + 5200] + (pad_temp_shared[index6 + 104] * placeholder_shared[index7]); compute[index5 + 5850] = compute[index5 + 5850] + (pad_temp_shared[index6 + 117] * placeholder_shared[index7]); compute[index5 + 6500] = compute[index5 + 6500] + (pad_temp_shared[index6 + 130] * placeholder_shared[index7]); compute[index5 + 7150] = compute[index5 + 7150] + (pad_temp_shared[index6 + 143] * placeholder_shared[index7]); compute[index5 + 7800] = compute[index5 + 7800] + (pad_temp_shared[index6 + 156] * placeholder_shared[index7]); } } } } } for (int ax1_inner = 0; ax1_inner < 50; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x * 84500 + (int)threadIdx.x * 8450 + ax1_inner * 169 + ax3_inner; int index9 = ax1_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x * 500 + (int)threadIdx.x * 50 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 13] = max(compute[index9 + 650] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 26] = max(compute[index9 + 1300] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 39] = max(compute[index9 + 1950] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 52] = max(compute[index9 + 2600] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 65] = max(compute[index9 + 3250] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 78] = max(compute[index9 + 3900] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 91] = max(compute[index9 + 4550] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 104] = max(compute[index9 + 5200] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 117] = max(compute[index9 + 5850] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 130] = max(compute[index9 + 6500] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 143] = max(compute[index9 + 7150] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 156] = max(compute[index9 + 7800] + placeholder2[index10], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmplxbztt2y
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[16900]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<16>; .reg .f32 %f<761>; .reg .b32 %r<87>; .reg .b64 %rd<37>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; ld.param.u64 %rd13, [default_function_kernel0_param_2]; ld.param.u64 %rd14, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd11; cvta.to.global.u64 %rd2, %rd12; cvta.to.global.u64 %rd3, %rd13; cvta.to.global.u64 %rd4, %rd14; add.u64 %rd5, %SPL, 0; mov.u32 %r34, 0; mov.u32 %r75, %r34; $L__BB0_1: mul.lo.s32 %r35, %r75, 13; mul.wide.s32 %rd16, %r35, 4; add.s64 %rd17, %rd5, %rd16; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd17+(0+i*4+j*676)], %r34; add.s32 %r75, %r75, 1; setp.ne.s32 %p1, %r75, 13; @%p1 bra $L__BB0_1; mov.u32 %r37, 0; mov.u32 %r76, %r37; $L__BB0_3: mad.lo.s32 %r38, %r76, 13, 845; mul.wide.s32 %rd18, %r38, 4; add.s64 %rd19, %rd5, %rd18; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd19+(0+i*4+j*676)], %r37; add.s32 %r76, %r76, 1; setp.ne.s32 %p2, %r76, 13; @%p2 bra $L__BB0_3; mov.u32 %r40, 0; mov.u32 %r77, %r40; $L__BB0_5: mad.lo.s32 %r41, %r77, 13, 1690; mul.wide.s32 %rd20, %r41, 4; add.s64 %rd21, %rd5, %rd20; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd21+(0+i*4+j*676)], %r40; add.s32 %r77, %r77, 1; setp.ne.s32 %p3, %r77, 13; @%p3 bra $L__BB0_5; mov.u32 %r43, 0; mov.u32 %r78, %r43; $L__BB0_7: mad.lo.s32 %r44, %r78, 13, 2535; mul.wide.s32 %rd22, %r44, 4; add.s64 %rd23, %rd5, %rd22; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd23+(0+i*4+j*676)], %r43; add.s32 %r78, %r78, 1; setp.ne.s32 %p4, %r78, 13; @%p4 bra $L__BB0_7; mov.u32 %r46, 0; mov.u32 %r79, %r46; $L__BB0_9: mad.lo.s32 %r47, %r79, 13, 3380; mul.wide.s32 %rd24, %r47, 4; add.s64 %rd25, %rd5, %rd24; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd25+(0+i*4+j*676)], %r46; add.s32 %r79, %r79, 1; setp.ne.s32 %p5, %r79, 13; @%p5 bra $L__BB0_9; mov.u32 %r80, 0; mov.u32 %r11, %tid.x; $L__BB0_11: mad.lo.s32 %r13, %r80, 40, %r11; setp.gt.s32 %p6, %r13, 86527; mul.wide.s32 %rd26, %r13, 4; add.s64 %rd6, %rd1, %rd26; shl.b32 %r50, %r13, 2; mov.u32 %r51, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r14, %r51, %r50; @%p6 bra $L__BB0_13; ld.global.nc.f32 %f11, [%rd6]; st.shared.f32 [%r14], %f11; $L__BB0_13: add.s32 %r52, %r13, 40; setp.gt.s32 %p7, %r52, 86527; @%p7 bra $L__BB0_15; ld.global.nc.f32 %f12, [%rd6+160]; st.shared.f32 [%r14+160], %f12; $L__BB0_15: add.s32 %r53, %r13, 80; setp.gt.s32 %p8, %r53, 86527; @%p8 bra $L__BB0_17; ld.global.nc.f32 %f13, [%rd6+320]; st.shared.f32 [%r14+320], %f13; $L__BB0_17: add.s32 %r54, %r13, 120; setp.gt.s32 %p9, %r54, 86527; @%p9 bra $L__BB0_19; ld.global.nc.f32 %f14, [%rd6+480]; st.shared.f32 [%r14+480], %f14; $L__BB0_19: add.s32 %r80, %r80, 4; setp.ne.s32 %p10, %r80, 2164; @%p10 bra $L__BB0_11; mov.u32 %r81, 0; mov.u32 %r58, _ZZ24default_function_kernel0E18placeholder_shared; $L__BB0_21: mad.lo.s32 %r56, %r81, 40, %r11; mul.wide.s32 %rd27, %r56, 4; add.s64 %rd28, %rd2, %rd27; ld.global.nc.f32 %f15, [%rd28]; shl.b32 %r57, %r56, 2; add.s32 %r59, %r58, %r57; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r59+(0+i*160)], %f(15+i*1); ld.global.nc.f32 %f(16+i*1), [%rd28+(160+i*160)]; st.shared.f32 [%r59+2400], %f30; add.s32 %r81, %r81, 16; setp.ne.s32 %p11, %r81, 12800; @%p11 bra $L__BB0_21; bar.sync 0; mul.lo.s32 %r18, %r11, 12800; add.s64 %rd7, %rd5, 1376; mov.u32 %r60, 0; mov.u32 %r82, %r60; $L__BB0_23: shl.b32 %r62, %r82, 1; add.s32 %r20, %r62, %r18; mad.lo.s32 %r64, %r82, 1352, %r51; add.s32 %r21, %r64, 4; mov.u32 %r83, %r60; $L__BB0_24: mad.lo.s32 %r66, %r83, 2560, %r20; shl.b32 %r67, %r66, 2; add.s32 %r69, %r58, %r67; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1+i*1), [%r69+(0+i*4)]; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 4, 1): ld.shared.f32 %f(3+i*4+j*1), [%r69+(2048+i*4+j*2048)]; mul.lo.s32 %r70, %r83, 845; mul.wide.s32 %rd29, %r70, 4; add.s64 %rd36, %rd7, %rd29; mov.u32 %r85, 676; mov.u32 %r84, %r21; $L__BB0_25: for.size.22 i in range(0, 13, 1): ld.shared.f32 %f(31+i*17), [%r84+(-4+i*4)]; ld.local.f32 %f(32+i*17), [%rd36+(-1376+i*4)]; fma.rn.f32 %f(33+i*17), %f(31+i*17), %f1, %f(32+i*17); ld.local.f32 %f(34+i*17), [%rd36+(-700+i*4)]; fma.rn.f32 %f(35+i*17), %f(31+i*17), %f3, %f(34+i*17); ld.local.f32 %f(36+i*17), [%rd36+(-24+i*4)]; fma.rn.f32 %f(37+i*17), %f(31+i*17), %f4, %f(36+i*17); ld.local.f32 %f(38+i*17), [%rd36+(652+i*4)]; fma.rn.f32 %f(39+i*17), %f(31+i*17), %f5, %f(38+i*17); ld.local.f32 %f(40+i*17), [%rd36+(1328+i*4)]; fma.rn.f32 %f(41+i*17), %f(31+i*17), %f6, %f(40+i*17); ld.shared.f32 %f(42+i*17), [%r84+(672+i*4)]; fma.rn.f32 %f(43+i*17), %f(42+i*17), %f2, %f(33+i*17); st.local.f32 [%rd36+(-1376+i*4)], %f(43+i*17); fma.rn.f32 %f(44+i*17), %f(42+i*17), %f7, %f(35+i*17); st.local.f32 [%rd36+(-700+i*4)], %f(44+i*17); fma.rn.f32 %f(45+i*17), %f(42+i*17), %f8, %f(37+i*17); st.local.f32 [%rd36+(-24+i*4)], %f(45+i*17); fma.rn.f32 %f(46+i*17), %f(42+i*17), %f9, %f(39+i*17); st.local.f32 [%rd36+(652+i*4)], %f(46+i*17); fma.rn.f32 %f(47+i*17), %f(42+i*17), %f10, %f(41+i*17); st.local.f32 [%rd36+(1328+i*4)], %f(47+i*17); add.s64 %rd36, %rd36, 52; for.size.1 i in range(0, 2, 1): add.s32 %r(84+i*1), %r(84+i*1), 52; setp.ne.s32 %p12, %r85, 1352; @%p12 bra $L__BB0_25; add.s32 %r83, %r83, 1; setp.lt.u32 %p13, %r83, 5; @%p13 bra $L__BB0_24; add.s32 %r82, %r82, 1; setp.lt.u32 %p14, %r82, 256; @%p14 bra $L__BB0_23; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(30+i*1), %r11, (25+i*4200); mov.u32 %r86, 0; $L__BB0_29: add.s32 %r72, %r86, %r30; mul.wide.s32 %rd30, %r72, 4; add.s64 %rd31, %rd4, %rd30; mul.lo.s32 %r73, %r86, 169; add.s32 %r74, %r73, %r31; mul.wide.s32 %rd32, %r73, 4; add.s64 %rd33, %rd5, %rd32; ld.local.f32 %f252, [%rd33]; ld.global.nc.f32 %f253, [%rd31]; add.f32 %f254, %f252, %f253; mov.f32 %f255, 0f00000000; max.f32 %f256, %f254, %f255; mul.wide.s32 %rd34, %r74, 4; add.s64 %rd35, %rd3, %rd34; for.size.4 i in range(0, 168, 1): st.global.f32 [%rd35+(0+i*4)], %f(256+i*3); ld.local.f32 %f(257+i*3), [%rd33+(4+i*4)]; add.f32 %f(258+i*3), %f(257+i*3), %f253; max.f32 %f(259+i*3), %f(258+i*3), %f255; st.global.f32 [%rd35+672], %f760; add.s32 %r86, %r86, 1; setp.ne.s32 %p15, %r86, 25; @%p15 bra $L__BB0_29; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4225]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[512000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 5; ++ff_outer_inner_init) { for (int yy_outer_inner_init = 0; yy_outer_inner_init < 13; ++yy_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 5; ++ff_inner_init) { int index0 = ff_outer_inner_init * 845 + ff_inner_init * 169 + yy_outer_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; } } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 2164; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 40 + (int)threadIdx.x < 86528) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 40 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index1]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 12800; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 40 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index2]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 256; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 5; ++ff_outer_inner) { for (int yy_outer_inner = 0; yy_outer_inner < 13; ++yy_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 5; ++ff_inner) { int index3 = ff_outer_inner * 845 + ff_inner * 169 + yy_outer_inner * 13 + xx_outer_inner; int index4 = rc_outer_inner * 338 + rc_inner * 169 + yy_outer_inner * 13 + xx_outer_inner; int index5 = (int)threadIdx.x * 12800 + ff_outer_inner * 2560 + ff_inner * 512 + rc_outer_inner * 2 + rc_inner; compute[index3] = compute[index3] + pad_temp_shared[index4] * placeholder_shared[index5]; } } } } } } for (int ax1_inner = 0; ax1_inner < 25; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index6 = (int)threadIdx.x * 4225 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index7 = ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index8 = (int)threadIdx.x * 25 + ax1_inner; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpgcubdad1
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<61>; .reg .b32 %r<37>; .reg .b64 %rd<36>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1280]; ld.param.u64 %rd18, [default_function_kernel0_param_0]; ld.param.u64 %rd19, [default_function_kernel0_param_1]; ld.param.u64 %rd20, [default_function_kernel0_param_2]; ld.param.u64 %rd17, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd21, %rd19; cvta.to.global.u64 %rd1, %rd20; mov.u32 %r1, %tid.x; shl.b32 %r11, %r1, 2; mov.u32 %r12, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r12, %r11; shl.b32 %r13, %r1, 4; mov.u32 %r14, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r14, %r13; for.size.2 i in range(0, 2, 1): or.b32 %r(15+i*2), %r11, (2+i*1); and.b32 %r(16+i*2), %r(15+i*2), (14+i*1); mul.hi.s32 %r19, %r1, 1626496491; shr.u32 %r20, %r19, 31; shr.s32 %r21, %r19, 6; add.s32 %r5, %r21, %r20; mul.lo.s32 %r22, %r5, 169; sub.s32 %r23, %r1, %r22; for.size.2 i in range(0, 2, 1): shl.b32 %r(24+i*1), %r(23+i*-18), (2+i*4); add.s32 %r(4+i*2), %r(12+i*2), %r(24+i*1); mov.u32 %r7, %ctaid.x; mul.lo.s32 %r26, %r7, 10240; or.b32 %r27, %r18, %r26; shl.b32 %r28, %r1, 7; and.b32 %r29, %r28, -512; add.s32 %r30, %r27, %r29; add.s32 %r31, %r30, -3; mul.wide.s32 %rd22, %r31, 4; add.s64 %rd32, %rd21, %rd22; add.s64 %rd35, %rd32, 12; or.b32 %r32, %r16, %r26; add.s32 %r33, %r32, %r29; mul.wide.s32 %rd23, %r33, 4; add.s64 %rd34, %rd21, %rd23; add.s64 %rd33, %rd32, 4; cvta.to.global.u64 %rd24, %rd18; mul.wide.s32 %rd25, %r1, 4; add.s64 %rd31, %rd24, %rd25; mov.f32 %f60, 0f00000000; mov.u32 %r36, 0; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 2703; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f4, [%rd31]; st.shared.f32 [%r2], %f4; $L__BB0_3: setp.gt.s32 %p2, %r1, 79; @%p2 bra $L__BB0_5; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(5+i*1), [%rd(32+i*1)+0]; st.shared.f32 [%r3+(0+i*4)], %f(5+i*1); $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(9+i*1), [%r(6+i*-2)+0]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(11+i*3), %f(10+i*3), %f(9+i*3), %f(60+i*-49); ld.shared.f32 %f(12+i*3), [%r6+(4+i*4)]; ld.shared.f32 %f(13+i*3), [%r4+(676+i*676)]; for.size.3 i in range(0, 13, 1): fma.rn.f32 %f(17+i*3), %f(16+i*3), %f(15+i*3), %f(14+i*3); ld.shared.f32 %f(18+i*3), [%r6+(12+i*4)]; ld.shared.f32 %f(19+i*3), [%r4+(2028+i*676)]; fma.rn.f32 %f60, %f55, %f54, %f53; for.size.1 i in range(0, 4, 1): add.s64 %rd(35+i*-1), %rd(35+i*-1), 64; add.s64 %rd31, %rd31, 10816; add.s32 %r36, %r36, 1; setp.ne.s32 %p3, %r36, 32; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd26, %rd17; mad.lo.s32 %r34, %r7, 20, %r5; mul.wide.s32 %rd27, %r34, 4; add.s64 %rd28, %rd26, %rd27; ld.global.nc.f32 %f56, [%rd28]; add.f32 %f57, %f60, %f56; mov.f32 %f58, 0f00000000; max.f32 %f59, %f57, %f58; mad.lo.s32 %r35, %r7, 3380, %r1; mul.wide.s32 %rd29, %r35, 4; add.s64 %rd30, %rd1, %rd29; st.global.f32 [%rd30], %f59; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[1]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[320]; compute[0] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 2704) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 80) { int index0 = (int)threadIdx.x; int index2 = (int)blockIdx.x * 10240 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4; placeholder_shared[index0 * 4] = placeholder1[index2]; } if ((int)threadIdx.x < 80) { int index3 = (int)threadIdx.x * 4 + 1; int index4 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 1 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 1 & 15); placeholder_shared[index3] = placeholder1[index4]; } if ((int)threadIdx.x < 80) { int index5 = (int)threadIdx.x * 4 + 2; int index6 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 2 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 2 & 15); placeholder_shared[index5] = placeholder1[index6]; } if ((int)threadIdx.x < 80) { int index7 = (int)threadIdx.x * 4 + 3; int index8 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 3 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + 3 & 15); placeholder_shared[index7] = placeholder1[index8]; } __syncthreads(); int index0 = (int)threadIdx.x; int index9 = (int)threadIdx.x / 169 * 16; compute[0] = compute[0] + pad_temp_shared[index0 % 169] * placeholder_shared[index9]; int index10 = (int)threadIdx.x % 169 + 169; compute[0] = compute[0] + (pad_temp_shared[index10] * placeholder_shared[index9 + 1]); int index11 = (int)threadIdx.x % 169 + 338; compute[0] = compute[0] + (pad_temp_shared[index11] * placeholder_shared[index9 + 2]); int index12 = (int)threadIdx.x % 169 + 507; compute[0] = compute[0] + (pad_temp_shared[index12] * placeholder_shared[index9 + 3]); int index13 = (int)threadIdx.x % 169 + 676; compute[0] = compute[0] + (pad_temp_shared[index13] * placeholder_shared[index9 + 4]); int index14 = (int)threadIdx.x % 169 + 845; compute[0] = compute[0] + (pad_temp_shared[index14] * placeholder_shared[index9 + 5]); int index15 = (int)threadIdx.x % 169 + 1014; compute[0] = compute[0] + (pad_temp_shared[index15] * placeholder_shared[index9 + 6]); int index16 = (int)threadIdx.x % 169 + 1183; compute[0] = compute[0] + (pad_temp_shared[index16] * placeholder_shared[index9 + 7]); int index17 = (int)threadIdx.x % 169 + 1352; compute[0] = compute[0] + (pad_temp_shared[index17] * placeholder_shared[index9 + 8]); int index18 = (int)threadIdx.x % 169 + 1521; compute[0] = compute[0] + (pad_temp_shared[index18] * placeholder_shared[index9 + 9]); int index19 = (int)threadIdx.x % 169 + 1690; compute[0] = compute[0] + (pad_temp_shared[index19] * placeholder_shared[index9 + 10]); int index20 = (int)threadIdx.x % 169 + 1859; compute[0] = compute[0] + (pad_temp_shared[index20] * placeholder_shared[index9 + 11]); int index21 = (int)threadIdx.x % 169 + 2028; compute[0] = compute[0] + (pad_temp_shared[index21] * placeholder_shared[index9 + 12]); int index22 = (int)threadIdx.x % 169 + 2197; compute[0] = compute[0] + (pad_temp_shared[index22] * placeholder_shared[index9 + 13]); int index23 = (int)threadIdx.x % 169 + 2366; compute[0] = compute[0] + (pad_temp_shared[index23] * placeholder_shared[index9 + 14]); int index24 = (int)threadIdx.x % 169 + 2535; compute[0] = compute[0] + (pad_temp_shared[index24] * placeholder_shared[index9 + 15]); } int index25 = (int)blockIdx.x * 3380 + (int)threadIdx.x; int index26 = (int)blockIdx.x * 20 + (int)threadIdx.x / 169; T_relu[index25] = max(compute[0] + placeholder2[index26], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpwh77r7zx
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[1352]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<623>; .reg .b32 %r<132>; .reg .b64 %rd<28>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1280]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd12, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd13, %rd9; cvta.to.global.u64 %rd1, %rd11; add.u64 %rd2, %SPL, 0; mov.u64 %rd15, 0; mov.u32 %r125, 0; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(52+i*104)], %r125; mov.f32 %f5, 0f00000000; for.size.2 i in range(0, 13, 1): st.local.v2.f32 [%rd2+(0+i*104)], {%f5, %f5}; st.local.u32 [%rd2+(56+i*104)], %r125; for.size.1 i in range(0, 12, 1): st.local.u32 [%rd2+(8+i*52)], %r125; for.size.1 i in range(0, 7, 1): st.local.u32 [%rd2+(684+i*104)], %r125; for.size.1 i in range(0, 6, 1): st.local.u32 [%rd2+(12+i*104)], %r125; for.size.2 i in range(0, 7, 1): st.local.v2.f32 [%rd2+(632+i*104)], {%f5, %f5}; st.local.u32 [%rd2+(688+i*104)], %r125; for.size.2 i in range(0, 6, 1): st.local.u32 [%rd2+(16+i*104)], %r125; st.local.v2.f32 [%rd2+(64+i*104)], {%f5, %f5}; for.size.1 i in range(0, 6, 1): st.local.u32 [%rd2+(640+i*52)], %r125; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1004+i*104)], %r125; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1160+i*52)], %r125; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1316+i*-1296)], %r125; for.size.1 i in range(0, 8, 1): st.local.u32 [%rd2+(124+i*104)], %r125; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd2+(952+i*104)], %rd15; for.size.3 i in range(0, 2, 1): st.local.u32 [%rd2+(1164+i*-1140)], %r125; st.local.v2.f32 [%rd2+(1264+i*-1192)], {%f5, %f5}; st.local.u32 [%rd2+(1320+i*-1192)], %r125; for.size.2 i in range(0, 11, 1): st.local.v2.f32 [%rd2+(176+i*104)], {%f5, %f5}; st.local.u32 [%rd2+(232+i*104)], %r125; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1324+i*-1296)], %r125; for.size.1 i in range(0, 8, 1): st.local.u32 [%rd2+(80+i*52)], %r125; for.size.1 i in range(0, 8, 1): st.local.u32 [%rd2+(548+i*104)], %r125; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd2+(84+i*104)], %r125; st.local.u64 [%rd2+496], %rd15; for.size.2 i in range(0, 8, 1): st.local.u32 [%rd2+(552+i*104)], %r125; st.local.v2.f32 [%rd2+(600+i*104)], {%f5, %f5}; for.size.2 i in range(0, 5, 1): st.local.v2.f32 [%rd2+(32+i*104)], {%f5, %f5}; st.local.u32 [%rd2+(88+i*104)], %r125; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(556+i*52)], %r125; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1284+i*-1192)], %r125; for.size.1 i in range(0, 10, 1): st.local.u32 [%rd2+(196+i*104)], %r125; for.size.2 i in range(0, 2, 1): st.local.u32 [%rd2+(1184+i*104)], %r125; st.local.v2.f32 [%rd2+(1232+i*104)], {%f5, %f5}; for.size.2 i in range(0, 11, 1): st.local.v2.f32 [%rd2+(40+i*104)], {%f5, %f5}; st.local.u32 [%rd2+(96+i*104)], %r125; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd2+(1188+i*52)], %r125; for.size.1 i in range(0, 26, 1): st.local.u32 [%rd2+(48+i*52)], %r125; cvta.to.global.u64 %rd3, %rd10; cvta.to.global.u64 %rd4, %rd12; mov.u32 %r1, %tid.x; mov.u32 %r2, %ctaid.x; for.size.2 i in range(0, 2, 1): shl.b32 %r(26+i*2), %r1, (8+i*-5); and.b32 %r(27+i*2), %r(26+i*2), (-512+i*520); mad.lo.s32 %r30, %r2, 10240, %r27; or.b32 %r3, %r30, %r29; shl.b32 %r4, %r1, 5; mov.u32 %r31, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r31, %r4; add.s32 %r6, %r1, 30; shl.b32 %r32, %r1, 2; mov.u32 %r33, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r34, %r33, %r32; add.s32 %r7, %r34, 80; add.s64 %rd5, %rd13, 80; $L__BB0_1: bar.sync 0; mad.lo.s32 %r36, %r125, 2704, %r1; mul.wide.s32 %rd16, %r36, 4; add.s64 %rd27, %rd5, %rd16; mov.u32 %r127, 20; mov.u32 %r126, %r7; mov.u32 %r128, %r6; bra.uni $L__BB0_2; $L__BB0_11: for.size.1 i in range(0, 2, 1): add.s32 %r(128+i*-1), %r(128+i*-1), 40; add.s32 %r126, %r126, 160; add.s64 %rd27, %rd27, 160; $L__BB0_2: add.s32 %r37, %r128, -30; setp.gt.s32 %p1, %r37, 2703; @%p1 bra $L__BB0_4; ld.global.nc.f32 %f6, [%rd27+-80]; st.shared.f32 [%r126+-80], %f6; $L__BB0_4: add.s32 %r38, %r128, -20; setp.gt.s32 %p2, %r38, 2703; @%p2 bra $L__BB0_6; ld.global.nc.f32 %f7, [%rd27+-40]; st.shared.f32 [%r126+-40], %f7; $L__BB0_6: add.s32 %r39, %r128, -10; setp.gt.s32 %p3, %r39, 2703; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f8, [%rd27]; st.shared.f32 [%r126], %f8; $L__BB0_8: setp.eq.s32 %p4, %r127, 2700; @%p4 bra $L__BB0_12; setp.gt.s32 %p5, %r128, 2703; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f9, [%rd27+40]; st.shared.f32 [%r126+40], %f9; bra.uni $L__BB0_11; $L__BB0_12: shl.b32 %r41, %r125, 4; add.s32 %r42, %r3, %r41; mul.wide.s32 %rd17, %r42, 4; add.s64 %rd18, %rd3, %rd17; for.size.1 i in range(0, 2, 1): ld.global.nc.v4.u32 {%r(43+i*8), %r(44+i*8), %r(45+i*8), %r(46+i*8)}, [%rd18+(0+i*16)]; mov.u32 %r40, 0; for.size.4 i in range(0, 3, 1): for.size.1 j in range(0, 2, 1): st.shared.v4.u32 [%r5+(0+i*320+j*16)], {%r(43+i*16+j*8), %r(44+i*16+j*8), %r(45+i*16+j*8), %r(46+i*16+j*8)}; for.size.1 j in range(0, 2, 1): ld.global.nc.v4.u32 {%r(59+i*16+j*8), %r(60+i*16+j*8), %r(61+i*16+j*8), %r(62+i*16+j*8)}, [%rd18+(10240+i*10240+j*16)]; for.size.1 i in range(0, 2, 1): st.shared.v4.u32 [%r5+(960+i*16)], {%r(91+i*8), %r(92+i*8), %r(93+i*8), %r(94+i*8)}; bar.sync 0; mov.u32 %r129, %r40; $L__BB0_13: mul.lo.s32 %r17, %r129, 338; for.size.2 i in range(0, 2, 1): shl.b32 %r(108+i*2), %r(129+i*-20), (1+i*1); add.s32 %r(109+i*3), %r(108+i*-77), %r(4+i*106); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): ld.shared.f32 %f(1+i*2+j*1), [%r112+(0+i*4+j*64)]; mov.u32 %r130, %r40; $L__BB0_14: mul.wide.s32 %rd19, %r130, 4; add.s64 %rd20, %rd2, %rd19; add.s32 %r113, %r130, %r17; shl.b32 %r114, %r113, 2; add.s32 %r116, %r33, %r114; for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(10+i*3), [%r116+(0+i*52)]; ld.local.f32 %f(11+i*3), [%rd20+(0+i*52)]; fma.rn.f32 %f(12+i*3), %f(10+i*3), %f1, %f(11+i*3); for.size.2 i in range(0, 13, 1): ld.local.f32 %f(49+i*2), [%rd20+(676+i*52)]; fma.rn.f32 %f(50+i*2), %f(10+i*3), %f2, %f(49+i*2); for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(75+i*2), [%r116+(676+i*52)]; fma.rn.f32 %f(76+i*2), %f(75+i*2), %f3, %f(12+i*3); st.local.f32 [%rd20+(0+i*52)], %f(76+i*2); for.size.2 i in range(0, 13, 1): fma.rn.f32 %f(101+i*1), %f(75+i*2), %f4, %f(50+i*2); st.local.f32 [%rd20+(676+i*52)], %f(101+i*1); add.s32 %r130, %r130, 1; setp.ne.s32 %p6, %r130, 13; @%p6 bra $L__BB0_14; add.s32 %r129, %r129, 1; setp.lt.u32 %p7, %r129, 8; @%p7 bra $L__BB0_13; add.s32 %r125, %r125, 1; setp.lt.u32 %p8, %r125, 32; @%p8 bra $L__BB0_1; shl.b32 %r118, %r1, 1; mad.lo.s32 %r22, %r2, 20, %r118; mul.lo.s32 %r119, %r1, 338; mad.lo.s32 %r23, %r2, 3380, %r119; mov.u32 %r131, 0; $L__BB0_18: add.s32 %r121, %r22, %r131; mul.wide.s32 %rd21, %r121, 4; add.s64 %rd22, %rd4, %rd21; mul.lo.s32 %r122, %r131, 169; add.s32 %r123, %r23, %r122; mul.wide.s32 %rd23, %r122, 4; add.s64 %rd24, %rd2, %rd23; ld.local.f32 %f114, [%rd24]; ld.global.nc.f32 %f115, [%rd22]; add.f32 %f116, %f114, %f115; mov.f32 %f117, 0f00000000; max.f32 %f118, %f116, %f117; mul.wide.s32 %rd25, %r123, 4; add.s64 %rd26, %rd1, %rd25; for.size.4 i in range(0, 168, 1): st.global.f32 [%rd26+(0+i*4)], %f(118+i*3); ld.local.f32 %f(119+i*3), [%rd24+(4+i*4)]; add.f32 %f(120+i*3), %f(119+i*3), %f115; max.f32 %f(121+i*3), %f(120+i*3), %f117; st.global.f32 [%rd26+672], %f622; add.s32 %r124, %r131, 1; mov.u32 %r131, 1; setp.lt.u32 %p9, %r124, 2; @%p9 bra $L__BB0_18; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[338]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[320]; for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_inner_init * 169 + yy_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 271; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 10 + (int)threadIdx.x < 2704) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 10 + (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 10 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 4; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { ulonglong4*(placeholder_shared + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 80 + (int)threadIdx.x * 8)[0] = ulonglong4*(placeholder1 + (int)blockIdx.x * 10240 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2560 + ((int)threadIdx.x >> 1) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 1) * 8)[0]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index3 = ff_inner * 169 + yy_inner * 13 + xx_outer_inner; int index4 = rc_outer_inner * 338 + rc_inner * 169 + yy_inner * 13 + xx_outer_inner; int index5 = (int)threadIdx.x * 32 + ff_inner * 16 + rc_outer_inner * 2 + rc_inner; compute[index3] = compute[index3] + pad_temp_shared[index4] * placeholder_shared[index5]; } } } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index6 = (int)blockIdx.x * 3380 + (int)threadIdx.x * 338 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index7 = ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index8 = (int)blockIdx.x * 20 + (int)threadIdx.x * 2 + ax1_inner; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmprljpowct
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[10400]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<18>; .reg .f32 %f<1010>; .reg .b32 %r<123>; .reg .b64 %rd<40>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd12, [default_function_kernel0_param_0]; ld.param.u64 %rd13, [default_function_kernel0_param_1]; ld.param.u64 %rd14, [default_function_kernel0_param_2]; ld.param.u64 %rd15, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd12; cvta.to.global.u64 %rd2, %rd13; cvta.to.global.u64 %rd3, %rd14; cvta.to.global.u64 %rd4, %rd15; add.u64 %rd5, %SPL, 0; mov.u32 %r43, 0; mov.u64 %rd38, %rd5; mov.u32 %r111, %r43; $L__BB0_1: for.size.2 i in range(0, 10, 1): for.size.1 j in range(0, 10, 1): st.local.u32 [%rd38+(0+i*52+j*1040)], %r43; add.s64 %rd38, %rd38, 4; add.s32 %r111, %r111, 1; setp.ne.s32 %p1, %r111, 13; @%p1 bra $L__BB0_1; mov.u32 %r45, 0; mov.u32 %r112, %r45; $L__BB0_3: add.s32 %r46, %r112, 130; mul.wide.s32 %rd17, %r46, 4; add.s64 %rd18, %rd5, %rd17; for.size.2 i in range(0, 10, 1): for.size.1 j in range(0, 10, 1): st.local.u32 [%rd18+(0+i*52+j*1040)], %r45; add.s32 %r112, %r112, 1; setp.ne.s32 %p2, %r112, 13; @%p2 bra $L__BB0_3; mov.u32 %r5, %tid.x; mul.hi.s32 %r49, %r5, 1321528399; shr.u32 %r50, %r49, 31; shr.s32 %r51, %r49, 2; add.s32 %r6, %r51, %r50; mul.lo.s32 %r52, %r6, 13; sub.s32 %r53, %r5, %r52; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(7+i*1), %r(53+i*-47), (13+i*307); shl.b32 %r54, %r5, 2; mov.u32 %r55, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r56, %r55, %r54; add.s32 %r9, %r56, 260; add.s64 %rd8, %rd1, 260; add.s32 %r10, %r6, 5; mov.u32 %r113, 0; $L__BB0_5: bar.sync 0; mad.lo.s32 %r58, %r113, 2704, %r5; mul.wide.s32 %rd19, %r58, 4; add.s64 %rd39, %rd8, %rd19; mov.u32 %r116, 5; mov.u32 %r114, %r5; mov.u32 %r115, %r10; mov.u32 %r117, %r9; $L__BB0_6: add.s32 %r59, %r115, -5; setp.gt.s32 %p3, %r59, 207; @%p3 bra $L__BB0_9; setp.gt.s32 %p4, %r114, 2703; @%p4 bra $L__BB0_9; ld.global.nc.f32 %f201, [%rd39+-260]; st.shared.f32 [%r117+-260], %f201; $L__BB0_9: setp.gt.s32 %p5, %r115, 207; @%p5 bra $L__BB0_12; add.s32 %r60, %r114, 65; setp.gt.s32 %p6, %r60, 2703; @%p6 bra $L__BB0_12; ld.global.nc.f32 %f202, [%rd39]; st.shared.f32 [%r117], %f202; $L__BB0_12: add.s32 %r117, %r117, 520; add.s64 %rd39, %rd39, 520; for.size.1 i in range(0, 2, 1): add.s32 %r(115+i*-1), %r(115+i*-1), (10+i*120); add.s32 %r116, %r116, 10; setp.ne.s32 %p7, %r116, 215; @%p7 bra $L__BB0_6; shl.b32 %r21, %r113, 4; mov.u32 %r118, 0; bra.uni $L__BB0_14; $L__BB0_23: add.s32 %r118, %r118, 4; $L__BB0_14: mad.lo.s32 %r23, %r118, 65, %r5; setp.gt.s32 %p8, %r23, 15999; shl.b32 %r62, %r23, 2; mov.u32 %r63, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r24, %r63, %r62; @%p8 bra $L__BB0_16; shl.b32 %r64, %r23, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(65+i*1), %r(64+i*-41), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(67+i*1), %r(66+i*1), %r(21+i*44); mul.wide.s32 %rd20, %r68, 4; add.s64 %rd21, %rd2, %rd20; ld.global.nc.f32 %f203, [%rd21]; st.shared.f32 [%r24], %f203; $L__BB0_16: add.s32 %r25, %r23, 65; setp.gt.s32 %p9, %r25, 15999; @%p9 bra $L__BB0_18; shl.b32 %r69, %r25, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(70+i*1), %r(69+i*-44), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(72+i*1), %r(71+i*1), %r(21+i*49); mul.wide.s32 %rd22, %r73, 4; add.s64 %rd23, %rd2, %rd22; ld.global.nc.f32 %f204, [%rd23]; st.shared.f32 [%r24+260], %f204; $L__BB0_18: add.s32 %r26, %r23, 130; setp.gt.s32 %p10, %r26, 15999; @%p10 bra $L__BB0_20; shl.b32 %r74, %r26, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(75+i*1), %r(74+i*-48), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(77+i*1), %r(76+i*1), %r(21+i*54); mul.wide.s32 %rd24, %r78, 4; add.s64 %rd25, %rd2, %rd24; ld.global.nc.f32 %f205, [%rd25]; st.shared.f32 [%r24+520], %f205; $L__BB0_20: add.s32 %r27, %r118, 3; setp.eq.s32 %p11, %r27, 247; @%p11 bra $L__BB0_24; mad.lo.s32 %r28, %r27, 65, %r5; setp.gt.s32 %p12, %r28, 15999; @%p12 bra $L__BB0_23; shl.b32 %r79, %r28, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(80+i*1), %r(79+i*-51), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(82+i*1), %r(81+i*1), %r(21+i*59); mul.wide.s32 %rd26, %r83, 4; add.s64 %rd27, %rd2, %rd26; ld.global.nc.f32 %f206, [%rd27]; st.shared.f32 [%r24+780], %f206; bra.uni $L__BB0_23; $L__BB0_24: bar.sync 0; mov.u32 %r119, 0; $L__BB0_25: mov.u32 %r120, 0; mov.u32 %r109, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r86, %r119, %r8; shl.b32 %r87, %r86, 2; add.s32 %r32, %r109, %r87; for.size.2 i in range(0, 10, 1): for.size.1 j in range(0, 10, 1): ld.shared.f32 %f(1+i*10+j*1), [%r32+(0+i*64+j*6400)]; $L__BB0_26: mad.lo.s32 %r103, %r119, 169, %r7; add.s32 %r89, %r103, %r120; shl.b32 %r90, %r89, 2; add.s32 %r92, %r55, %r90; mul.wide.s32 %rd28, %r120, 4; add.s64 %rd29, %rd5, %rd28; ld.shared.f32 %f207, [%r92]; for.size.4 i in range(0, 10, 1): for.size.3 j in range(0, 10, 1): ld.local.f32 %f(208+i*20+j*2), [%rd29+(0+i*52+j*1040)]; fma.rn.f32 %f(209+i*20+j*2), %f207, %f(1+i*10+j*1), %f(208+i*20+j*2); st.local.f32 [%rd29+(0+i*52+j*1040)], %f(209+i*20+j*2); add.s32 %r120, %r120, 1; setp.ne.s32 %p13, %r120, 13; @%p13 bra $L__BB0_26; ld.shared.f32 %f101, [%r32+640]; mov.u32 %r121, 0; for.size.1 i in range(0, 9, 1): ld.shared.f32 %f(102+i*1), [%r32+(7040+i*6400)]; for.size.2 i in range(0, 9, 1): for.size.1 j in range(0, 10, 1): ld.shared.f32 %f(111+i*10+j*1), [%r32+(704+i*64+j*6400)]; $L__BB0_28: mad.lo.s32 %r104, %r119, 169, %r7; add.s32 %r94, %r104, %r121; shl.b32 %r95, %r94, 2; add.s32 %r97, %r55, %r95; add.s32 %r98, %r121, 130; mul.wide.s32 %rd30, %r98, 4; add.s64 %rd31, %rd5, %rd30; ld.shared.f32 %f408, [%r97]; for.size.4 i in range(0, 10, 1): for.size.3 j in range(0, 10, 1): ld.local.f32 %f(409+i*20+j*2), [%rd31+(0+i*52+j*1040)]; fma.rn.f32 %f(410+i*20+j*2), %f408, %f(101+i*10+j*1), %f(409+i*20+j*2); st.local.f32 [%rd31+(0+i*52+j*1040)], %f(410+i*20+j*2); add.s32 %r121, %r121, 1; setp.ne.s32 %p14, %r121, 13; @%p14 bra $L__BB0_28; add.s32 %r119, %r119, 1; setp.lt.u32 %p15, %r119, 16; @%p15 bra $L__BB0_25; add.s32 %r113, %r113, 1; setp.lt.u32 %p16, %r113, 32; @%p16 bra $L__BB0_5; mul.hi.s32 %r108, %r5, 1321528399; shr.u32 %r107, %r108, 31; shr.s32 %r106, %r108, 2; add.s32 %r105, %r106, %r107; mul.lo.s32 %r39, %r105, 20; mad.lo.s32 %r40, %r105, 3380, %r7; mov.u32 %r122, 0; $L__BB0_32: add.s32 %r100, %r122, %r39; mul.wide.s32 %rd32, %r100, 4; add.s64 %rd33, %rd4, %rd32; mad.lo.s32 %r101, %r122, 169, %r40; for.size.1 i in range(0, 9, 1): ld.global.nc.f32 %f(609+i*1), [%rd33+(400+i*400)]; mul.lo.s32 %r102, %r122, 13; mul.wide.s32 %rd34, %r102, 4; add.s64 %rd35, %rd5, %rd34; ld.local.f32 %f618, [%rd35]; ld.global.nc.f32 %f619, [%rd33]; add.f32 %f620, %f618, %f619; mov.f32 %f621, 0f00000000; max.f32 %f622, %f620, %f621; mul.wide.s32 %rd36, %r101, 4; add.s64 %rd37, %rd3, %rd36; for.size.4 i in range(0, 9, 1): st.global.f32 [%rd37+(0+i*67600)], %f(622+i*3); ld.local.f32 %f(623+i*3), [%rd35+(1040+i*1040)]; add.f32 %f(624+i*3), %f(623+i*3), %f(609+i*1); max.f32 %f(625+i*3), %f(624+i*3), %f621; for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd37+(608400+i*4+j*-608396)], %f(649+i*30+j*3); ld.local.f32 %f(650+i*30+j*3), [%rd35+(4+i*4+j*1040)]; add.f32 %f(651+i*30+j*3), %f(650+i*30+j*3), %f(619+j*-10); max.f32 %f(652+i*30+j*3), %f(651+i*30+j*3), %f621; for.size.4 j in range(0, 8, 1): st.global.f32 [%rd37+(67604+i*4+j*67600)], %f(655+i*30+j*3); ld.local.f32 %f(656+i*30+j*3), [%rd35+(2084+i*4+j*1040)]; add.f32 %f(657+i*30+j*3), %f(656+i*30+j*3), %f(610+j*1); max.f32 %f(658+i*30+j*3), %f(657+i*30+j*3), %f621; st.global.f32 [%rd37+608448], %f1009; add.s32 %r122, %r122, 1; setp.ne.s32 %p17, %r122, 20; @%p17 bra $L__BB0_32; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2600]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 10; ++ff_inner_init) { int index0 = ff_outer_inner_init * 130 + ff_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 520] = 0.000000e+00f; compute[index0 + 780] = 0.000000e+00f; compute[index0 + 1040] = 0.000000e+00f; compute[index0 + 1300] = 0.000000e+00f; compute[index0 + 1560] = 0.000000e+00f; compute[index0 + 1820] = 0.000000e+00f; compute[index0 + 2080] = 0.000000e+00f; compute[index0 + 2340] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 42; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x / 13 < 208) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x < 2704) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 247; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x < 16000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int ff_inner = 0; ff_inner < 10; ++ff_inner) { int index5 = ff_outer_inner * 130 + ff_inner * 13 + xx_outer_inner; int index6 = rc_outer_inner * 169 + (int)threadIdx.x % 13 * 13 + xx_outer_inner; int index7 = (int)threadIdx.x / 13 * 320 + ff_outer_inner * 160 + ff_inner * 16 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 260] = compute[index5 + 260] + (pad_temp_shared[index6] * placeholder_shared[index7 + 1600]); compute[index5 + 520] = compute[index5 + 520] + (pad_temp_shared[index6] * placeholder_shared[index7 + 3200]); compute[index5 + 780] = compute[index5 + 780] + (pad_temp_shared[index6] * placeholder_shared[index7 + 4800]); compute[index5 + 1040] = compute[index5 + 1040] + (pad_temp_shared[index6] * placeholder_shared[index7 + 6400]); compute[index5 + 1300] = compute[index5 + 1300] + (pad_temp_shared[index6] * placeholder_shared[index7 + 8000]); compute[index5 + 1560] = compute[index5 + 1560] + (pad_temp_shared[index6] * placeholder_shared[index7 + 9600]); compute[index5 + 1820] = compute[index5 + 1820] + (pad_temp_shared[index6] * placeholder_shared[index7 + 11200]); compute[index5 + 2080] = compute[index5 + 2080] + (pad_temp_shared[index6] * placeholder_shared[index7 + 12800]); compute[index5 + 2340] = compute[index5 + 2340] + (pad_temp_shared[index6] * placeholder_shared[index7 + 14400]); } } } } } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)threadIdx.x / 13 * 3380 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + ax3_inner; int index9 = ax1_inner * 13 + ax3_inner; int index10 = (int)threadIdx.x / 13 * 20 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 16900] = max(compute[index9 + 260] + placeholder2[index10 + 100], 0.000000e+00f); T_relu[index8 + 33800] = max(compute[index9 + 520] + placeholder2[index10 + 200], 0.000000e+00f); T_relu[index8 + 50700] = max(compute[index9 + 780] + placeholder2[index10 + 300], 0.000000e+00f); T_relu[index8 + 67600] = max(compute[index9 + 1040] + placeholder2[index10 + 400], 0.000000e+00f); T_relu[index8 + 84500] = max(compute[index9 + 1300] + placeholder2[index10 + 500], 0.000000e+00f); T_relu[index8 + 101400] = max(compute[index9 + 1560] + placeholder2[index10 + 600], 0.000000e+00f); T_relu[index8 + 118300] = max(compute[index9 + 1820] + placeholder2[index10 + 700], 0.000000e+00f); T_relu[index8 + 135200] = max(compute[index9 + 2080] + placeholder2[index10 + 800], 0.000000e+00f); T_relu[index8 + 152100] = max(compute[index9 + 2340] + placeholder2[index10 + 900], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpsb_c9162
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[10400]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<22>; .reg .f32 %f<712>; .reg .b32 %r<123>; .reg .b64 %rd<52>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[128000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd15, [default_function_kernel0_param_0]; ld.param.u64 %rd16, [default_function_kernel0_param_1]; add.u64 %rd49, %SPL, 0; mov.u32 %r39, 0; mov.u64 %rd47, %rd49; mov.u32 %r111, %r39; $L__BB0_1: for.size.2 i in range(0, 20, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd47+(0+i*52+j*2080)], %r39; add.s64 %rd47, %rd47, 4; add.s32 %r111, %r111, 1; setp.ne.s32 %p1, %r111, 13; @%p1 bra $L__BB0_1; mov.u32 %r41, 0; mov.u32 %r112, %r41; $L__BB0_3: add.s32 %r42, %r112, 260; mul.wide.s32 %rd20, %r42, 4; add.s64 %rd21, %rd49, %rd20; for.size.2 i in range(0, 20, 1): for.size.1 j in range(0, 5, 1): st.local.u32 [%rd21+(0+i*52+j*2080)], %r41; add.s32 %r112, %r112, 1; setp.ne.s32 %p2, %r112, 13; @%p2 bra $L__BB0_3; mov.u32 %r45, %tid.x; mul.hi.s32 %r46, %r45, 1321528399; shr.u32 %r47, %r46, 31; shr.s32 %r48, %r46, 2; add.s32 %r5, %r48, %r47; mul.lo.s32 %r49, %r5, 13; sub.s32 %r50, %r45, %r49; mul.lo.s32 %r6, %r50, 13; mov.u32 %r113, 0; cvta.to.global.u64 %rd22, %rd15; $L__BB0_5: shl.b32 %r53, %r45, 2; mov.u32 %r54, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r55, %r54, %r53; add.s32 %r115, %r55, 520; bar.sync 0; mad.lo.s32 %r56, %r113, 5408, %r45; mul.wide.s32 %rd23, %r56, 4; add.s64 %rd24, %rd22, %rd23; add.s64 %rd48, %rd24, 520; add.s32 %r116, %r45, 65; mov.u32 %r114, 10; mov.u32 %r117, %r5; $L__BB0_6: setp.gt.s32 %p3, %r117, 415; @%p3 bra $L__BB0_9; add.s32 %r57, %r116, -65; setp.gt.s32 %p4, %r57, 5407; @%p4 bra $L__BB0_9; ld.global.nc.f32 %f201, [%rd48+-520]; st.shared.f32 [%r115+-520], %f201; $L__BB0_9: add.s32 %r58, %r117, 5; setp.gt.s32 %p5, %r58, 415; @%p5 bra $L__BB0_12; setp.gt.s32 %p6, %r116, 5407; @%p6 bra $L__BB0_12; ld.global.nc.f32 %f202, [%rd48+-260]; st.shared.f32 [%r115+-260], %f202; $L__BB0_12: add.s32 %r59, %r117, 10; setp.gt.s32 %p7, %r59, 415; @%p7 bra $L__BB0_15; add.s32 %r60, %r116, 65; setp.gt.s32 %p8, %r60, 5407; @%p8 bra $L__BB0_15; ld.global.nc.f32 %f203, [%rd48]; st.shared.f32 [%r115], %f203; $L__BB0_15: add.s32 %r61, %r117, 15; setp.gt.s32 %p9, %r61, 415; @%p9 bra $L__BB0_18; add.s32 %r62, %r116, 130; setp.gt.s32 %p10, %r62, 5407; @%p10 bra $L__BB0_18; ld.global.nc.f32 %f204, [%rd48+260]; st.shared.f32 [%r115+260], %f204; $L__BB0_18: for.size.1 i in range(0, 2, 1): add.s32 %r(117+i*-1), %r(117+i*-1), (20+i*240); add.s32 %r115, %r115, 1040; add.s64 %rd48, %rd48, 1040; add.s32 %r114, %r114, 20; setp.ne.s32 %p11, %r114, 430; @%p11 bra $L__BB0_6; shl.b32 %r18, %r113, 5; mov.u32 %r118, 0; bra.uni $L__BB0_20; $L__BB0_29: add.s32 %r118, %r118, 4; $L__BB0_20: mad.lo.s32 %r20, %r118, 65, %r45; setp.gt.s32 %p12, %r20, 31999; shl.b32 %r65, %r20, 2; mov.u32 %r66, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r21, %r66, %r65; @%p12 bra $L__BB0_22; shl.b32 %r67, %r20, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(68+i*1), %r(67+i*-47), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(70+i*1), %r(69+i*1), %r(18+i*50); cvta.to.global.u64 %rd25, %rd16; mul.wide.s32 %rd26, %r71, 4; add.s64 %rd27, %rd25, %rd26; ld.global.nc.f32 %f205, [%rd27]; st.shared.f32 [%r21], %f205; $L__BB0_22: add.s32 %r22, %r118, 1; setp.eq.s32 %p13, %r22, 493; @%p13 bra $L__BB0_30; mad.lo.s32 %r23, %r22, 65, %r45; setp.gt.s32 %p14, %r23, 31999; @%p14 bra $L__BB0_25; shl.b32 %r73, %r23, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(74+i*1), %r(73+i*-50), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(76+i*1), %r(75+i*1), %r(18+i*56); cvta.to.global.u64 %rd28, %rd16; mul.wide.s32 %rd29, %r77, 4; add.s64 %rd30, %rd28, %rd29; ld.global.nc.f32 %f206, [%rd30]; st.shared.f32 [%r21+260], %f206; $L__BB0_25: add.s32 %r24, %r20, 130; setp.gt.s32 %p15, %r24, 31999; @%p15 bra $L__BB0_27; shl.b32 %r78, %r24, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(79+i*1), %r(78+i*-54), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(81+i*1), %r(80+i*1), %r(18+i*61); cvta.to.global.u64 %rd31, %rd16; mul.wide.s32 %rd32, %r82, 4; add.s64 %rd33, %rd31, %rd32; ld.global.nc.f32 %f207, [%rd33]; st.shared.f32 [%r21+520], %f207; $L__BB0_27: add.s32 %r25, %r20, 195; setp.gt.s32 %p16, %r25, 31999; @%p16 bra $L__BB0_29; shl.b32 %r83, %r25, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(84+i*1), %r(83+i*-58), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(86+i*1), %r(85+i*1), %r(18+i*66); cvta.to.global.u64 %rd34, %rd16; mul.wide.s32 %rd35, %r87, 4; add.s64 %rd36, %rd34, %rd35; ld.global.nc.f32 %f208, [%rd36]; st.shared.f32 [%r21+780], %f208; bra.uni $L__BB0_29; $L__BB0_30: bar.sync 0; mov.u32 %r119, 0; $L__BB0_31: mov.u32 %r120, 0; $L__BB0_32: shl.b32 %r110, %r119, 1; mad.lo.s32 %r109, %r5, 1280, %r110; mov.u32 %r108, _ZZ24default_function_kernel0E18placeholder_shared; mad.lo.s32 %r92, %r120, 640, %r109; shl.b32 %r93, %r92, 2; add.s32 %r95, %r108, %r93; ld.shared.f32 %f1, [%r95]; mov.u32 %r121, 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(2+i*1), [%r95+(4+i*25596)]; for.size.1 i in range(0, 3, 1): ld.shared.f32 %f(4+i*1), [%r95+(51200+i*25600)]; for.size.2 i in range(0, 19, 1): for.size.1 j in range(0, 5, 1): ld.shared.f32 %f(7+i*5+j*1), [%r95+(128+i*128+j*25600)]; for.size.1 i in range(0, 4, 1): ld.shared.f32 %f(102+i*1), [%r95+(25604+i*25600)]; for.size.2 i in range(0, 19, 1): for.size.1 j in range(0, 5, 1): ld.shared.f32 %f(106+i*5+j*1), [%r95+(132+i*128+j*25600)]; $L__BB0_33: mov.u32 %r106, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r105, %r119, 338, %r6; mul.lo.s32 %r104, %r120, 260; add.s32 %r96, %r121, %r104; mul.wide.s32 %rd37, %r96, 4; add.s64 %rd38, %rd49, %rd37; add.s32 %r97, %r105, %r121; shl.b32 %r98, %r97, 2; add.s32 %r100, %r106, %r98; ld.shared.f32 %f209, [%r100]; for.size.2 i in range(0, 2, 1): ld.local.f32 %f(210+i*2), [%rd38+(0+i*2080)]; fma.rn.f32 %f(211+i*2), %f209, %f(1+i*2), %f(210+i*2); for.size.2 i in range(0, 3, 1): ld.local.f32 %f(214+i*2), [%rd38+(4160+i*2080)]; fma.rn.f32 %f(215+i*2), %f209, %f(4+i*1), %f(214+i*2); for.size.3 i in range(0, 19, 1): for.size.2 j in range(0, 5, 1): ld.local.f32 %f(220+i*10+j*2), [%rd38+(52+i*52+j*2080)]; fma.rn.f32 %f(221+i*10+j*2), %f209, %f(7+i*5+j*1), %f(220+i*10+j*2); ld.shared.f32 %f410, [%r100+676]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(411+i*1), %f410, %f(2+i*100), %f(211+i*2); st.local.f32 [%rd38+(0+i*2080)], %f(411+i*1); for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(413+i*1), %f410, %f(103+i*1), %f(215+i*2); st.local.f32 [%rd38+(4160+i*2080)], %f(413+i*1); for.size.3 i in range(0, 19, 1): for.size.2 j in range(0, 5, 1): fma.rn.f32 %f(416+i*5+j*1), %f410, %f(106+i*5+j*1), %f(221+i*10+j*2); st.local.f32 [%rd38+(52+i*52+j*2080)], %f(416+i*5+j*1); add.s32 %r121, %r121, 1; setp.ne.s32 %p17, %r121, 13; @%p17 bra $L__BB0_33; add.s32 %r120, %r120, 1; setp.lt.u32 %p18, %r120, 2; @%p18 bra $L__BB0_32; add.s32 %r119, %r119, 1; setp.lt.u32 %p19, %r119, 16; @%p19 bra $L__BB0_31; add.s32 %r113, %r113, 1; setp.lt.u32 %p20, %r113, 16; @%p20 bra $L__BB0_5; ld.param.u64 %rd46, [default_function_kernel0_param_2]; ld.param.u64 %rd45, [default_function_kernel0_param_3]; mul.lo.s32 %r102, %r5, 40; cvta.to.global.u64 %rd39, %rd45; mul.wide.s32 %rd40, %r102, 4; add.s64 %rd41, %rd39, %rd40; add.s64 %rd51, %rd41, 1600; mad.lo.s32 %r103, %r5, 6760, %r6; cvta.to.global.u64 %rd42, %rd46; mul.wide.s32 %rd43, %r103, 4; add.s64 %rd44, %rd42, %rd43; add.s64 %rd50, %rd44, 270424; mov.u32 %r122, 0; $L__BB0_38: for.size.1 i in range(0, 4, 1): ld.global.nc.f32 %f(511+i*1), [%rd51+(-800+i*800)]; ld.local.f32 %f515, [%rd49]; ld.global.nc.f32 %f516, [%rd51+-1600]; add.f32 %f517, %f515, %f516; mov.f32 %f518, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(519+i*3), %f(517+i*4), %f518; st.global.f32 [%rd50+(-270424+i*135200)], %f(519+i*3); ld.local.f32 %f(520+i*3), [%rd49+(2080+i*2080)]; add.f32 %f(521+i*3), %f(520+i*3), %f(511+i*1); for.size.4 i in range(0, 2, 1): max.f32 %f(525+i*3), %f(524+i*3), %f518; st.global.f32 [%rd50+(-24+i*135200)], %f(525+i*3); ld.local.f32 %f(526+i*3), [%rd49+(6240+i*2080)]; add.f32 %f(527+i*3), %f(526+i*3), %f(513+i*1); for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(531+i*15+j*3), %f(530+i*15+j*3), %f518; st.global.f32 [%rd50+(270376+i*4+j*-540796)], %f(531+i*15+j*3); ld.local.f32 %f(532+i*15+j*3), [%rd49+(4+i*4+j*2080)]; add.f32 %f(533+i*15+j*3), %f(532+i*15+j*3), %f(516+j*-5); for.size.4 j in range(0, 3, 1): max.f32 %f(537+i*15+j*3), %f(536+i*15+j*3), %f518; st.global.f32 [%rd50+(-135220+i*4+j*135200)], %f(537+i*15+j*3); ld.local.f32 %f(538+i*15+j*3), [%rd49+(4164+i*4+j*2080)]; add.f32 %f(539+i*15+j*3), %f(538+i*15+j*3), %f(512+j*1); max.f32 %f711, %f710, %f518; st.global.f32 [%rd50+270424], %f711; for.size.1 i in range(0, 2, 1): add.s64 %rd(51+i*-1), %rd(51+i*-1), (4+i*672); add.s64 %rd49, %rd49, 52; add.s32 %r122, %r122, 1; setp.ne.s32 %p21, %r122, 40; @%p21 bra $L__BB0_38; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2600]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[32000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 20; ++ff_inner_init) { int index0 = ff_outer_inner_init * 260 + ff_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 520] = 0.000000e+00f; compute[index0 + 1040] = 0.000000e+00f; compute[index0 + 1560] = 0.000000e+00f; compute[index0 + 2080] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 84; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x / 13 < 416) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x < 5408) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 493; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x < 32000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 65 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 20; ++ff_inner) { int index5 = ff_outer_inner * 260 + ff_inner * 13 + xx_outer_inner; int index6 = rc_outer_inner * 338 + rc_inner * 169 + (int)threadIdx.x % 13 * 13 + xx_outer_inner; int index7 = (int)threadIdx.x / 13 * 1280 + ff_outer_inner * 640 + ff_inner * 32 + rc_outer_inner * 2 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 520] = compute[index5 + 520] + (pad_temp_shared[index6] * placeholder_shared[index7 + 6400]); compute[index5 + 1040] = compute[index5 + 1040] + (pad_temp_shared[index6] * placeholder_shared[index7 + 12800]); compute[index5 + 1560] = compute[index5 + 1560] + (pad_temp_shared[index6] * placeholder_shared[index7 + 19200]); compute[index5 + 2080] = compute[index5 + 2080] + (pad_temp_shared[index6] * placeholder_shared[index7 + 25600]); } } } } } } for (int ax1_inner = 0; ax1_inner < 40; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)threadIdx.x / 13 * 6760 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + ax3_inner; int index9 = ax1_inner * 13 + ax3_inner; int index10 = (int)threadIdx.x / 13 * 40 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 33800] = max(compute[index9 + 520] + placeholder2[index10 + 200], 0.000000e+00f); T_relu[index8 + 67600] = max(compute[index9 + 1040] + placeholder2[index10 + 400], 0.000000e+00f); T_relu[index8 + 101400] = max(compute[index9 + 1560] + placeholder2[index10 + 600], 0.000000e+00f); T_relu[index8 + 135200] = max(compute[index9 + 2080] + placeholder2[index10 + 800], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmplxs41cn6
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[676]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<623>; .reg .b32 %r<55>; .reg .b64 %rd<22>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[409600]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd12, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd9; cvta.to.global.u64 %rd2, %rd11; add.u64 %rd3, %SPL, 0; mov.u32 %r50, 0; for.size.1 i in range(0, 169, 1): st.local.u32 [%rd3+(0+i*4)], %r50; cvta.to.global.u64 %rd5, %rd10; cvta.to.global.u64 %rd6, %rd12; mov.u32 %r1, %tid.x; mov.u32 %r23, _ZZ24default_function_kernel0E15pad_temp_shared; bra.uni $L__BB0_1; $L__BB0_10: add.s32 %r50, %r50, 4; $L__BB0_1: mad.lo.s32 %r3, %r50, 200, %r1; setp.gt.s32 %p1, %r3, 86527; mul.wide.s32 %rd14, %r3, 4; add.s64 %rd7, %rd1, %rd14; shl.b32 %r22, %r3, 2; add.s32 %r4, %r23, %r22; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f40, [%rd7]; st.shared.f32 [%r4], %f40; $L__BB0_3: add.s32 %r5, %r50, 1; setp.eq.s32 %p2, %r5, 433; @%p2 bra $L__BB0_11; mad.lo.s32 %r24, %r5, 200, %r1; setp.gt.s32 %p3, %r24, 86527; @%p3 bra $L__BB0_6; ld.global.nc.f32 %f41, [%rd7+800]; st.shared.f32 [%r4+800], %f41; $L__BB0_6: add.s32 %r25, %r3, 400; setp.gt.s32 %p4, %r25, 86527; @%p4 bra $L__BB0_8; ld.global.nc.f32 %f42, [%rd7+1600]; st.shared.f32 [%r4+1600], %f42; $L__BB0_8: add.s32 %r26, %r3, 600; setp.gt.s32 %p5, %r26, 86527; @%p5 bra $L__BB0_10; ld.global.nc.f32 %f43, [%rd7+2400]; st.shared.f32 [%r4+2400], %f43; bra.uni $L__BB0_10; $L__BB0_11: mov.u32 %r7, %ctaid.x; mad.lo.s32 %r8, %r7, 102400, %r1; mov.u32 %r51, 0; $L__BB0_12: mul.lo.s32 %r28, %r51, 200; add.s32 %r29, %r8, %r28; mul.wide.s32 %rd15, %r29, 4; add.s64 %rd16, %rd5, %rd15; ld.global.nc.f32 %f44, [%rd16]; add.s32 %r30, %r28, %r1; shl.b32 %r31, %r30, 2; mov.u32 %r32, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r33, %r32, %r31; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r33+(0+i*800)], %f(44+i*1); ld.global.nc.f32 %f(45+i*1), [%rd16+(800+i*800)]; st.shared.f32 [%r33+12000], %f59; add.s32 %r51, %r51, 16; setp.ne.s32 %p6, %r51, 512; @%p6 bra $L__BB0_12; bar.sync 0; shl.b32 %r11, %r1, 9; mov.u32 %r34, 0; mov.u32 %r52, %r34; $L__BB0_14: mul.lo.s32 %r13, %r52, 1352; shl.b32 %r36, %r52, 3; add.s32 %r14, %r36, %r11; mov.u32 %r53, %r34; $L__BB0_15: mul.lo.s32 %r38, %r53, 13; mul.wide.s32 %rd17, %r38, 4; add.s64 %rd8, %rd3, %rd17; add.s32 %r16, %r38, %r13; for.size.1 i in range(0, 13, 1): ld.local.f32 %f(622+i*-1), [%rd8+(0+i*4)]; mov.u32 %r54, 0; $L__BB0_16: mad.lo.s32 %r39, %r54, 169, %r16; shl.b32 %r40, %r39, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(42+i*1), %r(23+i*-9), %r(40+i*14); shl.b32 %r44, %r43, 2; add.s32 %r46, %r32, %r44; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(60+i*1), [%r(46+i*-4)+0]; for.size.2 i in range(0, 12, 1): fma.rn.f32 %f(62+i*2), %f(61+i*2), %f60, %f(622+i*-1); ld.shared.f32 %f(63+i*2), [%r42+(4+i*4)]; fma.rn.f32 %f86, %f85, %f60, %f610; ld.shared.f32 %f87, [%r46+4]; for.size.2 i in range(0, 13, 1): ld.shared.f32 %f(88+i*1), [%r42+(676+i*4)]; fma.rn.f32 %f(622+i*-1), %f(88+i*1), %f87, %f(62+i*2); add.s32 %r54, %r54, 2; setp.ne.s32 %p7, %r54, 8; @%p7 bra $L__BB0_16; for.size.1 i in range(0, 13, 1): st.local.f32 [%rd8+(0+i*4)], %f(622+i*-1); add.s32 %r53, %r53, 1; setp.lt.u32 %p8, %r53, 13; @%p8 bra $L__BB0_15; add.s32 %r52, %r52, 1; setp.lt.u32 %p9, %r52, 64; @%p9 bra $L__BB0_14; mad.lo.s32 %r47, %r7, 200, %r1; mul.wide.s32 %rd18, %r47, 4; add.s64 %rd19, %rd6, %rd18; mul.lo.s32 %r48, %r1, 169; mad.lo.s32 %r49, %r7, 33800, %r48; ld.local.f32 %f101, [%rd3]; ld.global.nc.f32 %f102, [%rd19]; add.f32 %f103, %f101, %f102; mov.f32 %f104, 0f00000000; max.f32 %f105, %f103, %f104; mul.wide.s32 %rd20, %r49, 4; add.s64 %rd21, %rd2, %rd20; for.size.4 i in range(0, 168, 1): st.global.f32 [%rd21+(0+i*4)], %f(105+i*3); ld.local.f32 %f(106+i*3), [%rd3+(4+i*4)]; add.f32 %f(107+i*3), %f(106+i*3), %f102; max.f32 %f(108+i*3), %f(107+i*3), %f104; st.global.f32 [%rd21+672], %f609; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[169]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[102400]; for (int yy_outer_inner_init = 0; yy_outer_inner_init < 13; ++yy_outer_inner_init) { int index0 = yy_outer_inner_init * 13; compute[index0] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; compute[index0 + 7] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 10] = 0.000000e+00f; compute[index0 + 11] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 433; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x < 86528) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index1]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 512; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 200 + (int)threadIdx.x; int index3 = (int)blockIdx.x * 102400 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 200 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int yy_outer_inner = 0; yy_outer_inner < 13; ++yy_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { int index4 = yy_outer_inner * 13; int index5 = rc_outer_inner * 1352 + rc_inner * 169 + yy_outer_inner * 13; int index6 = (int)threadIdx.x * 512 + rc_outer_inner * 8 + rc_inner; compute[index4] = compute[index4] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[index4 + 1] = compute[index4 + 1] + (pad_temp_shared[index5 + 1] * placeholder_shared[index6]); compute[index4 + 2] = compute[index4 + 2] + (pad_temp_shared[index5 + 2] * placeholder_shared[index6]); compute[index4 + 3] = compute[index4 + 3] + (pad_temp_shared[index5 + 3] * placeholder_shared[index6]); compute[index4 + 4] = compute[index4 + 4] + (pad_temp_shared[index5 + 4] * placeholder_shared[index6]); compute[index4 + 5] = compute[index4 + 5] + (pad_temp_shared[index5 + 5] * placeholder_shared[index6]); compute[index4 + 6] = compute[index4 + 6] + (pad_temp_shared[index5 + 6] * placeholder_shared[index6]); compute[index4 + 7] = compute[index4 + 7] + (pad_temp_shared[index5 + 7] * placeholder_shared[index6]); compute[index4 + 8] = compute[index4 + 8] + (pad_temp_shared[index5 + 8] * placeholder_shared[index6]); compute[index4 + 9] = compute[index4 + 9] + (pad_temp_shared[index5 + 9] * placeholder_shared[index6]); compute[index4 + 10] = compute[index4 + 10] + (pad_temp_shared[index5 + 10] * placeholder_shared[index6]); compute[index4 + 11] = compute[index4 + 11] + (pad_temp_shared[index5 + 11] * placeholder_shared[index6]); compute[index4 + 12] = compute[index4 + 12] + (pad_temp_shared[index5 + 12] * placeholder_shared[index6]); } } } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index7 = (int)blockIdx.x * 33800 + (int)threadIdx.x * 169 + ax2_inner * 13 + ax3_inner; int index8 = ax2_inner * 13 + ax3_inner; int index9 = (int)blockIdx.x * 200 + (int)threadIdx.x; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpryvkx9ry
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<10>; .reg .f32 %f<106>; .reg .b32 %r<41>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[104]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[4000]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; mul.hi.s32 %r12, %r1, 1321528399; shr.u32 %r13, %r12, 31; shr.s32 %r14, %r12, 2; add.s32 %r2, %r14, %r13; mul.lo.s32 %r15, %r2, 169; mov.u32 %r16, %ctaid.x; mul.hi.s32 %r17, %r16, 1321528399; shr.u32 %r18, %r17, 31; shr.s32 %r19, %r17, 2; add.s32 %r5, %r19, %r18; mul.lo.s32 %r20, %r5, 13; sub.s32 %r21, %r16, %r20; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(22+i*1), %r(21+i*-19), 13; sub.s32 %r24, %r1, %r23; for.size.1 i in range(0, 2, 1): add.s32 %r(25+i*-22), %r(15+i*10), %r(24+i*-2); shl.b32 %r26, %r1, 2; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r27, %r26; and.b32 %r28, %r1, 1; mov.u32 %r29, _ZZ24default_function_kernel0E18placeholder_shared; for.size.2 i in range(0, 2, 1): add.s32 %r(6+i*1), %r(29+i*-2), %r(26+i*4); shl.b32 %r(30+i*1), %r(24+i*-22), (2+i*1); add.s32 %r8, %r29, %r31; shl.b32 %r32, %r1, 8; and.b32 %r33, %r32, -512; mad.lo.s32 %r34, %r5, 256000, %r33; or.b32 %r35, %r34, %r28; for.size.1 i in range(0, 2, 1): add.s32 %r(36+i*1), %r(24+i*12), %r(22+i*-7); cvta.to.global.u64 %rd13, %rd9; for.size.2 i in range(0, 2, 1): mul.wide.s32 %rd(14+i*1), %r(37+i*-2), 4; add.s64 %rd(21+i*1), %rd(13+i*-1), %rd(14+i*1); mov.f32 %f102, 0f00000000; mov.u32 %r40, 0; for.size.1 i in range(0, 3, 1): mov.f32 %f(103+i*1), %f102; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 25; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f25, [%rd21]; st.shared.f32 [%r4], %f25; $L__BB0_3: setp.gt.s32 %p2, %r1, 999; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f26, [%rd22]; st.shared.f32 [%r6], %f26; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(27+i*1), [%r(8+i*-1)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(29+i*2), %f28, %f(27+i*3), %f(102+i*1); ld.shared.f32 %f(30+i*2), [%r8+(1000+i*1000)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(33+i*2), %f28, %f(32+i*2), %f(104+i*1); ld.shared.f32 %f(34+i*2), [%r8+(3000+i*-2996)]; ld.shared.f32 %f37, [%r7+52]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(5+i*1), %f37, %f(36+i*2), %f(29+i*2); ld.shared.f32 %f(38+i*1), [%r8+(1004+i*1000)]; fma.rn.f32 %f7, %f37, %f39, %f33; ld.shared.f32 %f40, [%r8+3004]; fma.rn.f32 %f8, %f37, %f40, %f35; bar.sync 0; @%p1 bra $L__BB0_7; ld.global.nc.f32 %f41, [%rd21+1352]; st.shared.f32 [%r4], %f41; $L__BB0_7: @%p2 bra $L__BB0_9; ld.global.nc.f32 %f42, [%rd22+8]; st.shared.f32 [%r6], %f42; $L__BB0_9: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(43+i*1), [%r(8+i*-1)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(45+i*2), %f44, %f(43+i*3), %f(5+i*1); ld.shared.f32 %f(46+i*2), [%r8+(1000+i*1000)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(49+i*2), %f44, %f(48+i*2), %f(7+i*1); ld.shared.f32 %f(50+i*2), [%r8+(3000+i*-2996)]; ld.shared.f32 %f53, [%r7+52]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(9+i*1), %f53, %f(52+i*2), %f(45+i*2); ld.shared.f32 %f(54+i*1), [%r8+(1004+i*1000)]; fma.rn.f32 %f11, %f53, %f55, %f49; ld.shared.f32 %f56, [%r8+3004]; fma.rn.f32 %f12, %f53, %f56, %f51; bar.sync 0; @%p1 bra $L__BB0_11; ld.global.nc.f32 %f57, [%rd21+2704]; st.shared.f32 [%r4], %f57; $L__BB0_11: @%p2 bra $L__BB0_13; ld.global.nc.f32 %f58, [%rd22+16]; st.shared.f32 [%r6], %f58; $L__BB0_13: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(59+i*1), [%r(8+i*-1)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(61+i*2), %f60, %f(59+i*3), %f(9+i*1); ld.shared.f32 %f(62+i*2), [%r8+(1000+i*1000)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(65+i*2), %f60, %f(64+i*2), %f(11+i*1); ld.shared.f32 %f(66+i*2), [%r8+(3000+i*-2996)]; ld.shared.f32 %f69, [%r7+52]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(13+i*1), %f69, %f(68+i*2), %f(61+i*2); ld.shared.f32 %f(70+i*1), [%r8+(1004+i*1000)]; fma.rn.f32 %f15, %f69, %f71, %f65; ld.shared.f32 %f72, [%r8+3004]; fma.rn.f32 %f16, %f69, %f72, %f67; bar.sync 0; @%p1 bra $L__BB0_15; ld.global.nc.f32 %f73, [%rd21+4056]; st.shared.f32 [%r4], %f73; $L__BB0_15: @%p2 bra $L__BB0_17; ld.global.nc.f32 %f74, [%rd22+24]; st.shared.f32 [%r6], %f74; $L__BB0_17: for.size.1 i in range(0, 2, 1): add.s64 %rd(22+i*-1), %rd(22+i*-1), (32+i*5376); bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(75+i*1), [%r(8+i*-1)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(77+i*2), %f76, %f(75+i*3), %f(13+i*1); ld.shared.f32 %f(78+i*2), [%r8+(1000+i*1000)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(81+i*2), %f76, %f(80+i*2), %f(15+i*1); ld.shared.f32 %f(82+i*2), [%r8+(3000+i*-2996)]; ld.shared.f32 %f85, [%r7+52]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(102+i*1), %f85, %f(84+i*2), %f(77+i*2); ld.shared.f32 %f(86+i*1), [%r8+(1004+i*1000)]; fma.rn.f32 %f104, %f85, %f87, %f81; ld.shared.f32 %f88, [%r8+3004]; fma.rn.f32 %f105, %f85, %f88, %f83; add.s32 %r40, %r40, 4; setp.ne.s32 %p9, %r40, 256; @%p9 bra $L__BB0_1; cvta.to.global.u64 %rd16, %rd8; mad.lo.s32 %r38, %r5, 500, %r2; mul.wide.s32 %rd17, %r38, 4; add.s64 %rd18, %rd16, %rd17; ld.global.nc.f32 %f89, [%rd18]; add.f32 %f90, %f102, %f89; mov.f32 %f91, 0f00000000; max.f32 %f92, %f90, %f91; mad.lo.s32 %r39, %r5, 84500, %r3; mul.wide.s32 %rd19, %r39, 4; add.s64 %rd20, %rd1, %rd19; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd20+(0+i*84500)], %f(92+i*3); ld.global.nc.f32 %f(93+i*3), [%rd18+(500+i*500)]; add.f32 %f(94+i*3), %f(103+i*1), %f(93+i*3); max.f32 %f(95+i*3), %f(94+i*3), %f91; st.global.f32 [%rd20+253500], %f101; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[26]; __shared__ float placeholder_shared[1000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 256; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 26) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 338 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 1000) { int index0 = (int)threadIdx.x; int index2 = (int)blockIdx.x / 13 * 256000 + ((int)threadIdx.x >> 1) * 512 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index0 = (int)threadIdx.x; int index3 = (int)threadIdx.x / 13 * 2; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 250]); compute[2] = compute[2] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 500]); compute[3] = compute[3] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 750]); int index4 = (int)threadIdx.x % 13 + 13; compute[0] = compute[0] + (pad_temp_shared[index4] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index3 + 251]); compute[2] = compute[2] + (pad_temp_shared[index4] * placeholder_shared[index3 + 501]); compute[3] = compute[3] + (pad_temp_shared[index4] * placeholder_shared[index3 + 751]); } int index5 = (int)blockIdx.x / 13 * 84500 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; int index6 = (int)blockIdx.x / 13 * 500 + (int)threadIdx.x / 13; T_relu[index5] = max(compute[0] + placeholder2[index6], 0.000000e+00f); T_relu[index5 + 21125] = max(compute[1] + placeholder2[index6 + 125], 0.000000e+00f); T_relu[index5 + 42250] = max(compute[2] + placeholder2[index6 + 250], 0.000000e+00f); T_relu[index5 + 63375] = max(compute[3] + placeholder2[index6 + 375], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpmpl3moqa
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<93>; .reg .b32 %r<96>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[8192]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; ld.param.u64 %rd10, [default_function_kernel0_param_2]; ld.param.u64 %rd11, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd8; cvta.to.global.u64 %rd1, %rd9; cvta.to.global.u64 %rd2, %rd10; cvta.to.global.u64 %rd3, %rd11; mov.u32 %r1, %ctaid.x; shl.b32 %r26, %r1, 12; mov.u32 %r2, %tid.x; shl.b32 %r27, %r2, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(28+i*1), %r(27+i*-25), (-512+i*767); bfi.b32 %r30, %r1, %r29, 12, 20; add.s32 %r3, %r30, %r28; shl.b32 %r31, %r2, 2; mov.u32 %r32, _ZZ24default_function_kernel0E18placeholder_shared; for.size.6 i in range(0, 2, 1): add.s32 %r(4+i*1), %r(32+i*5), %r(31+i*3); add.s32 %r(33+i*5), %r27, (676+i*676); and.b32 %r(34+i*5), %r(33+i*5), -512; add.s32 %r(35+i*5), %r2, (82+i*82); and.b32 %r(36+i*5), %r(35+i*5), 255; bfi.b32 %r(37+i*5), %r1, %r(36+i*5), 12, 20; for.size.6 i in range(0, 2, 1): add.s32 %r(6+i*1), %r(42+i*5), %r(39+i*5); add.s32 %r(43+i*5), %r27, (2028+i*676); and.b32 %r(44+i*5), %r(43+i*5), -512; add.s32 %r(45+i*5), %r2, (246+i*-174); and.b32 %r(46+i*5), %r(45+i*5), 255; bfi.b32 %r(47+i*5), %r1, %r(46+i*5), 12, 20; add.s32 %r8, %r52, %r49; for.size.2 i in range(0, 2, 1): add.s32 %r(53+i*2), %r(27+i*-25), (3380+i*-3226); and.b32 %r(54+i*2), %r(53+i*2), (-512+i*767); bfi.b32 %r57, %r1, %r56, 12, 20; add.s32 %r9, %r57, %r54; add.s32 %r58, %r27, 4056; and.b32 %r59, %r58, -512; for.size.1 i in range(0, 2, 1): add.s32 %r(60+i*1), %r(2+i*58), %r(26+i*33); add.s32 %r10, %r61, 236; mul.hi.s32 %r62, %r2, 1626496491; shr.u32 %r63, %r62, 31; shr.s32 %r64, %r62, 6; add.s32 %r12, %r64, %r63; mul.lo.s32 %r65, %r12, 169; sub.s32 %r11, %r2, %r65; shl.b32 %r13, %r12, 10; add.s64 %rd4, %rd12, 10816; mov.u32 %r66, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r67, %r66, %r31; add.s32 %r14, %r67, 10816; mov.u32 %r25, 0; mov.f32 %f89, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(90+i*1), %f89; mov.u32 %r92, %r25; $L__BB0_1: bar.sync 0; mad.lo.s32 %r69, %r92, 43264, %r2; mul.wide.s32 %rd13, %r69, 4; add.s64 %rd32, %rd4, %rd13; mov.u32 %r93, %r14; mov.u32 %r94, %r25; $L__BB0_2: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(17+i*1), [%rd32+(-10816+i*1352)]; st.shared.f32 [%r93+(-10816+i*1352)], %f(17+i*1); add.s64 %rd32, %rd32, 21632; for.size.1 i in range(0, 2, 1): add.s32 %r(93+i*1), %r(93+i*1), (21632+i*-21616); setp.ne.s32 %p1, %r94, 128; @%p1 bra $L__BB0_2; setp.gt.s32 %p2, %r2, 19; shl.b32 %r21, %r92, 8; for.size.5 i in range(0, 2, 1): add.s32 %r(70+i*1), %r(3+i*2), %r21; mul.wide.s32 %rd(14+i*2), %r(70+i*1), 4; add.s64 %rd(15+i*2), %rd1, %rd(14+i*2); ld.global.nc.f32 %f(33+i*1), [%rd(15+i*2)+0]; st.shared.f32 [%r4+(0+i*1352)], %f(33+i*1); for.size.5 i in range(0, 4, 1): add.s32 %r(72+i*1), %r(6+i*1), %r21; mul.wide.s32 %rd(18+i*2), %r(72+i*1), 4; add.s64 %rd(19+i*2), %rd1, %rd(18+i*2); ld.global.nc.f32 %f(35+i*1), [%rd(19+i*2)+0]; st.shared.f32 [%r4+(2704+i*1352)], %f(35+i*1); @%p2 bra $L__BB0_5; add.s32 %r76, %r10, %r21; mul.wide.s32 %rd26, %r76, 4; add.s64 %rd27, %rd1, %rd26; ld.global.nc.f32 %f39, [%rd27]; st.shared.f32 [%r4+8112], %f39; $L__BB0_5: bar.sync 0; mov.u32 %r95, 0; $L__BB0_6: mad.lo.s32 %r78, %r95, 338, %r11; for.size.2 i in range(0, 2, 1): shl.b32 %r(79+i*3), %r(78+i*17), (2+i*-1); add.s32 %r(81+i*2), %r(66+i*16), %r(79+i*-66); shl.b32 %r84, %r83, 2; add.s32 %r86, %r32, %r84; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(40+i*1), [%r(86+i*-5)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(42+i*2), %f41, %f(40+i*3), %f(92+i*-1); ld.shared.f32 %f(43+i*2), [%r86+(1024+i*-1020)]; ld.shared.f32 %f46, [%r81+676]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(47+i*2), %f46, %f(45+i*3), %f(42+i*2); ld.shared.f32 %f(48+i*2), [%r86+(1028+i*1020)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(51+i*2), %f41, %f(50+i*2), %f(90+i*-1); ld.shared.f32 %f(52+i*2), [%r86+(3072+i*-1020)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(55+i*2), %f46, %f(54+i*2), %f(51+i*2); ld.shared.f32 %f(56+i*2), [%r86+(3076+i*-3068)]; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(59+i*5), [%r81+(1352+i*676)]; fma.rn.f32 %f(60+i*32), %f(59+i*5), %f(58+i*5), %f(47+i*13); ld.shared.f32 %f(61+i*4), [%r86+(1032+i*4)]; fma.rn.f32 %f(62+i*29), %f(59+i*5), %f(61+i*4), %f(49+i*13); ld.shared.f32 %f(63+i*3), [%r86+(12+i*2044)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(67+i*2), %f59, %f(66+i*2), %f(55+i*2); ld.shared.f32 %f(68+i*2), [%r86+(3080+i*-1020)]; fma.rn.f32 %f90, %f64, %f70, %f67; ld.shared.f32 %f71, [%r86+3084]; fma.rn.f32 %f89, %f64, %f71, %f69; add.s32 %r95, %r95, 2; setp.ne.s32 %p3, %r95, 128; @%p3 bra $L__BB0_6; add.s32 %r92, %r92, 1; setp.lt.u32 %p4, %r92, 2; @%p4 bra $L__BB0_1; for.size.1 i in range(0, 2, 1): shl.b32 %r(87+i*1), %r(1+i*11), (3+i*-1); add.s32 %r89, %r88, %r87; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(90+i*1), %r(1+i*11), (1352+i*-676), %r(11+i*79); mul.wide.s32 %rd28, %r89, 4; add.s64 %rd29, %rd3, %rd28; ld.global.nc.f32 %f72, [%rd29]; add.f32 %f73, %f92, %f72; mov.f32 %f74, 0f00000000; max.f32 %f75, %f73, %f74; mul.wide.s32 %rd30, %r91, 4; add.s64 %rd31, %rd2, %rd30; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd31+(0+i*676)], %f(75+i*3); ld.global.nc.f32 %f(76+i*3), [%rd29+(4+i*4)]; add.f32 %f(77+i*3), %f(91+i*-1), %f(76+i*3); max.f32 %f(78+i*3), %f(77+i*3), %f74; st.global.f32 [%rd31+2028], %f84; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[2048]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 128; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 338 + (int)threadIdx.x; int index1 = rc_outer_outer * 43264 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 338 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } int index2 = (int)threadIdx.x; int index3 = (int)blockIdx.x * 4096 + ((int)threadIdx.x >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x & 255); placeholder_shared[index2] = placeholder1[index3]; int index4 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 338 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 82 & 255); placeholder_shared[index2 + 338] = placeholder1[index4]; int index5 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 676 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 164 & 255); placeholder_shared[index2 + 676] = placeholder1[index5]; int index6 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 1014 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 246 & 255); placeholder_shared[index2 + 1014] = placeholder1[index6]; int index7 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 1352 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 72 & 255); placeholder_shared[index2 + 1352] = placeholder1[index7]; int index8 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 1690 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 154 & 255); placeholder_shared[index2 + 1690] = placeholder1[index8]; if ((int)threadIdx.x < 20) { int index9 = (int)blockIdx.x * 4096 + ((int)threadIdx.x + 2028 >> 8) * 512 + rc_outer_outer * 256 + ((int)threadIdx.x + 236); placeholder_shared[index2 + 2028] = placeholder1[index9]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index10 = rc_outer_inner * 338 + (int)threadIdx.x % 169; int index11 = (int)threadIdx.x / 169 * 1024 + rc_outer_inner * 2; compute[0] = compute[0] + pad_temp_shared[index10] * placeholder_shared[index11]; compute[1] = compute[1] + (pad_temp_shared[index10] * placeholder_shared[index11 + 256]); compute[0] = compute[0] + (pad_temp_shared[index10 + 169] * placeholder_shared[index11 + 1]); compute[1] = compute[1] + (pad_temp_shared[index10 + 169] * placeholder_shared[index11 + 257]); compute[2] = compute[2] + (pad_temp_shared[index10] * placeholder_shared[index11 + 512]); compute[3] = compute[3] + (pad_temp_shared[index10] * placeholder_shared[index11 + 768]); compute[2] = compute[2] + (pad_temp_shared[index10 + 169] * placeholder_shared[index11 + 513]); compute[3] = compute[3] + (pad_temp_shared[index10 + 169] * placeholder_shared[index11 + 769]); } } for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index12 = (int)blockIdx.x * 1352 + (int)threadIdx.x / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index13 = ax1_inner; int index14 = (int)blockIdx.x * 8 + (int)threadIdx.x / 169 * 4 + ax1_inner; T_relu[index12] = max(compute[index13] + placeholder2[index14], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpuocqax87
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[135200]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<16>; .reg .f32 %f<362>; .reg .b32 %r<135>; .reg .b64 %rd<47>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd15, [default_function_kernel0_param_0]; ld.param.u64 %rd16, [default_function_kernel0_param_1]; ld.param.u64 %rd17, [default_function_kernel0_param_2]; ld.param.u64 %rd18, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd15; cvta.to.global.u64 %rd2, %rd16; cvta.to.global.u64 %rd3, %rd17; cvta.to.global.u64 %rd4, %rd18; add.u64 %rd5, %SPL, 0; mov.u32 %r38, 0; mov.u32 %r124, %r38; $L__BB0_1: mul.lo.s32 %r2, %r124, 130; mov.u32 %r125, %r38; $L__BB0_2: add.s32 %r40, %r125, %r2; mul.wide.s32 %rd20, %r40, 4; add.s64 %rd21, %rd5, %rd20; mov.u32 %r41, 0; for.size.2 i in range(0, 10, 1): for.size.1 j in range(0, 65, 1): st.local.u32 [%rd21+(0+i*52+j*2080)], %r41; add.s32 %r125, %r125, 1; setp.ne.s32 %p1, %r125, 13; @%p1 bra $L__BB0_2; add.s32 %r124, %r124, 1; setp.lt.u32 %p2, %r124, 4; @%p2 bra $L__BB0_1; mov.u32 %r6, %tid.x; mul.lo.s32 %r7, %r6, 640; mov.u32 %r42, 0; mov.u32 %r126, %r42; $L__BB0_5: bar.sync 0; mad.lo.s32 %r9, %r126, 2704, %r6; mov.u32 %r127, %r42; bra.uni $L__BB0_6; $L__BB0_15: add.s32 %r127, %r127, 4; $L__BB0_6: mul.lo.s32 %r11, %r127, 5; add.s32 %r44, %r11, %r6; setp.gt.s32 %p3, %r44, 2703; add.s32 %r45, %r9, %r11; mul.wide.s32 %rd22, %r45, 4; add.s64 %rd6, %rd1, %rd22; shl.b32 %r46, %r44, 2; mov.u32 %r47, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r12, %r47, %r46; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f19, [%rd6]; st.shared.f32 [%r12], %f19; $L__BB0_8: add.s32 %r13, %r127, 1; setp.eq.s32 %p4, %r13, 541; @%p4 bra $L__BB0_16; mad.lo.s32 %r48, %r13, 5, %r6; setp.gt.s32 %p5, %r48, 2703; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f20, [%rd6+20]; st.shared.f32 [%r12+20], %f20; $L__BB0_11: add.s32 %r50, %r44, 10; setp.gt.s32 %p6, %r50, 2703; @%p6 bra $L__BB0_13; ld.global.nc.f32 %f21, [%rd6+40]; st.shared.f32 [%r12+40], %f21; $L__BB0_13: add.s32 %r52, %r44, 15; setp.gt.s32 %p7, %r52, 2703; @%p7 bra $L__BB0_15; ld.global.nc.f32 %f22, [%rd6+60]; st.shared.f32 [%r12+60], %f22; bra.uni $L__BB0_15; $L__BB0_16: shl.b32 %r15, %r126, 4; mov.u32 %r128, 0; $L__BB0_17: mad.lo.s32 %r54, %r128, 5, %r6; shl.b32 %r55, %r54, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(56+i*1), %r(55+i*-1), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(58+i*1), %r(57+i*1), %r(15+i*41); mul.wide.s32 %rd23, %r59, 4; add.s64 %rd24, %rd2, %rd23; ld.global.nc.f32 %f23, [%rd24]; shl.b32 %r60, %r54, 2; mov.u32 %r61, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r62, %r61, %r60; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r62+(0+i*20)], %f(23+i*1); add.s32 %r(63+i*6), %r54, (5+i*5); shl.b32 %r(64+i*6), %r(63+i*6), 5; and.b32 %r(65+i*6), %r(64+i*6), -512; and.b32 %r(66+i*6), %r(63+i*6), 15; add.s32 %r(67+i*6), %r(66+i*6), %r15; add.s32 %r(68+i*6), %r(67+i*6), %r(65+i*6); mul.wide.s32 %rd(25+i*2), %r(68+i*6), 4; add.s64 %rd(26+i*2), %rd2, %rd(25+i*2); ld.global.nc.f32 %f(24+i*1), [%rd(26+i*2)+0]; st.shared.f32 [%r62+140], %f30; add.s32 %r128, %r128, 8; setp.ne.s32 %p8, %r128, 3200; @%p8 bra $L__BB0_17; bar.sync 0; mov.u32 %r105, 0; mov.u32 %r129, %r105; $L__BB0_19: mul.lo.s32 %r19, %r129, 169; add.s32 %r20, %r129, %r7; mov.u32 %r130, %r105; $L__BB0_20: mul.lo.s32 %r22, %r130, 130; mad.lo.s32 %r23, %r130, 160, %r20; mov.u32 %r107, 0; mov.u32 %r131, %r107; $L__BB0_21: for.size.1 i in range(0, 2, 1): add.s32 %r(25+i*84), %r131, %r(22+i*-3); shl.b32 %r110, %r109, 2; add.s32 %r112, %r47, %r110; for.size.1 i in range(0, 13, 1): ld.shared.f32 %f(1+i*1), [%r112+(0+i*52)]; mov.u32 %r132, %r107; $L__BB0_22: mad.lo.s32 %r113, %r132, 13, %r25; shl.b32 %r114, %r132, 4; add.s32 %r115, %r23, %r114; mul.wide.s32 %rd39, %r113, 4; add.s64 %rd40, %rd5, %rd39; shl.b32 %r116, %r115, 2; add.s32 %r118, %r61, %r116; for.size.40 i in range(0, 5, 1): ld.shared.f32 %f(31+i*27), [%r118+(0+i*12800)]; ld.local.f32 %f(32+i*27), [%rd40+(0+i*27040)]; fma.rn.f32 %f(33+i*27), %f1, %f(31+i*27), %f(32+i*27); st.local.f32 [%rd40+(0+i*27040)], %f(33+i*27); ld.local.f32 %f(34+i*27), [%rd40+(2080+i*27040)]; fma.rn.f32 %f(35+i*27), %f2, %f(31+i*27), %f(34+i*27); st.local.f32 [%rd40+(2080+i*27040)], %f(35+i*27); ld.local.f32 %f(36+i*27), [%rd40+(4160+i*27040)]; fma.rn.f32 %f(37+i*27), %f3, %f(31+i*27), %f(36+i*27); st.local.f32 [%rd40+(4160+i*27040)], %f(37+i*27); ld.local.f32 %f(38+i*27), [%rd40+(6240+i*27040)]; fma.rn.f32 %f(39+i*27), %f4, %f(31+i*27), %f(38+i*27); st.local.f32 [%rd40+(6240+i*27040)], %f(39+i*27); ld.local.f32 %f(40+i*27), [%rd40+(8320+i*27040)]; fma.rn.f32 %f(41+i*27), %f5, %f(31+i*27), %f(40+i*27); st.local.f32 [%rd40+(8320+i*27040)], %f(41+i*27); ld.local.f32 %f(42+i*27), [%rd40+(10400+i*27040)]; fma.rn.f32 %f(43+i*27), %f6, %f(31+i*27), %f(42+i*27); st.local.f32 [%rd40+(10400+i*27040)], %f(43+i*27); ld.local.f32 %f(44+i*27), [%rd40+(12480+i*27040)]; fma.rn.f32 %f(45+i*27), %f7, %f(31+i*27), %f(44+i*27); st.local.f32 [%rd40+(12480+i*27040)], %f(45+i*27); ld.local.f32 %f(46+i*27), [%rd40+(14560+i*27040)]; fma.rn.f32 %f(47+i*27), %f8, %f(31+i*27), %f(46+i*27); st.local.f32 [%rd40+(14560+i*27040)], %f(47+i*27); ld.local.f32 %f(48+i*27), [%rd40+(16640+i*27040)]; fma.rn.f32 %f(49+i*27), %f9, %f(31+i*27), %f(48+i*27); st.local.f32 [%rd40+(16640+i*27040)], %f(49+i*27); ld.local.f32 %f(50+i*27), [%rd40+(18720+i*27040)]; fma.rn.f32 %f(51+i*27), %f10, %f(31+i*27), %f(50+i*27); st.local.f32 [%rd40+(18720+i*27040)], %f(51+i*27); ld.local.f32 %f(52+i*27), [%rd40+(20800+i*27040)]; fma.rn.f32 %f(53+i*27), %f11, %f(31+i*27), %f(52+i*27); st.local.f32 [%rd40+(20800+i*27040)], %f(53+i*27); ld.local.f32 %f(54+i*27), [%rd40+(22880+i*27040)]; fma.rn.f32 %f(55+i*27), %f12, %f(31+i*27), %f(54+i*27); st.local.f32 [%rd40+(22880+i*27040)], %f(55+i*27); ld.local.f32 %f(56+i*27), [%rd40+(24960+i*27040)]; fma.rn.f32 %f(57+i*27), %f13, %f(31+i*27), %f(56+i*27); st.local.f32 [%rd40+(24960+i*27040)], %f(57+i*27); add.s32 %r132, %r132, 1; setp.ne.s32 %p9, %r132, 10; @%p9 bra $L__BB0_22; add.s32 %r131, %r131, 1; setp.lt.u32 %p10, %r131, 13; @%p10 bra $L__BB0_21; add.s32 %r130, %r130, 1; setp.lt.u32 %p11, %r130, 4; @%p11 bra $L__BB0_20; add.s32 %r129, %r129, 1; setp.lt.u32 %p12, %r129, 16; @%p12 bra $L__BB0_19; add.s32 %r126, %r126, 1; setp.lt.u32 %p13, %r126, 32; @%p13 bra $L__BB0_5; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(32+i*1), %r6, (40+i*6720); add.s64 %rd(7+i*1), %rd(3+i*2), (270712+i*-204152); mov.u32 %r119, 0; mov.u32 %r133, %r119; $L__BB0_28: add.s32 %r121, %r133, %r32; mul.wide.s32 %rd41, %r121, 4; add.s64 %rd42, %rd4, %rd41; for.size.1 i in range(0, 5, 1): ld.global.nc.f32 %f(14+i*1), [%rd42+(0+i*800)]; mad.lo.s32 %r122, %r133, 169, %r33; mul.wide.s32 %rd43, %r122, 4; add.s64 %rd46, %rd7, %rd43; mul.lo.s32 %r123, %r133, 13; mul.wide.s32 %rd44, %r123, 4; add.s64 %rd45, %rd8, %rd44; mov.u32 %r134, %r119; $L__BB0_29: ld.local.f32 %f166, [%rd45+-66560]; add.f32 %f167, %f166, %f14; mov.f32 %f168, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(169+i*3), %f(167+i*4), %f168; st.global.f32 [%rd46+(-270712+i*52)], %f(169+i*3); ld.local.f32 %f(170+i*3), [%rd45+(-64480+i*2080)]; add.f32 %f(171+i*3), %f(170+i*3), %f14; for.size.4 i in range(0, 10, 1): max.f32 %f(175+i*3), %f(174+i*3), %f168; st.global.f32 [%rd46+(-270608+i*52)], %f(175+i*3); ld.local.f32 %f(176+i*3), [%rd45+(-60320+i*2080)]; add.f32 %f(177+i*3), %f(176+i*3), %f14; for.size.10 i in range(0, 4, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(205+i*39+j*3), %f(204+i*39+j*3), %f168; st.global.f32 [%rd46+(-270088+i*135200+j*134576)], %f(205+i*39+j*3); ld.local.f32 %f(206+i*39+j*3), [%rd45+(-39520+i*27040+j*2080)]; add.f32 %f(207+i*39+j*3), %f(206+i*39+j*3), %f(15+j*1); for.size.4 j in range(0, 11, 1): max.f32 %f(211+i*39+j*3), %f(210+i*39+j*3), %f168; st.global.f32 [%rd46+(-135460+i*135200+j*52)], %f(211+i*39+j*3); ld.local.f32 %f(212+i*39+j*3), [%rd45+(-35360+i*27040+j*2080)]; add.f32 %f(213+i*39+j*3), %f(212+i*39+j*3), %f(15+j*1); max.f32 %f361, %f360, %f168; st.global.f32 [%rd46+270712], %f361; for.size.1 i in range(0, 2, 1): add.s64 %rd(46+i*-1), %rd(46+i*-1), 4; add.s32 %r134, %r134, 1; setp.ne.s32 %p14, %r134, 13; @%p14 bra $L__BB0_29; add.s32 %r133, %r133, 1; setp.lt.u32 %p15, %r133, 40; @%p15 bra $L__BB0_28; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[33800]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 4; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 10; ++ff_inner_init) { for (int vthread_s = 0; vthread_s < 65; ++vthread_s) { int index0 = vthread_s * 520 + ff_outer_inner_init * 130 + ff_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; } } } } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 541; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x < 2704) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 3200; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 4; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int ff_inner = 0; ff_inner < 10; ++ff_inner) { for (int vthread_s1 = 0; vthread_s1 < 65; ++vthread_s1) { int index5 = vthread_s1 * 520 + ff_outer_inner * 130 + ff_inner * 13 + xx_outer_inner; int index6 = rc_outer_inner * 169 + vthread_s1 % 13 * 13 + xx_outer_inner; int index7 = vthread_s1 / 13 * 3200 + (int)threadIdx.x * 640 + ff_outer_inner * 160 + ff_inner * 16 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } } for (int ax1_inner = 0; ax1_inner < 40; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { for (int vthread_s2 = 0; vthread_s2 < 65; ++vthread_s2) { int index8 = vthread_s2 / 13 * 33800 + (int)threadIdx.x * 6760 + ax1_inner * 169 + vthread_s2 % 13 * 13 + ax3_inner; int index9 = vthread_s2 * 520 + ax1_inner * 13 + ax3_inner; int index10 = vthread_s2 / 13 * 200 + (int)threadIdx.x * 40 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp4mky3o_9
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<208>; .reg .b32 %r<49>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[43264]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[5120]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd4, [default_function_kernel0_param_2]; ld.param.u64 %rd5, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd7; cvta.to.global.u64 %rd2, %rd6; mov.u32 %r1, %tid.x; shl.b32 %r15, %r1, 2; mov.u32 %r16, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r16, %r15; mov.u32 %r3, %ctaid.x; mul.lo.s32 %r17, %r3, 10240; shl.b32 %r18, %r1, 5; and.b32 %r19, %r18, -512; add.s32 %r20, %r19, %r17; and.b32 %r21, %r15, 60; or.b32 %r4, %r20, %r21; shl.b32 %r22, %r1, 4; mov.u32 %r23, _ZZ24default_function_kernel0E18placeholder_shared; for.size.4 i in range(0, 2, 1): add.s32 %r(5+i*1), %r(23+i*3), %r(22+i*-3); or.b32 %r(24+i*3), %r15, (1+i*1); and.b32 %r(25+i*3), %r(24+i*3), (61+i*1); or.b32 %r(26+i*3), %r(25+i*3), %r17; add.s32 %r7, %r29, %r19; or.b32 %r30, %r15, 3; and.b32 %r31, %r30, 63; or.b32 %r32, %r31, %r17; add.s32 %r8, %r32, %r19; mul.hi.s32 %r33, %r1, 1626496491; shr.u32 %r34, %r33, 31; shr.s32 %r35, %r33, 6; add.s32 %r10, %r35, %r34; mul.lo.s32 %r36, %r10, 169; sub.s32 %r37, %r1, %r36; for.size.2 i in range(0, 2, 1): shl.b32 %r(38+i*1), %r(37+i*-27), (2+i*6); add.s32 %r(9+i*2), %r(16+i*7), %r(38+i*1); mov.f32 %f207, 0f00000000; mov.u32 %r48, 0; $L__BB0_1: bar.sync 0; mad.lo.s32 %r40, %r48, 10816, %r1; mul.wide.s32 %rd8, %r40, 4; add.s64 %rd3, %rd2, %rd8; for.size.2 i in range(0, 3, 1): ld.global.nc.f32 %f(4+i*1), [%rd3+(0+i*13520)]; st.shared.f32 [%r2+(0+i*13520)], %f(4+i*1); setp.gt.s32 %p1, %r1, 675; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f7, [%rd3+40560]; st.shared.f32 [%r2+40560], %f7; $L__BB0_3: setp.gt.s32 %p2, %r1, 319; @%p2 bra $L__BB0_5; shl.b32 %r41, %r48, 6; for.size.5 i in range(0, 2, 1): add.s32 %r(42+i*1), %r(4+i*2), %r41; mul.wide.s32 %rd(9+i*2), %r(42+i*1), 4; add.s64 %rd(10+i*2), %rd1, %rd(9+i*2); ld.global.nc.f32 %f(8+i*1), [%rd(10+i*2)+0]; st.shared.f32 [%r5+(0+i*4)], %f(8+i*1); for.size.5 i in range(0, 2, 1): add.s32 %r(44+i*1), %r(7+i*1), %r41; mul.wide.s32 %rd(13+i*2), %r(44+i*1), 4; add.s64 %rd(14+i*2), %rd1, %rd(13+i*2); ld.global.nc.f32 %f(10+i*1), [%rd(14+i*2)+0]; st.shared.f32 [%r5+(8+i*4)], %f(10+i*1); $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(12+i*1), [%r(11+i*-2)+0]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(14+i*3), %f(13+i*3), %f(12+i*3), %f(207+i*-193); ld.shared.f32 %f(15+i*3), [%r11+(4+i*4)]; ld.shared.f32 %f(16+i*3), [%r9+(676+i*676)]; for.size.3 i in range(0, 61, 1): fma.rn.f32 %f(20+i*3), %f(19+i*3), %f(18+i*3), %f(17+i*3); ld.shared.f32 %f(21+i*3), [%r11+(12+i*4)]; ld.shared.f32 %f(22+i*3), [%r9+(2028+i*676)]; fma.rn.f32 %f207, %f202, %f201, %f200; add.s32 %r48, %r48, 1; setp.ne.s32 %p3, %r48, 8; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd5; mad.lo.s32 %r46, %r3, 20, %r10; mul.wide.s32 %rd18, %r46, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f203, [%rd19]; add.f32 %f204, %f207, %f203; mov.f32 %f205, 0f00000000; max.f32 %f206, %f204, %f205; mad.lo.s32 %r47, %r3, 3380, %r1; cvta.to.global.u64 %rd20, %rd4; mul.wide.s32 %rd21, %r47, 4; add.s64 %rd22, %rd20, %rd21; st.global.f32 [%rd22], %f206; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[1]; __shared__ float pad_temp_shared[10816]; __shared__ float placeholder_shared[1280]; compute[0] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 10816 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 3380] = placeholder[index1 + 3380]; pad_temp_shared[index0 + 6760] = placeholder[index1 + 6760]; if ((int)threadIdx.x < 676) { pad_temp_shared[index0 + 10140] = placeholder[index1 + 10140]; } if ((int)threadIdx.x < 320) { int index2 = (int)blockIdx.x * 10240 + ((int)threadIdx.x >> 4) * 512 + rc_outer_outer * 64 + ((int)threadIdx.x & 15) * 4; placeholder_shared[index0 * 4] = placeholder1[index2]; } if ((int)threadIdx.x < 320) { int index3 = (int)threadIdx.x * 4 + 1; int index4 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 1 >> 6) * 512 + rc_outer_outer * 64 + ((int)threadIdx.x * 4 + 1 & 63); placeholder_shared[index3] = placeholder1[index4]; } if ((int)threadIdx.x < 320) { int index5 = (int)threadIdx.x * 4 + 2; int index6 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 2 >> 6) * 512 + rc_outer_outer * 64 + ((int)threadIdx.x * 4 + 2 & 63); placeholder_shared[index5] = placeholder1[index6]; } if ((int)threadIdx.x < 320) { int index7 = (int)threadIdx.x * 4 + 3; int index8 = (int)blockIdx.x * 10240 + ((int)threadIdx.x * 4 + 3 >> 6) * 512 + rc_outer_outer * 64 + ((int)threadIdx.x * 4 + 3 & 63); placeholder_shared[index7] = placeholder1[index8]; } __syncthreads(); int index9 = (int)threadIdx.x / 169 * 64; compute[0] = compute[0] + pad_temp_shared[index0 % 169] * placeholder_shared[index9]; int index10 = (int)threadIdx.x % 169 + 169; compute[0] = compute[0] + (pad_temp_shared[index10] * placeholder_shared[index9 + 1]); int index11 = (int)threadIdx.x % 169 + 338; compute[0] = compute[0] + (pad_temp_shared[index11] * placeholder_shared[index9 + 2]); int index12 = (int)threadIdx.x % 169 + 507; compute[0] = compute[0] + (pad_temp_shared[index12] * placeholder_shared[index9 + 3]); int index13 = (int)threadIdx.x % 169 + 676; compute[0] = compute[0] + (pad_temp_shared[index13] * placeholder_shared[index9 + 4]); int index14 = (int)threadIdx.x % 169 + 845; compute[0] = compute[0] + (pad_temp_shared[index14] * placeholder_shared[index9 + 5]); int index15 = (int)threadIdx.x % 169 + 1014; compute[0] = compute[0] + (pad_temp_shared[index15] * placeholder_shared[index9 + 6]); int index16 = (int)threadIdx.x % 169 + 1183; compute[0] = compute[0] + (pad_temp_shared[index16] * placeholder_shared[index9 + 7]); int index17 = (int)threadIdx.x % 169 + 1352; compute[0] = compute[0] + (pad_temp_shared[index17] * placeholder_shared[index9 + 8]); int index18 = (int)threadIdx.x % 169 + 1521; compute[0] = compute[0] + (pad_temp_shared[index18] * placeholder_shared[index9 + 9]); int index19 = (int)threadIdx.x % 169 + 1690; compute[0] = compute[0] + (pad_temp_shared[index19] * placeholder_shared[index9 + 10]); int index20 = (int)threadIdx.x % 169 + 1859; compute[0] = compute[0] + (pad_temp_shared[index20] * placeholder_shared[index9 + 11]); int index21 = (int)threadIdx.x % 169 + 2028; compute[0] = compute[0] + (pad_temp_shared[index21] * placeholder_shared[index9 + 12]); int index22 = (int)threadIdx.x % 169 + 2197; compute[0] = compute[0] + (pad_temp_shared[index22] * placeholder_shared[index9 + 13]); int index23 = (int)threadIdx.x % 169 + 2366; compute[0] = compute[0] + (pad_temp_shared[index23] * placeholder_shared[index9 + 14]); int index24 = (int)threadIdx.x % 169 + 2535; compute[0] = compute[0] + (pad_temp_shared[index24] * placeholder_shared[index9 + 15]); int index25 = (int)threadIdx.x % 169 + 2704; compute[0] = compute[0] + (pad_temp_shared[index25] * placeholder_shared[index9 + 16]); int index26 = (int)threadIdx.x % 169 + 2873; compute[0] = compute[0] + (pad_temp_shared[index26] * placeholder_shared[index9 + 17]); int index27 = (int)threadIdx.x % 169 + 3042; compute[0] = compute[0] + (pad_temp_shared[index27] * placeholder_shared[index9 + 18]); int index28 = (int)threadIdx.x % 169 + 3211; compute[0] = compute[0] + (pad_temp_shared[index28] * placeholder_shared[index9 + 19]); int index29 = (int)threadIdx.x % 169 + 3380; compute[0] = compute[0] + (pad_temp_shared[index29] * placeholder_shared[index9 + 20]); int index30 = (int)threadIdx.x % 169 + 3549; compute[0] = compute[0] + (pad_temp_shared[index30] * placeholder_shared[index9 + 21]); int index31 = (int)threadIdx.x % 169 + 3718; compute[0] = compute[0] + (pad_temp_shared[index31] * placeholder_shared[index9 + 22]); int index32 = (int)threadIdx.x % 169 + 3887; compute[0] = compute[0] + (pad_temp_shared[index32] * placeholder_shared[index9 + 23]); int index33 = (int)threadIdx.x % 169 + 4056; compute[0] = compute[0] + (pad_temp_shared[index33] * placeholder_shared[index9 + 24]); int index34 = (int)threadIdx.x % 169 + 4225; compute[0] = compute[0] + (pad_temp_shared[index34] * placeholder_shared[index9 + 25]); int index35 = (int)threadIdx.x % 169 + 4394; compute[0] = compute[0] + (pad_temp_shared[index35] * placeholder_shared[index9 + 26]); int index36 = (int)threadIdx.x % 169 + 4563; compute[0] = compute[0] + (pad_temp_shared[index36] * placeholder_shared[index9 + 27]); int index37 = (int)threadIdx.x % 169 + 4732; compute[0] = compute[0] + (pad_temp_shared[index37] * placeholder_shared[index9 + 28]); int index38 = (int)threadIdx.x % 169 + 4901; compute[0] = compute[0] + (pad_temp_shared[index38] * placeholder_shared[index9 + 29]); int index39 = (int)threadIdx.x % 169 + 5070; compute[0] = compute[0] + (pad_temp_shared[index39] * placeholder_shared[index9 + 30]); int index40 = (int)threadIdx.x % 169 + 5239; compute[0] = compute[0] + (pad_temp_shared[index40] * placeholder_shared[index9 + 31]); int index41 = (int)threadIdx.x % 169 + 5408; compute[0] = compute[0] + (pad_temp_shared[index41] * placeholder_shared[index9 + 32]); int index42 = (int)threadIdx.x % 169 + 5577; compute[0] = compute[0] + (pad_temp_shared[index42] * placeholder_shared[index9 + 33]); int index43 = (int)threadIdx.x % 169 + 5746; compute[0] = compute[0] + (pad_temp_shared[index43] * placeholder_shared[index9 + 34]); int index44 = (int)threadIdx.x % 169 + 5915; compute[0] = compute[0] + (pad_temp_shared[index44] * placeholder_shared[index9 + 35]); int index45 = (int)threadIdx.x % 169 + 6084; compute[0] = compute[0] + (pad_temp_shared[index45] * placeholder_shared[index9 + 36]); int index46 = (int)threadIdx.x % 169 + 6253; compute[0] = compute[0] + (pad_temp_shared[index46] * placeholder_shared[index9 + 37]); int index47 = (int)threadIdx.x % 169 + 6422; compute[0] = compute[0] + (pad_temp_shared[index47] * placeholder_shared[index9 + 38]); int index48 = (int)threadIdx.x % 169 + 6591; compute[0] = compute[0] + (pad_temp_shared[index48] * placeholder_shared[index9 + 39]); int index49 = (int)threadIdx.x % 169 + 6760; compute[0] = compute[0] + (pad_temp_shared[index49] * placeholder_shared[index9 + 40]); int index50 = (int)threadIdx.x % 169 + 6929; compute[0] = compute[0] + (pad_temp_shared[index50] * placeholder_shared[index9 + 41]); int index51 = (int)threadIdx.x % 169 + 7098; compute[0] = compute[0] + (pad_temp_shared[index51] * placeholder_shared[index9 + 42]); int index52 = (int)threadIdx.x % 169 + 7267; compute[0] = compute[0] + (pad_temp_shared[index52] * placeholder_shared[index9 + 43]); int index53 = (int)threadIdx.x % 169 + 7436; compute[0] = compute[0] + (pad_temp_shared[index53] * placeholder_shared[index9 + 44]); int index54 = (int)threadIdx.x % 169 + 7605; compute[0] = compute[0] + (pad_temp_shared[index54] * placeholder_shared[index9 + 45]); int index55 = (int)threadIdx.x % 169 + 7774; compute[0] = compute[0] + (pad_temp_shared[index55] * placeholder_shared[index9 + 46]); int index56 = (int)threadIdx.x % 169 + 7943; compute[0] = compute[0] + (pad_temp_shared[index56] * placeholder_shared[index9 + 47]); int index57 = (int)threadIdx.x % 169 + 8112; compute[0] = compute[0] + (pad_temp_shared[index57] * placeholder_shared[index9 + 48]); int index58 = (int)threadIdx.x % 169 + 8281; compute[0] = compute[0] + (pad_temp_shared[index58] * placeholder_shared[index9 + 49]); int index59 = (int)threadIdx.x % 169 + 8450; compute[0] = compute[0] + (pad_temp_shared[index59] * placeholder_shared[index9 + 50]); int index60 = (int)threadIdx.x % 169 + 8619; compute[0] = compute[0] + (pad_temp_shared[index60] * placeholder_shared[index9 + 51]); int index61 = (int)threadIdx.x % 169 + 8788; compute[0] = compute[0] + (pad_temp_shared[index61] * placeholder_shared[index9 + 52]); int index62 = (int)threadIdx.x % 169 + 8957; compute[0] = compute[0] + (pad_temp_shared[index62] * placeholder_shared[index9 + 53]); int index63 = (int)threadIdx.x % 169 + 9126; compute[0] = compute[0] + (pad_temp_shared[index63] * placeholder_shared[index9 + 54]); int index64 = (int)threadIdx.x % 169 + 9295; compute[0] = compute[0] + (pad_temp_shared[index64] * placeholder_shared[index9 + 55]); int index65 = (int)threadIdx.x % 169 + 9464; compute[0] = compute[0] + (pad_temp_shared[index65] * placeholder_shared[index9 + 56]); int index66 = (int)threadIdx.x % 169 + 9633; compute[0] = compute[0] + (pad_temp_shared[index66] * placeholder_shared[index9 + 57]); int index67 = (int)threadIdx.x % 169 + 9802; compute[0] = compute[0] + (pad_temp_shared[index67] * placeholder_shared[index9 + 58]); int index68 = (int)threadIdx.x % 169 + 9971; compute[0] = compute[0] + (pad_temp_shared[index68] * placeholder_shared[index9 + 59]); int index69 = (int)threadIdx.x % 169 + 10140; compute[0] = compute[0] + (pad_temp_shared[index69] * placeholder_shared[index9 + 60]); int index70 = (int)threadIdx.x % 169 + 10309; compute[0] = compute[0] + (pad_temp_shared[index70] * placeholder_shared[index9 + 61]); int index71 = (int)threadIdx.x % 169 + 10478; compute[0] = compute[0] + (pad_temp_shared[index71] * placeholder_shared[index9 + 62]); int index72 = (int)threadIdx.x % 169 + 10647; compute[0] = compute[0] + (pad_temp_shared[index72] * placeholder_shared[index9 + 63]); } int index73 = (int)blockIdx.x * 3380 + (int)threadIdx.x; int index74 = (int)blockIdx.x * 20 + (int)threadIdx.x / 169; T_relu[index73] = max(compute[0] + placeholder2[index74], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpbdv83pqd
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[1300]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<9>; .reg .f32 %f<2693>; .reg .b32 %r<83>; .reg .b64 %rd<36>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[128000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.f32 %f2368, 0f00000000; mov.u32 %r78, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd1+(0+i*4+j*100)], %r78; mov.u32 %r18, %tid.x; mul.hi.s32 %r19, %r18, 1321528399; shr.u32 %r20, %r19, 31; shr.s32 %r21, %r19, 2; add.s32 %r2, %r21, %r20; mul.lo.s32 %r22, %r2, 13; sub.s32 %r23, %r18, %r22; mul.lo.s32 %r1, %r23, 13; cvta.to.global.u64 %rd16, %rd11; for.size.1 i in range(0, 324, 1): mov.f32 %f(2369+i*1), %f2368; $L__BB0_1: setp.gt.s32 %p1, %r18, 207; bar.sync 0; mad.lo.s32 %r25, %r78, 5408, %r18; mul.wide.s32 %rd17, %r25, 4; add.s64 %rd2, %rd16, %rd17; ld.global.nc.f32 %f1626, [%rd2]; shl.b32 %r26, %r18, 2; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r28, %r27, %r26; for.size.2 i in range(0, 9, 1): st.shared.f32 [%r28+(0+i*2080)], %f(1626+i*1); ld.global.nc.f32 %f(1627+i*1), [%rd2+(2080+i*2080)]; st.shared.f32 [%r28+18720], %f1635; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f1636, [%rd2+20800]; st.shared.f32 [%r28+20800], %f1636; $L__BB0_3: mov.u32 %r79, 0; $L__BB0_4: mad.lo.s32 %r5, %r79, 520, %r18; setp.gt.s32 %p2, %r5, 31999; @%p2 bra $L__BB0_6; shl.b32 %r37, %r5, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(38+i*1), %r(37+i*-32), (-512+i*543); bfi.b32 %r40, %r78, %r39, 5, 27; add.s32 %r41, %r40, %r38; cvta.to.global.u64 %rd18, %rd12; mul.wide.s32 %rd19, %r41, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f1637, [%rd20]; shl.b32 %r42, %r5, 2; mov.u32 %r43, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r44, %r43, %r42; st.shared.f32 [%r44], %f1637; $L__BB0_6: add.s32 %r6, %r5, 520; setp.gt.s32 %p3, %r6, 31999; @%p3 bra $L__BB0_8; shl.b32 %r45, %r6, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(46+i*1), %r(45+i*-39), (-512+i*543); bfi.b32 %r48, %r78, %r47, 5, 27; add.s32 %r49, %r48, %r46; cvta.to.global.u64 %rd21, %rd12; mul.wide.s32 %rd22, %r49, 4; add.s64 %rd23, %rd21, %rd22; ld.global.nc.f32 %f1638, [%rd23]; shl.b32 %r52, %r5, 2; mov.u32 %r53, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r54, %r53, %r52; st.shared.f32 [%r54+2080], %f1638; $L__BB0_8: add.s32 %r79, %r79, 2; setp.ne.s32 %p4, %r79, 62; @%p4 bra $L__BB0_4; bar.sync 0; mov.u32 %r80, 0; $L__BB0_10: mov.u32 %r81, 0; $L__BB0_11: mul.lo.s32 %r77, %r2, 800; shl.b32 %r76, %r80, 1; add.s32 %r75, %r76, %r77; mad.lo.s32 %r74, %r80, 338, %r1; mov.u32 %r73, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r59, %r81, 169, %r74; shl.b32 %r60, %r59, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(62+i*1), %r(73+i*2), %r(60+i*21); shl.b32 %r64, %r63, 2; mov.u32 %r65, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r66, %r65, %r64; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1639+i*1), [%r(66+i*-4)+0]; for.size.2 i in range(0, 12, 1): fma.rn.f32 %f(2692+i*-1), %f(1640+i*1), %f1639, %f(2692+i*-1); ld.shared.f32 %f(1641+i*1), [%r62+(4+i*4)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(2680+i*-13), %f1652, %f(1639+i*14), %f(2680+i*-13); ld.shared.f32 %f(1653+i*1), [%r66+(128+i*128)]; fma.rn.f32 %f(2679+i*-13), %f1640, %f(1653+i*1), %f(2679+i*-13); fma.rn.f32 %f(2678+i*-13), %f1641, %f(1653+i*1), %f(2678+i*-13); fma.rn.f32 %f(2677+i*-13), %f1642, %f(1653+i*1), %f(2677+i*-13); fma.rn.f32 %f(2676+i*-13), %f1643, %f(1653+i*1), %f(2676+i*-13); fma.rn.f32 %f(2675+i*-13), %f1644, %f(1653+i*1), %f(2675+i*-13); fma.rn.f32 %f(2674+i*-13), %f1645, %f(1653+i*1), %f(2674+i*-13); fma.rn.f32 %f(2673+i*-13), %f1646, %f(1653+i*1), %f(2673+i*-13); fma.rn.f32 %f(2672+i*-13), %f1647, %f(1653+i*1), %f(2672+i*-13); fma.rn.f32 %f(2671+i*-13), %f1648, %f(1653+i*1), %f(2671+i*-13); fma.rn.f32 %f(2670+i*-13), %f1649, %f(1653+i*1), %f(2670+i*-13); fma.rn.f32 %f(2669+i*-13), %f1650, %f(1653+i*1), %f(2669+i*-13); fma.rn.f32 %f(2668+i*-13), %f1651, %f(1653+i*1), %f(2668+i*-13); for.size.14 i in range(0, 22, 1): fma.rn.f32 %f(2654+i*-13), %f1652, %f(1654+i*1), %f(2654+i*-13); ld.shared.f32 %f(1655+i*1), [%r66+(384+i*128)]; fma.rn.f32 %f(2653+i*-13), %f1640, %f(1655+i*1), %f(2653+i*-13); fma.rn.f32 %f(2652+i*-13), %f1641, %f(1655+i*1), %f(2652+i*-13); fma.rn.f32 %f(2651+i*-13), %f1642, %f(1655+i*1), %f(2651+i*-13); fma.rn.f32 %f(2650+i*-13), %f1643, %f(1655+i*1), %f(2650+i*-13); fma.rn.f32 %f(2649+i*-13), %f1644, %f(1655+i*1), %f(2649+i*-13); fma.rn.f32 %f(2648+i*-13), %f1645, %f(1655+i*1), %f(2648+i*-13); fma.rn.f32 %f(2647+i*-13), %f1646, %f(1655+i*1), %f(2647+i*-13); fma.rn.f32 %f(2646+i*-13), %f1647, %f(1655+i*1), %f(2646+i*-13); fma.rn.f32 %f(2645+i*-13), %f1648, %f(1655+i*1), %f(2645+i*-13); fma.rn.f32 %f(2644+i*-13), %f1649, %f(1655+i*1), %f(2644+i*-13); fma.rn.f32 %f(2643+i*-13), %f1650, %f(1655+i*1), %f(2643+i*-13); fma.rn.f32 %f(2642+i*-13), %f1651, %f(1655+i*1), %f(2642+i*-13); fma.rn.f32 %f2368, %f1652, %f1676, %f2368; add.s32 %r67, %r81, 1; mov.u32 %r81, 1; setp.lt.u32 %p5, %r67, 2; @%p5 bra $L__BB0_11; add.s32 %r80, %r80, 1; setp.lt.u32 %p6, %r80, 16; @%p6 bra $L__BB0_10; add.s32 %r78, %r78, 1; setp.lt.u32 %p7, %r78, 16; @%p7 bra $L__BB0_1; ld.param.u64 %rd32, [default_function_kernel0_param_2]; ld.param.u64 %rd31, [default_function_kernel0_param_3]; add.u64 %rd34, %SPL, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.f32 [%rd34+(0+i*4+j*100)], %f(2692+i*-13+j*-1); mul.lo.s32 %r69, %r2, 25; cvta.to.global.u64 %rd24, %rd31; mul.wide.s32 %rd25, %r69, 4; add.s64 %rd35, %rd24, %rd25; mad.lo.s32 %r70, %r2, 4225, %r1; cvta.to.global.u64 %rd26, %rd32; mul.wide.s32 %rd27, %r70, 4; add.s64 %rd28, %rd26, %rd27; add.s64 %rd33, %rd28, 24; mov.u32 %r82, 0; $L__BB0_15: ld.global.nc.f32 %f1677, [%rd35]; ld.local.f32 %f1678, [%rd34]; add.f32 %f1679, %f1678, %f1677; mov.f32 %f1680, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(1681+i*3), %f(1679+i*4), %f1680; st.global.f32 [%rd33+(-24+i*4)], %f(1681+i*3); ld.local.f32 %f(1682+i*3), [%rd34+(100+i*100)]; add.f32 %f(1683+i*3), %f(1682+i*3), %f1677; for.size.4 i in range(0, 10, 1): max.f32 %f(1687+i*3), %f(1686+i*3), %f1680; st.global.f32 [%rd33+(-16+i*4)], %f(1687+i*3); ld.local.f32 %f(1688+i*3), [%rd34+(300+i*100)]; add.f32 %f(1689+i*3), %f(1688+i*3), %f1677; max.f32 %f1717, %f1716, %f1680; st.global.f32 [%rd33+24], %f1717; for.size.1 i in range(0, 2, 1): add.s64 %rd(35+i*-1), %rd(35+i*-1), 4; add.s64 %rd33, %rd33, 676; add.s32 %r82, %r82, 1; setp.ne.s32 %p8, %r82, 25; @%p8 bra $L__BB0_15; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[325]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[32000]; for (int ff_inner_init = 0; ff_inner_init < 25; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 25] = 0.000000e+00f; compute[index0 + 50] = 0.000000e+00f; compute[index0 + 75] = 0.000000e+00f; compute[index0 + 100] = 0.000000e+00f; compute[index0 + 125] = 0.000000e+00f; compute[index0 + 150] = 0.000000e+00f; compute[index0 + 175] = 0.000000e+00f; compute[index0 + 200] = 0.000000e+00f; compute[index0 + 225] = 0.000000e+00f; compute[index0 + 250] = 0.000000e+00f; compute[index0 + 275] = 0.000000e+00f; compute[index0 + 300] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 520] = placeholder[index2 + 520]; pad_temp_shared[index1 + 1040] = placeholder[index2 + 1040]; pad_temp_shared[index1 + 1560] = placeholder[index2 + 1560]; pad_temp_shared[index1 + 2080] = placeholder[index2 + 2080]; pad_temp_shared[index1 + 2600] = placeholder[index2 + 2600]; pad_temp_shared[index1 + 3120] = placeholder[index2 + 3120]; pad_temp_shared[index1 + 3640] = placeholder[index2 + 3640]; pad_temp_shared[index1 + 4160] = placeholder[index2 + 4160]; pad_temp_shared[index1 + 4680] = placeholder[index2 + 4680]; if ((int)threadIdx.x < 208) { pad_temp_shared[index1 + 5200] = placeholder[index2 + 5200]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 62; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x < 32000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 520 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 25; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 338 + rc_inner * 169 + (int)threadIdx.x % 13 * 13; int index7 = (int)threadIdx.x / 13 * 800 + ff_inner * 32 + rc_outer_inner * 2 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 25] = compute[index5 + 25] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 50] = compute[index5 + 50] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 75] = compute[index5 + 75] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 100] = compute[index5 + 100] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 125] = compute[index5 + 125] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 150] = compute[index5 + 150] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 175] = compute[index5 + 175] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 200] = compute[index5 + 200] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); compute[index5 + 225] = compute[index5 + 225] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 250] = compute[index5 + 250] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7]); compute[index5 + 275] = compute[index5 + 275] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7]); compute[index5 + 300] = compute[index5 + 300] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7]); } } } } for (int ax1_inner = 0; ax1_inner < 25; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 4225 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 25 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 1] = max(compute[index9 + 25] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 2] = max(compute[index9 + 50] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 3] = max(compute[index9 + 75] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 4] = max(compute[index9 + 100] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 5] = max(compute[index9 + 125] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 6] = max(compute[index9 + 150] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 7] = max(compute[index9 + 175] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 8] = max(compute[index9 + 200] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 9] = max(compute[index9 + 225] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 10] = max(compute[index9 + 250] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 11] = max(compute[index9 + 275] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 12] = max(compute[index9 + 300] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp9fpxo7ad
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<97>; .reg .b32 %r<18>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[16]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[16000]; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd11, [default_function_kernel0_param_1]; ld.param.u64 %rd12, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd13, %rd11; cvta.to.global.u64 %rd1, %rd12; mov.u32 %r1, %tid.x; mov.u32 %r8, %ctaid.x; mad.lo.s32 %r9, %r1, 169, %r8; shl.b32 %r10, %r1, 2; mov.u32 %r11, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r11, %r10; shr.u32 %r12, %r1, 2; and.b32 %r13, %r1, 3; mov.u32 %r14, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r14, %r10; shl.b32 %r15, %r1, 4; add.s32 %r4, %r14, %r15; cvt.s64.s32 %rd2, %r9; cvta.to.global.u64 %rd14, %rd10; mul.wide.s32 %rd15, %r9, 4; add.s64 %rd24, %rd14, %rd15; bfi.b32 %r16, %r12, %r13, 9, 23; mul.wide.s32 %rd16, %r16, 4; add.s64 %rd17, %rd13, %rd16; add.s64 %rd23, %rd17, 1024000; mov.f32 %f92, 0f00000000; mov.u32 %r17, 0; for.size.1 i in range(0, 4, 1): mov.f32 %f(93+i*1), %f92; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 3; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f16, [%rd24]; st.shared.f32 [%r2], %f16; $L__BB0_3: for.size.2 i in range(0, 20, 1): ld.global.nc.f32 %f(17+i*1), [%rd23+(-1024000+i*102400)]; st.shared.f32 [%r3+(0+i*800)], %f(17+i*1); bar.sync 0; for.size.11 i in range(0, 2, 1): ld.shared.f32 %f(37+i*11), [%r4+(0+i*4)]; ld.shared.f32 %f(38+i*11), [_ZZ24default_function_kernel0E15pad_temp_shared+(0+i*4)]; fma.rn.f32 %f(39+i*11), %f(38+i*11), %f(37+i*11), %f(92+i*-53); ld.shared.f32 %f(40+i*11), [%r4+(3200+i*4)]; fma.rn.f32 %f(41+i*11), %f(38+i*11), %f(40+i*11), %f(93+i*-52); ld.shared.f32 %f(42+i*11), [%r4+(6400+i*4)]; fma.rn.f32 %f(43+i*11), %f(38+i*11), %f(42+i*11), %f(94+i*-51); ld.shared.f32 %f(44+i*11), [%r4+(9600+i*4)]; fma.rn.f32 %f(45+i*11), %f(38+i*11), %f(44+i*11), %f(95+i*-50); ld.shared.f32 %f(46+i*11), [%r4+(12800+i*4)]; fma.rn.f32 %f(47+i*11), %f(38+i*11), %f(46+i*11), %f(96+i*-49); for.size.11 i in range(0, 2, 1): ld.shared.f32 %f(59+i*11), [%r4+(8+i*4)]; ld.shared.f32 %f(60+i*11), [_ZZ24default_function_kernel0E15pad_temp_shared+(8+i*4)]; fma.rn.f32 %f(61+i*31), %f(60+i*11), %f(59+i*11), %f(50+i*11); ld.shared.f32 %f(62+i*10), [%r4+(3208+i*4)]; fma.rn.f32 %f(63+i*30), %f(60+i*11), %f(62+i*10), %f(52+i*11); ld.shared.f32 %f(64+i*9), [%r4+(6408+i*4)]; fma.rn.f32 %f(65+i*29), %f(60+i*11), %f(64+i*9), %f(54+i*11); ld.shared.f32 %f(66+i*8), [%r4+(9608+i*4)]; fma.rn.f32 %f(67+i*28), %f(60+i*11), %f(66+i*8), %f(56+i*11); ld.shared.f32 %f(68+i*7), [%r4+(12808+i*4)]; fma.rn.f32 %f(69+i*27), %f(60+i*11), %f(68+i*7), %f(58+i*11); for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (2704+i*-2688); add.s32 %r17, %r17, 1; setp.ne.s32 %p2, %r17, 128; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd18, %rd9; mul.wide.s32 %rd19, %r1, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f76, [%rd20]; add.f32 %f77, %f92, %f76; mov.f32 %f78, 0f00000000; max.f32 %f79, %f77, %f78; shl.b64 %rd21, %rd2, 2; add.s64 %rd22, %rd1, %rd21; for.size.4 i in range(0, 4, 1): st.global.f32 [%rd22+(0+i*135200)], %f(79+i*3); ld.global.nc.f32 %f(80+i*3), [%rd20+(800+i*800)]; add.f32 %f(81+i*3), %f(93+i*1), %f(80+i*3); max.f32 %f(82+i*3), %f(81+i*3), %f78; st.global.f32 [%rd22+540800], %f91; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[5]; __shared__ float pad_temp_shared[4]; __shared__ float placeholder_shared[4000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 4) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 676 + (int)threadIdx.x * 169 + (int)blockIdx.x; pad_temp_shared[index0] = placeholder[index1]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 20; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x; int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 25600 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 4 + ((int)threadIdx.x & 3); placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { int index4 = rc_inner; int index5 = (int)threadIdx.x * 4 + rc_inner; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index5 + 800]); compute[2] = compute[2] + (pad_temp_shared[index4] * placeholder_shared[index5 + 1600]); compute[3] = compute[3] + (pad_temp_shared[index4] * placeholder_shared[index5 + 2400]); compute[4] = compute[4] + (pad_temp_shared[index4] * placeholder_shared[index5 + 3200]); } } int index0 = (int)threadIdx.x; int index6 = (int)threadIdx.x * 169 + (int)blockIdx.x; T_relu[index6] = max(compute[0] + placeholder2[index0], 0.000000e+00f); T_relu[index6 + 33800] = max(compute[1] + placeholder2[index0 + 200], 0.000000e+00f); T_relu[index6 + 67600] = max(compute[2] + placeholder2[index0 + 400], 0.000000e+00f); T_relu[index6 + 101400] = max(compute[3] + placeholder2[index0 + 600], 0.000000e+00f); T_relu[index6 + 135200] = max(compute[4] + placeholder2[index0 + 800], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmps9isacz8
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<9>; .reg .f32 %f<549>; .reg .b32 %r<154>; .reg .b64 %rd<39>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[13312]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd12, %rd8; cvta.to.global.u64 %rd2, %rd9; mov.u32 %r25, %ctaid.x; mul.hi.s32 %r26, %r25, 1321528399; shr.u32 %r27, %r26, 31; shr.s32 %r28, %r26, 2; add.s32 %r4, %r28, %r27; mul.lo.s32 %r29, %r4, 13; sub.s32 %r1, %r25, %r29; mov.u32 %r2, %tid.x; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(5+i*1), %r(4+i*-2), (128000+i*-126720); add.s64 %rd4, %rd12, 5200; mad.lo.s32 %r8, %r2, 13, %r1; shl.b32 %r30, %r2, 2; mov.u32 %r31, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r32, %r31, %r30; add.s32 %r9, %r32, 400; mov.u32 %r148, 0; mov.f32 %f419, 0f00000000; for.size.1 i in range(0, 64, 1): mov.f32 %f(420+i*1), %f419; $L__BB0_1: mov.u32 %r147, %tid.x; add.s32 %r151, %r147, 150; bar.sync 0; mad.lo.s32 %r34, %r148, 43264, %r8; mul.wide.s32 %rd13, %r34, 4; add.s64 %rd38, %rd4, %rd13; mov.u32 %r149, 100; mov.u32 %r150, %r9; bra.uni $L__BB0_2; $L__BB0_11: add.s32 %r151, %r151, 200; add.s64 %rd38, %rd38, 10400; for.size.1 i in range(0, 2, 1): add.s32 %r(150+i*-1), %r(150+i*-1), (800+i*-600); $L__BB0_2: add.s32 %r35, %r151, -150; setp.gt.s32 %p1, %r35, 3327; @%p1 bra $L__BB0_4; ld.global.nc.f32 %f261, [%rd38+-5200]; st.shared.f32 [%r150+-400], %f261; $L__BB0_4: add.s32 %r36, %r151, -100; setp.gt.s32 %p2, %r36, 3327; @%p2 bra $L__BB0_6; ld.global.nc.f32 %f262, [%rd38+-2600]; st.shared.f32 [%r150+-200], %f262; $L__BB0_6: add.s32 %r37, %r151, -50; setp.gt.s32 %p3, %r37, 3327; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f263, [%rd38]; st.shared.f32 [%r150], %f263; $L__BB0_8: setp.eq.s32 %p4, %r149, 3300; @%p4 bra $L__BB0_12; setp.gt.s32 %p5, %r151, 3327; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f264, [%rd38+2600]; st.shared.f32 [%r150+200], %f264; bra.uni $L__BB0_11; $L__BB0_12: shl.b32 %r39, %r148, 8; add.s32 %r18, %r39, %r5; mov.u32 %r152, 0; $L__BB0_13: mov.u32 %r135, %tid.x; shl.b32 %r134, %r135, 1; mad.lo.s32 %r40, %r152, 100, %r134; shl.b32 %r41, %r40, 2; mov.u32 %r42, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r43, %r42, %r41; for.size.10 i in range(0, 2, 1): shl.b32 %r(44+i*10), %r(40+i*13), 1; and.b32 %r(45+i*10), %r(44+i*10), -512; and.b32 %r(46+i*10), %r(40+i*13), 254; add.s32 %r(47+i*10), %r18, %r(46+i*10); add.s32 %r(48+i*10), %r(47+i*10), %r(45+i*10); mul.wide.s32 %rd(14+i*2), %r(48+i*10), 4; add.s64 %rd(15+i*2), %rd2, %rd(14+i*2); ld.global.nc.v2.u32 {%r(49+i*10), %r(50+i*10)}, [%rd(15+i*2)+0]; st.shared.v2.u32 [%r43+(0+i*400)], {%r(49+i*10), %r(50+i*10)}; add.s32 %r(53+i*10), %r40, (100+i*100); for.size.10 i in range(0, 5, 1): shl.b32 %r(64+i*10), %r(63+i*10), 1; and.b32 %r(65+i*10), %r(64+i*10), -512; and.b32 %r(66+i*10), %r(63+i*10), 254; add.s32 %r(67+i*10), %r18, %r(66+i*10); add.s32 %r(68+i*10), %r(67+i*10), %r(65+i*10); mul.wide.s32 %rd(18+i*2), %r(68+i*10), 4; add.s64 %rd(19+i*2), %rd2, %rd(18+i*2); ld.global.nc.v2.u32 {%r(69+i*10), %r(70+i*10)}, [%rd(19+i*2)+0]; st.shared.v2.u32 [%r43+(800+i*400)], {%r(69+i*10), %r(70+i*10)}; add.s32 %r(73+i*10), %r40, (300+i*100); shl.b32 %r114, %r113, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(115+i*1), %r(114+i*-1), (-512+i*766); for.size.1 i in range(0, 2, 1): add.s32 %r(117+i*1), %r(18+i*99), %r(116+i*-1); mul.wide.s32 %rd28, %r118, 4; add.s64 %rd29, %rd2, %rd28; ld.global.nc.v2.u32 {%r119, %r120}, [%rd29]; st.shared.v2.u32 [%r43+2800], {%r119, %r120}; add.s32 %r152, %r152, 8; setp.ne.s32 %p6, %r152, 640; @%p6 bra $L__BB0_13; bar.sync 0; mov.u32 %r153, 0; $L__BB0_15: mov.u32 %r137, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r136, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r124, %r153, %r6; shl.b32 %r125, %r124, 2; add.s32 %r127, %r136, %r125; mad.lo.s32 %r129, %r153, 52, %r137; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(265+i*1), [%r(129+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(483+i*-1), %f(265+i*2), %f266, %f(483+i*-1); ld.shared.f32 %f(267+i*1), [%r129+(4+i*4)]; for.size.2 i in range(0, 10, 1): fma.rn.f32 %f(481+i*-1), %f(268+i*1), %f266, %f(481+i*-1); ld.shared.f32 %f(269+i*1), [%r129+(12+i*4)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(471+i*-13), %f278, %f(266+i*13), %f(471+i*-13); ld.shared.f32 %f(279+i*1), [%r127+(1024+i*1024)]; fma.rn.f32 %f(470+i*-13), %f265, %f(279+i*1), %f(470+i*-13); fma.rn.f32 %f(469+i*-13), %f267, %f(279+i*1), %f(469+i*-13); fma.rn.f32 %f(468+i*-13), %f268, %f(279+i*1), %f(468+i*-13); fma.rn.f32 %f(467+i*-13), %f269, %f(279+i*1), %f(467+i*-13); fma.rn.f32 %f(466+i*-13), %f270, %f(279+i*1), %f(466+i*-13); fma.rn.f32 %f(465+i*-13), %f271, %f(279+i*1), %f(465+i*-13); fma.rn.f32 %f(464+i*-13), %f272, %f(279+i*1), %f(464+i*-13); fma.rn.f32 %f(463+i*-13), %f273, %f(279+i*1), %f(463+i*-13); fma.rn.f32 %f(462+i*-13), %f274, %f(279+i*1), %f(462+i*-13); fma.rn.f32 %f(461+i*-13), %f275, %f(279+i*1), %f(461+i*-13); fma.rn.f32 %f(460+i*-13), %f276, %f(279+i*1), %f(460+i*-13); fma.rn.f32 %f(459+i*-13), %f277, %f(279+i*1), %f(459+i*-13); for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(445+i*-13), %f278, %f(280+i*1), %f(445+i*-13); ld.shared.f32 %f(281+i*1), [%r127+(3072+i*1024)]; fma.rn.f32 %f(444+i*-13), %f265, %f(281+i*1), %f(444+i*-13); fma.rn.f32 %f(443+i*-13), %f267, %f(281+i*1), %f(443+i*-13); fma.rn.f32 %f(442+i*-13), %f268, %f(281+i*1), %f(442+i*-13); fma.rn.f32 %f(441+i*-13), %f269, %f(281+i*1), %f(441+i*-13); fma.rn.f32 %f(440+i*-13), %f270, %f(281+i*1), %f(440+i*-13); fma.rn.f32 %f(439+i*-13), %f271, %f(281+i*1), %f(439+i*-13); fma.rn.f32 %f(438+i*-13), %f272, %f(281+i*1), %f(438+i*-13); fma.rn.f32 %f(437+i*-13), %f273, %f(281+i*1), %f(437+i*-13); fma.rn.f32 %f(436+i*-13), %f274, %f(281+i*1), %f(436+i*-13); fma.rn.f32 %f(435+i*-13), %f275, %f(281+i*1), %f(435+i*-13); fma.rn.f32 %f(434+i*-13), %f276, %f(281+i*1), %f(434+i*-13); fma.rn.f32 %f(433+i*-13), %f277, %f(281+i*1), %f(433+i*-13); fma.rn.f32 %f419, %f278, %f282, %f419; add.s32 %r153, %r153, 1; setp.ne.s32 %p7, %r153, 256; @%p7 bra $L__BB0_15; add.s32 %r148, %r148, 1; setp.lt.u32 %p8, %r148, 2; @%p8 bra $L__BB0_1; mov.u32 %r145, %ctaid.x; mul.hi.s32 %r144, %r145, 1321528399; shr.s32 %r143, %r144, 2; shr.u32 %r142, %r144, 31; add.s32 %r141, %r143, %r142; ld.param.u64 %rd37, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd36, %rd37; ld.param.u64 %rd35, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd34, %rd35; mul.lo.s32 %r140, %r141, 13; sub.s32 %r139, %r145, %r140; mov.u32 %r138, %tid.x; mul.lo.s32 %r130, %r138, 5; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(131+i*1), %r(141+i*-3), (250+i*595), %r(130+i*9); mad.lo.s32 %r133, %r141, 42250, %r132; mul.wide.s32 %rd30, %r131, 4; add.s64 %rd31, %rd34, %rd30; ld.global.nc.f32 %f283, [%rd31]; add.f32 %f284, %f483, %f283; mov.f32 %f285, 0f00000000; max.f32 %f286, %f284, %f285; mul.wide.s32 %rd32, %r133, 4; add.s64 %rd33, %rd36, %rd32; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd33+(0+i*52)], %f(286+i*2); add.f32 %f(287+i*2), %f(482+i*-1), %f283; max.f32 %f(288+i*2), %f(287+i*2), %f285; for.size.40 i in range(0, 4, 1): st.global.f32 [%rd33+(624+i*676)], %f(310+i*27); ld.global.nc.f32 %f(311+i*27), [%rd31+(4+i*4)]; add.f32 %f(312+i*27), %f(470+i*-13), %f(311+i*27); max.f32 %f(313+i*27), %f(312+i*27), %f285; st.global.f32 [%rd33+(676+i*676)], %f(313+i*27); add.f32 %f(314+i*27), %f(469+i*-13), %f(311+i*27); max.f32 %f(315+i*27), %f(314+i*27), %f285; st.global.f32 [%rd33+(728+i*676)], %f(315+i*27); add.f32 %f(316+i*27), %f(468+i*-13), %f(311+i*27); max.f32 %f(317+i*27), %f(316+i*27), %f285; st.global.f32 [%rd33+(780+i*676)], %f(317+i*27); add.f32 %f(318+i*27), %f(467+i*-13), %f(311+i*27); max.f32 %f(319+i*27), %f(318+i*27), %f285; st.global.f32 [%rd33+(832+i*676)], %f(319+i*27); add.f32 %f(320+i*27), %f(466+i*-13), %f(311+i*27); max.f32 %f(321+i*27), %f(320+i*27), %f285; st.global.f32 [%rd33+(884+i*676)], %f(321+i*27); add.f32 %f(322+i*27), %f(465+i*-13), %f(311+i*27); max.f32 %f(323+i*27), %f(322+i*27), %f285; st.global.f32 [%rd33+(936+i*676)], %f(323+i*27); add.f32 %f(324+i*27), %f(464+i*-13), %f(311+i*27); max.f32 %f(325+i*27), %f(324+i*27), %f285; st.global.f32 [%rd33+(988+i*676)], %f(325+i*27); add.f32 %f(326+i*27), %f(463+i*-13), %f(311+i*27); max.f32 %f(327+i*27), %f(326+i*27), %f285; st.global.f32 [%rd33+(1040+i*676)], %f(327+i*27); add.f32 %f(328+i*27), %f(462+i*-13), %f(311+i*27); max.f32 %f(329+i*27), %f(328+i*27), %f285; st.global.f32 [%rd33+(1092+i*676)], %f(329+i*27); add.f32 %f(330+i*27), %f(461+i*-13), %f(311+i*27); max.f32 %f(331+i*27), %f(330+i*27), %f285; st.global.f32 [%rd33+(1144+i*676)], %f(331+i*27); add.f32 %f(332+i*27), %f(460+i*-13), %f(311+i*27); max.f32 %f(333+i*27), %f(332+i*27), %f285; st.global.f32 [%rd33+(1196+i*676)], %f(333+i*27); add.f32 %f(334+i*27), %f(459+i*-13), %f(311+i*27); max.f32 %f(335+i*27), %f(334+i*27), %f285; st.global.f32 [%rd33+(1248+i*676)], %f(335+i*27); add.f32 %f(336+i*27), %f(458+i*-13), %f(311+i*27); max.f32 %f(337+i*27), %f(336+i*27), %f285; st.global.f32 [%rd33+3328], %f418; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[65]; __shared__ float pad_temp_shared[3328]; __shared__ float placeholder_shared[64000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 5; ++ff_outer_inner_init) { for (int yy_outer_inner_init = 0; yy_outer_inner_init < 13; ++yy_outer_inner_init) { int index0 = ff_outer_inner_init * 13 + yy_outer_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 67; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 50 + (int)threadIdx.x < 3328) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 50 + (int)threadIdx.x; int index2 = rc_outer_outer * 43264 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 650 + (int)threadIdx.x * 13 + (int)blockIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 640; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { ((float2*)(placeholder_shared + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x * 2))[0] = ((float2*)(placeholder1 + (int)blockIdx.x / 13 * 128000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x * 2 >> 8) * 512 + rc_outer_outer * 256 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x * 2 & 255)))[0]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 256; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 5; ++ff_outer_inner) { for (int yy_outer_inner = 0; yy_outer_inner < 13; ++yy_outer_inner) { int index3 = ff_outer_inner * 13 + yy_outer_inner; int index4 = rc_outer_inner * 13 + yy_outer_inner; int index5 = (int)threadIdx.x * 1280 + ff_outer_inner * 256 + rc_outer_inner; compute[index3] = compute[index3] + pad_temp_shared[index4] * placeholder_shared[index5]; } } } } for (int ax1_inner = 0; ax1_inner < 5; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index6 = (int)blockIdx.x / 13 * 42250 + (int)threadIdx.x * 845 + ax1_inner * 169 + ax2_inner * 13 + (int)blockIdx.x % 13; int index7 = ax1_inner * 13 + ax2_inner; int index8 = (int)blockIdx.x / 13 * 250 + (int)threadIdx.x * 5 + ax1_inner; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpu5rdzo7q
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[800]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<1470>; .reg .b32 %r<83>; .reg .b64 %rd<44>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.u64 %rd16, 0; mov.f32 %f1070, 0f00000000; for.size.2 i in range(0, 3, 1): for.size.1 j in range(0, 5, 1): st.local.v4.f32 [%rd1+(0+i*16+j*160)], {%f1070, %f1070, %f1070, %f1070}; mov.u32 %r79, 0; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 4, 1): st.local.u32 [%rd1+(48+i*4+j*160)], %r79; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd1+(56+i*160)], %rd16; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd1+(688+i*-624)], {%f1070, %f1070, %f1070, %f1070}; for.size.1 i in range(0, 4, 1): st.local.v4.f32 [%rd1+(224+i*160)], {%f1070, %f1070, %f1070, %f1070}; for.size.1 i in range(0, 5, 1): st.local.v4.f32 [%rd1+(80+i*160)], {%f1070, %f1070, %f1070, %f1070}; for.size.4 i in range(0, 2, 1): for.size.1 j in range(0, 3, 1): st.local.v2.f32 [%rd1+(96+i*8+j*160)], {%f1070, %f1070}; for.size.1 j in range(0, 2, 1): st.local.u64 [%rd1+(576+i*8+j*160)], %rd16; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 5, 1): st.local.v4.f32 [%rd1+(112+i*16+j*160)], {%f1070, %f1070, %f1070, %f1070}; for.size.1 i in range(0, 5, 1): st.local.v2.f32 [%rd1+(144+i*160)], {%f1070, %f1070}; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd1+(152+i*160)], %rd16; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd1+(472+i*160)], {%f1070, %f1070}; mov.u32 %r19, %tid.x; mul.hi.s32 %r20, %r19, 1321528399; shr.u32 %r21, %r20, 31; shr.s32 %r22, %r20, 2; add.s32 %r1, %r22, %r21; mul.lo.s32 %r23, %r1, 13; sub.s32 %r2, %r19, %r23; cvta.to.global.u64 %rd17, %rd11; for.size.1 i in range(0, 199, 1): mov.f32 %f(1071+i*1), %f1070; $L__BB0_1: setp.gt.s32 %p1, %r19, 51; bar.sync 0; mov.u32 %r25, %ctaid.x; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(26+i*1), %r(25+i*-24), (13+i*156), %r(2+i*24); mad.lo.s32 %r28, %r79, 10816, %r27; mul.wide.s32 %rd18, %r28, 4; add.s64 %rd2, %rd17, %rd18; ld.global.nc.f32 %f801, [%rd2]; shl.b32 %r29, %r19, 2; mov.u32 %r30, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r31, %r30, %r29; for.size.2 i in range(0, 11, 1): st.shared.f32 [%r31+(0+i*260)], %f(801+i*1); ld.global.nc.f32 %f(802+i*1), [%rd2+(3380+i*3380)]; st.shared.f32 [%r31+2860], %f812; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f813, [%rd2+40560]; st.shared.f32 [%r31+3120], %f813; $L__BB0_3: mov.u32 %r80, 0; bra.uni $L__BB0_4; $L__BB0_13: add.s32 %r80, %r80, 4; $L__BB0_4: mad.lo.s32 %r5, %r80, 65, %r19; setp.gt.s32 %p2, %r5, 63999; shl.b32 %r38, %r5, 2; mov.u32 %r39, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r39, %r38; @%p2 bra $L__BB0_6; shl.b32 %r40, %r5, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(41+i*1), %r(40+i*-35), (-512+i*575); bfi.b32 %r43, %r79, %r42, 6, 26; add.s32 %r44, %r43, %r41; cvta.to.global.u64 %rd19, %rd12; mul.wide.s32 %rd20, %r44, 4; add.s64 %rd21, %rd19, %rd20; ld.global.nc.f32 %f814, [%rd21]; st.shared.f32 [%r6], %f814; $L__BB0_6: add.s32 %r7, %r80, 1; setp.eq.s32 %p3, %r7, 985; @%p3 bra $L__BB0_14; mad.lo.s32 %r8, %r7, 65, %r19; setp.gt.s32 %p4, %r8, 63999; @%p4 bra $L__BB0_9; shl.b32 %r46, %r8, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(47+i*1), %r(46+i*-38), (-512+i*575); bfi.b32 %r49, %r79, %r48, 6, 26; add.s32 %r50, %r49, %r47; cvta.to.global.u64 %rd22, %rd12; mul.wide.s32 %rd23, %r50, 4; add.s64 %rd24, %rd22, %rd23; ld.global.nc.f32 %f815, [%rd24]; st.shared.f32 [%r6+260], %f815; $L__BB0_9: add.s32 %r9, %r5, 130; setp.gt.s32 %p5, %r9, 63999; @%p5 bra $L__BB0_11; shl.b32 %r51, %r9, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(52+i*1), %r(51+i*-42), (-512+i*575); bfi.b32 %r54, %r79, %r53, 6, 26; add.s32 %r55, %r54, %r52; cvta.to.global.u64 %rd25, %rd12; mul.wide.s32 %rd26, %r55, 4; add.s64 %rd27, %rd25, %rd26; ld.global.nc.f32 %f816, [%rd27]; st.shared.f32 [%r6+520], %f816; $L__BB0_11: add.s32 %r10, %r5, 195; setp.gt.s32 %p6, %r10, 63999; @%p6 bra $L__BB0_13; shl.b32 %r56, %r10, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(57+i*1), %r(56+i*-46), (-512+i*575); bfi.b32 %r59, %r79, %r58, 6, 26; add.s32 %r60, %r59, %r57; cvta.to.global.u64 %rd28, %rd12; mul.wide.s32 %rd29, %r60, 4; add.s64 %rd30, %rd28, %rd29; ld.global.nc.f32 %f817, [%rd30]; st.shared.f32 [%r6+780], %f817; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r81, 0; $L__BB0_15: mul.lo.s32 %r78, %r1, 2560; mov.u32 %r77, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r75, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r62, %r81, 13, %r2; shl.b32 %r63, %r62, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(65+i*1), %r(75+i*6), %r(63+i*15); shl.b32 %r67, %r66, 2; add.s32 %r69, %r77, %r67; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(818+i*1), [%r(69+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1269+i*-1), %f819, %f(818+i*2), %f(1269+i*-1); ld.shared.f32 %f(820+i*1), [%r69+(51200+i*51200)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1267+i*-1), %f819, %f(821+i*1), %f(1267+i*-1); ld.shared.f32 %f(822+i*1), [%r69+(153600+i*51200)]; for.size.3 i in range(0, 39, 1): for.size.2 j in range(0, 5, 1): fma.rn.f32 %f(1265+i*-5+j*-1), %f819, %f(823+i*5+j*1), %f(1265+i*-5+j*-1); ld.shared.f32 %f(824+i*5+j*1), [%r69+(256+i*256+j*51200)]; fma.rn.f32 %f1070, %f819, %f1018, %f1070; add.s32 %r81, %r81, 1; setp.ne.s32 %p7, %r81, 64; @%p7 bra $L__BB0_15; add.s32 %r79, %r79, 1; setp.lt.u32 %p8, %r79, 8; @%p8 bra $L__BB0_1; ld.param.u64 %rd40, [default_function_kernel0_param_2]; mov.u32 %r76, %ctaid.x; ld.param.u64 %rd39, [default_function_kernel0_param_3]; add.u64 %rd43, %SPL, 0; mov.u32 %r82, 0; for.size.2 i in range(0, 3, 1): for.size.1 j in range(0, 5, 1): st.local.v4.f32 [%rd43+(0+i*16+j*160)], {%f(1269+i*-20+j*-1), %f(1264+i*-20+j*-1), %f(1259+i*-20+j*-1), %f(1254+i*-20+j*-1)}; for.size.2 i in range(0, 4, 1): for.size.1 j in range(0, 4, 1): st.local.f32 [%rd43+(48+i*4+j*160)], %f(1209+i*-5+j*-1); for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd43+(688+i*-624)], {%f(1205+i*-16), %f(1200+i*-16), %f(1195+i*-16), %f(1190+i*-16)}; for.size.1 i in range(0, 4, 1): st.local.v4.f32 [%rd43+(224+i*160)], {%f(1188+i*-1), %f(1183+i*-1), %f(1178+i*-1), %f(1173+i*-1)}; for.size.1 i in range(0, 5, 1): st.local.v4.f32 [%rd43+(80+i*160)], {%f(1169+i*-1), %f(1164+i*-1), %f(1159+i*-1), %f(1154+i*-1)}; for.size.6 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): st.local.f32 [%rd43+(576+i*8+j*160)], %f(1146+i*-10+j*-1); for.size.1 j in range(0, 3, 1): st.local.v2.f32 [%rd43+(96+i*8+j*160)], {%f(1149+i*-10+j*-1), %f(1144+i*-10+j*-1)}; for.size.1 j in range(0, 2, 1): st.local.f32 [%rd43+(580+i*8+j*160)], %f(1141+i*-10+j*-1); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 5, 1): st.local.v4.f32 [%rd43+(112+i*16+j*160)], {%f(1129+i*-20+j*-1), %f(1124+i*-20+j*-1), %f(1119+i*-20+j*-1), %f(1114+i*-20+j*-1)}; for.size.1 i in range(0, 5, 1): st.local.v2.f32 [%rd43+(144+i*160)], {%f(1089+i*-1), %f(1084+i*-1)}; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): st.local.f32 [%rd43+(152+i*4+j*160)], %f(1079+i*-5+j*-1); for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd43+(472+i*160)], {%f(1077+i*-1), %f(1072+i*-1)}; mul.lo.s32 %r71, %r1, 40; cvta.to.global.u64 %rd31, %rd39; mul.wide.s32 %rd32, %r71, 4; add.s64 %rd33, %rd31, %rd32; add.s64 %rd42, %rd33, 1604; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(72+i*2), %r(1+i*75), (6760+i*-6747), %r(2+i*70); cvta.to.global.u64 %rd34, %rd40; mul.wide.s32 %rd35, %r74, 4; add.s64 %rd36, %rd34, %rd35; add.s64 %rd41, %rd36, 271076; $L__BB0_18: ld.local.v2.f32 {%f1019, %f1020}, [%rd43]; ld.global.nc.f32 %f1023, [%rd42+-1604]; add.f32 %f1024, %f1019, %f1023; mov.f32 %f1025, 0f00000000; for.size.5 i in range(0, 2, 1): max.f32 %f(1026+i*7), %f(1024+i*8), %f1025; st.global.f32 [%rd41+(-271076+i*135200)], %f(1026+i*7); ld.local.v2.f32 {%f(1027+i*7), %f(1028+i*7)}, [%rd43+(160+i*160)]; ld.global.nc.f32 %f(1031+i*7), [%rd42+(-804+i*800)]; add.f32 %f(1032+i*7), %f(1027+i*7), %f(1031+i*7); for.size.5 i in range(0, 2, 1): max.f32 %f(1040+i*7), %f(1039+i*7), %f1025; st.global.f32 [%rd41+(-676+i*135200)], %f(1040+i*7); ld.local.v2.f32 {%f(1041+i*7), %f(1042+i*7)}, [%rd43+(480+i*160)]; ld.global.nc.f32 %f(1045+i*7), [%rd42+(796+i*800)]; add.f32 %f(1046+i*7), %f(1041+i*7), %f(1045+i*7); for.size.4 i in range(0, 2, 1): max.f32 %f(1054+i*3), %f(1053+i*3), %f1025; st.global.f32 [%rd41+(269724+i*-540124)], %f(1054+i*3); ld.global.nc.f32 %f(1055+i*3), [%rd42+(-1600+i*800)]; add.f32 %f(1056+i*3), %f(1020+i*8), %f(1055+i*3); for.size.4 i in range(0, 3, 1): max.f32 %f(1060+i*3), %f(1059+i*3), %f1025; st.global.f32 [%rd41+(-135200+i*135200)], %f(1060+i*3); ld.global.nc.f32 %f(1061+i*3), [%rd42+(0+i*800)]; add.f32 %f(1062+i*3), %f(1035+i*7), %f(1061+i*3); max.f32 %f1069, %f1068, %f1025; st.global.f32 [%rd41+270400], %f1069; for.size.1 i in range(0, 2, 1): add.s64 %rd(43+i*-1), %rd(43+i*-1), 8; add.s64 %rd41, %rd41, 1352; add.s32 %r82, %r82, 2; setp.ne.s32 %p9, %r82, 40; @%p9 bra $L__BB0_18; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[200]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; for (int ff_inner_init = 0; ff_inner_init < 40; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 40] = 0.000000e+00f; compute[index0 + 80] = 0.000000e+00f; compute[index0 + 120] = 0.000000e+00f; compute[index0 + 160] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 10816 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 65] = placeholder[index2 + 845]; pad_temp_shared[index1 + 130] = placeholder[index2 + 1690]; pad_temp_shared[index1 + 195] = placeholder[index2 + 2535]; pad_temp_shared[index1 + 260] = placeholder[index2 + 3380]; pad_temp_shared[index1 + 325] = placeholder[index2 + 4225]; pad_temp_shared[index1 + 390] = placeholder[index2 + 5070]; pad_temp_shared[index1 + 455] = placeholder[index2 + 5915]; pad_temp_shared[index1 + 520] = placeholder[index2 + 6760]; pad_temp_shared[index1 + 585] = placeholder[index2 + 7605]; pad_temp_shared[index1 + 650] = placeholder[index2 + 8450]; pad_temp_shared[index1 + 715] = placeholder[index2 + 9295]; if ((int)threadIdx.x < 52) { pad_temp_shared[index1 + 780] = placeholder[index2 + 10140]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 985; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x < 64000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 65 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 40; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 2560 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 40] = compute[index5 + 40] + (pad_temp_shared[index6] * placeholder_shared[index7 + 12800]); compute[index5 + 80] = compute[index5 + 80] + (pad_temp_shared[index6] * placeholder_shared[index7 + 25600]); compute[index5 + 120] = compute[index5 + 120] + (pad_temp_shared[index6] * placeholder_shared[index7 + 38400]); compute[index5 + 160] = compute[index5 + 160] + (pad_temp_shared[index6] * placeholder_shared[index7 + 51200]); } } } for (int ax1_inner = 0; ax1_inner < 40; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 6760 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 40 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 33800] = max(compute[index9 + 40] + placeholder2[index10 + 200], 0.000000e+00f); T_relu[index8 + 67600] = max(compute[index9 + 80] + placeholder2[index10 + 400], 0.000000e+00f); T_relu[index8 + 101400] = max(compute[index9 + 120] + placeholder2[index10 + 600], 0.000000e+00f); T_relu[index8 + 135200] = max(compute[index9 + 160] + placeholder2[index10 + 800], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpz4_y525l
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[1300]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<7>; .reg .f32 %f<2699>; .reg .b32 %r<61>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[5408]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[32000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.f32 %f2374, 0f00000000; mov.u32 %r57, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd1+(0+i*4+j*100)], %r57; mov.u32 %r13, %tid.x; mul.hi.s32 %r14, %r13, 1321528399; shr.u32 %r15, %r14, 31; shr.s32 %r16, %r14, 2; add.s32 %r2, %r16, %r15; mul.lo.s32 %r17, %r2, 13; sub.s32 %r18, %r13, %r17; mul.lo.s32 %r1, %r18, 13; cvta.to.global.u64 %rd16, %rd11; cvta.to.global.u64 %rd18, %rd12; for.size.1 i in range(0, 324, 1): mov.f32 %f(2375+i*1), %f2374; $L__BB0_1: setp.gt.s32 %p1, %r13, 311; bar.sync 0; mad.lo.s32 %r20, %r57, 1352, %r13; mul.wide.s32 %rd17, %r20, 4; add.s64 %rd2, %rd16, %rd17; ld.global.nc.f32 %f1626, [%rd2]; shl.b32 %r21, %r13, 2; mov.u32 %r22, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r23, %r22, %r21; st.shared.f32 [%r23], %f1626; ld.global.nc.f32 %f1627, [%rd2+2080]; st.shared.f32 [%r23+2080], %f1627; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f1628, [%rd2+4160]; st.shared.f32 [%r23+4160], %f1628; $L__BB0_3: setp.gt.s32 %p2, %r13, 499; @%p2 bra $L__BB0_5; for.size.1 i in range(0, 2, 1): shl.b32 %r(29+i*2), %r(57+i*-44), (3+i*7); add.s32 %r32, %r29, %r31; mul.wide.s32 %rd19, %r32, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f1629, [%rd20]; shl.b32 %r33, %r13, 6; mov.u32 %r34, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r35, %r34, %r33; for.size.2 i in range(0, 7, 1): st.shared.f32 [%r35+(0+i*4)], %f(1629+i*1); ld.global.nc.f32 %f(1630+i*1), [%rd20+(4+i*4)]; for.size.2 i in range(0, 8, 1): st.shared.f32 [%r35+(28+i*4)], %f(1636+i*1); ld.global.nc.f32 %f(1637+i*1), [%rd20+(2048+i*4)]; st.shared.f32 [%r35+60], %f1644; $L__BB0_5: bar.sync 0; mov.u32 %r58, 0; $L__BB0_6: mov.u32 %r59, 0; $L__BB0_7: shl.b32 %r56, %r58, 1; mad.lo.s32 %r55, %r2, 200, %r56; mov.u32 %r54, _ZZ24default_function_kernel0E15pad_temp_shared; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(53+i*-13), %r(58+i*1), (338+i*-169), %r(1+i*52); shl.b32 %r41, %r40, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(43+i*1), %r(54+i*1), %r(41+i*18); shl.b32 %r45, %r44, 2; mov.u32 %r46, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r47, %r46, %r45; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1645+i*1), [%r(47+i*-4)+0]; for.size.2 i in range(0, 12, 1): fma.rn.f32 %f(2698+i*-1), %f(1646+i*1), %f1645, %f(2698+i*-1); ld.shared.f32 %f(1647+i*1), [%r43+(4+i*4)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(2686+i*-13), %f1658, %f(1645+i*14), %f(2686+i*-13); ld.shared.f32 %f(1659+i*1), [%r47+(32+i*32)]; fma.rn.f32 %f(2685+i*-13), %f1646, %f(1659+i*1), %f(2685+i*-13); fma.rn.f32 %f(2684+i*-13), %f1647, %f(1659+i*1), %f(2684+i*-13); fma.rn.f32 %f(2683+i*-13), %f1648, %f(1659+i*1), %f(2683+i*-13); fma.rn.f32 %f(2682+i*-13), %f1649, %f(1659+i*1), %f(2682+i*-13); fma.rn.f32 %f(2681+i*-13), %f1650, %f(1659+i*1), %f(2681+i*-13); fma.rn.f32 %f(2680+i*-13), %f1651, %f(1659+i*1), %f(2680+i*-13); fma.rn.f32 %f(2679+i*-13), %f1652, %f(1659+i*1), %f(2679+i*-13); fma.rn.f32 %f(2678+i*-13), %f1653, %f(1659+i*1), %f(2678+i*-13); fma.rn.f32 %f(2677+i*-13), %f1654, %f(1659+i*1), %f(2677+i*-13); fma.rn.f32 %f(2676+i*-13), %f1655, %f(1659+i*1), %f(2676+i*-13); fma.rn.f32 %f(2675+i*-13), %f1656, %f(1659+i*1), %f(2675+i*-13); fma.rn.f32 %f(2674+i*-13), %f1657, %f(1659+i*1), %f(2674+i*-13); for.size.14 i in range(0, 22, 1): fma.rn.f32 %f(2660+i*-13), %f1658, %f(1660+i*1), %f(2660+i*-13); ld.shared.f32 %f(1661+i*1), [%r47+(96+i*32)]; fma.rn.f32 %f(2659+i*-13), %f1646, %f(1661+i*1), %f(2659+i*-13); fma.rn.f32 %f(2658+i*-13), %f1647, %f(1661+i*1), %f(2658+i*-13); fma.rn.f32 %f(2657+i*-13), %f1648, %f(1661+i*1), %f(2657+i*-13); fma.rn.f32 %f(2656+i*-13), %f1649, %f(1661+i*1), %f(2656+i*-13); fma.rn.f32 %f(2655+i*-13), %f1650, %f(1661+i*1), %f(2655+i*-13); fma.rn.f32 %f(2654+i*-13), %f1651, %f(1661+i*1), %f(2654+i*-13); fma.rn.f32 %f(2653+i*-13), %f1652, %f(1661+i*1), %f(2653+i*-13); fma.rn.f32 %f(2652+i*-13), %f1653, %f(1661+i*1), %f(2652+i*-13); fma.rn.f32 %f(2651+i*-13), %f1654, %f(1661+i*1), %f(2651+i*-13); fma.rn.f32 %f(2650+i*-13), %f1655, %f(1661+i*1), %f(2650+i*-13); fma.rn.f32 %f(2649+i*-13), %f1656, %f(1661+i*1), %f(2649+i*-13); fma.rn.f32 %f(2648+i*-13), %f1657, %f(1661+i*1), %f(2648+i*-13); fma.rn.f32 %f2374, %f1658, %f1682, %f2374; add.s32 %r48, %r59, 1; mov.u32 %r59, 1; setp.lt.u32 %p3, %r48, 2; @%p3 bra $L__BB0_7; add.s32 %r58, %r58, 1; setp.lt.u32 %p4, %r58, 4; @%p4 bra $L__BB0_6; add.s32 %r57, %r57, 1; setp.lt.u32 %p5, %r57, 64; @%p5 bra $L__BB0_1; ld.param.u64 %rd29, [default_function_kernel0_param_2]; ld.param.u64 %rd28, [default_function_kernel0_param_3]; add.u64 %rd31, %SPL, 0; for.size.2 i in range(0, 25, 1): for.size.1 j in range(0, 13, 1): st.local.f32 [%rd31+(0+i*4+j*100)], %f(2698+i*-13+j*-1); mul.lo.s32 %r50, %r2, 25; cvta.to.global.u64 %rd21, %rd28; mul.wide.s32 %rd22, %r50, 4; add.s64 %rd32, %rd21, %rd22; mad.lo.s32 %r51, %r2, 4225, %r1; cvta.to.global.u64 %rd23, %rd29; mul.wide.s32 %rd24, %r51, 4; add.s64 %rd25, %rd23, %rd24; add.s64 %rd30, %rd25, 24; mov.u32 %r60, 0; $L__BB0_11: ld.global.nc.f32 %f1683, [%rd32]; ld.local.f32 %f1684, [%rd31]; add.f32 %f1685, %f1684, %f1683; mov.f32 %f1686, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(1687+i*3), %f(1685+i*4), %f1686; st.global.f32 [%rd30+(-24+i*4)], %f(1687+i*3); ld.local.f32 %f(1688+i*3), [%rd31+(100+i*100)]; add.f32 %f(1689+i*3), %f(1688+i*3), %f1683; for.size.4 i in range(0, 10, 1): max.f32 %f(1693+i*3), %f(1692+i*3), %f1686; st.global.f32 [%rd30+(-16+i*4)], %f(1693+i*3); ld.local.f32 %f(1694+i*3), [%rd31+(300+i*100)]; add.f32 %f(1695+i*3), %f(1694+i*3), %f1683; max.f32 %f1723, %f1722, %f1686; st.global.f32 [%rd30+24], %f1723; for.size.1 i in range(0, 2, 1): add.s64 %rd(32+i*-1), %rd(32+i*-1), 4; add.s64 %rd30, %rd30, 676; add.s32 %r60, %r60, 1; setp.ne.s32 %p6, %r60, 25; @%p6 bra $L__BB0_11; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[325]; __shared__ float pad_temp_shared[1352]; __shared__ float placeholder_shared[8000]; for (int ff_inner_init = 0; ff_inner_init < 25; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 25] = 0.000000e+00f; compute[index0 + 50] = 0.000000e+00f; compute[index0 + 75] = 0.000000e+00f; compute[index0 + 100] = 0.000000e+00f; compute[index0 + 125] = 0.000000e+00f; compute[index0 + 150] = 0.000000e+00f; compute[index0 + 175] = 0.000000e+00f; compute[index0 + 200] = 0.000000e+00f; compute[index0 + 225] = 0.000000e+00f; compute[index0 + 250] = 0.000000e+00f; compute[index0 + 275] = 0.000000e+00f; compute[index0 + 300] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 1352 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 520] = placeholder[index2 + 520]; if ((int)threadIdx.x < 312) { pad_temp_shared[index1 + 1040] = placeholder[index2 + 1040]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; placeholder_shared[index1 * 16] = placeholder1[index3]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index4 = (int)threadIdx.x * 16 + 1; placeholder_shared[index4] = placeholder1[index3 + 1]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index5 = (int)threadIdx.x * 16 + 2; placeholder_shared[index5] = placeholder1[index3 + 2]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index6 = (int)threadIdx.x * 16 + 3; placeholder_shared[index6] = placeholder1[index3 + 3]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index7 = (int)threadIdx.x * 16 + 4; placeholder_shared[index7] = placeholder1[index3 + 4]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index8 = (int)threadIdx.x * 16 + 5; placeholder_shared[index8] = placeholder1[index3 + 5]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index9 = (int)threadIdx.x * 16 + 6; placeholder_shared[index9] = placeholder1[index3 + 6]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index10 = (int)threadIdx.x * 16 + 7; placeholder_shared[index10] = placeholder1[index3 + 7]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index11 = (int)threadIdx.x * 16 + 8; placeholder_shared[index11] = placeholder1[index3 + 512]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index12 = (int)threadIdx.x * 16 + 9; placeholder_shared[index12] = placeholder1[index3 + 513]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index13 = (int)threadIdx.x * 16 + 10; placeholder_shared[index13] = placeholder1[index3 + 514]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index14 = (int)threadIdx.x * 16 + 11; placeholder_shared[index14] = placeholder1[index3 + 515]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index15 = (int)threadIdx.x * 16 + 12; placeholder_shared[index15] = placeholder1[index3 + 516]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index16 = (int)threadIdx.x * 16 + 13; placeholder_shared[index16] = placeholder1[index3 + 517]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index17 = (int)threadIdx.x * 16 + 14; placeholder_shared[index17] = placeholder1[index3 + 518]; } if ((int)threadIdx.x < 500) { int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 8; int index18 = (int)threadIdx.x * 16 + 15; placeholder_shared[index18] = placeholder1[index3 + 519]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 25; ++ff_inner) { int index19 = ff_inner; int index20 = rc_outer_inner * 338 + rc_inner * 169 + (int)threadIdx.x % 13 * 13; int index21 = (int)threadIdx.x / 13 * 200 + ff_inner * 8 + rc_outer_inner * 2 + rc_inner; compute[index19] = compute[index19] + pad_temp_shared[index20] * placeholder_shared[index21]; compute[index19 + 25] = compute[index19 + 25] + (pad_temp_shared[index20 + 1] * placeholder_shared[index21]); compute[index19 + 50] = compute[index19 + 50] + (pad_temp_shared[index20 + 2] * placeholder_shared[index21]); compute[index19 + 75] = compute[index19 + 75] + (pad_temp_shared[index20 + 3] * placeholder_shared[index21]); compute[index19 + 100] = compute[index19 + 100] + (pad_temp_shared[index20 + 4] * placeholder_shared[index21]); compute[index19 + 125] = compute[index19 + 125] + (pad_temp_shared[index20 + 5] * placeholder_shared[index21]); compute[index19 + 150] = compute[index19 + 150] + (pad_temp_shared[index20 + 6] * placeholder_shared[index21]); compute[index19 + 175] = compute[index19 + 175] + (pad_temp_shared[index20 + 7] * placeholder_shared[index21]); compute[index19 + 200] = compute[index19 + 200] + (pad_temp_shared[index20 + 8] * placeholder_shared[index21]); compute[index19 + 225] = compute[index19 + 225] + (pad_temp_shared[index20 + 9] * placeholder_shared[index21]); compute[index19 + 250] = compute[index19 + 250] + (pad_temp_shared[index20 + 10] * placeholder_shared[index21]); compute[index19 + 275] = compute[index19 + 275] + (pad_temp_shared[index20 + 11] * placeholder_shared[index21]); compute[index19 + 300] = compute[index19 + 300] + (pad_temp_shared[index20 + 12] * placeholder_shared[index21]); } } } } for (int ax1_inner = 0; ax1_inner < 25; ++ax1_inner) { int index22 = (int)threadIdx.x / 13 * 4225 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13; int index23 = ax1_inner; int index24 = (int)threadIdx.x / 13 * 25 + ax1_inner; T_relu[index22] = max(compute[index23] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 1] = max(compute[index23 + 25] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 2] = max(compute[index23 + 50] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 3] = max(compute[index23 + 75] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 4] = max(compute[index23 + 100] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 5] = max(compute[index23 + 125] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 6] = max(compute[index23 + 150] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 7] = max(compute[index23 + 175] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 8] = max(compute[index23 + 200] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 9] = max(compute[index23 + 225] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 10] = max(compute[index23 + 250] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 11] = max(compute[index23 + 275] + placeholder2[index24], 0.000000e+00f); T_relu[index22 + 12] = max(compute[index23 + 300] + placeholder2[index24], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp9qc_yv4r
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<346>; .reg .b32 %r<47>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048000]; ld.param.u64 %rd12, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; ld.param.u64 %rd10, [default_function_kernel0_param_2]; ld.param.u64 %rd11, [default_function_kernel0_param_3]; mov.u32 %r19, %tid.x; cvt.s64.s32 %rd1, %r19; cvta.to.global.u64 %rd13, %rd12; mul.wide.s32 %rd14, %r19, 4; add.s64 %rd2, %rd13, %rd14; ld.global.nc.f32 %f81, [%rd2]; shl.b32 %r20, %r19, 2; mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r1, %r21, %r20; for.size.2 i in range(0, 19, 1): st.shared.f32 [%r1+(0+i*16900)], %f(81+i*1); ld.global.nc.f32 %f(82+i*1), [%rd2+(16900+i*16900)]; st.shared.f32 [%r1+321100], %f100; setp.gt.s32 %p1, %r19, 2027; @%p1 bra $L__BB0_2; ld.global.nc.f32 %f101, [%rd2+338000]; st.shared.f32 [%r1+338000], %f101; $L__BB0_2: cvt.u32.u64 %r23, %rd1; add.s32 %r43, %r23, 4225; mov.u32 %r44, 4225; cvta.to.global.u64 %rd15, %rd9; shl.b64 %rd16, %rd1, 2; add.s64 %rd17, %rd15, %rd16; add.s64 %rd22, %rd17, 16900; shl.b32 %r24, %r23, 2; mov.u32 %r25, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r26, %r25, %r24; add.s32 %r42, %r26, 16900; cvta.to.global.u64 %rd4, %rd11; cvta.to.global.u64 %rd5, %rd10; $L__BB0_3: add.s32 %r27, %r43, -4225; setp.gt.s32 %p2, %r27, 511999; add.s64 %rd7, %rd22, -16900; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f102, [%rd7]; add.s32 %r40, %r42, -16900; st.shared.f32 [%r40], %f102; $L__BB0_5: setp.gt.s32 %p3, %r43, 511999; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f103, [%rd7+16900]; add.s32 %r41, %r42, -16900; st.shared.f32 [%r41+16900], %f103; $L__BB0_7: for.size.1 i in range(0, 2, 1): add.s32 %r(44+i*-1), %r(44+i*-1), 8450; add.s64 %rd22, %rd22, 33800; add.s32 %r42, %r42, 33800; setp.ne.s32 %p4, %r44, 519675; @%p4 bra $L__BB0_3; bar.sync 0; mul.hi.s32 %r30, %r23, 1626496491; shr.u32 %r31, %r30, 31; shr.s32 %r32, %r30, 6; add.s32 %r12, %r32, %r31; mul.lo.s32 %r33, %r12, 169; sub.s32 %r11, %r23, %r33; shl.b32 %r34, %r11, 2; add.s32 %r46, %r21, %r34; mad.lo.s32 %r14, %r12, 40960, %r25; mov.f32 %f306, 0f00000000; mov.u32 %r45, 1024000; for.size.1 i in range(0, 39, 1): mov.f32 %f(307+i*1), %f306; $L__BB0_9: add.s32 %r37, %r14, %r45; ld.shared.f32 %f144, [%r37+-1024000]; for.size.2 i in range(0, 2, 1): ld.shared.f32 %f(145+i*1), [%r(46+i*-9)+0]; fma.rn.f32 %f(345+i*-1), %f145, %f(144+i*2), %f(345+i*-1); for.size.3 i in range(0, 19, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(147+i*2+j*1), [%r37+(-1021952+i*2048+j*1024000)]; fma.rn.f32 %f(343+i*-2+j*-1), %f145, %f(147+i*2+j*1), %f(343+i*-2+j*-1); for.size.1 i in range(0, 2, 1): add.s32 %r(46+i*-1), %r(46+i*-1), (676+i*-672); setp.ne.s32 %p5, %r45, 1026048; @%p5 bra $L__BB0_9; mul.lo.s32 %r38, %r12, 20; mad.lo.s32 %r39, %r12, 3380, %r11; mul.wide.s32 %rd18, %r38, 4; add.s64 %rd19, %rd4, %rd18; ld.global.nc.f32 %f185, [%rd19]; add.f32 %f186, %f345, %f185; mov.f32 %f187, 0f00000000; max.f32 %f188, %f186, %f187; mul.wide.s32 %rd20, %r39, 4; add.s64 %rd21, %rd5, %rd20; for.size.5 i in range(0, 19, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd21+(0+i*676+j*338000)], %f(188+i*6+j*3); ld.global.nc.f32 %f(189+i*6+j*3), [%rd19+(2000+i*4+j*-1996)]; add.f32 %f(190+i*6+j*3), %f(344+i*-2+j*-1), %f(189+i*6+j*3); max.f32 %f(191+i*6+j*3), %f(190+i*6+j*3), %f187; st.global.f32 [%rd21+12844], %f302; ld.global.nc.f32 %f303, [%rd19+2076]; add.f32 %f304, %f306, %f303; max.f32 %f305, %f304, %f187; st.global.f32 [%rd21+350844], %f305; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[40]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[512000]; compute[0] = 0.000000e+00f; compute[20] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[21] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[22] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[23] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[24] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[25] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[26] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[27] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[28] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[29] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[30] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[31] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[32] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[33] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[34] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[35] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[36] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[37] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[38] = 0.000000e+00f; compute[19] = 0.000000e+00f; compute[39] = 0.000000e+00f; int index0 = (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index0]; pad_temp_shared[index0 + 4225] = placeholder[index0 + 4225]; pad_temp_shared[index0 + 8450] = placeholder[index0 + 8450]; pad_temp_shared[index0 + 12675] = placeholder[index0 + 12675]; pad_temp_shared[index0 + 16900] = placeholder[index0 + 16900]; pad_temp_shared[index0 + 21125] = placeholder[index0 + 21125]; pad_temp_shared[index0 + 25350] = placeholder[index0 + 25350]; pad_temp_shared[index0 + 29575] = placeholder[index0 + 29575]; pad_temp_shared[index0 + 33800] = placeholder[index0 + 33800]; pad_temp_shared[index0 + 38025] = placeholder[index0 + 38025]; pad_temp_shared[index0 + 42250] = placeholder[index0 + 42250]; pad_temp_shared[index0 + 46475] = placeholder[index0 + 46475]; pad_temp_shared[index0 + 50700] = placeholder[index0 + 50700]; pad_temp_shared[index0 + 54925] = placeholder[index0 + 54925]; pad_temp_shared[index0 + 59150] = placeholder[index0 + 59150]; pad_temp_shared[index0 + 63375] = placeholder[index0 + 63375]; pad_temp_shared[index0 + 67600] = placeholder[index0 + 67600]; pad_temp_shared[index0 + 71825] = placeholder[index0 + 71825]; pad_temp_shared[index0 + 76050] = placeholder[index0 + 76050]; pad_temp_shared[index0 + 80275] = placeholder[index0 + 80275]; if ((int)threadIdx.x < 2028) { pad_temp_shared[index0 + 84500] = placeholder[index0 + 84500]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 122; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4225 + (int)threadIdx.x < 512000) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4225 + (int)threadIdx.x; placeholder_shared[index1] = placeholder1[index1]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 512; ++rc_outer_inner) { int index2 = rc_outer_inner * 169 + (int)threadIdx.x % 169; int index3 = (int)threadIdx.x / 169 * 10240 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index2] * placeholder_shared[index3]; compute[20] = compute[20] + (pad_temp_shared[index2] * placeholder_shared[index3 + 256000]); compute[1] = compute[1] + (pad_temp_shared[index2] * placeholder_shared[index3 + 512]); compute[21] = compute[21] + (pad_temp_shared[index2] * placeholder_shared[index3 + 256512]); compute[2] = compute[2] + (pad_temp_shared[index2] * placeholder_shared[index3 + 1024]); compute[22] = compute[22] + (pad_temp_shared[index2] * placeholder_shared[index3 + 257024]); compute[3] = compute[3] + (pad_temp_shared[index2] * placeholder_shared[index3 + 1536]); compute[23] = compute[23] + (pad_temp_shared[index2] * placeholder_shared[index3 + 257536]); compute[4] = compute[4] + (pad_temp_shared[index2] * placeholder_shared[index3 + 2048]); compute[24] = compute[24] + (pad_temp_shared[index2] * placeholder_shared[index3 + 258048]); compute[5] = compute[5] + (pad_temp_shared[index2] * placeholder_shared[index3 + 2560]); compute[25] = compute[25] + (pad_temp_shared[index2] * placeholder_shared[index3 + 258560]); compute[6] = compute[6] + (pad_temp_shared[index2] * placeholder_shared[index3 + 3072]); compute[26] = compute[26] + (pad_temp_shared[index2] * placeholder_shared[index3 + 259072]); compute[7] = compute[7] + (pad_temp_shared[index2] * placeholder_shared[index3 + 3584]); compute[27] = compute[27] + (pad_temp_shared[index2] * placeholder_shared[index3 + 259584]); compute[8] = compute[8] + (pad_temp_shared[index2] * placeholder_shared[index3 + 4096]); compute[28] = compute[28] + (pad_temp_shared[index2] * placeholder_shared[index3 + 260096]); compute[9] = compute[9] + (pad_temp_shared[index2] * placeholder_shared[index3 + 4608]); compute[29] = compute[29] + (pad_temp_shared[index2] * placeholder_shared[index3 + 260608]); compute[10] = compute[10] + (pad_temp_shared[index2] * placeholder_shared[index3 + 5120]); compute[30] = compute[30] + (pad_temp_shared[index2] * placeholder_shared[index3 + 261120]); compute[11] = compute[11] + (pad_temp_shared[index2] * placeholder_shared[index3 + 5632]); compute[31] = compute[31] + (pad_temp_shared[index2] * placeholder_shared[index3 + 261632]); compute[12] = compute[12] + (pad_temp_shared[index2] * placeholder_shared[index3 + 6144]); compute[32] = compute[32] + (pad_temp_shared[index2] * placeholder_shared[index3 + 262144]); compute[13] = compute[13] + (pad_temp_shared[index2] * placeholder_shared[index3 + 6656]); compute[33] = compute[33] + (pad_temp_shared[index2] * placeholder_shared[index3 + 262656]); compute[14] = compute[14] + (pad_temp_shared[index2] * placeholder_shared[index3 + 7168]); compute[34] = compute[34] + (pad_temp_shared[index2] * placeholder_shared[index3 + 263168]); compute[15] = compute[15] + (pad_temp_shared[index2] * placeholder_shared[index3 + 7680]); compute[35] = compute[35] + (pad_temp_shared[index2] * placeholder_shared[index3 + 263680]); compute[16] = compute[16] + (pad_temp_shared[index2] * placeholder_shared[index3 + 8192]); compute[36] = compute[36] + (pad_temp_shared[index2] * placeholder_shared[index3 + 264192]); compute[17] = compute[17] + (pad_temp_shared[index2] * placeholder_shared[index3 + 8704]); compute[37] = compute[37] + (pad_temp_shared[index2] * placeholder_shared[index3 + 264704]); compute[18] = compute[18] + (pad_temp_shared[index2] * placeholder_shared[index3 + 9216]); compute[38] = compute[38] + (pad_temp_shared[index2] * placeholder_shared[index3 + 265216]); compute[19] = compute[19] + (pad_temp_shared[index2] * placeholder_shared[index3 + 9728]); compute[39] = compute[39] + (pad_temp_shared[index2] * placeholder_shared[index3 + 265728]); } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { int index4 = (int)threadIdx.x / 169 * 3380 + ax1_inner * 169 + (int)threadIdx.x % 169; int index5 = ax1_inner; int index6 = (int)threadIdx.x / 169 * 20 + ax1_inner; T_relu[index4] = max(compute[index5] + placeholder2[index6], 0.000000e+00f); T_relu[index4 + 84500] = max(compute[index5 + 20] + placeholder2[index6 + 500], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmptbubjixh
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<9>; .reg .f32 %f<1023>; .reg .b32 %r<78>; .reg .b64 %rd<27>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd3, %rd6; mov.u32 %r1, %tid.x; mov.u32 %r2, %ctaid.x; mad.lo.s32 %r3, %r1, 208, %r2; mul.hi.s32 %r22, %r1, 1321528399; shr.u32 %r23, %r22, 31; shr.s32 %r24, %r22, 2; add.s32 %r5, %r24, %r23; mul.lo.s32 %r25, %r5, 13; sub.s32 %r4, %r1, %r25; mul.lo.s32 %r6, %r5, 6400; shl.b32 %r26, %r1, 6; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r7, %r27, %r26; mov.u32 %r75, 0; mov.f32 %f823, 0f00000000; for.size.1 i in range(0, 99, 1): mov.f32 %f(824+i*1), %f823; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 51; @%p1 bra $L__BB0_3; mad.lo.s32 %r28, %r75, 10816, %r3; mul.wide.s32 %rd9, %r28, 4; add.s64 %rd10, %rd1, %rd9; for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(401+i*1), [%rd10+(0+i*52)]; st.shared.f32 [%r7+(0+i*4)], %f(401+i*1); $L__BB0_3: mov.u32 %r76, 0; bra.uni $L__BB0_4; $L__BB0_13: add.s32 %r76, %r76, 4; $L__BB0_4: mad.lo.s32 %r11, %r76, 130, %r1; setp.gt.s32 %p2, %r11, 63999; shl.b32 %r30, %r11, 2; mov.u32 %r31, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r12, %r31, %r30; @%p2 bra $L__BB0_6; for.size.1 i in range(0, 2, 1): shl.b32 %r(64+i*-32), %r(75+i*-64), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(33+i*1), %r(32+i*-21), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(35+i*1), %r(34+i*1), %r(64+i*-31); mul.wide.s32 %rd11, %r36, 4; add.s64 %rd12, %rd3, %rd11; ld.global.nc.f32 %f417, [%rd12]; st.shared.f32 [%r12], %f417; $L__BB0_6: add.s32 %r13, %r76, 1; setp.eq.s32 %p3, %r13, 493; @%p3 bra $L__BB0_14; mad.lo.s32 %r14, %r13, 130, %r1; setp.gt.s32 %p4, %r14, 63999; @%p4 bra $L__BB0_9; for.size.1 i in range(0, 2, 1): shl.b32 %r(67+i*-30), %r(75+i*-61), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(38+i*1), %r(37+i*-23), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(40+i*1), %r(39+i*1), %r(67+i*-29); mul.wide.s32 %rd13, %r41, 4; add.s64 %rd14, %rd3, %rd13; ld.global.nc.f32 %f418, [%rd14]; st.shared.f32 [%r12+520], %f418; $L__BB0_9: add.s32 %r15, %r11, 260; setp.gt.s32 %p5, %r15, 63999; @%p5 bra $L__BB0_11; for.size.1 i in range(0, 2, 1): shl.b32 %r(66+i*-24), %r(75+i*-60), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(43+i*1), %r(42+i*-27), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(45+i*1), %r(44+i*1), %r(66+i*-23); mul.wide.s32 %rd15, %r46, 4; add.s64 %rd16, %rd3, %rd15; ld.global.nc.f32 %f419, [%rd16]; st.shared.f32 [%r12+1040], %f419; $L__BB0_11: add.s32 %r16, %r11, 390; setp.gt.s32 %p6, %r16, 63999; @%p6 bra $L__BB0_13; for.size.1 i in range(0, 2, 1): shl.b32 %r(65+i*-18), %r(75+i*-59), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(48+i*1), %r(47+i*-31), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(50+i*1), %r(49+i*1), %r(65+i*-17); mul.wide.s32 %rd17, %r51, 4; add.s64 %rd18, %rd3, %rd17; ld.global.nc.f32 %f420, [%rd18]; st.shared.f32 [%r12+1560], %f420; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r77, 0; $L__BB0_15: mov.u32 %r69, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r68, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r53, %r77, 13, %r4; shl.b32 %r54, %r53, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(56+i*1), %r(68+i*9), %r(54+i*-48); shl.b32 %r58, %r57, 2; add.s32 %r60, %r69, %r58; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(421+i*1), [%r(60+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(922+i*-1), %f422, %f(421+i*2), %f(922+i*-1); ld.shared.f32 %f(423+i*1), [%r60+(256+i*256)]; for.size.2 i in range(0, 97, 1): fma.rn.f32 %f(920+i*-1), %f422, %f(424+i*1), %f(920+i*-1); ld.shared.f32 %f(425+i*1), [%r60+(768+i*256)]; fma.rn.f32 %f823, %f422, %f521, %f823; add.s32 %r77, %r77, 1; setp.ne.s32 %p7, %r77, 64; @%p7 bra $L__BB0_15; add.s32 %r75, %r75, 1; setp.lt.u32 %p8, %r75, 8; @%p8 bra $L__BB0_1; ld.param.u64 %rd26, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd25, %rd26; ld.param.u64 %rd24, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd23, %rd24; mov.u32 %r74, %ctaid.x; mul.hi.s32 %r73, %r1, 1321528399; shr.u32 %r72, %r73, 31; shr.s32 %r71, %r73, 2; add.s32 %r70, %r71, %r72; mul.lo.s32 %r61, %r70, 100; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(62+i*1), %r(70+i*-66), (16900+i*-16887), %r(74+i*-12); mul.wide.s32 %rd19, %r61, 4; add.s64 %rd20, %rd23, %rd19; ld.global.nc.f32 %f522, [%rd20]; add.f32 %f523, %f922, %f522; mov.f32 %f524, 0f00000000; max.f32 %f525, %f523, %f524; mul.wide.s32 %rd21, %r63, 4; add.s64 %rd22, %rd25, %rd21; for.size.4 i in range(0, 99, 1): st.global.f32 [%rd22+(0+i*676)], %f(525+i*3); ld.global.nc.f32 %f(526+i*3), [%rd20+(4+i*4)]; add.f32 %f(527+i*3), %f(921+i*-1), %f(526+i*3); max.f32 %f(528+i*3), %f(527+i*3), %f524; st.global.f32 [%rd22+66924], %f822; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[100]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 25; ++ff_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 4; ++ff_inner_init) { int index0 = ff_outer_inner_init * 4 + ff_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_inner_s = 0; ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 16; ++ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) { if ((int)threadIdx.x * 16 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 832) { if ((int)threadIdx.x < 52) { int index1 = (int)threadIdx.x * 16 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s; int index2 = rc_outer_outer * 10816 + (int)threadIdx.x * 208 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s * 13 + (int)blockIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 493; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x < 64000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 25; ++ff_outer_inner) { for (int ff_inner = 0; ff_inner < 4; ++ff_inner) { int index5 = ff_outer_inner * 4 + ff_inner; int index6 = rc_outer_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 6400 + ff_outer_inner * 256 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } for (int ax1_inner = 0; ax1_inner < 100; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 16900 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 100 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpeyn9f7wr
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[67600]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<12>; .reg .f32 %f<446>; .reg .b32 %r<126>; .reg .b64 %rd<47>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd17, [default_function_kernel0_param_0]; ld.param.u64 %rd15, [default_function_kernel0_param_1]; ld.param.u64 %rd18, [default_function_kernel0_param_2]; ld.param.u64 %rd16, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd17; cvta.to.global.u64 %rd2, %rd18; add.u64 %rd3, %SPL, 0; mov.u32 %r32, 0; mov.u32 %r117, %r32; $L__BB0_1: mul.lo.s32 %r33, %r117, 169; mul.wide.s32 %rd20, %r33, 4; add.s64 %rd21, %rd3, %rd20; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(0+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(4+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(8+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(12+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(16+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(20+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(24+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(28+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(32+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(36+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(40+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(44+i*52+j*33800)], %r32; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd21+(48+i*52+j*33800)], %r32; add.s32 %r117, %r117, 1; setp.ne.s32 %p1, %r117, 50; @%p1 bra $L__BB0_1; cvta.to.global.u64 %rd4, %rd15; cvta.to.global.u64 %rd5, %rd16; mov.u32 %r3, %ctaid.x; mul.lo.s32 %r4, %r3, 256000; mov.u32 %r5, %tid.x; mul.lo.s32 %r6, %r5, 1600; mov.u32 %r35, 0; mov.u32 %r118, %r35; $L__BB0_3: bar.sync 0; mad.lo.s32 %r8, %r118, 5408, %r5; mov.u32 %r119, %r35; $L__BB0_4: mul.lo.s32 %r10, %r119, 5; add.s32 %r37, %r10, %r5; setp.gt.s32 %p2, %r37, 5407; add.s32 %r38, %r8, %r10; mul.wide.s32 %rd22, %r38, 4; add.s64 %rd6, %rd1, %rd22; shl.b32 %r39, %r37, 2; mov.u32 %r40, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r11, %r40, %r39; @%p2 bra $L__BB0_6; ld.global.nc.f32 %f19, [%rd6]; st.shared.f32 [%r11], %f19; $L__BB0_6: add.s32 %r42, %r37, 5; setp.gt.s32 %p3, %r42, 5407; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f20, [%rd6+20]; st.shared.f32 [%r11+20], %f20; $L__BB0_8: add.s32 %r119, %r119, 2; setp.ne.s32 %p4, %r119, 1082; @%p4 bra $L__BB0_4; shl.b32 %r44, %r118, 5; add.s32 %r13, %r44, %r4; mov.u32 %r120, 0; $L__BB0_10: mad.lo.s32 %r45, %r120, 5, %r5; shl.b32 %r46, %r45, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(47+i*1), %r(46+i*-1), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(49+i*1), %r(13+i*36), %r(48+i*-1); mul.wide.s32 %rd23, %r50, 4; add.s64 %rd24, %rd4, %rd23; ld.global.nc.f32 %f21, [%rd24]; shl.b32 %r51, %r45, 2; mov.u32 %r52, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r53, %r52, %r51; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r53+(0+i*20)], %f(21+i*1); add.s32 %r(54+i*6), %r45, (5+i*5); shl.b32 %r(55+i*6), %r(54+i*6), 4; and.b32 %r(56+i*6), %r(55+i*6), -512; and.b32 %r(57+i*6), %r(54+i*6), 31; add.s32 %r(58+i*6), %r13, %r(57+i*6); add.s32 %r(59+i*6), %r(58+i*6), %r(56+i*6); mul.wide.s32 %rd(25+i*2), %r(59+i*6), 4; add.s64 %rd(26+i*2), %rd4, %rd(25+i*2); ld.global.nc.f32 %f(22+i*1), [%rd(26+i*2)+0]; st.shared.f32 [%r53+140], %f28; add.s32 %r120, %r120, 8; setp.ne.s32 %p5, %r120, 3200; @%p5 bra $L__BB0_10; bar.sync 0; mov.u32 %r96, 0; mov.u32 %r121, %r96; $L__BB0_12: mul.lo.s32 %r17, %r121, 1352; shl.b32 %r98, %r121, 3; add.s32 %r18, %r98, %r6; mov.u32 %r122, %r96; $L__BB0_13: mul.lo.s32 %r20, %r122, 169; for.size.2 i in range(0, 2, 1): shl.b32 %r(100+i*2), %r(122+i*-21), (5+i*-3); add.s32 %r(101+i*3), %r(18+i*34), %r(100+i*2); ld.shared.f32 %f1, [%r104]; mov.u32 %r123, 0; for.size.2 i in range(0, 7, 1): for.size.1 j in range(0, 2, 1): ld.shared.f32 %f(2+i*2+j*1), [%r104+(32000+i*4+j*-31996)]; ld.shared.f32 %f16, [%r104+32028]; $L__BB0_14: add.s32 %r105, %r123, %r20; mul.wide.s32 %rd39, %r105, 4; add.s64 %rd40, %rd3, %rd39; add.s32 %r106, %r123, %r17; shl.b32 %r107, %r106, 2; add.s32 %r109, %r40, %r107; for.size.5 i in range(0, 13, 1): ld.shared.f32 %f(29+i*5), [%r109+(0+i*52)]; ld.local.f32 %f(30+i*5), [%rd40+(0+i*52)]; fma.rn.f32 %f(31+i*5), %f(29+i*5), %f1, %f(30+i*5); ld.local.f32 %f(32+i*5), [%rd40+(33800+i*52)]; fma.rn.f32 %f(33+i*5), %f(29+i*5), %f2, %f(32+i*5); for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(94+i*3), [%r109+(676+i*52)]; fma.rn.f32 %f(95+i*3), %f(94+i*3), %f3, %f(31+i*5); fma.rn.f32 %f(96+i*3), %f(94+i*3), %f4, %f(33+i*5); for.size.4 i in range(0, 5, 1): for.size.3 j in range(0, 13, 1): ld.shared.f32 %f(133+i*39+j*3), [%r109+(1352+i*676+j*52)]; fma.rn.f32 %f(134+i*39+j*3), %f(133+i*39+j*3), %f(5+j*2), %f(95+i*39+j*3); fma.rn.f32 %f(135+i*39+j*3), %f(133+i*39+j*3), %f(6+j*2), %f(96+i*39+j*3); for.size.5 i in range(0, 13, 1): ld.shared.f32 %f(328+i*3), [%r109+(4732+i*52)]; fma.rn.f32 %f(329+i*3), %f(328+i*3), %f15, %f(290+i*3); st.local.f32 [%rd40+(0+i*52)], %f(329+i*3); fma.rn.f32 %f(330+i*3), %f(328+i*3), %f16, %f(291+i*3); st.local.f32 [%rd40+(33800+i*52)], %f(330+i*3); add.s32 %r123, %r123, 1; setp.ne.s32 %p6, %r123, 13; @%p6 bra $L__BB0_14; add.s32 %r122, %r122, 1; setp.lt.u32 %p7, %r122, 50; @%p7 bra $L__BB0_13; add.s32 %r121, %r121, 1; setp.lt.u32 %p8, %r121, 4; @%p8 bra $L__BB0_12; add.s32 %r118, %r118, 1; setp.lt.u32 %p9, %r118, 16; @%p9 bra $L__BB0_3; mul.lo.s32 %r111, %r5, 50; mad.lo.s32 %r26, %r3, 500, %r111; for.size.1 i in range(0, 2, 1): add.s64 %rd(7+i*1), %rd(3+i*-1), (33848+i*135200); mul.lo.s32 %r112, %r5, 8450; mad.lo.s32 %r27, %r3, 84500, %r112; mov.u32 %r110, 0; mov.u32 %r124, %r110; $L__BB0_19: add.s32 %r114, %r26, %r124; mul.wide.s32 %rd41, %r114, 4; add.s64 %rd42, %rd5, %rd41; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(17+i*1), [%rd42+(0+i*1000)]; mul.lo.s32 %r115, %r124, 169; mul.wide.s32 %rd43, %r115, 4; add.s64 %rd46, %rd7, %rd43; add.s32 %r116, %r27, %r115; mul.wide.s32 %rd44, %r116, 4; add.s64 %rd45, %rd8, %rd44; mov.u32 %r125, %r110; $L__BB0_20: ld.local.f32 %f367, [%rd46+-33848]; add.f32 %f368, %f367, %f17; mov.f32 %f369, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(370+i*3), %f(368+i*4), %f369; st.global.f32 [%rd45+(-169048+i*169000)], %f(370+i*3); ld.local.f32 %f(371+i*3), [%rd46+(-48+i*-33796)]; add.f32 %f(372+i*3), %f(371+i*3), %f(18+i*-1); for.size.5 i in range(0, 11, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(376+i*6+j*3), %f(375+i*6+j*3), %f369; st.global.f32 [%rd45+(-169044+i*4+j*169000)], %f(376+i*6+j*3); ld.local.f32 %f(377+i*6+j*3), [%rd46+(-44+i*4+j*-33796)]; add.f32 %f(378+i*6+j*3), %f(377+i*6+j*3), %f(18+j*-1); max.f32 %f442, %f441, %f369; st.global.f32 [%rd45+-169000], %f442; ld.local.f32 %f443, [%rd46]; add.f32 %f444, %f443, %f18; max.f32 %f445, %f444, %f369; st.global.f32 [%rd45], %f445; for.size.1 i in range(0, 2, 1): add.s64 %rd(46+i*-1), %rd(46+i*-1), 52; add.s32 %r125, %r125, 1; setp.ne.s32 %p10, %r125, 13; @%p10 bra $L__BB0_20; add.s32 %r124, %r124, 1; setp.lt.u32 %p11, %r124, 50; @%p11 bra $L__BB0_19; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16900]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[16000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 50; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_outer_inner_init * 169 + yy_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 8450] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 1082; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x < 5408) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 3200; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 256000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 50; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = ff_outer_inner * 169 + yy_inner * 13 + xx_outer_inner; int index6 = rc_outer_inner * 1352 + rc_inner * 169 + yy_inner * 13 + xx_outer_inner; int index7 = (int)threadIdx.x * 1600 + ff_outer_inner * 32 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 8450] = compute[index5 + 8450] + (pad_temp_shared[index6] * placeholder_shared[index7 + 8000]); } } } } } } for (int ax1_inner = 0; ax1_inner < 50; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x * 84500 + (int)threadIdx.x * 8450 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index9 = ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x * 500 + (int)threadIdx.x * 50 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 42250] = max(compute[index9 + 8450] + placeholder2[index10 + 250], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmptxroi8al
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<10>; .reg .f32 %f<1214>; .reg .b32 %r<90>; .reg .b64 %rd<28>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd2, [default_function_kernel0_param_0]; ld.param.u64 %rd3, [default_function_kernel0_param_1]; mov.u32 %r22, %tid.x; mul.hi.s32 %r23, %r22, 1321528399; shr.u32 %r24, %r23, 31; shr.s32 %r25, %r23, 2; add.s32 %r1, %r25, %r24; mov.u32 %r26, %ctaid.x; mul.lo.s32 %r27, %r1, 13; sub.s32 %r2, %r22, %r27; mad.lo.s32 %r3, %r26, 13, %r2; mov.u32 %r86, 0; mov.f32 %f1114, 0f00000000; cvta.to.global.u64 %rd6, %rd2; for.size.1 i in range(0, 99, 1): mov.f32 %f(1115+i*1), %f1114; $L__BB0_1: setp.gt.s32 %p1, %r22, 51; bar.sync 0; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(29+i*1), %r(1+i*85), (169+i*10647), %r(3+i*26); mul.wide.s32 %rd7, %r30, 4; add.s64 %rd1, %rd6, %rd7; ld.global.nc.f32 %f501, [%rd1]; shl.b32 %r31, %r22, 2; mov.u32 %r32, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r33, %r32, %r31; for.size.2 i in range(0, 5, 1): st.shared.f32 [%r33+(0+i*520)], %f(501+i*1); ld.global.nc.f32 %f(502+i*1), [%rd1+(6760+i*6760)]; st.shared.f32 [%r33+2600], %f506; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f507, [%rd1+40560]; st.shared.f32 [%r33+3120], %f507; $L__BB0_3: mov.u32 %r87, 0; bra.uni $L__BB0_4; $L__BB0_13: add.s32 %r87, %r87, 4; $L__BB0_4: mad.lo.s32 %r6, %r87, 130, %r22; setp.gt.s32 %p2, %r6, 63999; shl.b32 %r40, %r6, 2; mov.u32 %r41, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r41, %r40; @%p2 bra $L__BB0_6; shl.b32 %r42, %r6, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(43+i*1), %r(42+i*-36), (-512+i*575); bfi.b32 %r45, %r86, %r44, 6, 26; add.s32 %r46, %r45, %r43; cvta.to.global.u64 %rd8, %rd3; mul.wide.s32 %rd9, %r46, 4; add.s64 %rd10, %rd8, %rd9; ld.global.nc.f32 %f508, [%rd10]; st.shared.f32 [%r7], %f508; $L__BB0_6: add.s32 %r8, %r87, 1; setp.eq.s32 %p3, %r8, 493; @%p3 bra $L__BB0_14; mad.lo.s32 %r9, %r8, 130, %r22; setp.gt.s32 %p4, %r9, 63999; @%p4 bra $L__BB0_9; shl.b32 %r48, %r9, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(49+i*1), %r(48+i*-39), (-512+i*575); bfi.b32 %r51, %r86, %r50, 6, 26; add.s32 %r52, %r51, %r49; cvta.to.global.u64 %rd11, %rd3; mul.wide.s32 %rd12, %r52, 4; add.s64 %rd13, %rd11, %rd12; ld.global.nc.f32 %f509, [%rd13]; st.shared.f32 [%r7+520], %f509; $L__BB0_9: add.s32 %r10, %r6, 260; setp.gt.s32 %p5, %r10, 63999; @%p5 bra $L__BB0_11; shl.b32 %r53, %r10, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(54+i*1), %r(53+i*-43), (-512+i*575); bfi.b32 %r56, %r86, %r55, 6, 26; add.s32 %r57, %r56, %r54; cvta.to.global.u64 %rd14, %rd3; mul.wide.s32 %rd15, %r57, 4; add.s64 %rd16, %rd14, %rd15; ld.global.nc.f32 %f510, [%rd16]; st.shared.f32 [%r7+1040], %f510; $L__BB0_11: add.s32 %r11, %r6, 390; setp.gt.s32 %p6, %r11, 63999; @%p6 bra $L__BB0_13; shl.b32 %r58, %r11, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(59+i*1), %r(58+i*-47), (-512+i*575); bfi.b32 %r61, %r86, %r60, 6, 26; add.s32 %r62, %r61, %r59; cvta.to.global.u64 %rd17, %rd3; mul.wide.s32 %rd18, %r62, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f511, [%rd19]; st.shared.f32 [%r7+1560], %f511; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r88, 0; $L__BB0_15: mov.u32 %r89, 0; $L__BB0_16: mov.u32 %r83, _ZZ24default_function_kernel0E18placeholder_shared; mul.lo.s32 %r82, %r1, 1280; shl.b32 %r81, %r88, 2; add.s32 %r80, %r81, %r82; mov.u32 %r79, _ZZ24default_function_kernel0E15pad_temp_shared; mul.lo.s32 %r78, %r1, 13; sub.s32 %r77, %r22, %r78; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(76+i*-10), %r(88+i*1), (52+i*-39), %r(77+i*-1); shl.b32 %r67, %r66, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(69+i*1), %r(79+i*1), %r(67+i*22); shl.b32 %r71, %r70, 2; add.s32 %r73, %r83, %r71; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(512+i*1), [%r(73+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1213+i*-20), %f513, %f(512+i*2), %f(1213+i*-20); ld.shared.f32 %f(514+i*1), [%r73+(51200+i*51200)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1173+i*-20), %f513, %f(515+i*1), %f(1173+i*-20); ld.shared.f32 %f(516+i*1), [%r73+(153600+i*51200)]; for.size.6 i in range(0, 19, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(1133+i*-1+j*79), %f513, %f(517+i*5+j*1), %f(1133+i*-1+j*79); ld.shared.f32 %f(518+i*5+j*1), [%r73+(256+i*256+j*51200)]; for.size.2 j in range(0, 3, 1): fma.rn.f32 %f(1192+i*-1+j*-20), %f513, %f(519+i*5+j*1), %f(1192+i*-1+j*-20); ld.shared.f32 %f(520+i*5+j*1), [%r73+(102656+i*256+j*51200)]; fma.rn.f32 %f1114, %f513, %f612, %f1114; add.s32 %r89, %r89, 1; setp.ne.s32 %p7, %r89, 4; @%p7 bra $L__BB0_16; add.s32 %r88, %r88, 1; setp.lt.u32 %p8, %r88, 16; @%p8 bra $L__BB0_15; add.s32 %r86, %r86, 1; setp.lt.u32 %p9, %r86, 8; @%p9 bra $L__BB0_1; ld.param.u64 %rd27, [default_function_kernel0_param_2]; ld.param.u64 %rd26, [default_function_kernel0_param_3]; mul.lo.s32 %r74, %r1, 20; mad.lo.s32 %r75, %r1, 3380, %r3; cvta.to.global.u64 %rd20, %rd26; mul.wide.s32 %rd21, %r74, 4; add.s64 %rd22, %rd20, %rd21; ld.global.nc.f32 %f613, [%rd22]; add.f32 %f614, %f1213, %f613; mov.f32 %f615, 0f00000000; max.f32 %f616, %f614, %f615; cvta.to.global.u64 %rd23, %rd27; mul.wide.s32 %rd24, %r75, 4; add.s64 %rd25, %rd23, %rd24; for.size.4 i in range(0, 4, 1): st.global.f32 [%rd25+(0+i*135200)], %f(616+i*3); ld.global.nc.f32 %f(617+i*3), [%rd22+(800+i*800)]; add.f32 %f(618+i*3), %f(1193+i*-20), %f(617+i*3); max.f32 %f(619+i*3), %f(618+i*3), %f615; for.size.10 i in range(0, 19, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd25+(540800+i*676+j*-540124)], %f(628+i*15+j*3); ld.global.nc.f32 %f(629+i*15+j*3), [%rd22+(4+i*4+j*800)]; add.f32 %f(630+i*15+j*3), %f(1212+i*-1+j*-20), %f(629+i*15+j*3); max.f32 %f(631+i*15+j*3), %f(630+i*15+j*3), %f615; for.size.4 j in range(0, 3, 1): st.global.f32 [%rd25+(135876+i*676+j*135200)], %f(634+i*15+j*3); ld.global.nc.f32 %f(635+i*15+j*3), [%rd22+(1604+i*4+j*800)]; add.f32 %f(636+i*15+j*3), %f(1172+i*-1+j*-20), %f(635+i*15+j*3); max.f32 %f(637+i*15+j*3), %f(636+i*15+j*3), %f615; st.global.f32 [%rd25+553644], %f913; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[100]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; for (int ff_inner_init = 0; ff_inner_init < 20; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 20] = 0.000000e+00f; compute[index0 + 40] = 0.000000e+00f; compute[index0 + 60] = 0.000000e+00f; compute[index0 + 80] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 10816 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 130] = placeholder[index2 + 1690]; pad_temp_shared[index1 + 260] = placeholder[index2 + 3380]; pad_temp_shared[index1 + 390] = placeholder[index2 + 5070]; pad_temp_shared[index1 + 520] = placeholder[index2 + 6760]; pad_temp_shared[index1 + 650] = placeholder[index2 + 8450]; if ((int)threadIdx.x < 52) { pad_temp_shared[index1 + 780] = placeholder[index2 + 10140]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 493; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x < 64000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int ff_inner = 0; ff_inner < 20; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 52 + rc_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 1280 + ff_inner * 64 + rc_outer_inner * 4 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 20] = compute[index5 + 20] + (pad_temp_shared[index6] * placeholder_shared[index7 + 12800]); compute[index5 + 40] = compute[index5 + 40] + (pad_temp_shared[index6] * placeholder_shared[index7 + 25600]); compute[index5 + 60] = compute[index5 + 60] + (pad_temp_shared[index6] * placeholder_shared[index7 + 38400]); compute[index5 + 80] = compute[index5 + 80] + (pad_temp_shared[index6] * placeholder_shared[index7 + 51200]); } } } } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 3380 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 20 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 33800] = max(compute[index9 + 20] + placeholder2[index10 + 200], 0.000000e+00f); T_relu[index8 + 67600] = max(compute[index9 + 40] + placeholder2[index10 + 400], 0.000000e+00f); T_relu[index8 + 101400] = max(compute[index9 + 60] + placeholder2[index10 + 600], 0.000000e+00f); T_relu[index8 + 135200] = max(compute[index9 + 80] + placeholder2[index10 + 800], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp0tn6c6kp
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<21>; .reg .f32 %f<172>; .reg .b32 %r<74>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[26624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[204800]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd10, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd11, %rd9; cvta.to.global.u64 %rd1, %rd10; mov.u32 %r1, %ctaid.x; mul.hi.s32 %r31, %r1, 1321528399; shr.u32 %r32, %r31, 31; shr.s32 %r33, %r31, 2; add.s32 %r34, %r33, %r32; mul.lo.s32 %r35, %r34, 13; sub.s32 %r2, %r1, %r35; mov.u32 %r3, %tid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(67+i*-31), %r3, (2+i*2); mov.u32 %r37, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r38, %r37, %r36; add.s32 %r65, %r38, 1612; mad.lo.s32 %r39, %r3, 52, %r2; mul.wide.s32 %rd12, %r39, 4; add.s64 %rd13, %rd11, %rd12; add.s64 %rd20, %rd13, 20956; mov.u32 %r68, 18; mov.u32 %r66, %r3; bra.uni $L__BB0_1; $L__BB0_26: for.size.1 i in range(0, 2, 1): add.s32 %r(67+i*-1), %r(67+i*-1), (800+i*-600); add.s32 %r65, %r65, 3200; add.s64 %rd20, %rd20, 41600; $L__BB0_1: setp.gt.s32 %p1, %r67, 6655; @%p1 bra $L__BB0_4; setp.gt.s32 %p2, %r66, 1663; @%p2 bra $L__BB0_4; ld.global.nc.f32 %f40, [%rd20+-20956]; st.shared.f32 [%r65+-1612], %f40; $L__BB0_4: setp.gt.s32 %p3, %r67, 6654; @%p3 bra $L__BB0_7; setp.gt.s32 %p4, %r66, 1663; @%p4 bra $L__BB0_7; ld.global.nc.f32 %f41, [%rd20+-20904]; st.shared.f32 [%r65+-1608], %f41; $L__BB0_7: setp.gt.s32 %p5, %r67, 6653; @%p5 bra $L__BB0_10; setp.gt.s32 %p6, %r66, 1663; @%p6 bra $L__BB0_10; ld.global.nc.f32 %f42, [%rd20+-20852]; st.shared.f32 [%r65+-1604], %f42; $L__BB0_10: setp.gt.s32 %p7, %r67, 6652; @%p7 bra $L__BB0_13; setp.gt.s32 %p8, %r66, 1663; @%p8 bra $L__BB0_13; ld.global.nc.f32 %f43, [%rd20+-20800]; st.shared.f32 [%r65+-1600], %f43; $L__BB0_13: add.s32 %r68, %r68, -2; setp.eq.s32 %p9, %r68, 0; @%p9 bra $L__BB0_27; add.s32 %r11, %r67, 400; setp.gt.s32 %p10, %r11, 6655; @%p10 bra $L__BB0_17; add.s32 %r40, %r66, 100; setp.gt.s32 %p11, %r40, 1663; @%p11 bra $L__BB0_17; ld.global.nc.f32 %f44, [%rd20+-156]; st.shared.f32 [%r65+-12], %f44; $L__BB0_17: setp.gt.s32 %p12, %r11, 6654; @%p12 bra $L__BB0_20; add.s32 %r41, %r66, 100; setp.gt.s32 %p13, %r41, 1663; @%p13 bra $L__BB0_20; ld.global.nc.f32 %f45, [%rd20+-104]; st.shared.f32 [%r65+-8], %f45; $L__BB0_20: setp.gt.s32 %p14, %r11, 6653; @%p14 bra $L__BB0_23; add.s32 %r42, %r66, 100; setp.gt.s32 %p15, %r42, 1663; @%p15 bra $L__BB0_23; ld.global.nc.f32 %f46, [%rd20+-52]; st.shared.f32 [%r65+-4], %f46; $L__BB0_23: setp.gt.s32 %p16, %r11, 6652; @%p16 bra $L__BB0_26; add.s32 %r43, %r66, 100; setp.gt.s32 %p17, %r43, 1663; @%p17 bra $L__BB0_26; ld.global.nc.f32 %f47, [%rd20]; st.shared.f32 [%r65], %f47; bra.uni $L__BB0_26; $L__BB0_27: cvta.to.global.u64 %rd5, %rd7; cvta.to.global.u64 %rd6, %rd8; mad.lo.s32 %r16, %r34, 51200, %r3; mov.u32 %r69, 0; $L__BB0_28: mul.lo.s32 %r48, %r69, 100; add.s32 %r49, %r16, %r48; mul.wide.s32 %rd14, %r49, 4; add.s64 %rd15, %rd5, %rd14; ld.global.nc.f32 %f48, [%rd15]; add.s32 %r50, %r48, %r3; shl.b32 %r51, %r50, 2; mov.u32 %r52, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r53, %r52, %r51; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r53+(0+i*400)], %f(48+i*1); ld.global.nc.f32 %f(49+i*1), [%rd15+(400+i*400)]; st.shared.f32 [%r53+6000], %f63; add.s32 %r69, %r69, 16; setp.ne.s32 %p18, %r69, 512; @%p18 bra $L__BB0_28; bar.sync 0; shl.b32 %r55, %r3, 11; add.s32 %r57, %r52, %r55; add.s32 %r19, %r57, 4; mov.f32 %f159, 0f00000000; mov.u32 %r70, 0; for.size.1 i in range(0, 12, 1): mov.f32 %f(160+i*1), %f159; $L__BB0_30: mov.u32 %r73, 4; shl.b32 %r59, %r70, 6; add.s32 %r72, %r19, %r59; mad.lo.s32 %r61, %r70, 832, %r37; add.s32 %r71, %r61, 52; $L__BB0_31: for.size.27 i in range(0, 2, 1): ld.shared.f32 %f(77+i*27), [%r72+(-4+i*4)]; ld.shared.f32 %f(78+i*27), [%r71+(-52+i*52)]; fma.rn.f32 %f(79+i*80), %f(78+i*27), %f(77+i*27), %f(159+i*-80); ld.shared.f32 %f(80+i*26), [%r71+(-48+i*52)]; fma.rn.f32 %f(81+i*79), %f(77+i*27), %f(80+i*26), %f(160+i*-79); ld.shared.f32 %f(82+i*25), [%r71+(-44+i*52)]; fma.rn.f32 %f(83+i*78), %f(82+i*25), %f(77+i*27), %f(161+i*-78); ld.shared.f32 %f(84+i*24), [%r71+(-40+i*52)]; fma.rn.f32 %f(85+i*77), %f(77+i*27), %f(84+i*24), %f(162+i*-77); ld.shared.f32 %f(86+i*23), [%r71+(-36+i*52)]; fma.rn.f32 %f(87+i*76), %f(86+i*23), %f(77+i*27), %f(163+i*-76); ld.shared.f32 %f(88+i*22), [%r71+(-32+i*52)]; fma.rn.f32 %f(89+i*75), %f(77+i*27), %f(88+i*22), %f(164+i*-75); ld.shared.f32 %f(90+i*21), [%r71+(-28+i*52)]; fma.rn.f32 %f(91+i*74), %f(90+i*21), %f(77+i*27), %f(165+i*-74); ld.shared.f32 %f(92+i*20), [%r71+(-24+i*52)]; fma.rn.f32 %f(93+i*73), %f(77+i*27), %f(92+i*20), %f(166+i*-73); ld.shared.f32 %f(94+i*19), [%r71+(-20+i*52)]; fma.rn.f32 %f(95+i*72), %f(94+i*19), %f(77+i*27), %f(167+i*-72); ld.shared.f32 %f(96+i*18), [%r71+(-16+i*52)]; fma.rn.f32 %f(97+i*71), %f(77+i*27), %f(96+i*18), %f(168+i*-71); ld.shared.f32 %f(98+i*17), [%r71+(-12+i*52)]; fma.rn.f32 %f(99+i*70), %f(98+i*17), %f(77+i*27), %f(169+i*-70); ld.shared.f32 %f(100+i*16), [%r71+(-8+i*52)]; fma.rn.f32 %f(101+i*69), %f(77+i*27), %f(100+i*16), %f(170+i*-69); ld.shared.f32 %f(102+i*15), [%r71+(-4+i*52)]; fma.rn.f32 %f(103+i*68), %f(102+i*15), %f(77+i*27), %f(171+i*-68); for.size.1 i in range(0, 2, 1): add.s32 %r(72+i*-1), %r(72+i*-1), (8+i*96); add.s32 %r73, %r73, 8; setp.ne.s32 %p19, %r73, 68; @%p19 bra $L__BB0_31; add.s32 %r70, %r70, 1; setp.lt.u32 %p20, %r70, 32; @%p20 bra $L__BB0_30; mad.lo.s32 %r62, %r34, 100, %r3; mul.wide.s32 %rd16, %r62, 4; add.s64 %rd17, %rd6, %rd16; ld.global.nc.f32 %f118, [%rd17]; add.f32 %f119, %f159, %f118; mov.f32 %f120, 0f00000000; max.f32 %f121, %f119, %f120; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(63+i*1), %r(3+i*31), (169+i*16731), %r(2+i*61); mul.wide.s32 %rd18, %r64, 4; add.s64 %rd19, %rd1, %rd18; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd19+(0+i*52)], %f(121+i*2); add.f32 %f(122+i*2), %f(160+i*1), %f118; max.f32 %f(123+i*2), %f(122+i*2), %f120; st.global.f32 [%rd19+624], %f145; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[6656]; __shared__ float placeholder_shared[51200]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 17; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4 < 6656) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 1664) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5200 + (int)threadIdx.x * 52 + (int)blockIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; } } if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4 < 6655) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 1664) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5200 + (int)threadIdx.x * 52 + (int)blockIdx.x % 13; pad_temp_shared[index0 + 1] = placeholder[index1 + 13]; } } if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4 < 6654) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 1664) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5200 + (int)threadIdx.x * 52 + (int)blockIdx.x % 13; pad_temp_shared[index0 + 2] = placeholder[index1 + 26]; } } if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4 < 6653) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 1664) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 400 + (int)threadIdx.x * 4; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5200 + (int)threadIdx.x * 52 + (int)blockIdx.x % 13; pad_temp_shared[index0 + 3] = placeholder[index1 + 39]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 512; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; int index3 = (int)blockIdx.x / 13 * 51200 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 16; ++rc_inner) { int index4 = rc_outer_inner * 208 + rc_inner * 13; int index5 = (int)threadIdx.x * 512 + rc_outer_inner * 16 + rc_inner; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4 + 1] * placeholder_shared[index5]); compute[2] = compute[2] + (pad_temp_shared[index4 + 2] * placeholder_shared[index5]); compute[3] = compute[3] + (pad_temp_shared[index4 + 3] * placeholder_shared[index5]); compute[4] = compute[4] + (pad_temp_shared[index4 + 4] * placeholder_shared[index5]); compute[5] = compute[5] + (pad_temp_shared[index4 + 5] * placeholder_shared[index5]); compute[6] = compute[6] + (pad_temp_shared[index4 + 6] * placeholder_shared[index5]); compute[7] = compute[7] + (pad_temp_shared[index4 + 7] * placeholder_shared[index5]); compute[8] = compute[8] + (pad_temp_shared[index4 + 8] * placeholder_shared[index5]); compute[9] = compute[9] + (pad_temp_shared[index4 + 9] * placeholder_shared[index5]); compute[10] = compute[10] + (pad_temp_shared[index4 + 10] * placeholder_shared[index5]); compute[11] = compute[11] + (pad_temp_shared[index4 + 11] * placeholder_shared[index5]); compute[12] = compute[12] + (pad_temp_shared[index4 + 12] * placeholder_shared[index5]); } } int index6 = (int)blockIdx.x / 13 * 16900 + (int)threadIdx.x * 169 + (int)blockIdx.x % 13; int index7 = (int)blockIdx.x / 13 * 100 + (int)threadIdx.x; T_relu[index6] = max(compute[0] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 13] = max(compute[1] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 26] = max(compute[2] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 39] = max(compute[3] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 52] = max(compute[4] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 65] = max(compute[5] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 78] = max(compute[6] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 91] = max(compute[7] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 104] = max(compute[8] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 117] = max(compute[9] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 130] = max(compute[10] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 143] = max(compute[11] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 156] = max(compute[12] + placeholder2[index7], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp9xupp66c
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<177>; .reg .b32 %r<36>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[2704]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[3200]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; shl.b32 %r12, %r1, 2; mov.u32 %r13, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r13, %r12; shl.b32 %r14, %r1, 4; mov.u32 %r15, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r15, %r14; mul.hi.s32 %r16, %r1, 1321528399; shr.u32 %r17, %r16, 31; shr.s32 %r18, %r16, 2; add.s32 %r6, %r18, %r17; mul.lo.s32 %r19, %r6, 13; sub.s32 %r4, %r1, %r19; for.size.2 i in range(0, 2, 1): shl.b32 %r(20+i*1), %r(4+i*2), (2+i*2); add.s32 %r(5+i*2), %r(13+i*2), %r(20+i*1); mov.u32 %r8, %ctaid.x; shl.b32 %r22, %r1, 9; mad.lo.s32 %r23, %r8, 102400, %r22; mul.wide.s32 %rd13, %r23, 4; add.s64 %rd22, %rd12, %rd13; cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r1, 4; add.s64 %rd21, %rd14, %rd15; mov.u32 %r35, 0; mov.f32 %f164, 0f00000000; for.size.1 i in range(0, 12, 1): mov.f32 %f(165+i*1), %f164; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 675; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f40, [%rd21]; st.shared.f32 [%r2], %f40; $L__BB0_3: setp.gt.s32 %p2, %r1, 199; @%p2 bra $L__BB0_5; ld.global.nc.v4.u32 {%r24, %r25, %r26, %r27}, [%rd22]; st.shared.v4.u32 [%r3], {%r24, %r25, %r26, %r27}; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(41+i*1), [%r(7+i*-2)+0]; for.size.2 i in range(0, 12, 1): fma.rn.f32 %f(43+i*2), %f(42+i*2), %f41, %f(176+i*-1); ld.shared.f32 %f(44+i*2), [%r5+(52+i*52)]; for.size.27 i in range(0, 2, 1): fma.rn.f32 %f(67+i*27), %f(66+i*27), %f(41+i*27), %f(164+i*-97); ld.shared.f32 %f(68+i*27), [%r7+(4+i*4)]; ld.shared.f32 %f(69+i*27), [%r5+(676+i*676)]; fma.rn.f32 %f(70+i*27), %f(69+i*27), %f(68+i*27), %f(43+i*27); ld.shared.f32 %f(71+i*27), [%r5+(728+i*676)]; fma.rn.f32 %f(72+i*27), %f(71+i*27), %f(68+i*27), %f(45+i*27); ld.shared.f32 %f(73+i*27), [%r5+(780+i*676)]; fma.rn.f32 %f(74+i*27), %f(73+i*27), %f(68+i*27), %f(47+i*27); ld.shared.f32 %f(75+i*27), [%r5+(832+i*676)]; fma.rn.f32 %f(76+i*27), %f(75+i*27), %f(68+i*27), %f(49+i*27); ld.shared.f32 %f(77+i*27), [%r5+(884+i*676)]; fma.rn.f32 %f(78+i*27), %f(77+i*27), %f(68+i*27), %f(51+i*27); ld.shared.f32 %f(79+i*27), [%r5+(936+i*676)]; fma.rn.f32 %f(80+i*27), %f(79+i*27), %f(68+i*27), %f(53+i*27); ld.shared.f32 %f(81+i*27), [%r5+(988+i*676)]; fma.rn.f32 %f(82+i*27), %f(81+i*27), %f(68+i*27), %f(55+i*27); ld.shared.f32 %f(83+i*27), [%r5+(1040+i*676)]; fma.rn.f32 %f(84+i*27), %f(83+i*27), %f(68+i*27), %f(57+i*27); ld.shared.f32 %f(85+i*27), [%r5+(1092+i*676)]; fma.rn.f32 %f(86+i*27), %f(85+i*27), %f(68+i*27), %f(59+i*27); ld.shared.f32 %f(87+i*27), [%r5+(1144+i*676)]; fma.rn.f32 %f(88+i*27), %f(87+i*27), %f(68+i*27), %f(61+i*27); ld.shared.f32 %f(89+i*27), [%r5+(1196+i*676)]; fma.rn.f32 %f(90+i*27), %f(89+i*27), %f(68+i*27), %f(63+i*27); ld.shared.f32 %f(91+i*27), [%r5+(1248+i*676)]; fma.rn.f32 %f(92+i*27), %f(91+i*27), %f(68+i*27), %f(65+i*27); ld.shared.f32 %f(93+i*27), [%r5+(1300+i*676)]; fma.rn.f32 %f121, %f120, %f95, %f94; ld.shared.f32 %f122, [%r7+12]; for.size.2 i in range(0, 13, 1): ld.shared.f32 %f(123+i*1), [%r5+(2028+i*52)]; fma.rn.f32 %f(176+i*-1), %f(123+i*1), %f122, %f(97+i*2); for.size.1 i in range(0, 2, 1): add.s64 %rd(22+i*-1), %rd(22+i*-1), (16+i*2688); add.s32 %r35, %r35, 1; setp.ne.s32 %p3, %r35, 128; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd16, %rd8; mad.lo.s32 %r32, %r8, 200, %r6; mul.wide.s32 %rd17, %r32, 4; add.s64 %rd18, %rd16, %rd17; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(33+i*1), %r(8+i*-2), (33800+i*-33631), %r(4+i*29); ld.global.nc.f32 %f136, [%rd18]; add.f32 %f137, %f176, %f136; mov.f32 %f138, 0f00000000; max.f32 %f139, %f137, %f138; mul.wide.s32 %rd19, %r34, 4; add.s64 %rd20, %rd1, %rd19; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd20+(0+i*52)], %f(139+i*2); add.f32 %f(140+i*2), %f(175+i*-1), %f136; max.f32 %f(141+i*2), %f(140+i*2), %f138; st.global.f32 [%rd20+624], %f163; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[676]; __shared__ float placeholder_shared[800]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 676 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 200) { ((float4*)(placeholder_shared + (int)threadIdx.x * 4))[0] = ((float4*)(placeholder1 + (int)blockIdx.x * 102400 + (int)threadIdx.x * 512 + rc_outer_outer * 4))[0]; } __syncthreads(); int index0 = (int)threadIdx.x; int index2 = (int)threadIdx.x / 13 * 4; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index2]; int index3 = (int)threadIdx.x % 13 + 13; compute[1] = compute[1] + pad_temp_shared[index3] * placeholder_shared[index2]; int index4 = (int)threadIdx.x % 13 + 26; compute[2] = compute[2] + pad_temp_shared[index4] * placeholder_shared[index2]; int index5 = (int)threadIdx.x % 13 + 39; compute[3] = compute[3] + pad_temp_shared[index5] * placeholder_shared[index2]; int index6 = (int)threadIdx.x % 13 + 52; compute[4] = compute[4] + pad_temp_shared[index6] * placeholder_shared[index2]; int index7 = (int)threadIdx.x % 13 + 65; compute[5] = compute[5] + pad_temp_shared[index7] * placeholder_shared[index2]; int index8 = (int)threadIdx.x % 13 + 78; compute[6] = compute[6] + pad_temp_shared[index8] * placeholder_shared[index2]; int index9 = (int)threadIdx.x % 13 + 91; compute[7] = compute[7] + pad_temp_shared[index9] * placeholder_shared[index2]; int index10 = (int)threadIdx.x % 13 + 104; compute[8] = compute[8] + pad_temp_shared[index10] * placeholder_shared[index2]; int index11 = (int)threadIdx.x % 13 + 117; compute[9] = compute[9] + pad_temp_shared[index11] * placeholder_shared[index2]; int index12 = (int)threadIdx.x % 13 + 130; compute[10] = compute[10] + pad_temp_shared[index12] * placeholder_shared[index2]; int index13 = (int)threadIdx.x % 13 + 143; compute[11] = compute[11] + pad_temp_shared[index13] * placeholder_shared[index2]; int index14 = (int)threadIdx.x % 13 + 156; compute[12] = compute[12] + pad_temp_shared[index14] * placeholder_shared[index2]; int index15 = (int)threadIdx.x % 13 + 169; compute[0] = compute[0] + (pad_temp_shared[index15] * placeholder_shared[index2 + 1]); int index16 = (int)threadIdx.x % 13 + 182; compute[1] = compute[1] + (pad_temp_shared[index16] * placeholder_shared[index2 + 1]); int index17 = (int)threadIdx.x % 13 + 195; compute[2] = compute[2] + (pad_temp_shared[index17] * placeholder_shared[index2 + 1]); int index18 = (int)threadIdx.x % 13 + 208; compute[3] = compute[3] + (pad_temp_shared[index18] * placeholder_shared[index2 + 1]); int index19 = (int)threadIdx.x % 13 + 221; compute[4] = compute[4] + (pad_temp_shared[index19] * placeholder_shared[index2 + 1]); int index20 = (int)threadIdx.x % 13 + 234; compute[5] = compute[5] + (pad_temp_shared[index20] * placeholder_shared[index2 + 1]); int index21 = (int)threadIdx.x % 13 + 247; compute[6] = compute[6] + (pad_temp_shared[index21] * placeholder_shared[index2 + 1]); int index22 = (int)threadIdx.x % 13 + 260; compute[7] = compute[7] + (pad_temp_shared[index22] * placeholder_shared[index2 + 1]); int index23 = (int)threadIdx.x % 13 + 273; compute[8] = compute[8] + (pad_temp_shared[index23] * placeholder_shared[index2 + 1]); int index24 = (int)threadIdx.x % 13 + 286; compute[9] = compute[9] + (pad_temp_shared[index24] * placeholder_shared[index2 + 1]); int index25 = (int)threadIdx.x % 13 + 299; compute[10] = compute[10] + (pad_temp_shared[index25] * placeholder_shared[index2 + 1]); int index26 = (int)threadIdx.x % 13 + 312; compute[11] = compute[11] + (pad_temp_shared[index26] * placeholder_shared[index2 + 1]); int index27 = (int)threadIdx.x % 13 + 325; compute[12] = compute[12] + (pad_temp_shared[index27] * placeholder_shared[index2 + 1]); int index28 = (int)threadIdx.x % 13 + 338; compute[0] = compute[0] + (pad_temp_shared[index28] * placeholder_shared[index2 + 2]); int index29 = (int)threadIdx.x % 13 + 351; compute[1] = compute[1] + (pad_temp_shared[index29] * placeholder_shared[index2 + 2]); int index30 = (int)threadIdx.x % 13 + 364; compute[2] = compute[2] + (pad_temp_shared[index30] * placeholder_shared[index2 + 2]); int index31 = (int)threadIdx.x % 13 + 377; compute[3] = compute[3] + (pad_temp_shared[index31] * placeholder_shared[index2 + 2]); int index32 = (int)threadIdx.x % 13 + 390; compute[4] = compute[4] + (pad_temp_shared[index32] * placeholder_shared[index2 + 2]); int index33 = (int)threadIdx.x % 13 + 403; compute[5] = compute[5] + (pad_temp_shared[index33] * placeholder_shared[index2 + 2]); int index34 = (int)threadIdx.x % 13 + 416; compute[6] = compute[6] + (pad_temp_shared[index34] * placeholder_shared[index2 + 2]); int index35 = (int)threadIdx.x % 13 + 429; compute[7] = compute[7] + (pad_temp_shared[index35] * placeholder_shared[index2 + 2]); int index36 = (int)threadIdx.x % 13 + 442; compute[8] = compute[8] + (pad_temp_shared[index36] * placeholder_shared[index2 + 2]); int index37 = (int)threadIdx.x % 13 + 455; compute[9] = compute[9] + (pad_temp_shared[index37] * placeholder_shared[index2 + 2]); int index38 = (int)threadIdx.x % 13 + 468; compute[10] = compute[10] + (pad_temp_shared[index38] * placeholder_shared[index2 + 2]); int index39 = (int)threadIdx.x % 13 + 481; compute[11] = compute[11] + (pad_temp_shared[index39] * placeholder_shared[index2 + 2]); int index40 = (int)threadIdx.x % 13 + 494; compute[12] = compute[12] + (pad_temp_shared[index40] * placeholder_shared[index2 + 2]); int index41 = (int)threadIdx.x % 13 + 507; compute[0] = compute[0] + (pad_temp_shared[index41] * placeholder_shared[index2 + 3]); int index42 = (int)threadIdx.x % 13 + 520; compute[1] = compute[1] + (pad_temp_shared[index42] * placeholder_shared[index2 + 3]); int index43 = (int)threadIdx.x % 13 + 533; compute[2] = compute[2] + (pad_temp_shared[index43] * placeholder_shared[index2 + 3]); int index44 = (int)threadIdx.x % 13 + 546; compute[3] = compute[3] + (pad_temp_shared[index44] * placeholder_shared[index2 + 3]); int index45 = (int)threadIdx.x % 13 + 559; compute[4] = compute[4] + (pad_temp_shared[index45] * placeholder_shared[index2 + 3]); int index46 = (int)threadIdx.x % 13 + 572; compute[5] = compute[5] + (pad_temp_shared[index46] * placeholder_shared[index2 + 3]); int index47 = (int)threadIdx.x % 13 + 585; compute[6] = compute[6] + (pad_temp_shared[index47] * placeholder_shared[index2 + 3]); int index48 = (int)threadIdx.x % 13 + 598; compute[7] = compute[7] + (pad_temp_shared[index48] * placeholder_shared[index2 + 3]); int index49 = (int)threadIdx.x % 13 + 611; compute[8] = compute[8] + (pad_temp_shared[index49] * placeholder_shared[index2 + 3]); int index50 = (int)threadIdx.x % 13 + 624; compute[9] = compute[9] + (pad_temp_shared[index50] * placeholder_shared[index2 + 3]); int index51 = (int)threadIdx.x % 13 + 637; compute[10] = compute[10] + (pad_temp_shared[index51] * placeholder_shared[index2 + 3]); int index52 = (int)threadIdx.x % 13 + 650; compute[11] = compute[11] + (pad_temp_shared[index52] * placeholder_shared[index2 + 3]); int index53 = (int)threadIdx.x % 13 + 663; compute[12] = compute[12] + (pad_temp_shared[index53] * placeholder_shared[index2 + 3]); } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index54 = (int)blockIdx.x * 33800 + (int)threadIdx.x / 13 * 169 + ax2_inner * 13 + (int)threadIdx.x % 13; int index55 = ax2_inner; int index56 = (int)blockIdx.x * 200 + (int)threadIdx.x / 13; T_relu[index54] = max(compute[index55] + placeholder2[index56], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpq_3z6ail
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[67600]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<12>; .reg .f32 %f<1615>; .reg .b32 %r<148>; .reg .b64 %rd<51>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd13, [default_function_kernel0_param_0]; ld.param.u64 %rd14, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.u32 %r28, 0; mov.u32 %r139, %r28; $L__BB0_1: mul.lo.s32 %r29, %r139, 169; mul.wide.s32 %rd18, %r29, 4; add.s64 %rd19, %rd1, %rd18; for.size.2 i in range(0, 169, 1): for.size.1 j in range(0, 2, 1): st.local.u32 [%rd19+(0+i*4+j*33800)], %r28; add.s32 %r139, %r139, 1; setp.ne.s32 %p1, %r139, 50; @%p1 bra $L__BB0_1; mov.u32 %r31, 0; cvta.to.global.u64 %rd20, %rd13; cvta.to.global.u64 %rd22, %rd14; mov.u32 %r140, %r31; $L__BB0_3: bar.sync 0; mov.u32 %r33, %tid.x; mad.lo.s32 %r4, %r140, 5408, %r33; mov.u32 %r141, %r31; $L__BB0_4: mov.u32 %r126, %tid.x; mul.lo.s32 %r6, %r141, 5; add.s32 %r35, %r6, %r126; setp.gt.s32 %p2, %r35, 5407; add.s32 %r36, %r4, %r6; mul.wide.s32 %rd21, %r36, 4; add.s64 %rd2, %rd20, %rd21; shl.b32 %r37, %r35, 2; mov.u32 %r38, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r7, %r38, %r37; @%p2 bra $L__BB0_6; ld.global.nc.f32 %f1017, [%rd2]; st.shared.f32 [%r7], %f1017; $L__BB0_6: add.s32 %r41, %r35, 5; setp.gt.s32 %p3, %r41, 5407; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f1018, [%rd2+20]; st.shared.f32 [%r7+20], %f1018; $L__BB0_8: add.s32 %r141, %r141, 2; setp.ne.s32 %p4, %r141, 1082; @%p4 bra $L__BB0_4; shl.b32 %r43, %r140, 5; mov.u32 %r44, %ctaid.x; mad.lo.s32 %r9, %r44, 256000, %r43; mov.u32 %r142, 0; $L__BB0_10: mov.u32 %r127, %tid.x; mad.lo.s32 %r46, %r142, 5, %r127; shl.b32 %r47, %r46, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(48+i*1), %r(47+i*-1), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(50+i*1), %r(9+i*41), %r(49+i*-1); mul.wide.s32 %rd23, %r51, 4; add.s64 %rd24, %rd22, %rd23; ld.global.nc.f32 %f1019, [%rd24]; shl.b32 %r52, %r46, 2; mov.u32 %r53, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r54, %r53, %r52; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r54+(0+i*20)], %f(1019+i*1); add.s32 %r(55+i*6), %r46, (5+i*5); shl.b32 %r(56+i*6), %r(55+i*6), 4; and.b32 %r(57+i*6), %r(56+i*6), -512; and.b32 %r(58+i*6), %r(55+i*6), 31; add.s32 %r(59+i*6), %r9, %r(58+i*6); add.s32 %r(60+i*6), %r(59+i*6), %r(57+i*6); mul.wide.s32 %rd(25+i*2), %r(60+i*6), 4; add.s64 %rd(26+i*2), %rd22, %rd(25+i*2); ld.global.nc.f32 %f(1020+i*1), [%rd(26+i*2)+0]; st.shared.f32 [%r54+140], %f1026; add.s32 %r142, %r142, 8; setp.ne.s32 %p5, %r142, 3200; @%p5 bra $L__BB0_10; bar.sync 0; mov.u32 %r143, 0; $L__BB0_12: mov.u32 %r144, 0; $L__BB0_13: mov.u32 %r131, %tid.x; mul.lo.s32 %r103, %r144, 169; mul.wide.s32 %rd39, %r103, 4; add.s64 %rd3, %rd1, %rd39; for.size.2 i in range(0, 169, 1): for.size.1 j in range(0, 2, 1): ld.local.f32 %f(1614+i*-2+j*-1), [%rd3+(0+i*4+j*33800)]; mov.u32 %r145, 0; $L__BB0_14: shl.b32 %r137, %r143, 3; mov.u32 %r136, %tid.x; shl.b32 %r135, %r144, 5; mad.lo.s32 %r134, %r136, 1600, %r137; add.s32 %r133, %r134, %r135; mov.u32 %r123, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r122, _ZZ24default_function_kernel0E18placeholder_shared; mul.lo.s32 %r121, %r143, 1352; mad.lo.s32 %r104, %r145, 169, %r121; for.size.2 i in range(0, 2, 1): add.s32 %r(105+i*3), %r(133+i*-11), %r(145+i*-39); shl.b32 %r(106+i*3), %r(105+i*-1), 2; add.s32 %r111, %r123, %r109; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1027+i*1), [%r(111+i*-3)+0]; fma.rn.f32 %f1614, %f1027, %f1028, %f1614; ld.shared.f32 %f1029, [%r108+32000]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(1613+i*-2), %f(1027+i*3), %f1029, %f(1613+i*-2); ld.shared.f32 %f(1030+i*1), [%r111+(4+i*4)]; fma.rn.f32 %f(1612+i*-2), %f(1030+i*1), %f1028, %f(1612+i*-2); for.size.3 i in range(0, 166, 1): fma.rn.f32 %f(1609+i*-2), %f(1031+i*1), %f1029, %f(1609+i*-2); ld.shared.f32 %f(1032+i*1), [%r111+(12+i*4)]; fma.rn.f32 %f(1608+i*-2), %f(1032+i*1), %f1028, %f(1608+i*-2); fma.rn.f32 %f1277, %f1197, %f1029, %f1277; add.s32 %r145, %r145, 1; setp.ne.s32 %p6, %r145, 8; @%p6 bra $L__BB0_14; mul.lo.s32 %r138, %r144, 169; mul.wide.s32 %rd48, %r138, 4; add.s64 %rd47, %rd1, %rd48; for.size.2 i in range(0, 169, 1): for.size.1 j in range(0, 2, 1): st.local.f32 [%rd47+(0+i*4+j*33800)], %f(1614+i*-2+j*-1); add.s32 %r144, %r144, 1; setp.lt.u32 %p7, %r144, 50; @%p7 bra $L__BB0_13; add.s32 %r143, %r143, 1; setp.lt.u32 %p8, %r143, 4; @%p8 bra $L__BB0_12; add.s32 %r140, %r140, 1; setp.lt.u32 %p9, %r140, 16; @%p9 bra $L__BB0_3; ld.param.u64 %rd46, [default_function_kernel0_param_3]; ld.param.u64 %rd45, [default_function_kernel0_param_2]; mov.u32 %r125, %ctaid.x; mov.u32 %r124, %tid.x; mul.lo.s32 %r115, %r124, 50; mad.lo.s32 %r22, %r125, 500, %r115; add.s64 %rd4, %rd1, 33848; cvta.to.global.u64 %rd40, %rd45; add.s64 %rd5, %rd40, 169048; mul.lo.s32 %r116, %r124, 8450; mad.lo.s32 %r23, %r125, 84500, %r116; cvta.to.global.u64 %rd6, %rd46; mov.u32 %r112, 0; mov.u32 %r146, %r112; $L__BB0_19: add.s32 %r118, %r22, %r146; mul.wide.s32 %rd41, %r118, 4; add.s64 %rd42, %rd6, %rd41; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(1015+i*1), [%rd42+(0+i*1000)]; mul.lo.s32 %r119, %r146, 169; mul.wide.s32 %rd43, %r119, 4; add.s64 %rd50, %rd4, %rd43; add.s32 %r120, %r23, %r119; mul.wide.s32 %rd44, %r120, 4; add.s64 %rd49, %rd5, %rd44; mov.u32 %r147, %r112; $L__BB0_20: ld.local.f32 %f1198, [%rd50+-33848]; add.f32 %f1199, %f1198, %f1015; mov.f32 %f1200, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(1201+i*3), %f(1199+i*4), %f1200; st.global.f32 [%rd49+(-169048+i*169000)], %f(1201+i*3); ld.local.f32 %f(1202+i*3), [%rd50+(-48+i*-33796)]; add.f32 %f(1203+i*3), %f(1202+i*3), %f(1016+i*-1); for.size.5 i in range(0, 11, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(1207+i*6+j*3), %f(1206+i*6+j*3), %f1200; st.global.f32 [%rd49+(-169044+i*4+j*169000)], %f(1207+i*6+j*3); ld.local.f32 %f(1208+i*6+j*3), [%rd50+(-44+i*4+j*-33796)]; add.f32 %f(1209+i*6+j*3), %f(1208+i*6+j*3), %f(1016+j*-1); max.f32 %f1273, %f1272, %f1200; st.global.f32 [%rd49+-169000], %f1273; ld.local.f32 %f1274, [%rd50]; add.f32 %f1275, %f1274, %f1016; max.f32 %f1276, %f1275, %f1200; st.global.f32 [%rd49], %f1276; for.size.1 i in range(0, 2, 1): add.s64 %rd(50+i*-1), %rd(50+i*-1), 52; add.s32 %r147, %r147, 1; setp.ne.s32 %p10, %r147, 13; @%p10 bra $L__BB0_20; add.s32 %r146, %r146, 1; setp.lt.u32 %p11, %r146, 50; @%p11 bra $L__BB0_19; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16900]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[16000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 50; ++ff_outer_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_outer_inner_init * 169 + yy_inner_init * 13; compute[index0] = 0.000000e+00f; compute[index0 + 8450] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 8451] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 8452] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 8453] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 8454] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 8455] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; compute[index0 + 8456] = 0.000000e+00f; compute[index0 + 7] = 0.000000e+00f; compute[index0 + 8457] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 8458] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 8459] = 0.000000e+00f; compute[index0 + 10] = 0.000000e+00f; compute[index0 + 8460] = 0.000000e+00f; compute[index0 + 11] = 0.000000e+00f; compute[index0 + 8461] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; compute[index0 + 8462] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 1082; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x < 5408) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 5 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 3200; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 256000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 5 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 50; ++ff_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = ff_outer_inner * 169 + yy_inner * 13; int index6 = rc_outer_inner * 1352 + rc_inner * 169 + yy_inner * 13; int index7 = (int)threadIdx.x * 1600 + ff_outer_inner * 32 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 8450] = compute[index5 + 8450] + (pad_temp_shared[index6] * placeholder_shared[index7 + 8000]); compute[index5 + 1] = compute[index5 + 1] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 8451] = compute[index5 + 8451] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 8000]); compute[index5 + 2] = compute[index5 + 2] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 8452] = compute[index5 + 8452] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 8000]); compute[index5 + 3] = compute[index5 + 3] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 8453] = compute[index5 + 8453] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 8000]); compute[index5 + 4] = compute[index5 + 4] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 8454] = compute[index5 + 8454] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7 + 8000]); compute[index5 + 5] = compute[index5 + 5] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 8455] = compute[index5 + 8455] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7 + 8000]); compute[index5 + 6] = compute[index5 + 6] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 8456] = compute[index5 + 8456] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7 + 8000]); compute[index5 + 7] = compute[index5 + 7] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 8457] = compute[index5 + 8457] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7 + 8000]); compute[index5 + 8] = compute[index5 + 8] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); compute[index5 + 8458] = compute[index5 + 8458] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7 + 8000]); compute[index5 + 9] = compute[index5 + 9] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 8459] = compute[index5 + 8459] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7 + 8000]); compute[index5 + 10] = compute[index5 + 10] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7]); compute[index5 + 8460] = compute[index5 + 8460] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7 + 8000]); compute[index5 + 11] = compute[index5 + 11] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7]); compute[index5 + 8461] = compute[index5 + 8461] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7 + 8000]); compute[index5 + 12] = compute[index5 + 12] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7]); compute[index5 + 8462] = compute[index5 + 8462] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7 + 8000]); } } } } } for (int ax1_inner = 0; ax1_inner < 50; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x * 84500 + (int)threadIdx.x * 8450 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index9 = ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x * 500 + (int)threadIdx.x * 50 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 42250] = max(compute[index9 + 8450] + placeholder2[index10 + 250], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmppnxkj_ut
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<179>; .reg .b32 %r<87>; .reg .b64 %rd<31>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[64]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[32000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; mov.u32 %r20, %ctaid.x; mul.hi.s32 %r21, %r20, 1626496491; shr.u32 %r22, %r21, 31; shr.s32 %r23, %r21, 6; add.s32 %r5, %r23, %r22; mul.lo.s32 %r24, %r5, 169; sub.s32 %r2, %r20, %r24; mad.lo.s32 %r3, %r1, 169, %r2; shl.b32 %r25, %r1, 2; mov.u32 %r26, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r26, %r25; mul.lo.s32 %r6, %r5, 256000; shl.b32 %r27, %r1, 7; mov.u32 %r28, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r28, %r27; add.s32 %r8, %r1, 875; add.s32 %r9, %r28, %r25; mov.u32 %r83, 0; mov.f32 %f175, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(176+i*1), %f175; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 15; @%p1 bra $L__BB0_3; mad.lo.s32 %r29, %r83, 2704, %r3; mul.wide.s32 %rd9, %r29, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f13, [%rd10]; st.shared.f32 [%r4], %f13; $L__BB0_3: shl.b32 %r31, %r83, 4; add.s32 %r11, %r31, %r6; mov.u32 %r86, 2000; mov.u32 %r84, %r1; mov.u32 %r85, %r8; $L__BB0_4: shl.b32 %r32, %r84, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(33+i*1), %r(32+i*52), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(35+i*1), %r(11+i*24), %r(34+i*-1); mul.wide.s32 %rd11, %r36, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f14, [%rd12]; add.s32 %r37, %r9, %r86; for.size.10 i in range(0, 4, 1): st.shared.f32 [%r37+(-2000+i*500)], %f(14+i*1); add.s32 %r(38+i*6), %r84, (125+i*125); shl.b32 %r(39+i*6), %r(38+i*6), 5; and.b32 %r(40+i*6), %r(39+i*6), -512; and.b32 %r(41+i*6), %r(38+i*6), 15; add.s32 %r(42+i*6), %r11, %r(41+i*6); add.s32 %r(43+i*6), %r(42+i*6), %r(40+i*6); mul.wide.s32 %rd(13+i*2), %r(43+i*6), 4; add.s64 %rd(14+i*2), %rd1, %rd(13+i*2); ld.global.nc.f32 %f(15+i*1), [%rd(14+i*2)+0]; for.size.10 i in range(0, 2, 1): st.shared.f32 [%r37+(0+i*500)], %f(18+i*1); add.s32 %r(62+i*6), %r85, (-250+i*125); shl.b32 %r(63+i*6), %r(62+i*6), 5; and.b32 %r(64+i*6), %r(63+i*6), -512; and.b32 %r(65+i*6), %r(62+i*6), 15; add.s32 %r(66+i*6), %r11, %r(65+i*6); add.s32 %r(67+i*6), %r(66+i*6), %r(64+i*6); mul.wide.s32 %rd(21+i*2), %r(67+i*6), 4; add.s64 %rd(22+i*2), %rd1, %rd(21+i*2); ld.global.nc.f32 %f(19+i*1), [%rd(22+i*2)+0]; st.shared.f32 [%r37+1000], %f20; shl.b32 %r74, %r85, 5; for.size.1 i in range(0, 2, 1): and.b32 %r(75+i*1), %r(74+i*11), (-512+i*527); for.size.1 i in range(0, 2, 1): add.s32 %r(77+i*1), %r(11+i*66), %r(76+i*-1); mul.wide.s32 %rd25, %r78, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f21, [%rd26]; st.shared.f32 [%r37+1500], %f21; for.size.1 i in range(0, 2, 1): add.s32 %r(85+i*-1), %r(85+i*-1), 1000; add.s32 %r86, %r86, 4000; setp.ne.s32 %p2, %r86, 34000; @%p2 bra $L__BB0_4; bar.sync 0; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(22+i*5), [%r7+(0+i*4)]; ld.shared.f32 %f(23+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(0+i*4)]; fma.rn.f32 %f(24+i*5), %f(23+i*5), %f(22+i*5), %f(178+i*-154); ld.shared.f32 %f(25+i*5), [%r7+(16000+i*4)]; fma.rn.f32 %f(26+i*5), %f(23+i*5), %f(25+i*5), %f(176+i*-150); for.size.5 i in range(0, 6, 1): ld.shared.f32 %f(32+i*5), [%r7+(8+i*4)]; ld.shared.f32 %f(33+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(8+i*4)]; fma.rn.f32 %f(34+i*5), %f(33+i*5), %f(32+i*5), %f(29+i*5); ld.shared.f32 %f(35+i*5), [%r7+(16008+i*4)]; fma.rn.f32 %f(36+i*5), %f(33+i*5), %f(35+i*5), %f(31+i*5); for.size.2 i in range(0, 2, 1): ld.shared.f32 %f(62+i*2), [%r7+(64+i*16000)]; fma.rn.f32 %f(63+i*2), %f23, %f(62+i*2), %f(177+i*-2); for.size.3 i in range(0, 7, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(66+i*4+j*2), [%r7+(68+i*4+j*16000)]; fma.rn.f32 %f(67+i*4+j*2), %f(28+j*5), %f(66+i*4+j*2), %f(63+i*4+j*2); for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(94+i*5), [%r7+(32+i*4)]; ld.shared.f32 %f(95+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(32+i*4)]; fma.rn.f32 %f(96+i*5), %f(95+i*5), %f(94+i*5), %f(59+i*37); ld.shared.f32 %f(97+i*5), [%r7+(16032+i*4)]; fma.rn.f32 %f(98+i*5), %f(95+i*5), %f(97+i*5), %f(61+i*37); for.size.5 i in range(0, 5, 1): ld.shared.f32 %f(104+i*5), [%r7+(40+i*4)]; ld.shared.f32 %f(105+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(40+i*4)]; fma.rn.f32 %f(106+i*5), %f(105+i*5), %f(104+i*5), %f(101+i*5); ld.shared.f32 %f(107+i*5), [%r7+(16040+i*4)]; fma.rn.f32 %f(108+i*5), %f(105+i*5), %f(107+i*5), %f(103+i*5); ld.shared.f32 %f129, [%r7+60]; ld.shared.f32 %f130, [_ZZ24default_function_kernel0E15pad_temp_shared+60]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(178+i*-2), %f130, %f(129+i*2), %f(126+i*2); ld.shared.f32 %f(131+i*1), [%r7+(16060+i*-15964)]; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(133+i*4+j*2), %f(95+j*5), %f(132+i*4+j*2), %f(91+i*42+j*2); ld.shared.f32 %f(134+i*4+j*2), [%r7+(16096+i*4+j*-15996)]; for.size.3 i in range(0, 5, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(141+i*4+j*2), %f(105+j*5), %f(140+i*4+j*2), %f(137+i*4+j*2); ld.shared.f32 %f(142+i*4+j*2), [%r7+(16104+i*4+j*-15996)]; fma.rn.f32 %f177, %f130, %f160, %f157; ld.shared.f32 %f161, [%r7+16124]; fma.rn.f32 %f175, %f130, %f161, %f159; add.s32 %r83, %r83, 1; setp.lt.u32 %p3, %r83, 32; @%p3 bra $L__BB0_1; shl.b32 %r79, %r1, 1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(80+i*1), %r(5+i*-4), (500+i*-162), %r(79+i*-77); mad.lo.s32 %r82, %r5, 84500, %r81; mul.wide.s32 %rd27, %r80, 4; add.s64 %rd28, %rd4, %rd27; ld.global.nc.f32 %f162, [%rd28]; add.f32 %f163, %f178, %f162; mov.f32 %f164, 0f00000000; max.f32 %f165, %f163, %f164; mul.wide.s32 %rd29, %r82, 4; add.s64 %rd30, %rd3, %rd29; for.size.4 i in range(0, 2, 1): st.global.f32 [%rd30+(0+i*169000)], %f(165+i*3); ld.global.nc.f32 %f(166+i*3), [%rd28+(1000+i*-996)]; add.f32 %f(167+i*3), %f(176+i*1), %f(166+i*3); max.f32 %f(168+i*3), %f(167+i*3), %f164; st.global.f32 [%rd30+676], %f171; ld.global.nc.f32 %f172, [%rd28+1004]; add.f32 %f173, %f175, %f172; max.f32 %f174, %f173, %f164; st.global.f32 [%rd30+169676], %f174; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[16]; __shared__ float placeholder_shared[8000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { int index0 = ff_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 16) { int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + (int)threadIdx.x * 169 + (int)blockIdx.x % 169; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 64; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 125 + (int)threadIdx.x; int index4 = (int)blockIdx.x / 169 * 256000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 125 + (int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 125 + (int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { int index5 = ff_outer_inner; int index6 = rc_outer_inner * 8 + rc_inner; int index7 = (int)threadIdx.x * 32 + ff_outer_inner * 16 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 2] = compute[index5 + 2] + (pad_temp_shared[index6] * placeholder_shared[index7 + 4000]); } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index8 = (int)blockIdx.x / 169 * 84500 + (int)threadIdx.x * 338 + ax1_inner * 169 + (int)blockIdx.x % 169; int index9 = ax1_inner; int index10 = (int)blockIdx.x / 169 * 500 + (int)threadIdx.x * 2 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 42250] = max(compute[index9 + 2] + placeholder2[index10 + 250], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpnqqnl2o0
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<336>; .reg .b32 %r<31>; .reg .b64 %rd<20>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[43264]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; mov.u32 %r1, %tid.x; shl.b32 %r11, %r1, 2; mov.u32 %r12, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r12, %r11; shr.u32 %r13, %r1, 6; and.b32 %r14, %r1, 63; bfi.b32 %r3, %r13, %r14, 9, 23; mov.u32 %r15, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r15, %r11; mul.hi.s32 %r16, %r1, 1626496491; shr.u32 %r17, %r16, 31; shr.s32 %r18, %r16, 6; add.s32 %r6, %r18, %r17; mul.lo.s32 %r19, %r6, 169; sub.s32 %r20, %r1, %r19; for.size.2 i in range(0, 2, 1): shl.b32 %r(21+i*1), %r(20+i*-14), (2+i*6); add.s32 %r(5+i*2), %r(12+i*3), %r(21+i*1); mov.f32 %f334, 0f00000000; mov.u32 %r30, 0; mov.f32 %f335, %f334; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 10815; @%p1 bra $L__BB0_3; mad.lo.s32 %r23, %r30, 10816, %r1; mul.wide.s32 %rd8, %r23, 4; add.s64 %rd9, %rd2, %rd8; ld.global.nc.f32 %f7, [%rd9]; st.shared.f32 [%r2], %f7; $L__BB0_3: setp.gt.s32 %p2, %r1, 63999; @%p2 bra $L__BB0_5; shl.b32 %r24, %r30, 6; add.s32 %r25, %r3, %r24; mul.wide.s32 %rd10, %r25, 4; add.s64 %rd11, %rd1, %rd10; ld.global.nc.f32 %f8, [%rd11]; st.shared.f32 [%r4], %f8; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(9+i*1), [%r(7+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(11+i*2), %f10, %f(9+i*3), %f(334+i*1); ld.shared.f32 %f(12+i*2), [%r7+(128000+i*-127996)]; for.size.5 i in range(0, 62, 1): ld.shared.f32 %f(15+i*5), [%r5+(676+i*676)]; fma.rn.f32 %f(16+i*5), %f(15+i*5), %f(14+i*5), %f(11+i*5); ld.shared.f32 %f(17+i*5), [%r7+(128004+i*4)]; fma.rn.f32 %f(18+i*5), %f(15+i*5), %f(17+i*5), %f(13+i*5); ld.shared.f32 %f(19+i*5), [%r7+(8+i*4)]; ld.shared.f32 %f325, [%r5+42588]; fma.rn.f32 %f334, %f325, %f324, %f321; ld.shared.f32 %f326, [%r7+128252]; fma.rn.f32 %f335, %f325, %f326, %f323; add.s32 %r30, %r30, 1; setp.ne.s32 %p3, %r30, 8; @%p3 bra $L__BB0_1; ld.param.u64 %rd19, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd18, %rd19; mul.hi.s32 %r29, %r1, 1626496491; shr.u32 %r28, %r29, 31; shr.s32 %r27, %r29, 6; add.s32 %r26, %r27, %r28; ld.param.u64 %rd17, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd17; mul.wide.s32 %rd13, %r26, 4; add.s64 %rd14, %rd12, %rd13; ld.global.nc.f32 %f327, [%rd14]; add.f32 %f328, %f334, %f327; mov.f32 %f329, 0f00000000; max.f32 %f330, %f328, %f329; mul.wide.s32 %rd15, %r1, 4; add.s64 %rd16, %rd18, %rd15; st.global.f32 [%rd16], %f330; ld.global.nc.f32 %f331, [%rd14+2000]; add.f32 %f332, %f335, %f331; max.f32 %f333, %f332, %f329; st.global.f32 [%rd16+338000], %f333; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2]; __shared__ float pad_temp_shared[10816]; __shared__ float placeholder_shared[64000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 10816) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 10816 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 64000) { int index0 = (int)threadIdx.x; int index2 = ((int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + ((int)threadIdx.x & 63); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index0 = (int)threadIdx.x; int index3 = (int)threadIdx.x / 169 * 64; compute[0] = compute[0] + pad_temp_shared[index0 % 169] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 32000]); int index4 = (int)threadIdx.x % 169 + 169; compute[0] = compute[0] + (pad_temp_shared[index4] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index3 + 32001]); int index5 = (int)threadIdx.x % 169 + 338; compute[0] = compute[0] + (pad_temp_shared[index5] * placeholder_shared[index3 + 2]); compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index3 + 32002]); int index6 = (int)threadIdx.x % 169 + 507; compute[0] = compute[0] + (pad_temp_shared[index6] * placeholder_shared[index3 + 3]); compute[1] = compute[1] + (pad_temp_shared[index6] * placeholder_shared[index3 + 32003]); int index7 = (int)threadIdx.x % 169 + 676; compute[0] = compute[0] + (pad_temp_shared[index7] * placeholder_shared[index3 + 4]); compute[1] = compute[1] + (pad_temp_shared[index7] * placeholder_shared[index3 + 32004]); int index8 = (int)threadIdx.x % 169 + 845; compute[0] = compute[0] + (pad_temp_shared[index8] * placeholder_shared[index3 + 5]); compute[1] = compute[1] + (pad_temp_shared[index8] * placeholder_shared[index3 + 32005]); int index9 = (int)threadIdx.x % 169 + 1014; compute[0] = compute[0] + (pad_temp_shared[index9] * placeholder_shared[index3 + 6]); compute[1] = compute[1] + (pad_temp_shared[index9] * placeholder_shared[index3 + 32006]); int index10 = (int)threadIdx.x % 169 + 1183; compute[0] = compute[0] + (pad_temp_shared[index10] * placeholder_shared[index3 + 7]); compute[1] = compute[1] + (pad_temp_shared[index10] * placeholder_shared[index3 + 32007]); int index11 = (int)threadIdx.x % 169 + 1352; compute[0] = compute[0] + (pad_temp_shared[index11] * placeholder_shared[index3 + 8]); compute[1] = compute[1] + (pad_temp_shared[index11] * placeholder_shared[index3 + 32008]); int index12 = (int)threadIdx.x % 169 + 1521; compute[0] = compute[0] + (pad_temp_shared[index12] * placeholder_shared[index3 + 9]); compute[1] = compute[1] + (pad_temp_shared[index12] * placeholder_shared[index3 + 32009]); int index13 = (int)threadIdx.x % 169 + 1690; compute[0] = compute[0] + (pad_temp_shared[index13] * placeholder_shared[index3 + 10]); compute[1] = compute[1] + (pad_temp_shared[index13] * placeholder_shared[index3 + 32010]); int index14 = (int)threadIdx.x % 169 + 1859; compute[0] = compute[0] + (pad_temp_shared[index14] * placeholder_shared[index3 + 11]); compute[1] = compute[1] + (pad_temp_shared[index14] * placeholder_shared[index3 + 32011]); int index15 = (int)threadIdx.x % 169 + 2028; compute[0] = compute[0] + (pad_temp_shared[index15] * placeholder_shared[index3 + 12]); compute[1] = compute[1] + (pad_temp_shared[index15] * placeholder_shared[index3 + 32012]); int index16 = (int)threadIdx.x % 169 + 2197; compute[0] = compute[0] + (pad_temp_shared[index16] * placeholder_shared[index3 + 13]); compute[1] = compute[1] + (pad_temp_shared[index16] * placeholder_shared[index3 + 32013]); int index17 = (int)threadIdx.x % 169 + 2366; compute[0] = compute[0] + (pad_temp_shared[index17] * placeholder_shared[index3 + 14]); compute[1] = compute[1] + (pad_temp_shared[index17] * placeholder_shared[index3 + 32014]); int index18 = (int)threadIdx.x % 169 + 2535; compute[0] = compute[0] + (pad_temp_shared[index18] * placeholder_shared[index3 + 15]); compute[1] = compute[1] + (pad_temp_shared[index18] * placeholder_shared[index3 + 32015]); int index19 = (int)threadIdx.x % 169 + 2704; compute[0] = compute[0] + (pad_temp_shared[index19] * placeholder_shared[index3 + 16]); compute[1] = compute[1] + (pad_temp_shared[index19] * placeholder_shared[index3 + 32016]); int index20 = (int)threadIdx.x % 169 + 2873; compute[0] = compute[0] + (pad_temp_shared[index20] * placeholder_shared[index3 + 17]); compute[1] = compute[1] + (pad_temp_shared[index20] * placeholder_shared[index3 + 32017]); int index21 = (int)threadIdx.x % 169 + 3042; compute[0] = compute[0] + (pad_temp_shared[index21] * placeholder_shared[index3 + 18]); compute[1] = compute[1] + (pad_temp_shared[index21] * placeholder_shared[index3 + 32018]); int index22 = (int)threadIdx.x % 169 + 3211; compute[0] = compute[0] + (pad_temp_shared[index22] * placeholder_shared[index3 + 19]); compute[1] = compute[1] + (pad_temp_shared[index22] * placeholder_shared[index3 + 32019]); int index23 = (int)threadIdx.x % 169 + 3380; compute[0] = compute[0] + (pad_temp_shared[index23] * placeholder_shared[index3 + 20]); compute[1] = compute[1] + (pad_temp_shared[index23] * placeholder_shared[index3 + 32020]); int index24 = (int)threadIdx.x % 169 + 3549; compute[0] = compute[0] + (pad_temp_shared[index24] * placeholder_shared[index3 + 21]); compute[1] = compute[1] + (pad_temp_shared[index24] * placeholder_shared[index3 + 32021]); int index25 = (int)threadIdx.x % 169 + 3718; compute[0] = compute[0] + (pad_temp_shared[index25] * placeholder_shared[index3 + 22]); compute[1] = compute[1] + (pad_temp_shared[index25] * placeholder_shared[index3 + 32022]); int index26 = (int)threadIdx.x % 169 + 3887; compute[0] = compute[0] + (pad_temp_shared[index26] * placeholder_shared[index3 + 23]); compute[1] = compute[1] + (pad_temp_shared[index26] * placeholder_shared[index3 + 32023]); int index27 = (int)threadIdx.x % 169 + 4056; compute[0] = compute[0] + (pad_temp_shared[index27] * placeholder_shared[index3 + 24]); compute[1] = compute[1] + (pad_temp_shared[index27] * placeholder_shared[index3 + 32024]); int index28 = (int)threadIdx.x % 169 + 4225; compute[0] = compute[0] + (pad_temp_shared[index28] * placeholder_shared[index3 + 25]); compute[1] = compute[1] + (pad_temp_shared[index28] * placeholder_shared[index3 + 32025]); int index29 = (int)threadIdx.x % 169 + 4394; compute[0] = compute[0] + (pad_temp_shared[index29] * placeholder_shared[index3 + 26]); compute[1] = compute[1] + (pad_temp_shared[index29] * placeholder_shared[index3 + 32026]); int index30 = (int)threadIdx.x % 169 + 4563; compute[0] = compute[0] + (pad_temp_shared[index30] * placeholder_shared[index3 + 27]); compute[1] = compute[1] + (pad_temp_shared[index30] * placeholder_shared[index3 + 32027]); int index31 = (int)threadIdx.x % 169 + 4732; compute[0] = compute[0] + (pad_temp_shared[index31] * placeholder_shared[index3 + 28]); compute[1] = compute[1] + (pad_temp_shared[index31] * placeholder_shared[index3 + 32028]); int index32 = (int)threadIdx.x % 169 + 4901; compute[0] = compute[0] + (pad_temp_shared[index32] * placeholder_shared[index3 + 29]); compute[1] = compute[1] + (pad_temp_shared[index32] * placeholder_shared[index3 + 32029]); int index33 = (int)threadIdx.x % 169 + 5070; compute[0] = compute[0] + (pad_temp_shared[index33] * placeholder_shared[index3 + 30]); compute[1] = compute[1] + (pad_temp_shared[index33] * placeholder_shared[index3 + 32030]); int index34 = (int)threadIdx.x % 169 + 5239; compute[0] = compute[0] + (pad_temp_shared[index34] * placeholder_shared[index3 + 31]); compute[1] = compute[1] + (pad_temp_shared[index34] * placeholder_shared[index3 + 32031]); int index35 = (int)threadIdx.x % 169 + 5408; compute[0] = compute[0] + (pad_temp_shared[index35] * placeholder_shared[index3 + 32]); compute[1] = compute[1] + (pad_temp_shared[index35] * placeholder_shared[index3 + 32032]); int index36 = (int)threadIdx.x % 169 + 5577; compute[0] = compute[0] + (pad_temp_shared[index36] * placeholder_shared[index3 + 33]); compute[1] = compute[1] + (pad_temp_shared[index36] * placeholder_shared[index3 + 32033]); int index37 = (int)threadIdx.x % 169 + 5746; compute[0] = compute[0] + (pad_temp_shared[index37] * placeholder_shared[index3 + 34]); compute[1] = compute[1] + (pad_temp_shared[index37] * placeholder_shared[index3 + 32034]); int index38 = (int)threadIdx.x % 169 + 5915; compute[0] = compute[0] + (pad_temp_shared[index38] * placeholder_shared[index3 + 35]); compute[1] = compute[1] + (pad_temp_shared[index38] * placeholder_shared[index3 + 32035]); int index39 = (int)threadIdx.x % 169 + 6084; compute[0] = compute[0] + (pad_temp_shared[index39] * placeholder_shared[index3 + 36]); compute[1] = compute[1] + (pad_temp_shared[index39] * placeholder_shared[index3 + 32036]); int index40 = (int)threadIdx.x % 169 + 6253; compute[0] = compute[0] + (pad_temp_shared[index40] * placeholder_shared[index3 + 37]); compute[1] = compute[1] + (pad_temp_shared[index40] * placeholder_shared[index3 + 32037]); int index41 = (int)threadIdx.x % 169 + 6422; compute[0] = compute[0] + (pad_temp_shared[index41] * placeholder_shared[index3 + 38]); compute[1] = compute[1] + (pad_temp_shared[index41] * placeholder_shared[index3 + 32038]); int index42 = (int)threadIdx.x % 169 + 6591; compute[0] = compute[0] + (pad_temp_shared[index42] * placeholder_shared[index3 + 39]); compute[1] = compute[1] + (pad_temp_shared[index42] * placeholder_shared[index3 + 32039]); int index43 = (int)threadIdx.x % 169 + 6760; compute[0] = compute[0] + (pad_temp_shared[index43] * placeholder_shared[index3 + 40]); compute[1] = compute[1] + (pad_temp_shared[index43] * placeholder_shared[index3 + 32040]); int index44 = (int)threadIdx.x % 169 + 6929; compute[0] = compute[0] + (pad_temp_shared[index44] * placeholder_shared[index3 + 41]); compute[1] = compute[1] + (pad_temp_shared[index44] * placeholder_shared[index3 + 32041]); int index45 = (int)threadIdx.x % 169 + 7098; compute[0] = compute[0] + (pad_temp_shared[index45] * placeholder_shared[index3 + 42]); compute[1] = compute[1] + (pad_temp_shared[index45] * placeholder_shared[index3 + 32042]); int index46 = (int)threadIdx.x % 169 + 7267; compute[0] = compute[0] + (pad_temp_shared[index46] * placeholder_shared[index3 + 43]); compute[1] = compute[1] + (pad_temp_shared[index46] * placeholder_shared[index3 + 32043]); int index47 = (int)threadIdx.x % 169 + 7436; compute[0] = compute[0] + (pad_temp_shared[index47] * placeholder_shared[index3 + 44]); compute[1] = compute[1] + (pad_temp_shared[index47] * placeholder_shared[index3 + 32044]); int index48 = (int)threadIdx.x % 169 + 7605; compute[0] = compute[0] + (pad_temp_shared[index48] * placeholder_shared[index3 + 45]); compute[1] = compute[1] + (pad_temp_shared[index48] * placeholder_shared[index3 + 32045]); int index49 = (int)threadIdx.x % 169 + 7774; compute[0] = compute[0] + (pad_temp_shared[index49] * placeholder_shared[index3 + 46]); compute[1] = compute[1] + (pad_temp_shared[index49] * placeholder_shared[index3 + 32046]); int index50 = (int)threadIdx.x % 169 + 7943; compute[0] = compute[0] + (pad_temp_shared[index50] * placeholder_shared[index3 + 47]); compute[1] = compute[1] + (pad_temp_shared[index50] * placeholder_shared[index3 + 32047]); int index51 = (int)threadIdx.x % 169 + 8112; compute[0] = compute[0] + (pad_temp_shared[index51] * placeholder_shared[index3 + 48]); compute[1] = compute[1] + (pad_temp_shared[index51] * placeholder_shared[index3 + 32048]); int index52 = (int)threadIdx.x % 169 + 8281; compute[0] = compute[0] + (pad_temp_shared[index52] * placeholder_shared[index3 + 49]); compute[1] = compute[1] + (pad_temp_shared[index52] * placeholder_shared[index3 + 32049]); int index53 = (int)threadIdx.x % 169 + 8450; compute[0] = compute[0] + (pad_temp_shared[index53] * placeholder_shared[index3 + 50]); compute[1] = compute[1] + (pad_temp_shared[index53] * placeholder_shared[index3 + 32050]); int index54 = (int)threadIdx.x % 169 + 8619; compute[0] = compute[0] + (pad_temp_shared[index54] * placeholder_shared[index3 + 51]); compute[1] = compute[1] + (pad_temp_shared[index54] * placeholder_shared[index3 + 32051]); int index55 = (int)threadIdx.x % 169 + 8788; compute[0] = compute[0] + (pad_temp_shared[index55] * placeholder_shared[index3 + 52]); compute[1] = compute[1] + (pad_temp_shared[index55] * placeholder_shared[index3 + 32052]); int index56 = (int)threadIdx.x % 169 + 8957; compute[0] = compute[0] + (pad_temp_shared[index56] * placeholder_shared[index3 + 53]); compute[1] = compute[1] + (pad_temp_shared[index56] * placeholder_shared[index3 + 32053]); int index57 = (int)threadIdx.x % 169 + 9126; compute[0] = compute[0] + (pad_temp_shared[index57] * placeholder_shared[index3 + 54]); compute[1] = compute[1] + (pad_temp_shared[index57] * placeholder_shared[index3 + 32054]); int index58 = (int)threadIdx.x % 169 + 9295; compute[0] = compute[0] + (pad_temp_shared[index58] * placeholder_shared[index3 + 55]); compute[1] = compute[1] + (pad_temp_shared[index58] * placeholder_shared[index3 + 32055]); int index59 = (int)threadIdx.x % 169 + 9464; compute[0] = compute[0] + (pad_temp_shared[index59] * placeholder_shared[index3 + 56]); compute[1] = compute[1] + (pad_temp_shared[index59] * placeholder_shared[index3 + 32056]); int index60 = (int)threadIdx.x % 169 + 9633; compute[0] = compute[0] + (pad_temp_shared[index60] * placeholder_shared[index3 + 57]); compute[1] = compute[1] + (pad_temp_shared[index60] * placeholder_shared[index3 + 32057]); int index61 = (int)threadIdx.x % 169 + 9802; compute[0] = compute[0] + (pad_temp_shared[index61] * placeholder_shared[index3 + 58]); compute[1] = compute[1] + (pad_temp_shared[index61] * placeholder_shared[index3 + 32058]); int index62 = (int)threadIdx.x % 169 + 9971; compute[0] = compute[0] + (pad_temp_shared[index62] * placeholder_shared[index3 + 59]); compute[1] = compute[1] + (pad_temp_shared[index62] * placeholder_shared[index3 + 32059]); int index63 = (int)threadIdx.x % 169 + 10140; compute[0] = compute[0] + (pad_temp_shared[index63] * placeholder_shared[index3 + 60]); compute[1] = compute[1] + (pad_temp_shared[index63] * placeholder_shared[index3 + 32060]); int index64 = (int)threadIdx.x % 169 + 10309; compute[0] = compute[0] + (pad_temp_shared[index64] * placeholder_shared[index3 + 61]); compute[1] = compute[1] + (pad_temp_shared[index64] * placeholder_shared[index3 + 32061]); int index65 = (int)threadIdx.x % 169 + 10478; compute[0] = compute[0] + (pad_temp_shared[index65] * placeholder_shared[index3 + 62]); compute[1] = compute[1] + (pad_temp_shared[index65] * placeholder_shared[index3 + 32062]); int index66 = (int)threadIdx.x % 169 + 10647; compute[0] = compute[0] + (pad_temp_shared[index66] * placeholder_shared[index3 + 63]); compute[1] = compute[1] + (pad_temp_shared[index66] * placeholder_shared[index3 + 32063]); } int index0 = (int)threadIdx.x; T_relu[index0] = max(compute[0] + placeholder2[index0 / 169], 0.000000e+00f); int index67 = (int)threadIdx.x / 169 + 500; T_relu[index0 + 84500] = max(compute[1] + placeholder2[index67], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpcud9jkeu
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<37>; .reg .f32 %f<1012>; .reg .b32 %r<74>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[208]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[16000]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd12, %rd9; cvta.to.global.u64 %rd2, %rd8; mov.u32 %r1, %tid.x; mul.hi.s32 %r26, %r1, 1321528399; shr.u32 %r27, %r26, 31; shr.s32 %r28, %r26, 2; add.s32 %r2, %r28, %r27; mov.u32 %r29, %ctaid.x; mul.lo.s32 %r30, %r2, 13; sub.s32 %r3, %r1, %r30; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(31+i*-27), %r(29+i*-27), (13+i*156), %r(3+i*28); shl.b32 %r32, %r1, 2; mov.u32 %r33, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r5, %r33, %r32; for.size.1 i in range(0, 2, 1): shl.b32 %r(6+i*29), %r(2+i*-1), (2+i*6); mov.u32 %r36, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r37, %r36, %r35; add.s32 %r9, %r37, 16; add.s64 %rd4, %rd12, 2048; mov.u32 %r68, 0; mov.f32 %f812, 0f00000000; for.size.1 i in range(0, 99, 1): mov.f32 %f(813+i*1), %f812; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 51; @%p1 bra $L__BB0_3; mad.lo.s32 %r38, %r68, 676, %r4; mul.wide.s32 %rd13, %r38, 4; add.s64 %rd14, %rd2, %rd13; ld.global.nc.f32 %f401, [%rd14]; st.shared.f32 [%r5], %f401; $L__BB0_3: for.size.1 i in range(0, 2, 1): shl.b32 %r(72+i*-6), %r1, (4+i*2); or.b32 %r70, %r66, 7; for.size.1 i in range(0, 2, 1): shl.b32 %r(64+i*-24), %r(1+i*67), (13+i*-11); add.s32 %r41, %r64, %r40; mul.wide.s32 %rd15, %r41, 4; add.s64 %rd24, %rd4, %rd15; mov.u32 %r71, 7; mov.u32 %r69, %r9; $L__BB0_4: setp.gt.s32 %p2, %r72, 999; @%p2 bra $L__BB0_7; setp.gt.s32 %p3, %r1, 62; add.s32 %r42, %r70, -7; setp.gt.s32 %p4, %r42, 3999; or.pred %p5, %p3, %p4; @%p5 bra $L__BB0_7; ld.global.nc.f32 %f402, [%rd24+-2048]; st.shared.f32 [%r69+-16], %f402; $L__BB0_7: @%p2 bra $L__BB0_10; setp.gt.s32 %p7, %r1, 62; add.s32 %r43, %r70, -6; setp.gt.s32 %p8, %r43, 3999; or.pred %p9, %p7, %p8; @%p9 bra $L__BB0_10; ld.global.nc.f32 %f403, [%rd24+-2044]; st.shared.f32 [%r69+-12], %f403; $L__BB0_10: @%p2 bra $L__BB0_13; setp.gt.s32 %p11, %r1, 62; add.s32 %r44, %r70, -5; setp.gt.s32 %p12, %r44, 3999; or.pred %p13, %p11, %p12; @%p13 bra $L__BB0_13; ld.global.nc.f32 %f404, [%rd24+-2040]; st.shared.f32 [%r69+-8], %f404; $L__BB0_13: @%p2 bra $L__BB0_16; setp.gt.s32 %p15, %r1, 62; add.s32 %r45, %r70, -4; setp.gt.s32 %p16, %r45, 3999; or.pred %p17, %p15, %p16; @%p17 bra $L__BB0_16; ld.global.nc.f32 %f405, [%rd24+-2036]; st.shared.f32 [%r69+-4], %f405; $L__BB0_16: add.s32 %r17, %r72, 1; setp.gt.s32 %p18, %r17, 999; @%p18 bra $L__BB0_19; setp.gt.s32 %p19, %r1, 62; add.s32 %r46, %r70, -3; setp.gt.s32 %p20, %r46, 3999; or.pred %p21, %p19, %p20; @%p21 bra $L__BB0_19; ld.global.nc.f32 %f406, [%rd24]; st.shared.f32 [%r69], %f406; $L__BB0_19: @%p18 bra $L__BB0_22; setp.gt.s32 %p23, %r1, 62; add.s32 %r47, %r70, -2; setp.gt.s32 %p24, %r47, 3999; or.pred %p25, %p23, %p24; @%p25 bra $L__BB0_22; ld.global.nc.f32 %f407, [%rd24+4]; st.shared.f32 [%r69+4], %f407; $L__BB0_22: @%p18 bra $L__BB0_25; setp.gt.s32 %p27, %r1, 62; add.s32 %r48, %r70, -1; setp.gt.s32 %p28, %r48, 3999; or.pred %p29, %p27, %p28; @%p29 bra $L__BB0_25; ld.global.nc.f32 %f408, [%rd24+8]; st.shared.f32 [%r69+8], %f408; $L__BB0_25: @%p18 bra $L__BB0_28; for.size.1 i in range(0, 2, 1): setp.gt.s32 %p(31+i*1), %r(1+i*69), (62+i*3937); or.pred %p33, %p31, %p32; @%p33 bra $L__BB0_28; ld.global.nc.f32 %f409, [%rd24+12]; st.shared.f32 [%r69+12], %f409; $L__BB0_28: for.size.1 i in range(0, 2, 1): add.s32 %r(72+i*-2), %r(72+i*-2), (2+i*6); add.s32 %r69, %r69, 32; add.s64 %rd24, %rd24, 4096; add.s32 %r71, %r71, 8; setp.ne.s32 %p34, %r71, 71; @%p34 bra $L__BB0_4; bar.sync 0; mov.u32 %r73, 0; $L__BB0_30: mov.u32 %r59, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r58, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r50, %r73, 13, %r3; shl.b32 %r51, %r50, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(53+i*1), %r(58+i*15), %r(51+i*-45); shl.b32 %r55, %r54, 2; add.s32 %r57, %r59, %r55; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(410+i*1), [%r(57+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(911+i*-1), %f411, %f(410+i*2), %f(911+i*-1); ld.shared.f32 %f(412+i*1), [%r57+(160+i*160)]; for.size.2 i in range(0, 97, 1): fma.rn.f32 %f(909+i*-1), %f411, %f(413+i*1), %f(909+i*-1); ld.shared.f32 %f(414+i*1), [%r57+(480+i*160)]; fma.rn.f32 %f812, %f411, %f510, %f812; add.s32 %r73, %r73, 1; setp.ne.s32 %p35, %r73, 4; @%p35 bra $L__BB0_30; add.s32 %r68, %r68, 1; setp.lt.u32 %p36, %r68, 128; @%p36 bra $L__BB0_1; ld.param.u64 %rd23, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd22, %rd23; ld.param.u64 %rd21, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd20, %rd21; mul.hi.s32 %r63, %r1, 1321528399; shr.u32 %r62, %r63, 31; shr.s32 %r61, %r63, 2; add.s32 %r60, %r61, %r62; mul.wide.s32 %rd16, %r60, 4; add.s64 %rd17, %rd20, %rd16; ld.global.nc.f32 %f511, [%rd17]; add.f32 %f512, %f911, %f511; mov.f32 %f513, 0f00000000; max.f32 %f514, %f512, %f513; mul.wide.s32 %rd18, %r4, 4; add.s64 %rd19, %rd22, %rd18; for.size.4 i in range(0, 99, 1): st.global.f32 [%rd19+(0+i*6760)], %f(514+i*3); ld.global.nc.f32 %f(515+i*3), [%rd17+(40+i*40)]; add.f32 %f(516+i*3), %f(910+i*-1), %f(515+i*3); max.f32 %f(517+i*3), %f(516+i*3), %f513; st.global.f32 [%rd19+669240], %f811; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[100]; __shared__ float pad_temp_shared[52]; __shared__ float placeholder_shared[4000]; for (int vthread_s = 0; vthread_s < 100; ++vthread_s) { int index0 = vthread_s; compute[index0] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 52) { int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 676 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_inner_s = 0; ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 64; ++ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) { if ((int)threadIdx.x * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_inner_s >> 2) < 1000) { if ((int)threadIdx.x * 64 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 4000) { if ((int)threadIdx.x < 63) { int index3 = (int)threadIdx.x * 64 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s; int index4 = (int)threadIdx.x * 8192 + (ax0_ax1_fused_ax2_fused_ax3_fused_inner_s >> 2) * 512 + rc_outer_outer * 4 + (ax0_ax1_fused_ax2_fused_ax3_fused_inner_s & 3); placeholder_shared[index3] = placeholder1[index4]; } } } } __syncthreads(); for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int vthread_s1 = 0; vthread_s1 < 100; ++vthread_s1) { int index5 = vthread_s1; int index6 = rc_inner * 13 + (int)threadIdx.x % 13; int index7 = vthread_s1 * 40 + (int)threadIdx.x / 13 * 4 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } for (int vthread_s2 = 0; vthread_s2 < 100; ++vthread_s2) { int index8 = vthread_s2 * 1690 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index9 = vthread_s2; int index10 = vthread_s2 * 10 + (int)threadIdx.x / 13; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpbe969css
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<61>; .reg .b32 %r<49>; .reg .b64 %rd<36>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[832]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[12800]; ld.param.u64 %rd18, [default_function_kernel0_param_0]; ld.param.u64 %rd19, [default_function_kernel0_param_1]; ld.param.u64 %rd20, [default_function_kernel0_param_2]; ld.param.u64 %rd17, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd21, %rd19; cvta.to.global.u64 %rd1, %rd20; mov.u32 %r1, %tid.x; mul.hi.s32 %r12, %r1, 1321528399; shr.u32 %r13, %r12, 31; shr.s32 %r14, %r12, 2; add.s32 %r2, %r14, %r13; mul.lo.s32 %r15, %r2, 169; mov.u32 %r16, %ctaid.x; mul.hi.s32 %r17, %r16, 1321528399; shr.u32 %r18, %r17, 31; shr.s32 %r19, %r17, 2; add.s32 %r5, %r19, %r18; mul.lo.s32 %r20, %r5, 13; sub.s32 %r21, %r16, %r20; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(22+i*1), %r(21+i*-19), 13; sub.s32 %r24, %r1, %r23; for.size.1 i in range(0, 2, 1): add.s32 %r(25+i*-22), %r(15+i*10), %r(24+i*-2); shl.b32 %r26, %r1, 2; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; for.size.2 i in range(0, 2, 1): add.s32 %r(4+i*2), %r27, %r(26+i*2); shl.b32 %r(28+i*1), %r(24+i*-23), (2+i*2); mov.u32 %r30, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r30, %r29; for.size.2 i in range(0, 2, 1): or.b32 %r(31+i*2), %r26, (2+i*1); and.b32 %r(32+i*2), %r(31+i*2), (14+i*1); shl.b32 %r35, %r2, 6; add.s32 %r8, %r30, %r35; mul.lo.s32 %r36, %r5, 102400; or.b32 %r37, %r34, %r36; shl.b32 %r38, %r1, 7; and.b32 %r39, %r38, -512; add.s32 %r40, %r37, %r39; add.s32 %r41, %r40, -3; mul.wide.s32 %rd22, %r41, 4; add.s64 %rd32, %rd21, %rd22; add.s64 %rd35, %rd32, 12; or.b32 %r42, %r32, %r36; add.s32 %r43, %r42, %r39; mul.wide.s32 %rd23, %r43, 4; add.s64 %rd34, %rd21, %rd23; add.s64 %rd33, %rd32, 4; for.size.1 i in range(0, 2, 1): add.s32 %r(44+i*1), %r(24+i*20), %r(22+i*-7); cvta.to.global.u64 %rd24, %rd18; mul.wide.s32 %rd25, %r45, 4; add.s64 %rd31, %rd24, %rd25; mov.f32 %f60, 0f00000000; mov.u32 %r48, 0; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 207; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f4, [%rd31]; st.shared.f32 [%r4], %f4; $L__BB0_3: setp.gt.s32 %p2, %r1, 799; @%p2 bra $L__BB0_5; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(5+i*1), [%rd(32+i*1)+0]; st.shared.f32 [%r7+(0+i*4)], %f(5+i*1); $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(9+i*1), [%r(8+i*-2)+0]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(11+i*3), %f(10+i*3), %f(9+i*3), %f(60+i*-49); ld.shared.f32 %f(12+i*3), [%r8+(4+i*4)]; ld.shared.f32 %f(13+i*3), [%r6+(52+i*52)]; for.size.3 i in range(0, 13, 1): fma.rn.f32 %f(17+i*3), %f(16+i*3), %f(15+i*3), %f(14+i*3); ld.shared.f32 %f(18+i*3), [%r8+(12+i*4)]; ld.shared.f32 %f(19+i*3), [%r6+(156+i*52)]; fma.rn.f32 %f60, %f55, %f54, %f53; for.size.1 i in range(0, 4, 1): add.s64 %rd(35+i*-1), %rd(35+i*-1), 64; add.s64 %rd31, %rd31, 10816; add.s32 %r48, %r48, 1; setp.ne.s32 %p3, %r48, 32; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd26, %rd17; mad.lo.s32 %r46, %r5, 200, %r2; mul.wide.s32 %rd27, %r46, 4; add.s64 %rd28, %rd26, %rd27; ld.global.nc.f32 %f56, [%rd28]; add.f32 %f57, %f60, %f56; mov.f32 %f58, 0f00000000; max.f32 %f59, %f57, %f58; mad.lo.s32 %r47, %r5, 33800, %r3; mul.wide.s32 %rd29, %r47, 4; add.s64 %rd30, %rd1, %rd29; st.global.f32 [%rd30], %f59; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[1]; __shared__ float pad_temp_shared[208]; __shared__ float placeholder_shared[3200]; compute[0] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 208) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_inner_s = 0; ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 4; ++ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) { if ((int)threadIdx.x * 4 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 3200) { if ((int)threadIdx.x < 800) { int index2 = (int)threadIdx.x * 4 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s; int index3 = (int)blockIdx.x / 13 * 102400 + ((int)threadIdx.x * 4 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 4 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s & 15); placeholder_shared[index2] = placeholder1[index3]; } } } __syncthreads(); for (int rc_inner = 0; rc_inner < 16; ++rc_inner) { int index4 = rc_inner * 13 + (int)threadIdx.x % 13; int index5 = (int)threadIdx.x / 13 * 16 + rc_inner; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; } } int index6 = (int)blockIdx.x / 13 * 33800 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; int index7 = (int)blockIdx.x / 13 * 200 + (int)threadIdx.x / 13; T_relu[index6] = max(compute[0] + placeholder2[index7], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp21x5437p
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<181>; .reg .b32 %r<133>; .reg .b64 %rd<65>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1664]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[128000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; mov.u32 %r2, %ctaid.x; mad.lo.s32 %r3, %r1, 13, %r2; shl.b32 %r39, %r1, 2; mov.u32 %r40, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r40, %r39; shr.u32 %r41, %r1, 5; and.b32 %r42, %r1, 31; bfi.b32 %r5, %r41, %r42, 9, 23; mov.u32 %r43, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r43, %r39; for.size.5 i in range(0, 3, 1): add.s32 %r(44+i*4), %r1, (1000+i*1000); shr.u32 %r(45+i*4), %r(44+i*4), 5; add.s32 %r(46+i*4), %r1, (8+i*8); and.b32 %r(47+i*4), %r(46+i*4), 31; bfi.b32 %r(7+i*1), %r(45+i*4), %r(47+i*4), 9, 23; for.size.4 i in range(0, 7, 1): for.size.3 j in range(0, 3, 1): add.s32 %r(56+i*6+j*2), %r1, (5000+i*4000+j*1000); shr.u32 %r(57+i*6+j*2), %r(56+i*6+j*2), 5; bfi.b32 %r(10+i*3+j*1), %r(57+i*6+j*2), %r(47+j*4), 9, 23; shl.b32 %r98, %r1, 7; add.s32 %r31, %r43, %r98; mov.u32 %r130, 0; mov.f32 %f168, 0f00000000; for.size.1 i in range(0, 12, 1): mov.f32 %f(169+i*1), %f168; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 415; @%p1 bra $L__BB0_3; mad.lo.s32 %r99, %r130, 5408, %r3; mul.wide.s32 %rd9, %r99, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f53, [%rd10]; st.shared.f32 [%r4], %f53; $L__BB0_3: shl.b32 %r102, %r130, 5; add.s32 %r103, %r5, %r102; mul.wide.s32 %rd11, %r103, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f54, [%rd12]; mov.u32 %r132, 4; for.size.5 i in range(0, 3, 1): st.shared.f32 [%r6+(0+i*4000)], %f(54+i*1); add.s32 %r(104+i*1), %r(7+i*1), %r102; mul.wide.s32 %rd(13+i*2), %r(104+i*1), 4; add.s64 %rd(14+i*2), %rd1, %rd(13+i*2); ld.global.nc.f32 %f(55+i*1), [%rd(14+i*2)+0]; for.size.17 i in range(0, 7, 1): st.shared.f32 [%r6+(12000+i*16000)], %f(57+i*4); ld.global.nc.f32 %f(58+i*4), [%rd12+(256000+i*256000)]; st.shared.f32 [%r6+(16000+i*16000)], %f(58+i*4); add.s32 %r(107+i*3), %r(10+i*3), %r102; mul.wide.s32 %rd(19+i*6), %r(107+i*3), 4; add.s64 %rd(20+i*6), %rd1, %rd(19+i*6); ld.global.nc.f32 %f(59+i*4), [%rd(20+i*6)+0]; st.shared.f32 [%r6+(20000+i*16000)], %f(59+i*4); add.s32 %r(108+i*3), %r(11+i*3), %r102; mul.wide.s32 %rd(21+i*6), %r(108+i*3), 4; add.s64 %rd(22+i*6), %rd1, %rd(21+i*6); ld.global.nc.f32 %f(60+i*4), [%rd(22+i*6)+0]; st.shared.f32 [%r6+(24000+i*16000)], %f(60+i*4); add.s32 %r(109+i*3), %r(12+i*3), %r102; mul.wide.s32 %rd(23+i*6), %r(109+i*3), 4; add.s64 %rd(24+i*6), %rd1, %rd(23+i*6); ld.global.nc.f32 %f(61+i*4), [%rd(24+i*6)+0]; st.shared.f32 [%r6+124000], %f85; bar.sync 0; mov.u32 %r131, %r40; $L__BB0_4: add.s32 %r128, %r31, %r132; for.size.27 i in range(0, 2, 1): ld.shared.f32 %f(86+i*27), [%r128+(-4+i*4)]; ld.shared.f32 %f(87+i*27), [%r131+(0+i*52)]; fma.rn.f32 %f(88+i*92), %f(87+i*27), %f(86+i*27), %f(180+i*-92); ld.shared.f32 %f(89+i*26), [%r131+(4+i*52)]; fma.rn.f32 %f(90+i*89), %f(89+i*26), %f(86+i*27), %f(179+i*-89); ld.shared.f32 %f(91+i*25), [%r131+(8+i*52)]; fma.rn.f32 %f(92+i*86), %f(91+i*25), %f(86+i*27), %f(178+i*-86); ld.shared.f32 %f(93+i*24), [%r131+(12+i*52)]; fma.rn.f32 %f(94+i*83), %f(93+i*24), %f(86+i*27), %f(177+i*-83); ld.shared.f32 %f(95+i*23), [%r131+(16+i*52)]; fma.rn.f32 %f(96+i*80), %f(95+i*23), %f(86+i*27), %f(176+i*-80); ld.shared.f32 %f(97+i*22), [%r131+(20+i*52)]; fma.rn.f32 %f(98+i*77), %f(97+i*22), %f(86+i*27), %f(175+i*-77); ld.shared.f32 %f(99+i*21), [%r131+(24+i*52)]; fma.rn.f32 %f(100+i*74), %f(99+i*21), %f(86+i*27), %f(174+i*-74); ld.shared.f32 %f(101+i*20), [%r131+(28+i*52)]; fma.rn.f32 %f(102+i*71), %f(101+i*20), %f(86+i*27), %f(173+i*-71); ld.shared.f32 %f(103+i*19), [%r131+(32+i*52)]; fma.rn.f32 %f(104+i*68), %f(103+i*19), %f(86+i*27), %f(172+i*-68); ld.shared.f32 %f(105+i*18), [%r131+(36+i*52)]; fma.rn.f32 %f(106+i*65), %f(105+i*18), %f(86+i*27), %f(171+i*-65); ld.shared.f32 %f(107+i*17), [%r131+(40+i*52)]; fma.rn.f32 %f(108+i*62), %f(107+i*17), %f(86+i*27), %f(170+i*-62); ld.shared.f32 %f(109+i*16), [%r131+(44+i*52)]; fma.rn.f32 %f(110+i*59), %f(109+i*16), %f(86+i*27), %f(169+i*-59); ld.shared.f32 %f(111+i*15), [%r131+(48+i*52)]; fma.rn.f32 %f(112+i*56), %f(111+i*15), %f(86+i*27), %f(168+i*-56); for.size.1 i in range(0, 2, 1): add.s32 %r(131+i*1), %r(131+i*1), (104+i*-96); setp.ne.s32 %p2, %r132, 132; @%p2 bra $L__BB0_4; add.s32 %r130, %r130, 1; setp.lt.u32 %p3, %r130, 16; @%p3 bra $L__BB0_1; mul.wide.s32 %rd61, %r1, 4; add.s64 %rd62, %rd4, %rd61; mad.lo.s32 %r129, %r1, 169, %r2; ld.global.nc.f32 %f127, [%rd62]; add.f32 %f128, %f180, %f127; mov.f32 %f129, 0f00000000; max.f32 %f130, %f128, %f129; mul.wide.s32 %rd63, %r129, 4; add.s64 %rd64, %rd3, %rd63; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd64+(0+i*52)], %f(130+i*2); add.f32 %f(131+i*2), %f(179+i*-1), %f127; max.f32 %f(132+i*2), %f(131+i*2), %f129; st.global.f32 [%rd64+624], %f154; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[416]; __shared__ float placeholder_shared[32000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 416) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 5408 + (int)threadIdx.x * 13 + (int)blockIdx.x; pad_temp_shared[index0] = placeholder[index1]; } int index0 = (int)threadIdx.x; int index2 = ((int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index0] = placeholder1[index2]; int index3 = ((int)threadIdx.x + 1000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 1000] = placeholder1[index3]; int index4 = ((int)threadIdx.x + 2000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 2000] = placeholder1[index4]; int index5 = ((int)threadIdx.x + 3000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 3000] = placeholder1[index5]; placeholder_shared[index0 + 4000] = placeholder1[index2 + 64000]; int index6 = ((int)threadIdx.x + 5000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 5000] = placeholder1[index6]; int index7 = ((int)threadIdx.x + 6000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 6000] = placeholder1[index7]; int index8 = ((int)threadIdx.x + 7000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 7000] = placeholder1[index8]; placeholder_shared[index0 + 8000] = placeholder1[index2 + 128000]; int index9 = ((int)threadIdx.x + 9000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 9000] = placeholder1[index9]; int index10 = ((int)threadIdx.x + 10000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 10000] = placeholder1[index10]; int index11 = ((int)threadIdx.x + 11000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 11000] = placeholder1[index11]; placeholder_shared[index0 + 12000] = placeholder1[index2 + 192000]; int index12 = ((int)threadIdx.x + 13000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 13000] = placeholder1[index12]; int index13 = ((int)threadIdx.x + 14000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 14000] = placeholder1[index13]; int index14 = ((int)threadIdx.x + 15000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 15000] = placeholder1[index14]; placeholder_shared[index0 + 16000] = placeholder1[index2 + 256000]; int index15 = ((int)threadIdx.x + 17000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 17000] = placeholder1[index15]; int index16 = ((int)threadIdx.x + 18000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 18000] = placeholder1[index16]; int index17 = ((int)threadIdx.x + 19000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 19000] = placeholder1[index17]; placeholder_shared[index0 + 20000] = placeholder1[index2 + 320000]; int index18 = ((int)threadIdx.x + 21000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 21000] = placeholder1[index18]; int index19 = ((int)threadIdx.x + 22000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 22000] = placeholder1[index19]; int index20 = ((int)threadIdx.x + 23000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 23000] = placeholder1[index20]; placeholder_shared[index0 + 24000] = placeholder1[index2 + 384000]; int index21 = ((int)threadIdx.x + 25000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 25000] = placeholder1[index21]; int index22 = ((int)threadIdx.x + 26000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 26000] = placeholder1[index22]; int index23 = ((int)threadIdx.x + 27000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 27000] = placeholder1[index23]; placeholder_shared[index0 + 28000] = placeholder1[index2 + 448000]; int index24 = ((int)threadIdx.x + 29000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index0 + 29000] = placeholder1[index24]; int index25 = ((int)threadIdx.x + 30000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index0 + 30000] = placeholder1[index25]; int index26 = ((int)threadIdx.x + 31000 >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index0 + 31000] = placeholder1[index26]; __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { int index27 = rc_outer_inner * 13; int index28 = (int)threadIdx.x * 32 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index27] * placeholder_shared[index28]; compute[1] = compute[1] + (pad_temp_shared[index27 + 1] * placeholder_shared[index28]); compute[2] = compute[2] + (pad_temp_shared[index27 + 2] * placeholder_shared[index28]); compute[3] = compute[3] + (pad_temp_shared[index27 + 3] * placeholder_shared[index28]); compute[4] = compute[4] + (pad_temp_shared[index27 + 4] * placeholder_shared[index28]); compute[5] = compute[5] + (pad_temp_shared[index27 + 5] * placeholder_shared[index28]); compute[6] = compute[6] + (pad_temp_shared[index27 + 6] * placeholder_shared[index28]); compute[7] = compute[7] + (pad_temp_shared[index27 + 7] * placeholder_shared[index28]); compute[8] = compute[8] + (pad_temp_shared[index27 + 8] * placeholder_shared[index28]); compute[9] = compute[9] + (pad_temp_shared[index27 + 9] * placeholder_shared[index28]); compute[10] = compute[10] + (pad_temp_shared[index27 + 10] * placeholder_shared[index28]); compute[11] = compute[11] + (pad_temp_shared[index27 + 11] * placeholder_shared[index28]); compute[12] = compute[12] + (pad_temp_shared[index27 + 12] * placeholder_shared[index28]); } } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index0 = (int)threadIdx.x; int index29 = (int)threadIdx.x * 169 + ax2_inner * 13 + (int)blockIdx.x; int index30 = ax2_inner; T_relu[index29] = max(compute[index30] + placeholder2[index0], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp3zucuv2_
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<8>; .reg .f32 %f<142>; .reg .b32 %r<99>; .reg .b64 %rd<27>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[26624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[204800]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd10, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd9; cvta.to.global.u64 %rd2, %rd10; mov.u32 %r1, %ctaid.x; mul.hi.s32 %r29, %r1, 1321528399; shr.u32 %r30, %r29, 31; shr.s32 %r31, %r29, 2; add.s32 %r32, %r31, %r30; mul.lo.s32 %r33, %r32, 13; sub.s32 %r34, %r1, %r33; mul.lo.s32 %r2, %r34, 13; mov.u32 %r3, %tid.x; mov.u32 %r92, 200; shl.b32 %r35, %r3, 2; mov.u32 %r97, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r37, %r97, %r35; for.size.1 i in range(0, 2, 1): add.s32 %r(94+i*-1), %r(37+i*-34), (800+i*-500); mov.u32 %r91, %r3; bra.uni $L__BB0_1; $L__BB0_10: for.size.1 i in range(0, 2, 1): add.s32 %r(94+i*-1), %r(94+i*-1), (1600+i*-1200); for.size.1 i in range(0, 2, 1): add.s32 %r(92+i*-1), %r(92+i*-1), 400; $L__BB0_1: setp.gt.s32 %p1, %r91, 6655; @%p1 bra $L__BB0_3; mul.hi.s32 %r38, %r91, 1321528399; shr.u32 %r39, %r38, 31; shr.s32 %r40, %r38, 2; add.s32 %r41, %r40, %r39; mul.lo.s32 %r42, %r41, 13; sub.s32 %r43, %r91, %r42; add.s32 %r44, %r43, %r2; mad.lo.s32 %r45, %r41, 169, %r44; mul.wide.s32 %rd11, %r45, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f27, [%rd12]; add.s32 %r87, %r94, -800; st.shared.f32 [%r87], %f27; $L__BB0_3: add.s32 %r11, %r93, -200; setp.gt.s32 %p2, %r11, 6655; @%p2 bra $L__BB0_5; mul.hi.s32 %r46, %r11, 1321528399; shr.u32 %r47, %r46, 31; shr.s32 %r48, %r46, 2; add.s32 %r49, %r48, %r47; mul.lo.s32 %r50, %r49, 13; sub.s32 %r51, %r11, %r50; add.s32 %r52, %r51, %r2; mad.lo.s32 %r53, %r49, 169, %r52; mul.wide.s32 %rd13, %r53, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f28, [%rd14]; add.s32 %r88, %r94, -800; st.shared.f32 [%r88+400], %f28; $L__BB0_5: add.s32 %r12, %r93, -100; setp.gt.s32 %p3, %r12, 6655; @%p3 bra $L__BB0_7; mul.hi.s32 %r54, %r12, 1321528399; shr.u32 %r55, %r54, 31; shr.s32 %r56, %r54, 2; add.s32 %r57, %r56, %r55; mul.lo.s32 %r58, %r57, 13; sub.s32 %r59, %r12, %r58; add.s32 %r60, %r59, %r2; mad.lo.s32 %r61, %r57, 169, %r60; mul.wide.s32 %rd15, %r61, 4; add.s64 %rd16, %rd1, %rd15; ld.global.nc.f32 %f29, [%rd16]; add.s32 %r89, %r94, -800; st.shared.f32 [%r89+800], %f29; $L__BB0_7: setp.eq.s32 %p4, %r92, 6600; @%p4 bra $L__BB0_11; setp.gt.s32 %p5, %r93, 6655; @%p5 bra $L__BB0_10; mul.hi.s32 %r62, %r93, 1321528399; shr.u32 %r63, %r62, 31; shr.s32 %r64, %r62, 2; add.s32 %r65, %r64, %r63; mul.lo.s32 %r66, %r65, 13; sub.s32 %r67, %r93, %r66; add.s32 %r68, %r67, %r2; mad.lo.s32 %r69, %r65, 169, %r68; mul.wide.s32 %rd17, %r69, 4; add.s64 %rd18, %rd1, %rd17; ld.global.nc.f32 %f30, [%rd18]; add.s32 %r90, %r94, -800; st.shared.f32 [%r90+1200], %f30; bra.uni $L__BB0_10; $L__BB0_11: cvta.to.global.u64 %rd19, %rd7; cvta.to.global.u64 %rd3, %rd8; mad.lo.s32 %r74, %r32, 51200, %r3; mul.wide.s32 %rd20, %r74, 4; add.s64 %rd21, %rd19, %rd20; add.s64 %rd26, %rd21, 3200; mov.u32 %r76, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r77, %r76, %r35; add.s32 %r95, %r77, 3200; mov.u32 %r96, 0; $L__BB0_12: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(31+i*1), [%rd26+(-3200+i*400)]; st.shared.f32 [%r95+(-3200+i*400)], %f(31+i*1); add.s64 %rd26, %rd26, 6400; for.size.1 i in range(0, 2, 1): add.s32 %r(95+i*1), %r(95+i*1), (6400+i*-6384); setp.ne.s32 %p6, %r96, 512; @%p6 bra $L__BB0_12; bar.sync 0; shl.b32 %r80, %r3, 11; add.s32 %r23, %r76, %r80; mov.f32 %f129, 0f00000000; mov.u32 %r98, 4; for.size.1 i in range(0, 12, 1): mov.f32 %f(130+i*1), %f129; $L__BB0_14: add.s32 %r82, %r23, %r98; for.size.27 i in range(0, 2, 1): ld.shared.f32 %f(60+i*27), [%r82+(-4+i*4)]; ld.shared.f32 %f(61+i*27), [%r97+(0+i*52)]; fma.rn.f32 %f(62+i*79), %f(61+i*27), %f(60+i*27), %f(141+i*-79); ld.shared.f32 %f(63+i*26), [%r97+(4+i*52)]; fma.rn.f32 %f(64+i*76), %f(63+i*26), %f(60+i*27), %f(140+i*-76); ld.shared.f32 %f(65+i*25), [%r97+(8+i*52)]; fma.rn.f32 %f(66+i*73), %f(65+i*25), %f(60+i*27), %f(139+i*-73); ld.shared.f32 %f(67+i*24), [%r97+(12+i*52)]; fma.rn.f32 %f(68+i*70), %f(67+i*24), %f(60+i*27), %f(138+i*-70); ld.shared.f32 %f(69+i*23), [%r97+(16+i*52)]; fma.rn.f32 %f(70+i*67), %f(69+i*23), %f(60+i*27), %f(137+i*-67); ld.shared.f32 %f(71+i*22), [%r97+(20+i*52)]; fma.rn.f32 %f(72+i*64), %f(71+i*22), %f(60+i*27), %f(136+i*-64); ld.shared.f32 %f(73+i*21), [%r97+(24+i*52)]; fma.rn.f32 %f(74+i*61), %f(73+i*21), %f(60+i*27), %f(135+i*-61); ld.shared.f32 %f(75+i*20), [%r97+(28+i*52)]; fma.rn.f32 %f(76+i*58), %f(75+i*20), %f(60+i*27), %f(134+i*-58); ld.shared.f32 %f(77+i*19), [%r97+(32+i*52)]; fma.rn.f32 %f(78+i*55), %f(77+i*19), %f(60+i*27), %f(133+i*-55); ld.shared.f32 %f(79+i*18), [%r97+(36+i*52)]; fma.rn.f32 %f(80+i*52), %f(79+i*18), %f(60+i*27), %f(132+i*-52); ld.shared.f32 %f(81+i*17), [%r97+(40+i*52)]; fma.rn.f32 %f(82+i*49), %f(81+i*17), %f(60+i*27), %f(131+i*-49); ld.shared.f32 %f(83+i*16), [%r97+(44+i*52)]; fma.rn.f32 %f(84+i*46), %f(83+i*16), %f(60+i*27), %f(130+i*-46); ld.shared.f32 %f(85+i*15), [%r97+(48+i*52)]; fma.rn.f32 %f(86+i*43), %f(85+i*15), %f(60+i*27), %f(129+i*-43); for.size.1 i in range(0, 2, 1): add.s32 %r(97+i*1), %r(97+i*1), (104+i*-96); setp.ne.s32 %p7, %r98, 2052; @%p7 bra $L__BB0_14; mad.lo.s32 %r83, %r32, 100, %r3; mul.wide.s32 %rd22, %r83, 4; add.s64 %rd23, %rd3, %rd22; mul.lo.s32 %r84, %r3, 169; mad.lo.s32 %r85, %r32, 16900, %r84; add.s32 %r86, %r85, %r2; ld.global.nc.f32 %f101, [%rd23]; add.f32 %f102, %f141, %f101; mov.f32 %f103, 0f00000000; max.f32 %f104, %f102, %f103; mul.wide.s32 %rd24, %r86, 4; add.s64 %rd25, %rd2, %rd24; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd25+(0+i*4)], %f(104+i*2); add.f32 %f(105+i*2), %f(140+i*-1), %f101; max.f32 %f(106+i*2), %f(105+i*2), %f103; st.global.f32 [%rd25+48], %f128; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[6656]; __shared__ float placeholder_shared[51200]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 67; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 6656) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x; int index1 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x) / 13 * 169 + (int)blockIdx.x % 13 * 13 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x) % 13; pad_temp_shared[index0] = placeholder[index1]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 512; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; int index3 = (int)blockIdx.x / 13 * 51200 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 256; ++rc_outer_inner) { int index4 = rc_outer_inner * 26; int index5 = (int)threadIdx.x * 512 + rc_outer_inner * 2; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4 + 1] * placeholder_shared[index5]); compute[2] = compute[2] + (pad_temp_shared[index4 + 2] * placeholder_shared[index5]); compute[3] = compute[3] + (pad_temp_shared[index4 + 3] * placeholder_shared[index5]); compute[4] = compute[4] + (pad_temp_shared[index4 + 4] * placeholder_shared[index5]); compute[5] = compute[5] + (pad_temp_shared[index4 + 5] * placeholder_shared[index5]); compute[6] = compute[6] + (pad_temp_shared[index4 + 6] * placeholder_shared[index5]); compute[7] = compute[7] + (pad_temp_shared[index4 + 7] * placeholder_shared[index5]); compute[8] = compute[8] + (pad_temp_shared[index4 + 8] * placeholder_shared[index5]); compute[9] = compute[9] + (pad_temp_shared[index4 + 9] * placeholder_shared[index5]); compute[10] = compute[10] + (pad_temp_shared[index4 + 10] * placeholder_shared[index5]); compute[11] = compute[11] + (pad_temp_shared[index4 + 11] * placeholder_shared[index5]); compute[12] = compute[12] + (pad_temp_shared[index4 + 12] * placeholder_shared[index5]); compute[0] = compute[0] + (pad_temp_shared[index4 + 13] * placeholder_shared[index5 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4 + 14] * placeholder_shared[index5 + 1]); compute[2] = compute[2] + (pad_temp_shared[index4 + 15] * placeholder_shared[index5 + 1]); compute[3] = compute[3] + (pad_temp_shared[index4 + 16] * placeholder_shared[index5 + 1]); compute[4] = compute[4] + (pad_temp_shared[index4 + 17] * placeholder_shared[index5 + 1]); compute[5] = compute[5] + (pad_temp_shared[index4 + 18] * placeholder_shared[index5 + 1]); compute[6] = compute[6] + (pad_temp_shared[index4 + 19] * placeholder_shared[index5 + 1]); compute[7] = compute[7] + (pad_temp_shared[index4 + 20] * placeholder_shared[index5 + 1]); compute[8] = compute[8] + (pad_temp_shared[index4 + 21] * placeholder_shared[index5 + 1]); compute[9] = compute[9] + (pad_temp_shared[index4 + 22] * placeholder_shared[index5 + 1]); compute[10] = compute[10] + (pad_temp_shared[index4 + 23] * placeholder_shared[index5 + 1]); compute[11] = compute[11] + (pad_temp_shared[index4 + 24] * placeholder_shared[index5 + 1]); compute[12] = compute[12] + (pad_temp_shared[index4 + 25] * placeholder_shared[index5 + 1]); } for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index6 = (int)blockIdx.x / 13 * 16900 + (int)threadIdx.x * 169 + (int)blockIdx.x % 13 * 13 + ax3_inner; int index7 = ax3_inner; int index8 = (int)blockIdx.x / 13 * 100 + (int)threadIdx.x; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp2onmogqt
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<9>; .reg .f32 %f<209>; .reg .b32 %r<73>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd8, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd7; cvta.to.global.u64 %rd2, %rd8; cvta.to.global.u64 %rd3, %rd6; cvta.to.global.u64 %rd4, %rd9; mov.u32 %r1, %tid.x; mul.hi.s32 %r24, %r1, 1321528399; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 2; add.s32 %r2, %r26, %r25; mov.u32 %r27, %ctaid.x; mul.lo.s32 %r28, %r2, 13; sub.s32 %r3, %r1, %r28; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(4+i*1), %r(27+i*-25), (13+i*156), %r(3+i*1); shl.b32 %r29, %r1, 2; mov.u32 %r30, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r6, %r30, %r29; mul.lo.s32 %r7, %r2, 1280; mov.u32 %r31, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r32, %r31, %r29; add.s32 %r8, %r32, 5200; mov.u32 %r68, 0; mov.f32 %f169, 0f00000000; for.size.1 i in range(0, 19, 1): mov.f32 %f(170+i*1), %f169; $L__BB0_1: bar.sync 0; mad.lo.s32 %r33, %r68, 10816, %r5; mul.wide.s32 %rd10, %r33, 4; add.s64 %rd5, %rd3, %rd10; ld.global.nc.f32 %f81, [%rd5]; st.shared.f32 [%r6], %f81; setp.gt.s32 %p1, %r1, 181; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f82, [%rd5+33800]; st.shared.f32 [%r6+2600], %f82; $L__BB0_3: shl.b32 %r11, %r68, 6; mov.u32 %r70, 1300; mov.u32 %r69, %r1; mov.u32 %r71, %r8; bra.uni $L__BB0_4; $L__BB0_13: for.size.1 i in range(0, 2, 1): add.s32 %r(71+i*-1), %r(71+i*-1), (10400+i*-7800); add.s32 %r69, %r69, 2600; $L__BB0_4: setp.gt.s32 %p2, %r69, 63999; @%p2 bra $L__BB0_6; shl.b32 %r35, %r69, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(36+i*1), %r(35+i*34), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(38+i*1), %r(37+i*1), %r(11+i*25); mul.wide.s32 %rd11, %r39, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f83, [%rd12]; st.shared.f32 [%r71+-5200], %f83; $L__BB0_6: add.s32 %r15, %r69, 650; setp.gt.s32 %p3, %r15, 63999; @%p3 bra $L__BB0_8; shl.b32 %r40, %r15, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(41+i*1), %r(40+i*-25), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(43+i*1), %r(42+i*1), %r(11+i*30); mul.wide.s32 %rd13, %r44, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f84, [%rd14]; st.shared.f32 [%r71+-2600], %f84; $L__BB0_8: add.s32 %r16, %r69, 1300; setp.gt.s32 %p4, %r16, 63999; @%p4 bra $L__BB0_10; shl.b32 %r45, %r16, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(46+i*1), %r(45+i*-29), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(48+i*1), %r(47+i*1), %r(11+i*35); mul.wide.s32 %rd15, %r49, 4; add.s64 %rd16, %rd1, %rd15; ld.global.nc.f32 %f85, [%rd16]; st.shared.f32 [%r71], %f85; $L__BB0_10: setp.eq.s32 %p5, %r70, 63700; @%p5 bra $L__BB0_14; add.s32 %r50, %r69, 1950; setp.gt.s32 %p6, %r50, 63999; @%p6 bra $L__BB0_13; shl.b32 %r52, %r50, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(53+i*1), %r(52+i*-2), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(55+i*1), %r(54+i*1), %r(11+i*42); mul.wide.s32 %rd17, %r56, 4; add.s64 %rd18, %rd1, %rd17; ld.global.nc.f32 %f86, [%rd18]; st.shared.f32 [%r71+2600], %f86; bra.uni $L__BB0_13; $L__BB0_14: bar.sync 0; mov.u32 %r72, 0; $L__BB0_15: mad.lo.s32 %r58, %r72, 13, %r3; shl.b32 %r59, %r58, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(61+i*1), %r(30+i*42), %r(59+i*-52); shl.b32 %r63, %r62, 2; add.s32 %r65, %r31, %r63; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(87+i*1), [%r(65+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(188+i*-1), %f88, %f(87+i*2), %f(188+i*-1); ld.shared.f32 %f(89+i*1), [%r65+(256+i*256)]; for.size.2 i in range(0, 17, 1): fma.rn.f32 %f(186+i*-1), %f88, %f(90+i*1), %f(186+i*-1); ld.shared.f32 %f(91+i*1), [%r65+(768+i*256)]; fma.rn.f32 %f169, %f88, %f107, %f169; add.s32 %r72, %r72, 1; setp.ne.s32 %p7, %r72, 64; @%p7 bra $L__BB0_15; add.s32 %r68, %r68, 1; setp.lt.u32 %p8, %r68, 8; @%p8 bra $L__BB0_1; mul.lo.s32 %r66, %r2, 20; mad.lo.s32 %r67, %r2, 3380, %r4; mul.wide.s32 %rd19, %r66, 4; add.s64 %rd20, %rd4, %rd19; ld.global.nc.f32 %f108, [%rd20]; add.f32 %f109, %f188, %f108; mov.f32 %f110, 0f00000000; max.f32 %f111, %f109, %f110; mul.wide.s32 %rd21, %r67, 4; add.s64 %rd22, %rd2, %rd21; for.size.4 i in range(0, 19, 1): st.global.f32 [%rd22+(0+i*676)], %f(111+i*3); ld.global.nc.f32 %f(112+i*3), [%rd20+(4+i*4)]; add.f32 %f(113+i*3), %f(187+i*-1), %f(112+i*3); max.f32 %f(114+i*3), %f(113+i*3), %f110; st.global.f32 [%rd22+12844], %f168; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[20]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; for (int ff_inner_init = 0; ff_inner_init < 20; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 10816 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; if ((int)threadIdx.x < 182) { pad_temp_shared[index1 + 650] = placeholder[index2 + 8450]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 99; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 650 + (int)threadIdx.x < 64000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 650 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 650 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 650 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 20; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 1280 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 3380 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 20 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpde301nm2
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<95>; .reg .b32 %r<94>; .reg .b64 %rd<24>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[64]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[12800]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; mov.u32 %r10, %ctaid.x; mul.hi.s32 %r11, %r10, 1626496491; shr.u32 %r12, %r11, 31; shr.s32 %r13, %r11, 6; add.s32 %r5, %r13, %r12; mul.lo.s32 %r14, %r5, 169; sub.s32 %r15, %r10, %r14; mad.lo.s32 %r2, %r1, 169, %r15; shl.b32 %r16, %r1, 2; mov.u32 %r17, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r17, %r16; shl.b32 %r18, %r1, 4; mov.u32 %r19, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r19, %r18; and.b32 %r20, %r16, 12; shl.b32 %r21, %r1, 6; add.s32 %r6, %r19, %r21; cvta.to.global.u64 %rd13, %rd9; mul.wide.s32 %rd14, %r2, 4; add.s64 %rd23, %rd13, %rd14; mul.lo.s32 %r22, %r5, 102400; or.b32 %r23, %r20, %r22; shl.b32 %r24, %r1, 7; and.b32 %r25, %r24, -512; add.s32 %r26, %r23, %r25; mul.wide.s32 %rd15, %r26, 4; add.s64 %rd16, %rd12, %rd15; add.s64 %rd22, %rd16, 204800; mov.f32 %f93, 0f00000000; mov.u32 %r93, 0; mov.f32 %f94, %f93; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 15; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f7, [%rd23]; st.shared.f32 [%r3], %f7; $L__BB0_3: for.size.2 i in range(0, 8, 1): ld.global.nc.v4.u32 {%r(27+i*8), %r(28+i*8), %r(29+i*8), %r(30+i*8)}, [%rd22+(-204800+i*51200)]; st.shared.v4.u32 [%r4+(0+i*1600)], {%r(27+i*8), %r(28+i*8), %r(29+i*8), %r(30+i*8)}; bar.sync 0; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(8+i*5), [%r6+(0+i*4)]; ld.shared.f32 %f(9+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(0+i*4)]; fma.rn.f32 %f(10+i*5), %f(9+i*5), %f(8+i*5), %f(93+i*-83); ld.shared.f32 %f(11+i*5), [%r6+(6400+i*4)]; fma.rn.f32 %f(12+i*5), %f(9+i*5), %f(11+i*5), %f(94+i*-82); for.size.5 i in range(0, 13, 1): ld.shared.f32 %f(18+i*5), [%r6+(8+i*4)]; ld.shared.f32 %f(19+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(8+i*4)]; fma.rn.f32 %f(20+i*5), %f(19+i*5), %f(18+i*5), %f(15+i*5); ld.shared.f32 %f(21+i*5), [%r6+(6408+i*4)]; fma.rn.f32 %f(22+i*5), %f(19+i*5), %f(21+i*5), %f(17+i*5); ld.shared.f32 %f83, [%r6+60]; ld.shared.f32 %f84, [_ZZ24default_function_kernel0E15pad_temp_shared+60]; fma.rn.f32 %f93, %f84, %f83, %f80; ld.shared.f32 %f85, [%r6+6460]; fma.rn.f32 %f94, %f84, %f85, %f82; for.size.1 i in range(0, 2, 1): add.s64 %rd(23+i*-1), %rd(23+i*-1), (10816+i*-10752); add.s32 %r93, %r93, 1; setp.ne.s32 %p2, %r93, 32; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd8; mad.lo.s32 %r91, %r5, 200, %r1; mul.wide.s32 %rd18, %r91, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f86, [%rd19]; add.f32 %f87, %f93, %f86; mov.f32 %f88, 0f00000000; max.f32 %f89, %f87, %f88; mad.lo.s32 %r92, %r5, 33800, %r2; mul.wide.s32 %rd20, %r92, 4; add.s64 %rd21, %rd1, %rd20; st.global.f32 [%rd21], %f89; ld.global.nc.f32 %f90, [%rd19+400]; add.f32 %f91, %f94, %f90; max.f32 %f92, %f91, %f88; st.global.f32 [%rd21+67600], %f92; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2]; __shared__ float pad_temp_shared[16]; __shared__ float placeholder_shared[3200]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 16) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } ((float4*)(placeholder_shared + (int)threadIdx.x * 4))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 400))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 12800))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 800))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 25600))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 1200))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 38400))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 1600))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 51200))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 2000))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 64000))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 2400))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 76800))[0]; ((float4*)(placeholder_shared + (int)threadIdx.x * 4 + 2800))[0] = ((float4*)(placeholder1 + (int)blockIdx.x / 169 * 102400 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 3) * 4 + 89600))[0]; __syncthreads(); int index0 = (int)threadIdx.x; compute[0] = compute[0] + pad_temp_shared[0] * placeholder_shared[index0 * 16]; int index2 = (int)threadIdx.x * 16 + 1600; compute[1] = compute[1] + pad_temp_shared[0] * placeholder_shared[index2]; int index3 = (int)threadIdx.x * 16 + 1; compute[0] = compute[0] + pad_temp_shared[1] * placeholder_shared[index3]; int index4 = (int)threadIdx.x * 16 + 1601; compute[1] = compute[1] + pad_temp_shared[1] * placeholder_shared[index4]; int index5 = (int)threadIdx.x * 16 + 2; compute[0] = compute[0] + pad_temp_shared[2] * placeholder_shared[index5]; int index6 = (int)threadIdx.x * 16 + 1602; compute[1] = compute[1] + pad_temp_shared[2] * placeholder_shared[index6]; int index7 = (int)threadIdx.x * 16 + 3; compute[0] = compute[0] + pad_temp_shared[3] * placeholder_shared[index7]; int index8 = (int)threadIdx.x * 16 + 1603; compute[1] = compute[1] + pad_temp_shared[3] * placeholder_shared[index8]; int index9 = (int)threadIdx.x * 16 + 4; compute[0] = compute[0] + pad_temp_shared[4] * placeholder_shared[index9]; int index10 = (int)threadIdx.x * 16 + 1604; compute[1] = compute[1] + pad_temp_shared[4] * placeholder_shared[index10]; int index11 = (int)threadIdx.x * 16 + 5; compute[0] = compute[0] + pad_temp_shared[5] * placeholder_shared[index11]; int index12 = (int)threadIdx.x * 16 + 1605; compute[1] = compute[1] + pad_temp_shared[5] * placeholder_shared[index12]; int index13 = (int)threadIdx.x * 16 + 6; compute[0] = compute[0] + pad_temp_shared[6] * placeholder_shared[index13]; int index14 = (int)threadIdx.x * 16 + 1606; compute[1] = compute[1] + pad_temp_shared[6] * placeholder_shared[index14]; int index15 = (int)threadIdx.x * 16 + 7; compute[0] = compute[0] + pad_temp_shared[7] * placeholder_shared[index15]; int index16 = (int)threadIdx.x * 16 + 1607; compute[1] = compute[1] + pad_temp_shared[7] * placeholder_shared[index16]; int index17 = (int)threadIdx.x * 16 + 8; compute[0] = compute[0] + pad_temp_shared[8] * placeholder_shared[index17]; int index18 = (int)threadIdx.x * 16 + 1608; compute[1] = compute[1] + pad_temp_shared[8] * placeholder_shared[index18]; int index19 = (int)threadIdx.x * 16 + 9; compute[0] = compute[0] + pad_temp_shared[9] * placeholder_shared[index19]; int index20 = (int)threadIdx.x * 16 + 1609; compute[1] = compute[1] + pad_temp_shared[9] * placeholder_shared[index20]; int index21 = (int)threadIdx.x * 16 + 10; compute[0] = compute[0] + pad_temp_shared[10] * placeholder_shared[index21]; int index22 = (int)threadIdx.x * 16 + 1610; compute[1] = compute[1] + pad_temp_shared[10] * placeholder_shared[index22]; int index23 = (int)threadIdx.x * 16 + 11; compute[0] = compute[0] + pad_temp_shared[11] * placeholder_shared[index23]; int index24 = (int)threadIdx.x * 16 + 1611; compute[1] = compute[1] + pad_temp_shared[11] * placeholder_shared[index24]; int index25 = (int)threadIdx.x * 16 + 12; compute[0] = compute[0] + pad_temp_shared[12] * placeholder_shared[index25]; int index26 = (int)threadIdx.x * 16 + 1612; compute[1] = compute[1] + pad_temp_shared[12] * placeholder_shared[index26]; int index27 = (int)threadIdx.x * 16 + 13; compute[0] = compute[0] + pad_temp_shared[13] * placeholder_shared[index27]; int index28 = (int)threadIdx.x * 16 + 1613; compute[1] = compute[1] + pad_temp_shared[13] * placeholder_shared[index28]; int index29 = (int)threadIdx.x * 16 + 14; compute[0] = compute[0] + pad_temp_shared[14] * placeholder_shared[index29]; int index30 = (int)threadIdx.x * 16 + 1614; compute[1] = compute[1] + pad_temp_shared[14] * placeholder_shared[index30]; int index31 = (int)threadIdx.x * 16 + 15; compute[0] = compute[0] + pad_temp_shared[15] * placeholder_shared[index31]; int index32 = (int)threadIdx.x * 16 + 1615; compute[1] = compute[1] + pad_temp_shared[15] * placeholder_shared[index32]; } int index33 = (int)blockIdx.x / 169 * 33800 + (int)threadIdx.x * 169 + (int)blockIdx.x % 169; int index34 = (int)blockIdx.x / 169 * 200 + (int)threadIdx.x; T_relu[index33] = max(compute[0] + placeholder2[index34], 0.000000e+00f); T_relu[index33 + 16900] = max(compute[1] + placeholder2[index34 + 100], 0.000000e+00f); }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmph1f429rm
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[2080]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<7>; .reg .f32 %f<706>; .reg .b32 %r<51>; .reg .b64 %rd<24>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[2704]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[16000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd7, [default_function_kernel0_param_0]; ld.param.u64 %rd8, [default_function_kernel0_param_1]; ld.param.u64 %rd9, [default_function_kernel0_param_2]; ld.param.u64 %rd10, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd8; cvta.to.global.u64 %rd2, %rd9; add.u64 %rd3, %SPL, 0; mov.u64 %rd12, 0; mov.f32 %f14, 0f00000000; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(0+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 10, 1): st.local.v2.f32 [%rd3+(16+i*208)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(24+i*208)], %rd12; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(856+i*208)], {%f14, %f14}; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(32+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(48+i*208)], %rd12; mov.u32 %r47, 0; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 7, 1): st.local.u32 [%rd3+(56+i*4+j*208)], %r47; st.local.u64 [%rd3+1512], %rd12; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(1712+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 8, 1): st.local.v2.f32 [%rd3+(480+i*208)], {%f14, %f14}; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(64+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(488+i*208)], %rd12; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(80+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(96+i*208)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(1344+i*208)], %rd12; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd3+(1352+i*208)], %r47; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(104+i*208)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd3+(1356+i*208)], %r47; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 10, 1): st.local.v4.f32 [%rd3+(112+i*16+j*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 10, 1): st.local.v2.f32 [%rd3+(144+i*208)], {%f14, %f14}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(152+i*208)], %rd12; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(984+i*208)], {%f14, %f14}; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(160+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(176+i*208)], %rd12; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 8, 1): st.local.u32 [%rd3+(184+i*4+j*208)], %r47; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(1840+i*208)], {%f14, %f14, %f14, %f14}; for.size.1 i in range(0, 8, 1): st.local.v2.f32 [%rd3+(608+i*208)], {%f14, %f14}; st.local.u32 [%rd3+616], %r47; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(192+i*208)], {%f14, %f14, %f14, %f14}; st.local.u32 [%rd3+620], %r47; for.size.1 i in range(0, 7, 1): st.local.u64 [%rd3+(824+i*208)], %rd12; cvta.to.global.u64 %rd4, %rd7; cvta.to.global.u64 %rd5, %rd10; mov.u32 %r1, %tid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(2+i*19), %r1, (2+i*2); mov.u32 %r22, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r22, %r21; shl.b32 %r4, %r1, 9; mov.u32 %r23, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r23, %r21; mul.hi.s32 %r24, %r1, 1321528399; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 2; add.s32 %r7, %r26, %r25; mul.lo.s32 %r27, %r7, 13; sub.s32 %r6, %r1, %r27; shl.b32 %r8, %r7, 4; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 168; @%p1 bra $L__BB0_3; mad.lo.s32 %r28, %r47, 676, %r2; mul.wide.s32 %rd13, %r28, 4; add.s64 %rd14, %rd4, %rd13; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(15+i*1), [%rd14+(0+i*4)]; st.shared.f32 [%r3+(0+i*4)], %f(15+i*1); $L__BB0_3: setp.gt.s32 %p2, %r1, 24; shl.b32 %r29, %r47, 2; add.s32 %r30, %r29, %r4; mul.wide.s32 %rd15, %r30, 4; add.s64 %rd6, %rd1, %rd15; for.size.3 i in range(0, 3, 1): for.size.2 j in range(0, 4, 1): ld.global.nc.f32 %f(19+i*4+j*1), [%rd6+(0+i*665600+j*4)]; st.shared.f32 [%r5+(0+i*5200+j*4)], %f(19+i*4+j*1); @%p2 bra $L__BB0_5; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(31+i*1), [%rd6+(1996800+i*4)]; st.shared.f32 [%r5+(15600+i*4)], %f(31+i*1); $L__BB0_5: bar.sync 0; mov.u32 %r31, 0; mov.u32 %r48, %r31; $L__BB0_6: mad.lo.s32 %r33, %r48, 169, %r6; add.s32 %r11, %r48, %r8; shl.b32 %r34, %r33, 2; add.s32 %r36, %r22, %r34; for.size.1 i in range(0, 13, 1): ld.shared.f32 %f(1+i*1), [%r36+(0+i*52)]; mov.u32 %r49, %r31; $L__BB0_7: for.size.2 i in range(0, 2, 1): shl.b32 %r(37+i*2), %r(49+i*-11), 2; add.s32 %r(38+i*3), %r(11+i*12), %r(37+i*2); for.size.1 i in range(0, 9, 1): ld.shared.f32 %f(35+i*1), [%r41+(1600+i*1600)]; mul.lo.s32 %r42, %r49, 13; mul.wide.s32 %rd16, %r42, 4; add.s64 %rd17, %rd3, %rd16; ld.shared.f32 %f44, [%r41]; for.size.8 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): ld.local.f32 %f(45+i*20+j*2), [%rd17+(0+i*4+j*208)]; fma.rn.f32 %f(46+i*20+j*2), %f(1+j*1), %f(44+j*-9), %f(45+i*20+j*2); st.local.f32 [%rd17+(0+i*4+j*208)], %f(46+i*20+j*2); for.size.3 j in range(0, 8, 1): ld.local.f32 %f(49+i*20+j*2), [%rd17+(416+i*4+j*208)]; fma.rn.f32 %f(50+i*20+j*2), %f(1+j*1), %f(36+j*1), %f(49+i*20+j*2); st.local.f32 [%rd17+(416+i*4+j*208)], %f(50+i*20+j*2); add.s32 %r49, %r49, 1; setp.ne.s32 %p3, %r49, 4; @%p3 bra $L__BB0_7; add.s32 %r48, %r48, 1; setp.lt.u32 %p4, %r48, 4; @%p4 bra $L__BB0_6; add.s32 %r47, %r47, 1; setp.lt.u32 %p5, %r47, 128; @%p5 bra $L__BB0_1; shl.b32 %r16, %r7, 2; mad.lo.s32 %r17, %r7, 676, %r6; mov.u32 %r50, 0; $L__BB0_11: add.s32 %r44, %r50, %r16; mul.wide.s32 %rd18, %r44, 4; add.s64 %rd19, %rd5, %rd18; mad.lo.s32 %r45, %r50, 169, %r17; for.size.1 i in range(0, 9, 1): ld.global.nc.f32 %f(305+i*1), [%rd19+(400+i*400)]; mul.lo.s32 %r46, %r50, 13; mul.wide.s32 %rd20, %r46, 4; add.s64 %rd21, %rd3, %rd20; ld.local.f32 %f314, [%rd21]; ld.global.nc.f32 %f315, [%rd19]; add.f32 %f316, %f314, %f315; mov.f32 %f317, 0f00000000; max.f32 %f318, %f316, %f317; mul.wide.s32 %rd22, %r45, 4; add.s64 %rd23, %rd2, %rd22; for.size.4 i in range(0, 9, 1): st.global.f32 [%rd23+(0+i*67600)], %f(318+i*3); ld.local.f32 %f(319+i*3), [%rd21+(208+i*208)]; add.f32 %f(320+i*3), %f(319+i*3), %f(305+i*1); max.f32 %f(321+i*3), %f(320+i*3), %f317; for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd23+(608400+i*52+j*-608348)], %f(345+i*30+j*3); ld.local.f32 %f(346+i*30+j*3), [%rd21+(4+i*4+j*208)]; add.f32 %f(347+i*30+j*3), %f(346+i*30+j*3), %f(315+j*-10); max.f32 %f(348+i*30+j*3), %f(347+i*30+j*3), %f317; for.size.4 j in range(0, 8, 1): st.global.f32 [%rd23+(67652+i*52+j*67600)], %f(351+i*30+j*3); ld.local.f32 %f(352+i*30+j*3), [%rd21+(420+i*4+j*208)]; add.f32 %f(353+i*30+j*3), %f(352+i*30+j*3), %f(306+j*1); max.f32 %f(354+i*30+j*3), %f(353+i*30+j*3), %f317; st.global.f32 [%rd23+609024], %f705; add.s32 %r50, %r50, 1; setp.ne.s32 %p6, %r50, 4; @%p6 bra $L__BB0_11; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[520]; __shared__ float pad_temp_shared[676]; __shared__ float placeholder_shared[4000]; for (int ff_inner_init = 0; ff_inner_init < 4; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 52] = 0.000000e+00f; compute[index0 + 104] = 0.000000e+00f; compute[index0 + 156] = 0.000000e+00f; compute[index0 + 208] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 312] = 0.000000e+00f; compute[index0 + 364] = 0.000000e+00f; compute[index0 + 416] = 0.000000e+00f; compute[index0 + 468] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 169) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 676 + (int)threadIdx.x * 4; pad_temp_shared[index1] = placeholder[index2]; } if ((int)threadIdx.x < 169) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 676 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 1] = placeholder[index2 + 1]; } if ((int)threadIdx.x < 169) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 676 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 2] = placeholder[index2 + 2]; } if ((int)threadIdx.x < 169) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 676 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 3] = placeholder[index2 + 3]; } int index1 = (int)threadIdx.x * 4; int index3 = (int)threadIdx.x * 512 + rc_outer_outer * 4; placeholder_shared[index1] = placeholder1[index3]; placeholder_shared[index1 + 1] = placeholder1[index3 + 1]; placeholder_shared[index1 + 2] = placeholder1[index3 + 2]; placeholder_shared[index1 + 3] = placeholder1[index3 + 3]; placeholder_shared[index1 + 1300] = placeholder1[index3 + 166400]; placeholder_shared[index1 + 1301] = placeholder1[index3 + 166401]; placeholder_shared[index1 + 1302] = placeholder1[index3 + 166402]; placeholder_shared[index1 + 1303] = placeholder1[index3 + 166403]; placeholder_shared[index1 + 2600] = placeholder1[index3 + 332800]; placeholder_shared[index1 + 2601] = placeholder1[index3 + 332801]; placeholder_shared[index1 + 2602] = placeholder1[index3 + 332802]; placeholder_shared[index1 + 2603] = placeholder1[index3 + 332803]; if ((int)threadIdx.x < 25) { placeholder_shared[index1 + 3900] = placeholder1[index3 + 499200]; placeholder_shared[index1 + 3901] = placeholder1[index3 + 499201]; placeholder_shared[index1 + 3902] = placeholder1[index3 + 499202]; placeholder_shared[index1 + 3903] = placeholder1[index3 + 499203]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 4; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index4 = ff_inner * 13 + yy_inner; int index5 = rc_outer_inner * 169 + yy_inner * 13 + (int)threadIdx.x % 13; int index6 = (int)threadIdx.x / 13 * 16 + ff_inner * 4 + rc_outer_inner; compute[index4] = compute[index4] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[index4 + 52] = compute[index4 + 52] + (pad_temp_shared[index5] * placeholder_shared[index6 + 400]); compute[index4 + 104] = compute[index4 + 104] + (pad_temp_shared[index5] * placeholder_shared[index6 + 800]); compute[index4 + 156] = compute[index4 + 156] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1200]); compute[index4 + 208] = compute[index4 + 208] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1600]); compute[index4 + 260] = compute[index4 + 260] + (pad_temp_shared[index5] * placeholder_shared[index6 + 2000]); compute[index4 + 312] = compute[index4 + 312] + (pad_temp_shared[index5] * placeholder_shared[index6 + 2400]); compute[index4 + 364] = compute[index4 + 364] + (pad_temp_shared[index5] * placeholder_shared[index6 + 2800]); compute[index4 + 416] = compute[index4 + 416] + (pad_temp_shared[index5] * placeholder_shared[index6 + 3200]); compute[index4 + 468] = compute[index4 + 468] + (pad_temp_shared[index5] * placeholder_shared[index6 + 3600]); } } } } for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index7 = (int)threadIdx.x / 13 * 676 + ax1_inner * 169 + ax2_inner * 13 + (int)threadIdx.x % 13; int index8 = ax1_inner * 13 + ax2_inner; int index9 = (int)threadIdx.x / 13 * 4 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); T_relu[index7 + 16900] = max(compute[index8 + 52] + placeholder2[index9 + 100], 0.000000e+00f); T_relu[index7 + 33800] = max(compute[index8 + 104] + placeholder2[index9 + 200], 0.000000e+00f); T_relu[index7 + 50700] = max(compute[index8 + 156] + placeholder2[index9 + 300], 0.000000e+00f); T_relu[index7 + 67600] = max(compute[index8 + 208] + placeholder2[index9 + 400], 0.000000e+00f); T_relu[index7 + 84500] = max(compute[index8 + 260] + placeholder2[index9 + 500], 0.000000e+00f); T_relu[index7 + 101400] = max(compute[index8 + 312] + placeholder2[index9 + 600], 0.000000e+00f); T_relu[index7 + 118300] = max(compute[index8 + 364] + placeholder2[index9 + 700], 0.000000e+00f); T_relu[index7 + 135200] = max(compute[index8 + 416] + placeholder2[index9 + 800], 0.000000e+00f); T_relu[index7 + 152100] = max(compute[index8 + 468] + placeholder2[index9 + 900], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpquuzo03x
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 4 .b8 __local_depot0[84500]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<15>; .reg .f32 %f<358>; .reg .b32 %r<90>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[2704]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[16000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd15, [default_function_kernel0_param_0]; ld.param.u64 %rd16, [default_function_kernel0_param_1]; ld.param.u64 %rd17, [default_function_kernel0_param_2]; ld.param.u64 %rd18, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd15; cvta.to.global.u64 %rd2, %rd16; cvta.to.global.u64 %rd3, %rd17; cvta.to.global.u64 %rd4, %rd18; add.u64 %rd5, %SPL, 0; mov.u32 %r35, 0; mov.u32 %r80, %r35; $L__BB0_1: mul.lo.s32 %r2, %r80, 13; mov.u32 %r81, %r35; $L__BB0_2: add.s32 %r37, %r81, %r2; mul.wide.s32 %rd20, %r37, 4; add.s64 %rd21, %rd5, %rd20; mov.u32 %r38, 0; for.size.1 i in range(0, 65, 1): st.local.u32 [%rd21+(0+i*1300)], %r38; add.s32 %r81, %r81, 1; setp.ne.s32 %p1, %r81, 13; @%p1 bra $L__BB0_2; add.s32 %r80, %r80, 1; setp.lt.u32 %p2, %r80, 25; @%p2 bra $L__BB0_1; mov.u32 %r6, %tid.x; shr.u32 %r40, %r6, 2; and.b32 %r41, %r6, 3; bfi.b32 %r7, %r40, %r41, 9, 23; mul.lo.s32 %r8, %r6, 100; mov.u32 %r39, 0; mov.u32 %r82, %r39; $L__BB0_5: bar.sync 0; mad.lo.s32 %r10, %r82, 676, %r6; mov.u32 %r83, %r39; bra.uni $L__BB0_6; $L__BB0_15: add.s32 %r83, %r83, 4; $L__BB0_6: shl.b32 %r12, %r83, 3; add.s32 %r43, %r12, %r6; setp.gt.s32 %p3, %r43, 675; add.s32 %r44, %r10, %r12; mul.wide.s32 %rd22, %r44, 4; add.s64 %rd6, %rd1, %rd22; shl.b32 %r45, %r43, 2; mov.u32 %r46, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r13, %r46, %r45; @%p3 bra $L__BB0_8; ld.global.nc.f32 %f11, [%rd6]; st.shared.f32 [%r13], %f11; $L__BB0_8: add.s32 %r14, %r83, 1; setp.eq.s32 %p4, %r14, 85; @%p4 bra $L__BB0_16; shl.b32 %r47, %r14, 3; add.s32 %r48, %r47, %r6; setp.gt.s32 %p5, %r48, 675; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f12, [%rd6+32]; st.shared.f32 [%r13+32], %f12; $L__BB0_11: add.s32 %r50, %r43, 16; setp.gt.s32 %p6, %r50, 675; @%p6 bra $L__BB0_13; ld.global.nc.f32 %f13, [%rd6+64]; st.shared.f32 [%r13+64], %f13; $L__BB0_13: add.s32 %r52, %r43, 24; setp.gt.s32 %p7, %r52, 675; @%p7 bra $L__BB0_15; ld.global.nc.f32 %f14, [%rd6+96]; st.shared.f32 [%r13+96], %f14; bra.uni $L__BB0_15; $L__BB0_16: shl.b32 %r54, %r82, 2; add.s32 %r16, %r7, %r54; mov.u32 %r84, 0; $L__BB0_17: shl.b32 %r55, %r84, 10; add.s32 %r56, %r16, %r55; mul.wide.s32 %rd23, %r56, 4; add.s64 %rd24, %rd2, %rd23; ld.global.nc.f32 %f15, [%rd24]; shl.b32 %r57, %r84, 3; add.s32 %r58, %r57, %r6; shl.b32 %r59, %r58, 2; mov.u32 %r60, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r61, %r60, %r59; for.size.2 i in range(0, 3, 1): st.shared.f32 [%r61+(0+i*32)], %f(15+i*1); ld.global.nc.f32 %f(16+i*1), [%rd24+(4096+i*4096)]; st.shared.f32 [%r61+96], %f18; add.s32 %r84, %r84, 4; setp.ne.s32 %p8, %r84, 500; @%p8 bra $L__BB0_17; bar.sync 0; mov.u32 %r62, 0; mov.u32 %r85, %r62; $L__BB0_19: mul.lo.s32 %r20, %r85, 169; add.s32 %r21, %r85, %r8; mov.u32 %r86, %r62; $L__BB0_20: mul.lo.s32 %r23, %r86, 13; for.size.2 i in range(0, 2, 1): shl.b32 %r(65+i*2), %r(86+i*-20), 2; add.s32 %r(66+i*3), %r(21+i*39), %r(65+i*2); ld.shared.f32 %f1, [%r69]; mov.u32 %r87, 0; for.size.1 i in range(0, 4, 1): ld.shared.f32 %f(2+i*1), [%r69+(3200+i*3200)]; $L__BB0_21: mad.lo.s32 %r70, %r87, 13, %r20; add.s32 %r71, %r87, %r23; mul.wide.s32 %rd25, %r71, 4; add.s64 %rd26, %rd5, %rd25; shl.b32 %r72, %r70, 2; add.s32 %r74, %r46, %r72; for.size.4 i in range(0, 13, 1): ld.shared.f32 %f(19+i*3), [%r74+(0+i*4)]; ld.local.f32 %f(20+i*3), [%rd26+(0+i*1300)]; fma.rn.f32 %f(21+i*3), %f(19+i*3), %f1, %f(20+i*3); st.local.f32 [%rd26+(0+i*1300)], %f(21+i*3); for.size.4 i in range(0, 4, 1): for.size.3 j in range(0, 13, 1): ld.local.f32 %f(58+i*26+j*2), [%rd26+(16900+i*16900+j*1300)]; fma.rn.f32 %f(59+i*26+j*2), %f(19+j*3), %f(2+j*1), %f(58+i*26+j*2); st.local.f32 [%rd26+(16900+i*16900+j*1300)], %f(59+i*26+j*2); add.s32 %r87, %r87, 1; setp.ne.s32 %p9, %r87, 13; @%p9 bra $L__BB0_21; add.s32 %r86, %r86, 1; setp.lt.u32 %p10, %r86, 25; @%p10 bra $L__BB0_20; add.s32 %r85, %r85, 1; setp.lt.u32 %p11, %r85, 4; @%p11 bra $L__BB0_19; add.s32 %r82, %r82, 1; setp.lt.u32 %p12, %r82, 128; @%p12 bra $L__BB0_5; mul.lo.s32 %r29, %r6, 25; for.size.1 i in range(0, 2, 1): add.s64 %rd(7+i*1), %rd(5+i*-2), (41600+i*228824); mul.lo.s32 %r30, %r6, 4225; mov.u32 %r75, 0; mov.u32 %r88, %r75; $L__BB0_26: add.s32 %r77, %r88, %r29; mul.wide.s32 %rd27, %r77, 4; add.s64 %rd28, %rd4, %rd27; for.size.1 i in range(0, 5, 1): ld.global.nc.f32 %f(6+i*1), [%rd28+(0+i*800)]; mul.lo.s32 %r78, %r88, 13; mul.wide.s32 %rd29, %r78, 4; add.s64 %rd32, %rd7, %rd29; mad.lo.s32 %r79, %r88, 169, %r30; mul.wide.s32 %rd30, %r79, 4; add.s64 %rd31, %rd8, %rd30; mov.u32 %r89, %r75; $L__BB0_27: ld.local.f32 %f162, [%rd32+-41600]; add.f32 %f163, %f162, %f6; mov.f32 %f164, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(165+i*3), %f(163+i*4), %f164; st.global.f32 [%rd31+(-270424+i*4)], %f(165+i*3); ld.local.f32 %f(166+i*3), [%rd32+(-40300+i*1300)]; add.f32 %f(167+i*3), %f(166+i*3), %f6; for.size.4 i in range(0, 10, 1): max.f32 %f(171+i*3), %f(170+i*3), %f164; st.global.f32 [%rd31+(-270416+i*4)], %f(171+i*3); ld.local.f32 %f(172+i*3), [%rd32+(-37700+i*1300)]; add.f32 %f(173+i*3), %f(172+i*3), %f6; for.size.10 i in range(0, 4, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(201+i*39+j*3), %f(200+i*39+j*3), %f164; st.global.f32 [%rd31+(-270376+i*135200+j*135152)], %f(201+i*39+j*3); ld.local.f32 %f(202+i*39+j*3), [%rd32+(-24700+i*16900+j*1300)]; add.f32 %f(203+i*39+j*3), %f(202+i*39+j*3), %f(7+j*1); for.size.4 j in range(0, 11, 1): max.f32 %f(207+i*39+j*3), %f(206+i*39+j*3), %f164; st.global.f32 [%rd31+(-135220+i*135200+j*4)], %f(207+i*39+j*3); ld.local.f32 %f(208+i*39+j*3), [%rd32+(-22100+i*16900+j*1300)]; add.f32 %f(209+i*39+j*3), %f(208+i*39+j*3), %f(7+j*1); max.f32 %f357, %f356, %f164; st.global.f32 [%rd31+270424], %f357; for.size.1 i in range(0, 2, 1): add.s64 %rd(32+i*-1), %rd(32+i*-1), (4+i*48); add.s32 %r89, %r89, 1; setp.ne.s32 %p13, %r89, 13; @%p13 bra $L__BB0_27; add.s32 %r88, %r88, 1; setp.lt.u32 %p14, %r88, 25; @%p14 bra $L__BB0_26; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[21125]; __shared__ float pad_temp_shared[676]; __shared__ float placeholder_shared[4000]; for (int ff_inner_init = 0; ff_inner_init < 25; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { for (int vthread_s = 0; vthread_s < 65; ++vthread_s) { int index0 = vthread_s * 325 + ff_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 85; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 8 + (int)threadIdx.x < 676) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 8 + (int)threadIdx.x; int index2 = rc_outer_outer * 676 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 8 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 500; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 8 + (int)threadIdx.x; int index4 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 1024 + ((int)threadIdx.x >> 2) * 512 + rc_outer_outer * 4 + ((int)threadIdx.x & 3); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 25; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { for (int vthread_s1 = 0; vthread_s1 < 65; ++vthread_s1) { int index5 = vthread_s1 * 325 + ff_inner * 13 + yy_inner; int index6 = rc_outer_inner * 169 + yy_inner * 13 + vthread_s1 % 13; int index7 = vthread_s1 / 13 * 800 + (int)threadIdx.x * 100 + ff_inner * 4 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } for (int ax1_inner = 0; ax1_inner < 25; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int vthread_s2 = 0; vthread_s2 < 65; ++vthread_s2) { int index8 = vthread_s2 / 13 * 33800 + (int)threadIdx.x * 4225 + ax1_inner * 169 + ax2_inner * 13 + vthread_s2 % 13; int index9 = vthread_s2 * 325 + ax1_inner * 13 + ax2_inner; int index10 = vthread_s2 / 13 * 200 + (int)threadIdx.x * 25 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp2_tcj55b
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<222>; .reg .b32 %r<43>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[409600]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; ld.param.u64 %rd6, [default_function_kernel0_param_2]; ld.param.u64 %rd7, [default_function_kernel0_param_3]; mov.u32 %r1, %tid.x; cvta.to.global.u64 %rd9, %rd8; mul.wide.s32 %rd10, %r1, 4; add.s64 %rd1, %rd9, %rd10; ld.global.nc.f32 %f40, [%rd1]; shl.b32 %r18, %r1, 2; mov.u32 %r19, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r19, %r18; for.size.2 i in range(0, 32, 1): st.shared.f32 [%r2+(0+i*10400)], %f(40+i*1); ld.global.nc.f32 %f(41+i*1), [%rd1+(10400+i*10400)]; st.shared.f32 [%r2+332800], %f72; setp.gt.s32 %p1, %r1, 727; @%p1 bra $L__BB0_2; ld.global.nc.f32 %f73, [%rd1+343200]; st.shared.f32 [%r2+343200], %f73; $L__BB0_2: mov.u32 %r3, %ctaid.x; mad.lo.s32 %r20, %r3, 102400, %r1; cvta.to.global.u64 %rd11, %rd5; mul.wide.s32 %rd12, %r20, 4; add.s64 %rd2, %rd11, %rd12; ld.global.nc.f32 %f74, [%rd2]; mov.u32 %r22, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r22, %r18; for.size.2 i in range(0, 38, 1): st.shared.f32 [%r4+(0+i*10400)], %f(74+i*1); ld.global.nc.f32 %f(75+i*1), [%rd2+(10400+i*10400)]; st.shared.f32 [%r4+395200], %f112; setp.gt.s32 %p2, %r1, 999; @%p2 bra $L__BB0_4; ld.global.nc.f32 %f113, [%rd2+405600]; st.shared.f32 [%r4+405600], %f113; $L__BB0_4: bar.sync 0; mul.hi.s32 %r24, %r1, 1321528399; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 2; add.s32 %r5, %r26, %r25; mul.lo.s32 %r27, %r5, 13; sub.s32 %r28, %r1, %r27; shl.b32 %r29, %r5, 11; add.s32 %r31, %r22, %r29; add.s32 %r6, %r31, 4; mad.lo.s32 %r33, %r28, 52, %r19; add.s32 %r7, %r33, 724; cvta.to.global.u64 %rd3, %rd6; cvta.to.global.u64 %rd4, %rd7; mov.u32 %r39, 0; mov.f32 %f209, 0f00000000; for.size.1 i in range(0, 12, 1): mov.f32 %f(210+i*1), %f209; $L__BB0_5: shl.b32 %r35, %r39, 5; add.s32 %r41, %r6, %r35; mad.lo.s32 %r40, %r39, 5408, %r7; mov.u32 %r42, 4; $L__BB0_6: for.size.27 i in range(0, 2, 1): ld.shared.f32 %f(127+i*27), [%r41+(-4+i*4)]; ld.shared.f32 %f(128+i*27), [%r40+(-724+i*676)]; fma.rn.f32 %f(129+i*92), %f(128+i*27), %f(127+i*27), %f(221+i*-92); ld.shared.f32 %f(130+i*26), [%r40+(-720+i*676)]; fma.rn.f32 %f(131+i*89), %f(130+i*26), %f(127+i*27), %f(220+i*-89); ld.shared.f32 %f(132+i*25), [%r40+(-716+i*676)]; fma.rn.f32 %f(133+i*86), %f(132+i*25), %f(127+i*27), %f(219+i*-86); ld.shared.f32 %f(134+i*24), [%r40+(-712+i*676)]; fma.rn.f32 %f(135+i*83), %f(134+i*24), %f(127+i*27), %f(218+i*-83); ld.shared.f32 %f(136+i*23), [%r40+(-708+i*676)]; fma.rn.f32 %f(137+i*80), %f(136+i*23), %f(127+i*27), %f(217+i*-80); ld.shared.f32 %f(138+i*22), [%r40+(-704+i*676)]; fma.rn.f32 %f(139+i*77), %f(138+i*22), %f(127+i*27), %f(216+i*-77); ld.shared.f32 %f(140+i*21), [%r40+(-700+i*676)]; fma.rn.f32 %f(141+i*74), %f(140+i*21), %f(127+i*27), %f(215+i*-74); ld.shared.f32 %f(142+i*20), [%r40+(-696+i*676)]; fma.rn.f32 %f(143+i*71), %f(142+i*20), %f(127+i*27), %f(214+i*-71); ld.shared.f32 %f(144+i*19), [%r40+(-692+i*676)]; fma.rn.f32 %f(145+i*68), %f(144+i*19), %f(127+i*27), %f(213+i*-68); ld.shared.f32 %f(146+i*18), [%r40+(-688+i*676)]; fma.rn.f32 %f(147+i*65), %f(146+i*18), %f(127+i*27), %f(212+i*-65); ld.shared.f32 %f(148+i*17), [%r40+(-684+i*676)]; fma.rn.f32 %f(149+i*62), %f(148+i*17), %f(127+i*27), %f(211+i*-62); ld.shared.f32 %f(150+i*16), [%r40+(-680+i*676)]; fma.rn.f32 %f(151+i*59), %f(150+i*16), %f(127+i*27), %f(210+i*-59); ld.shared.f32 %f(152+i*15), [%r40+(-676+i*676)]; fma.rn.f32 %f(153+i*56), %f(152+i*15), %f(127+i*27), %f(209+i*-56); for.size.1 i in range(0, 2, 1): add.s32 %r(41+i*-1), %r(41+i*-1), (8+i*1344); add.s32 %r42, %r42, 8; setp.ne.s32 %p3, %r42, 36; @%p3 bra $L__BB0_6; add.s32 %r39, %r39, 1; setp.lt.u32 %p4, %r39, 64; @%p4 bra $L__BB0_5; mad.lo.s32 %r36, %r3, 200, %r5; mul.wide.s32 %rd13, %r36, 4; add.s64 %rd14, %rd4, %rd13; mul.lo.s32 %r37, %r1, 13; mad.lo.s32 %r38, %r3, 33800, %r37; ld.global.nc.f32 %f168, [%rd14]; add.f32 %f169, %f221, %f168; mov.f32 %f170, 0f00000000; max.f32 %f171, %f169, %f170; mul.wide.s32 %rd15, %r38, 4; add.s64 %rd16, %rd3, %rd15; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd16+(0+i*4)], %f(171+i*2); add.f32 %f(172+i*2), %f(220+i*-1), %f168; max.f32 %f(173+i*2), %f(172+i*2), %f170; st.global.f32 [%rd16+48], %f195; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[102400]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; int index0 = (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index0]; pad_temp_shared[index0 + 2600] = placeholder[index0 + 2600]; pad_temp_shared[index0 + 5200] = placeholder[index0 + 5200]; pad_temp_shared[index0 + 7800] = placeholder[index0 + 7800]; pad_temp_shared[index0 + 10400] = placeholder[index0 + 10400]; pad_temp_shared[index0 + 13000] = placeholder[index0 + 13000]; pad_temp_shared[index0 + 15600] = placeholder[index0 + 15600]; pad_temp_shared[index0 + 18200] = placeholder[index0 + 18200]; pad_temp_shared[index0 + 20800] = placeholder[index0 + 20800]; pad_temp_shared[index0 + 23400] = placeholder[index0 + 23400]; pad_temp_shared[index0 + 26000] = placeholder[index0 + 26000]; pad_temp_shared[index0 + 28600] = placeholder[index0 + 28600]; pad_temp_shared[index0 + 31200] = placeholder[index0 + 31200]; pad_temp_shared[index0 + 33800] = placeholder[index0 + 33800]; pad_temp_shared[index0 + 36400] = placeholder[index0 + 36400]; pad_temp_shared[index0 + 39000] = placeholder[index0 + 39000]; pad_temp_shared[index0 + 41600] = placeholder[index0 + 41600]; pad_temp_shared[index0 + 44200] = placeholder[index0 + 44200]; pad_temp_shared[index0 + 46800] = placeholder[index0 + 46800]; pad_temp_shared[index0 + 49400] = placeholder[index0 + 49400]; pad_temp_shared[index0 + 52000] = placeholder[index0 + 52000]; pad_temp_shared[index0 + 54600] = placeholder[index0 + 54600]; pad_temp_shared[index0 + 57200] = placeholder[index0 + 57200]; pad_temp_shared[index0 + 59800] = placeholder[index0 + 59800]; pad_temp_shared[index0 + 62400] = placeholder[index0 + 62400]; pad_temp_shared[index0 + 65000] = placeholder[index0 + 65000]; pad_temp_shared[index0 + 67600] = placeholder[index0 + 67600]; pad_temp_shared[index0 + 70200] = placeholder[index0 + 70200]; pad_temp_shared[index0 + 72800] = placeholder[index0 + 72800]; pad_temp_shared[index0 + 75400] = placeholder[index0 + 75400]; pad_temp_shared[index0 + 78000] = placeholder[index0 + 78000]; pad_temp_shared[index0 + 80600] = placeholder[index0 + 80600]; pad_temp_shared[index0 + 83200] = placeholder[index0 + 83200]; if ((int)threadIdx.x < 728) { pad_temp_shared[index0 + 85800] = placeholder[index0 + 85800]; } int index1 = (int)blockIdx.x * 102400 + (int)threadIdx.x; placeholder_shared[index0] = placeholder1[index1]; placeholder_shared[index0 + 2600] = placeholder1[index1 + 2600]; placeholder_shared[index0 + 5200] = placeholder1[index1 + 5200]; placeholder_shared[index0 + 7800] = placeholder1[index1 + 7800]; placeholder_shared[index0 + 10400] = placeholder1[index1 + 10400]; placeholder_shared[index0 + 13000] = placeholder1[index1 + 13000]; placeholder_shared[index0 + 15600] = placeholder1[index1 + 15600]; placeholder_shared[index0 + 18200] = placeholder1[index1 + 18200]; placeholder_shared[index0 + 20800] = placeholder1[index1 + 20800]; placeholder_shared[index0 + 23400] = placeholder1[index1 + 23400]; placeholder_shared[index0 + 26000] = placeholder1[index1 + 26000]; placeholder_shared[index0 + 28600] = placeholder1[index1 + 28600]; placeholder_shared[index0 + 31200] = placeholder1[index1 + 31200]; placeholder_shared[index0 + 33800] = placeholder1[index1 + 33800]; placeholder_shared[index0 + 36400] = placeholder1[index1 + 36400]; placeholder_shared[index0 + 39000] = placeholder1[index1 + 39000]; placeholder_shared[index0 + 41600] = placeholder1[index1 + 41600]; placeholder_shared[index0 + 44200] = placeholder1[index1 + 44200]; placeholder_shared[index0 + 46800] = placeholder1[index1 + 46800]; placeholder_shared[index0 + 49400] = placeholder1[index1 + 49400]; placeholder_shared[index0 + 52000] = placeholder1[index1 + 52000]; placeholder_shared[index0 + 54600] = placeholder1[index1 + 54600]; placeholder_shared[index0 + 57200] = placeholder1[index1 + 57200]; placeholder_shared[index0 + 59800] = placeholder1[index1 + 59800]; placeholder_shared[index0 + 62400] = placeholder1[index1 + 62400]; placeholder_shared[index0 + 65000] = placeholder1[index1 + 65000]; placeholder_shared[index0 + 67600] = placeholder1[index1 + 67600]; placeholder_shared[index0 + 70200] = placeholder1[index1 + 70200]; placeholder_shared[index0 + 72800] = placeholder1[index1 + 72800]; placeholder_shared[index0 + 75400] = placeholder1[index1 + 75400]; placeholder_shared[index0 + 78000] = placeholder1[index1 + 78000]; placeholder_shared[index0 + 80600] = placeholder1[index1 + 80600]; placeholder_shared[index0 + 83200] = placeholder1[index1 + 83200]; placeholder_shared[index0 + 85800] = placeholder1[index1 + 85800]; placeholder_shared[index0 + 88400] = placeholder1[index1 + 88400]; placeholder_shared[index0 + 91000] = placeholder1[index1 + 91000]; placeholder_shared[index0 + 93600] = placeholder1[index1 + 93600]; placeholder_shared[index0 + 96200] = placeholder1[index1 + 96200]; placeholder_shared[index0 + 98800] = placeholder1[index1 + 98800]; if ((int)threadIdx.x < 1000) { placeholder_shared[index0 + 101400] = placeholder1[index1 + 101400]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { int index2 = rc_outer_inner * 1352 + rc_inner * 169 + (int)threadIdx.x % 13 * 13; int index3 = (int)threadIdx.x / 13 * 512 + rc_outer_inner * 8 + rc_inner; compute[0] = compute[0] + pad_temp_shared[index2] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index2 + 1] * placeholder_shared[index3]); compute[2] = compute[2] + (pad_temp_shared[index2 + 2] * placeholder_shared[index3]); compute[3] = compute[3] + (pad_temp_shared[index2 + 3] * placeholder_shared[index3]); compute[4] = compute[4] + (pad_temp_shared[index2 + 4] * placeholder_shared[index3]); compute[5] = compute[5] + (pad_temp_shared[index2 + 5] * placeholder_shared[index3]); compute[6] = compute[6] + (pad_temp_shared[index2 + 6] * placeholder_shared[index3]); compute[7] = compute[7] + (pad_temp_shared[index2 + 7] * placeholder_shared[index3]); compute[8] = compute[8] + (pad_temp_shared[index2 + 8] * placeholder_shared[index3]); compute[9] = compute[9] + (pad_temp_shared[index2 + 9] * placeholder_shared[index3]); compute[10] = compute[10] + (pad_temp_shared[index2 + 10] * placeholder_shared[index3]); compute[11] = compute[11] + (pad_temp_shared[index2 + 11] * placeholder_shared[index3]); compute[12] = compute[12] + (pad_temp_shared[index2 + 12] * placeholder_shared[index3]); } } for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index4 = (int)blockIdx.x * 33800 + (int)threadIdx.x * 13 + ax3_inner; int index5 = ax3_inner; int index6 = (int)blockIdx.x * 200 + (int)threadIdx.x / 13; T_relu[index4] = max(compute[index5] + placeholder2[index6], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp248d5s34
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<257>; .reg .b32 %r<37>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[26624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048000]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; ld.param.u64 %rd6, [default_function_kernel0_param_2]; ld.param.u64 %rd7, [default_function_kernel0_param_3]; mov.u32 %r1, %tid.x; mov.u32 %r2, %ctaid.x; mad.lo.s32 %r15, %r1, 13, %r2; cvta.to.global.u64 %rd9, %rd8; mul.wide.s32 %rd10, %r15, 4; add.s64 %rd1, %rd9, %rd10; ld.global.nc.f32 %f79, [%rd1]; shl.b32 %r16, %r1, 2; mov.u32 %r17, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r17, %r16; for.size.2 i in range(0, 12, 1): st.shared.f32 [%r3+(0+i*2000)], %f(79+i*1); ld.global.nc.f32 %f(80+i*1), [%rd1+(26000+i*26000)]; st.shared.f32 [%r3+24000], %f91; setp.gt.s32 %p1, %r1, 155; @%p1 bra $L__BB0_2; ld.global.nc.f32 %f92, [%rd1+338000]; st.shared.f32 [%r3+26000], %f92; $L__BB0_2: cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd6; cvta.to.global.u64 %rd4, %rd5; mov.u32 %r33, 0; mov.u32 %r21, _ZZ24default_function_kernel0E18placeholder_shared; $L__BB0_3: mad.lo.s32 %r19, %r33, 500, %r1; mul.wide.s32 %rd11, %r19, 4; add.s64 %rd12, %rd4, %rd11; ld.global.nc.f32 %f93, [%rd12]; shl.b32 %r20, %r19, 2; add.s32 %r22, %r21, %r20; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r22+(0+i*2000)], %f(93+i*1); ld.global.nc.f32 %f(94+i*1), [%rd12+(2000+i*2000)]; st.shared.f32 [%r22+30000], %f108; add.s32 %r33, %r33, 16; setp.ne.s32 %p2, %r33, 1024; @%p2 bra $L__BB0_3; bar.sync 0; shl.b32 %r24, %r1, 12; add.s32 %r6, %r21, %r24; mov.u32 %r34, 0; mov.f32 %f231, 0f00000000; for.size.1 i in range(0, 25, 1): mov.f32 %f(232+i*1), %f231; $L__BB0_5: shl.b32 %r27, %r34, 5; add.s32 %r8, %r6, %r27; mad.lo.s32 %r29, %r34, 416, %r17; add.s32 %r35, %r29, 24; mov.u32 %r36, 2048; $L__BB0_6: add.s32 %r30, %r8, %r36; ld.shared.f32 %f135, [%r30+-2048]; for.size.2 i in range(0, 13, 1): ld.shared.f32 %f(136+i*1), [%r35+(-24+i*4)]; fma.rn.f32 %f(256+i*-1), %f(136+i*1), %f135, %f(256+i*-1); ld.shared.f32 %f149, [%r30]; for.size.1 i in range(0, 13, 1): fma.rn.f32 %f(243+i*-1), %f(136+i*1), %f149, %f(243+i*-1); for.size.1 i in range(0, 2, 1): add.s32 %r(35+i*1), %r(35+i*1), (52+i*-48); setp.ne.s32 %p3, %r36, 2080; @%p3 bra $L__BB0_6; add.s32 %r34, %r34, 1; setp.lt.u32 %p4, %r34, 64; @%p4 bra $L__BB0_5; shl.b32 %r31, %r1, 1; mad.lo.s32 %r32, %r1, 338, %r2; mul.wide.s32 %rd13, %r31, 4; add.s64 %rd14, %rd2, %rd13; ld.global.nc.f32 %f150, [%rd14]; add.f32 %f151, %f256, %f150; mov.f32 %f152, 0f00000000; max.f32 %f153, %f151, %f152; mul.wide.s32 %rd15, %r32, 4; add.s64 %rd16, %rd3, %rd15; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd16+(0+i*52)], %f(153+i*2); add.f32 %f(154+i*2), %f(255+i*-1), %f150; max.f32 %f(155+i*2), %f(154+i*2), %f152; st.global.f32 [%rd16+624], %f177; ld.global.nc.f32 %f178, [%rd14+4]; for.size.3 i in range(0, 13, 1): add.f32 %f(179+i*2), %f(243+i*-1), %f178; max.f32 %f(180+i*2), %f(179+i*2), %f152; st.global.f32 [%rd16+(676+i*52)], %f(180+i*2); ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[26]; __shared__ float pad_temp_shared[6656]; __shared__ float placeholder_shared[512000]; for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = ff_inner_init * 13; compute[index0] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; compute[index0 + 7] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 10] = 0.000000e+00f; compute[index0 + 11] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; } int index1 = (int)threadIdx.x; int index2 = (int)threadIdx.x * 13 + (int)blockIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 500] = placeholder[index2 + 6500]; pad_temp_shared[index1 + 1000] = placeholder[index2 + 13000]; pad_temp_shared[index1 + 1500] = placeholder[index2 + 19500]; pad_temp_shared[index1 + 2000] = placeholder[index2 + 26000]; pad_temp_shared[index1 + 2500] = placeholder[index2 + 32500]; pad_temp_shared[index1 + 3000] = placeholder[index2 + 39000]; pad_temp_shared[index1 + 3500] = placeholder[index2 + 45500]; pad_temp_shared[index1 + 4000] = placeholder[index2 + 52000]; pad_temp_shared[index1 + 4500] = placeholder[index2 + 58500]; pad_temp_shared[index1 + 5000] = placeholder[index2 + 65000]; pad_temp_shared[index1 + 5500] = placeholder[index2 + 71500]; pad_temp_shared[index1 + 6000] = placeholder[index2 + 78000]; if ((int)threadIdx.x < 156) { pad_temp_shared[index1 + 6500] = placeholder[index2 + 84500]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 1024; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 500 + (int)threadIdx.x; placeholder_shared[index3] = placeholder1[index3]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index4 = ff_inner * 13; int index5 = rc_outer_inner * 104 + rc_inner * 13; int index6 = (int)threadIdx.x * 1024 + ff_inner * 512 + rc_outer_inner * 8 + rc_inner; compute[index4] = compute[index4] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[index4 + 1] = compute[index4 + 1] + (pad_temp_shared[index5 + 1] * placeholder_shared[index6]); compute[index4 + 2] = compute[index4 + 2] + (pad_temp_shared[index5 + 2] * placeholder_shared[index6]); compute[index4 + 3] = compute[index4 + 3] + (pad_temp_shared[index5 + 3] * placeholder_shared[index6]); compute[index4 + 4] = compute[index4 + 4] + (pad_temp_shared[index5 + 4] * placeholder_shared[index6]); compute[index4 + 5] = compute[index4 + 5] + (pad_temp_shared[index5 + 5] * placeholder_shared[index6]); compute[index4 + 6] = compute[index4 + 6] + (pad_temp_shared[index5 + 6] * placeholder_shared[index6]); compute[index4 + 7] = compute[index4 + 7] + (pad_temp_shared[index5 + 7] * placeholder_shared[index6]); compute[index4 + 8] = compute[index4 + 8] + (pad_temp_shared[index5 + 8] * placeholder_shared[index6]); compute[index4 + 9] = compute[index4 + 9] + (pad_temp_shared[index5 + 9] * placeholder_shared[index6]); compute[index4 + 10] = compute[index4 + 10] + (pad_temp_shared[index5 + 10] * placeholder_shared[index6]); compute[index4 + 11] = compute[index4 + 11] + (pad_temp_shared[index5 + 11] * placeholder_shared[index6]); compute[index4 + 12] = compute[index4 + 12] + (pad_temp_shared[index5 + 12] * placeholder_shared[index6]); } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index7 = (int)threadIdx.x * 338 + ax1_inner * 169 + ax2_inner * 13 + (int)blockIdx.x; int index8 = ax1_inner * 13 + ax2_inner; int index9 = (int)threadIdx.x * 2 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmppqgcaqsm
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[2080]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<8>; .reg .f32 %f<981>; .reg .b32 %r<55>; .reg .b64 %rd<26>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1352]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[8000]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd9, [default_function_kernel0_param_1]; ld.param.u64 %rd10, [default_function_kernel0_param_2]; ld.param.u64 %rd11, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd9; cvta.to.global.u64 %rd2, %rd10; add.u64 %rd3, %SPL, 0; mov.u64 %rd13, 0; mov.f32 %f27, 0f00000000; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(0+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 10, 1): st.local.v2.f32 [%rd3+(16+i*208)], {%f27, %f27}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(24+i*208)], %rd13; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(856+i*208)], {%f27, %f27}; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(32+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(48+i*208)], %rd13; mov.u32 %r51, 0; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 7, 1): st.local.u32 [%rd3+(56+i*4+j*208)], %r51; st.local.u64 [%rd3+1512], %rd13; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(1712+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 8, 1): st.local.v2.f32 [%rd3+(480+i*208)], {%f27, %f27}; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(64+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(488+i*208)], %rd13; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(80+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(96+i*208)], {%f27, %f27}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(1344+i*208)], %rd13; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd3+(1352+i*208)], %r51; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(104+i*208)], {%f27, %f27}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd3+(1356+i*208)], %r51; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 10, 1): st.local.v4.f32 [%rd3+(112+i*16+j*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 10, 1): st.local.v2.f32 [%rd3+(144+i*208)], {%f27, %f27}; for.size.1 i in range(0, 4, 1): st.local.u64 [%rd3+(152+i*208)], %rd13; for.size.1 i in range(0, 6, 1): st.local.v2.f32 [%rd3+(984+i*208)], {%f27, %f27}; for.size.1 i in range(0, 10, 1): st.local.v4.f32 [%rd3+(160+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 8, 1): st.local.u64 [%rd3+(176+i*208)], %rd13; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 8, 1): st.local.u32 [%rd3+(184+i*4+j*208)], %r51; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(1840+i*208)], {%f27, %f27, %f27, %f27}; for.size.1 i in range(0, 8, 1): st.local.v2.f32 [%rd3+(608+i*208)], {%f27, %f27}; st.local.u32 [%rd3+616], %r51; for.size.1 i in range(0, 2, 1): st.local.v4.f32 [%rd3+(192+i*208)], {%f27, %f27, %f27, %f27}; st.local.u32 [%rd3+620], %r51; for.size.1 i in range(0, 7, 1): st.local.u64 [%rd3+(824+i*208)], %rd13; cvta.to.global.u64 %rd4, %rd8; cvta.to.global.u64 %rd5, %rd11; mov.u32 %r1, %tid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(2+i*20), %r1, (2+i*2); mov.u32 %r23, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r23, %r22; shl.b32 %r4, %r1, 10; mov.u32 %r24, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r24, %r22; mul.hi.s32 %r25, %r1, 1321528399; shr.u32 %r26, %r25, 31; shr.s32 %r27, %r25, 2; add.s32 %r7, %r27, %r26; mul.lo.s32 %r28, %r7, 13; sub.s32 %r6, %r1, %r28; for.size.1 i in range(0, 2, 1): shl.b32 %r(8+i*21), %r(7+i*-1), (3+i*-1); add.s32 %r9, %r23, %r29; or.b32 %r10, %r8, 1; $L__BB0_1: bar.sync 0; mad.lo.s32 %r30, %r51, 338, %r2; mul.wide.s32 %rd14, %r30, 4; add.s64 %rd6, %rd4, %rd14; setp.gt.s32 %p1, %r1, 84; @%p1 bra $L__BB0_3; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(28+i*1), [%rd6+(0+i*4)]; st.shared.f32 [%r3+(0+i*4)], %f(28+i*1); $L__BB0_3: setp.gt.s32 %p2, %r1, 83; @%p2 bra $L__BB0_5; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(30+i*1), [%rd6+(8+i*4)]; st.shared.f32 [%r3+(8+i*4)], %f(30+i*1); $L__BB0_5: setp.gt.s32 %p3, %r1, 174; shl.b32 %r31, %r51, 1; add.s32 %r32, %r31, %r4; mul.wide.s32 %rd15, %r32, 4; add.s64 %rd7, %rd1, %rd15; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.global.nc.f32 %f(32+i*2+j*1), [%rd7+(0+i*2048+j*4)]; st.shared.f32 [%r5+(0+i*8+j*4)], %f(32+i*2+j*1); @%p3 bra $L__BB0_7; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.global.nc.f32 %f(36+i*2+j*1), [%rd7+(1331200+i*2048+j*4)]; st.shared.f32 [%r5+(5200+i*8+j*4)], %f(36+i*2+j*1); $L__BB0_7: bar.sync 0; ld.shared.f32 %f1, [%r9]; mov.u32 %r52, 0; for.size.1 i in range(0, 12, 1): ld.shared.f32 %f(2+i*1), [%r9+(52+i*52)]; $L__BB0_8: for.size.2 i in range(0, 2, 1): shl.b32 %r(34+i*2), %r(52+i*-17), (1+i*1); add.s32 %r(35+i*3), %r(8+i*16), %r(34+i*2); for.size.1 i in range(0, 9, 1): ld.shared.f32 %f(40+i*1), [%r38+(800+i*800)]; mul.lo.s32 %r39, %r52, 13; mul.wide.s32 %rd16, %r39, 4; add.s64 %rd17, %rd3, %rd16; ld.shared.f32 %f49, [%r38]; for.size.8 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): ld.local.f32 %f(50+i*20+j*2), [%rd17+(0+i*4+j*208)]; fma.rn.f32 %f(51+i*20+j*2), %f(1+j*1), %f(49+j*-9), %f(50+i*20+j*2); st.local.f32 [%rd17+(0+i*4+j*208)], %f(51+i*20+j*2); for.size.3 j in range(0, 8, 1): ld.local.f32 %f(54+i*20+j*2), [%rd17+(416+i*4+j*208)]; fma.rn.f32 %f(55+i*20+j*2), %f(1+j*1), %f(41+j*1), %f(54+i*20+j*2); st.local.f32 [%rd17+(416+i*4+j*208)], %f(55+i*20+j*2); add.s32 %r52, %r52, 1; setp.ne.s32 %p4, %r52, 4; @%p4 bra $L__BB0_8; ld.shared.f32 %f14, [%r9+676]; mov.u32 %r53, 0; for.size.1 i in range(0, 12, 1): ld.shared.f32 %f(15+i*1), [%r9+(728+i*52)]; $L__BB0_10: for.size.2 i in range(0, 2, 1): shl.b32 %r(41+i*2), %r(53+i*-11), (1+i*1); add.s32 %r(42+i*3), %r(10+i*14), %r(41+i*2); for.size.1 i in range(0, 9, 1): ld.shared.f32 %f(310+i*1), [%r45+(800+i*800)]; mul.lo.s32 %r46, %r53, 13; mul.wide.s32 %rd18, %r46, 4; add.s64 %rd19, %rd3, %rd18; ld.shared.f32 %f319, [%r45]; for.size.8 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): ld.local.f32 %f(320+i*20+j*2), [%rd19+(0+i*4+j*208)]; fma.rn.f32 %f(321+i*20+j*2), %f(14+j*1), %f(319+j*-9), %f(320+i*20+j*2); st.local.f32 [%rd19+(0+i*4+j*208)], %f(321+i*20+j*2); for.size.3 j in range(0, 8, 1): ld.local.f32 %f(324+i*20+j*2), [%rd19+(416+i*4+j*208)]; fma.rn.f32 %f(325+i*20+j*2), %f(14+j*1), %f(311+j*1), %f(324+i*20+j*2); st.local.f32 [%rd19+(416+i*4+j*208)], %f(325+i*20+j*2); add.s32 %r53, %r53, 1; setp.ne.s32 %p5, %r53, 4; @%p5 bra $L__BB0_10; add.s32 %r51, %r51, 1; setp.lt.u32 %p6, %r51, 256; @%p6 bra $L__BB0_1; shl.b32 %r17, %r7, 2; mad.lo.s32 %r18, %r7, 676, %r6; mov.u32 %r54, 0; $L__BB0_13: add.s32 %r48, %r54, %r17; mul.wide.s32 %rd20, %r48, 4; add.s64 %rd21, %rd5, %rd20; mad.lo.s32 %r49, %r54, 169, %r18; for.size.1 i in range(0, 9, 1): ld.global.nc.f32 %f(580+i*1), [%rd21+(400+i*400)]; mul.lo.s32 %r50, %r54, 13; mul.wide.s32 %rd22, %r50, 4; add.s64 %rd23, %rd3, %rd22; ld.local.f32 %f589, [%rd23]; ld.global.nc.f32 %f590, [%rd21]; add.f32 %f591, %f589, %f590; mov.f32 %f592, 0f00000000; max.f32 %f593, %f591, %f592; mul.wide.s32 %rd24, %r49, 4; add.s64 %rd25, %rd2, %rd24; for.size.4 i in range(0, 9, 1): st.global.f32 [%rd25+(0+i*67600)], %f(593+i*3); ld.local.f32 %f(594+i*3), [%rd23+(208+i*208)]; add.f32 %f(595+i*3), %f(594+i*3), %f(580+i*1); max.f32 %f(596+i*3), %f(595+i*3), %f592; for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd25+(608400+i*52+j*-608348)], %f(620+i*30+j*3); ld.local.f32 %f(621+i*30+j*3), [%rd23+(4+i*4+j*208)]; add.f32 %f(622+i*30+j*3), %f(621+i*30+j*3), %f(590+j*-10); max.f32 %f(623+i*30+j*3), %f(622+i*30+j*3), %f592; for.size.4 j in range(0, 8, 1): st.global.f32 [%rd25+(67652+i*52+j*67600)], %f(626+i*30+j*3); ld.local.f32 %f(627+i*30+j*3), [%rd23+(420+i*4+j*208)]; add.f32 %f(628+i*30+j*3), %f(627+i*30+j*3), %f(581+j*1); max.f32 %f(629+i*30+j*3), %f(628+i*30+j*3), %f592; st.global.f32 [%rd25+609024], %f980; add.s32 %r54, %r54, 1; setp.ne.s32 %p7, %r54, 4; @%p7 bra $L__BB0_13; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[520]; __shared__ float pad_temp_shared[338]; __shared__ float placeholder_shared[2000]; for (int ff_inner_init = 0; ff_inner_init < 4; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 52] = 0.000000e+00f; compute[index0 + 104] = 0.000000e+00f; compute[index0 + 156] = 0.000000e+00f; compute[index0 + 208] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 312] = 0.000000e+00f; compute[index0 + 364] = 0.000000e+00f; compute[index0 + 416] = 0.000000e+00f; compute[index0 + 468] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 256; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 85) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 338 + (int)threadIdx.x * 4; pad_temp_shared[index1] = placeholder[index2]; } if ((int)threadIdx.x < 85) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 338 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 1] = placeholder[index2 + 1]; } if ((int)threadIdx.x < 84) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 338 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 2] = placeholder[index2 + 2]; } if ((int)threadIdx.x < 84) { int index1 = (int)threadIdx.x * 4; int index2 = rc_outer_outer * 338 + (int)threadIdx.x * 4; pad_temp_shared[index1 + 3] = placeholder[index2 + 3]; } int index1 = (int)threadIdx.x * 4; int index3 = (int)threadIdx.x * 1024 + rc_outer_outer * 2; placeholder_shared[index1] = placeholder1[index3]; placeholder_shared[index1 + 1] = placeholder1[index3 + 1]; placeholder_shared[index1 + 2] = placeholder1[index3 + 512]; placeholder_shared[index1 + 3] = placeholder1[index3 + 513]; if ((int)threadIdx.x < 175) { placeholder_shared[index1 + 1300] = placeholder1[index3 + 332800]; } if ((int)threadIdx.x < 175) { placeholder_shared[index1 + 1301] = placeholder1[index3 + 332801]; } if ((int)threadIdx.x < 175) { placeholder_shared[index1 + 1302] = placeholder1[index3 + 333312]; } if ((int)threadIdx.x < 175) { placeholder_shared[index1 + 1303] = placeholder1[index3 + 333313]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 4; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index4 = ff_inner * 13 + yy_inner; int index5 = rc_outer_inner * 169 + yy_inner * 13 + (int)threadIdx.x % 13; int index6 = (int)threadIdx.x / 13 * 8 + ff_inner * 2 + rc_outer_inner; compute[index4] = compute[index4] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[index4 + 52] = compute[index4 + 52] + (pad_temp_shared[index5] * placeholder_shared[index6 + 200]); compute[index4 + 104] = compute[index4 + 104] + (pad_temp_shared[index5] * placeholder_shared[index6 + 400]); compute[index4 + 156] = compute[index4 + 156] + (pad_temp_shared[index5] * placeholder_shared[index6 + 600]); compute[index4 + 208] = compute[index4 + 208] + (pad_temp_shared[index5] * placeholder_shared[index6 + 800]); compute[index4 + 260] = compute[index4 + 260] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1000]); compute[index4 + 312] = compute[index4 + 312] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1200]); compute[index4 + 364] = compute[index4 + 364] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1400]); compute[index4 + 416] = compute[index4 + 416] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1600]); compute[index4 + 468] = compute[index4 + 468] + (pad_temp_shared[index5] * placeholder_shared[index6 + 1800]); } } } } for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index7 = (int)threadIdx.x / 13 * 676 + ax1_inner * 169 + ax2_inner * 13 + (int)threadIdx.x % 13; int index8 = ax1_inner * 13 + ax2_inner; int index9 = (int)threadIdx.x / 13 * 4 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); T_relu[index7 + 16900] = max(compute[index8 + 52] + placeholder2[index9 + 100], 0.000000e+00f); T_relu[index7 + 33800] = max(compute[index8 + 104] + placeholder2[index9 + 200], 0.000000e+00f); T_relu[index7 + 50700] = max(compute[index8 + 156] + placeholder2[index9 + 300], 0.000000e+00f); T_relu[index7 + 67600] = max(compute[index8 + 208] + placeholder2[index9 + 400], 0.000000e+00f); T_relu[index7 + 84500] = max(compute[index8 + 260] + placeholder2[index9 + 500], 0.000000e+00f); T_relu[index7 + 101400] = max(compute[index8 + 312] + placeholder2[index9 + 600], 0.000000e+00f); T_relu[index7 + 118300] = max(compute[index8 + 364] + placeholder2[index9 + 700], 0.000000e+00f); T_relu[index7 + 135200] = max(compute[index8 + 416] + placeholder2[index9 + 800], 0.000000e+00f); T_relu[index7 + 152100] = max(compute[index8 + 468] + placeholder2[index9 + 900], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp6rwuh1d4
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<178>; .reg .b32 %r<31>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[2048]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048000]; ld.param.u64 %rd4, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; ld.param.u64 %rd6, [default_function_kernel0_param_2]; ld.param.u64 %rd7, [default_function_kernel0_param_3]; mov.u32 %r1, %tid.x; setp.gt.s32 %p1, %r1, 31; @%p1 bra $L__BB0_2; mov.u32 %r10, %ctaid.x; mad.lo.s32 %r11, %r1, 2704, %r10; cvta.to.global.u64 %rd8, %rd4; mul.wide.s32 %rd9, %r11, 4; add.s64 %rd10, %rd8, %rd9; ld.global.nc.f32 %f21, [%rd10]; shl.b32 %r12, %r1, 6; mov.u32 %r13, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r14, %r13, %r12; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r14+(0+i*4)], %f(21+i*1); ld.global.nc.f32 %f(22+i*1), [%rd10+(676+i*676)]; st.shared.f32 [%r14+60], %f36; $L__BB0_2: shl.b32 %r16, %r1, 2; mov.u32 %r17, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r18, %r17, %r16; add.s32 %r28, %r18, 3200; cvta.to.global.u64 %rd11, %rd5; mul.wide.s32 %rd12, %r1, 4; add.s64 %rd13, %rd11, %rd12; add.s64 %rd20, %rd13, 3200; mov.u32 %r29, 0; $L__BB0_3: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(37+i*1), [%rd20+(-3200+i*400)]; st.shared.f32 [%r28+(-3200+i*400)], %f(37+i*1); add.s32 %r28, %r28, 6400; add.s64 %rd20, %rd20, 6400; add.s32 %r29, %r29, 16; setp.ne.s32 %p2, %r29, 5120; @%p2 bra $L__BB0_3; bar.sync 0; mad.lo.s32 %r7, %r1, 10240, %r17; mov.f32 %f168, 0f00000000; mov.u32 %r30, 1024000; for.size.1 i in range(0, 9, 1): mov.f32 %f(169+i*1), %f168; $L__BB0_5: mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; for.size.1 i in range(0, 2, 1): add.s32 %r(22+i*1), %r(21+i*-14), %r30; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(63+i*1), [%r(23+i*-1)+-1024000]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(65+i*2), %f64, %f(63+i*3), %f(177+i*-1); ld.shared.f32 %f(66+i*2), [%r23+(0+i*-1021952)]; for.size.3 i in range(0, 3, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(69+i*4+j*2), %f64, %f(68+i*4+j*2), %f(175+i*-2+j*-1); ld.shared.f32 %f(70+i*4+j*2), [%r23+(2048+i*2048+j*-1021952)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(81+i*2), %f64, %f(80+i*2), %f(169+i*-1); ld.shared.f32 %f(82+i*2), [%r23+(8192+i*-1032188)]; for.size.21 i in range(0, 2, 1): ld.shared.f32 %f(85+i*21), [%r22+(-1023996+i*4)]; fma.rn.f32 %f(86+i*21), %f(85+i*21), %f(84+i*21), %f(65+i*21); ld.shared.f32 %f(87+i*21), [%r23+(4+i*4)]; fma.rn.f32 %f(88+i*21), %f(85+i*21), %f(87+i*21), %f(67+i*21); ld.shared.f32 %f(89+i*21), [%r23+(-1021948+i*4)]; fma.rn.f32 %f(90+i*21), %f(85+i*21), %f(89+i*21), %f(69+i*21); ld.shared.f32 %f(91+i*21), [%r23+(2052+i*4)]; fma.rn.f32 %f(92+i*21), %f(85+i*21), %f(91+i*21), %f(71+i*21); ld.shared.f32 %f(93+i*21), [%r23+(-1019900+i*4)]; fma.rn.f32 %f(94+i*21), %f(85+i*21), %f(93+i*21), %f(73+i*21); ld.shared.f32 %f(95+i*21), [%r23+(4100+i*4)]; fma.rn.f32 %f(96+i*21), %f(85+i*21), %f(95+i*21), %f(75+i*21); ld.shared.f32 %f(97+i*21), [%r23+(-1017852+i*4)]; fma.rn.f32 %f(98+i*21), %f(85+i*21), %f(97+i*21), %f(77+i*21); ld.shared.f32 %f(99+i*21), [%r23+(6148+i*4)]; fma.rn.f32 %f(100+i*21), %f(85+i*21), %f(99+i*21), %f(79+i*21); ld.shared.f32 %f(101+i*21), [%r23+(-1015804+i*4)]; fma.rn.f32 %f(102+i*21), %f(85+i*21), %f(101+i*21), %f(81+i*21); ld.shared.f32 %f(103+i*21), [%r23+(8196+i*4)]; fma.rn.f32 %f(104+i*21), %f(85+i*21), %f(103+i*21), %f(83+i*21); ld.shared.f32 %f(105+i*21), [%r23+(-1023992+i*4)]; ld.shared.f32 %f127, [%r22+-1023988]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(177+i*-1), %f127, %f(126+i*2), %f(107+i*2); ld.shared.f32 %f(128+i*1), [%r23+(12+i*-1021952)]; for.size.3 i in range(0, 3, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(175+i*-2+j*-1), %f127, %f(129+i*2+j*1), %f(111+i*4+j*2); ld.shared.f32 %f(130+i*2+j*1), [%r23+(2060+i*2048+j*-1021952)]; fma.rn.f32 %f169, %f127, %f135, %f123; ld.shared.f32 %f136, [%r23+8204]; fma.rn.f32 %f168, %f127, %f136, %f125; add.s32 %r30, %r30, 16; setp.ne.s32 %p3, %r30, 1026048; @%p3 bra $L__BB0_5; mul.lo.s32 %r25, %r1, 5; mov.u32 %r26, %ctaid.x; mad.lo.s32 %r27, %r1, 845, %r26; cvta.to.global.u64 %rd14, %rd7; mul.wide.s32 %rd15, %r25, 4; add.s64 %rd16, %rd14, %rd15; ld.global.nc.f32 %f137, [%rd16]; add.f32 %f138, %f177, %f137; mov.f32 %f139, 0f00000000; max.f32 %f140, %f138, %f139; cvta.to.global.u64 %rd17, %rd6; mul.wide.s32 %rd18, %r27, 4; add.s64 %rd19, %rd17, %rd18; for.size.5 i in range(0, 4, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd19+(0+i*676+j*338000)], %f(140+i*6+j*3); ld.global.nc.f32 %f(141+i*6+j*3), [%rd16+(2000+i*4+j*-1996)]; add.f32 %f(142+i*6+j*3), %f(176+i*-2+j*-1), %f(141+i*6+j*3); max.f32 %f(143+i*6+j*3), %f(142+i*6+j*3), %f139; st.global.f32 [%rd19+2704], %f164; ld.global.nc.f32 %f165, [%rd16+2016]; add.f32 %f166, %f168, %f165; max.f32 %f167, %f166, %f139; st.global.f32 [%rd19+340704], %f167; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[10]; __shared__ float pad_temp_shared[512]; __shared__ float placeholder_shared[512000]; compute[0] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[9] = 0.000000e+00f; if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 1] = placeholder[index1 + 169]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 2] = placeholder[index1 + 338]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 3] = placeholder[index1 + 507]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 4] = placeholder[index1 + 676]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 5] = placeholder[index1 + 845]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 6] = placeholder[index1 + 1014]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 7] = placeholder[index1 + 1183]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 8] = placeholder[index1 + 1352]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 9] = placeholder[index1 + 1521]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 10] = placeholder[index1 + 1690]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 11] = placeholder[index1 + 1859]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 12] = placeholder[index1 + 2028]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 13] = placeholder[index1 + 2197]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 14] = placeholder[index1 + 2366]; } if ((int)threadIdx.x < 32) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 2704 + (int)blockIdx.x; pad_temp_shared[index0 + 15] = placeholder[index1 + 2535]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 5120; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index2]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index3 = rc_outer_inner * 4; int index4 = (int)threadIdx.x * 2560 + rc_outer_inner * 4; compute[0] = compute[0] + pad_temp_shared[index3] * placeholder_shared[index4]; compute[5] = compute[5] + (pad_temp_shared[index3] * placeholder_shared[index4 + 256000]); compute[1] = compute[1] + (pad_temp_shared[index3] * placeholder_shared[index4 + 512]); compute[6] = compute[6] + (pad_temp_shared[index3] * placeholder_shared[index4 + 256512]); compute[2] = compute[2] + (pad_temp_shared[index3] * placeholder_shared[index4 + 1024]); compute[7] = compute[7] + (pad_temp_shared[index3] * placeholder_shared[index4 + 257024]); compute[3] = compute[3] + (pad_temp_shared[index3] * placeholder_shared[index4 + 1536]); compute[8] = compute[8] + (pad_temp_shared[index3] * placeholder_shared[index4 + 257536]); compute[4] = compute[4] + (pad_temp_shared[index3] * placeholder_shared[index4 + 2048]); compute[9] = compute[9] + (pad_temp_shared[index3] * placeholder_shared[index4 + 258048]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 1]); compute[5] = compute[5] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 256001]); compute[1] = compute[1] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 513]); compute[6] = compute[6] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 256513]); compute[2] = compute[2] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 1025]); compute[7] = compute[7] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 257025]); compute[3] = compute[3] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 1537]); compute[8] = compute[8] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 257537]); compute[4] = compute[4] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 2049]); compute[9] = compute[9] + (pad_temp_shared[index3 + 1] * placeholder_shared[index4 + 258049]); compute[0] = compute[0] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 2]); compute[5] = compute[5] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 256002]); compute[1] = compute[1] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 514]); compute[6] = compute[6] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 256514]); compute[2] = compute[2] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 1026]); compute[7] = compute[7] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 257026]); compute[3] = compute[3] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 1538]); compute[8] = compute[8] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 257538]); compute[4] = compute[4] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 2050]); compute[9] = compute[9] + (pad_temp_shared[index3 + 2] * placeholder_shared[index4 + 258050]); compute[0] = compute[0] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 3]); compute[5] = compute[5] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 256003]); compute[1] = compute[1] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 515]); compute[6] = compute[6] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 256515]); compute[2] = compute[2] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 1027]); compute[7] = compute[7] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 257027]); compute[3] = compute[3] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 1539]); compute[8] = compute[8] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 257539]); compute[4] = compute[4] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 2051]); compute[9] = compute[9] + (pad_temp_shared[index3 + 3] * placeholder_shared[index4 + 258051]); } for (int ax1_inner = 0; ax1_inner < 5; ++ax1_inner) { int index5 = (int)threadIdx.x * 845 + ax1_inner * 169 + (int)blockIdx.x; int index6 = ax1_inner; int index7 = (int)threadIdx.x * 5 + ax1_inner; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); T_relu[index5 + 84500] = max(compute[index6 + 5] + placeholder2[index7 + 500], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpk4pdvukn
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<15>; .reg .f32 %f<817>; .reg .b32 %r<90>; .reg .b64 %rd<27>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[6656]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[512000]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd3, %rd7; mov.u32 %r1, %tid.x; mul.hi.s32 %r25, %r1, 1321528399; shr.u32 %r26, %r25, 31; shr.s32 %r27, %r25, 2; add.s32 %r2, %r27, %r26; mov.u32 %r28, %ctaid.x; mul.lo.s32 %r29, %r2, 13; sub.s32 %r3, %r1, %r29; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(4+i*1), %r(28+i*-26), (13+i*156), %r(3+i*1); mul.lo.s32 %r6, %r2, 6400; shl.b32 %r30, %r1, 2; mov.u32 %r31, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r7, %r31, %r30; mov.u32 %r32, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r33, %r32, %r30; add.s32 %r9, %r33, 2080; mov.u32 %r85, 0; mov.f32 %f717, 0f00000000; for.size.1 i in range(0, 49, 1): mov.f32 %f(718+i*1), %f717; $L__BB0_1: bar.sync 0; mad.lo.s32 %r34, %r85, 21632, %r5; mul.wide.s32 %rd10, %r34, 4; add.s64 %rd5, %rd1, %rd10; setp.gt.s32 %p1, %r1, 1663; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f201, [%rd5]; st.shared.f32 [%r7], %f201; $L__BB0_3: setp.gt.s32 %p2, %r1, 1403; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f202, [%rd5+13520]; st.shared.f32 [%r7+1040], %f202; $L__BB0_5: setp.gt.s32 %p3, %r1, 1143; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f203, [%rd5+27040]; st.shared.f32 [%r7+2080], %f203; $L__BB0_7: setp.gt.s32 %p4, %r1, 883; @%p4 bra $L__BB0_9; ld.global.nc.f32 %f204, [%rd5+40560]; st.shared.f32 [%r7+3120], %f204; $L__BB0_9: setp.gt.s32 %p5, %r1, 623; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f205, [%rd5+54080]; st.shared.f32 [%r7+4160], %f205; $L__BB0_11: setp.gt.s32 %p6, %r1, 363; @%p6 bra $L__BB0_13; ld.global.nc.f32 %f206, [%rd5+67600]; st.shared.f32 [%r7+5200], %f206; $L__BB0_13: setp.gt.s32 %p7, %r1, 103; @%p7 bra $L__BB0_15; ld.global.nc.f32 %f207, [%rd5+81120]; st.shared.f32 [%r7+6240], %f207; $L__BB0_15: add.s32 %r88, %r1, 780; mov.u32 %r86, 496; mov.u32 %r87, %r9; bra.uni $L__BB0_16; $L__BB0_25: for.size.1 i in range(0, 2, 1): add.s32 %r(88+i*-1), %r(88+i*-1), (1040+i*3120); $L__BB0_16: add.s32 %r36, %r88, -780; setp.gt.s32 %p8, %r36, 127999; @%p8 bra $L__BB0_18; for.size.1 i in range(0, 2, 1): shl.b32 %r(70+i*-32), %r(85+i*-49), (7+i*-5); for.size.1 i in range(0, 2, 1): and.b32 %r(39+i*1), %r(38+i*-2), (-512+i*639); for.size.1 i in range(0, 2, 1): add.s32 %r(41+i*1), %r(40+i*1), %r(70+i*-31); mul.wide.s32 %rd11, %r42, 4; add.s64 %rd12, %rd3, %rd11; ld.global.nc.f32 %f208, [%rd12]; st.shared.f32 [%r87+-2080], %f208; $L__BB0_18: add.s32 %r86, %r86, -4; setp.eq.s32 %p9, %r86, 0; @%p9 bra $L__BB0_26; add.s32 %r17, %r88, -520; setp.gt.s32 %p10, %r17, 127999; @%p10 bra $L__BB0_21; for.size.1 i in range(0, 2, 1): shl.b32 %r(73+i*-30), %r(85+i*-68), (7+i*-5); for.size.1 i in range(0, 2, 1): and.b32 %r(44+i*1), %r(43+i*-26), (-512+i*639); for.size.1 i in range(0, 2, 1): add.s32 %r(46+i*1), %r(45+i*1), %r(73+i*-29); mul.wide.s32 %rd13, %r47, 4; add.s64 %rd14, %rd3, %rd13; ld.global.nc.f32 %f209, [%rd14]; st.shared.f32 [%r87+-1040], %f209; $L__BB0_21: add.s32 %r18, %r88, -260; setp.gt.s32 %p11, %r18, 127999; @%p11 bra $L__BB0_23; for.size.1 i in range(0, 2, 1): shl.b32 %r(72+i*-24), %r(85+i*-67), (7+i*-5); for.size.1 i in range(0, 2, 1): and.b32 %r(49+i*1), %r(48+i*-30), (-512+i*639); for.size.1 i in range(0, 2, 1): add.s32 %r(51+i*1), %r(50+i*1), %r(72+i*-23); mul.wide.s32 %rd15, %r52, 4; add.s64 %rd16, %rd3, %rd15; ld.global.nc.f32 %f210, [%rd16]; st.shared.f32 [%r87], %f210; $L__BB0_23: setp.gt.s32 %p12, %r88, 127999; @%p12 bra $L__BB0_25; for.size.1 i in range(0, 2, 1): shl.b32 %r(71+i*-18), %r(85+i*3), (7+i*-5); for.size.1 i in range(0, 2, 1): and.b32 %r(54+i*1), %r(53+i*35), (-512+i*639); for.size.1 i in range(0, 2, 1): add.s32 %r(56+i*1), %r(55+i*1), %r(71+i*-17); mul.wide.s32 %rd17, %r57, 4; add.s64 %rd18, %rd3, %rd17; ld.global.nc.f32 %f211, [%rd18]; st.shared.f32 [%r87+1040], %f211; bra.uni $L__BB0_25; $L__BB0_26: bar.sync 0; mov.u32 %r89, 0; $L__BB0_27: mov.u32 %r75, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r74, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r59, %r89, 52, %r3; for.size.2 i in range(0, 2, 1): shl.b32 %r(60+i*2), %r(89+i*-30), 2; add.s32 %r(61+i*3), %r(60+i*14), %r(6+i*56); shl.b32 %r65, %r61, 2; add.s32 %r67, %r75, %r65; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(212+i*1), [%r(67+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(214+i*2), %f213, %f(212+i*3), %f(766+i*-1); ld.shared.f32 %f(215+i*2), [%r67+(512+i*-508)]; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(218+i*5), [%r64+(52+i*52)]; fma.rn.f32 %f(219+i*5), %f(218+i*5), %f(217+i*5), %f(214+i*5); ld.shared.f32 %f(220+i*5), [%r67+(516+i*4)]; fma.rn.f32 %f(221+i*5), %f(218+i*5), %f(220+i*5), %f(216+i*5); ld.shared.f32 %f(222+i*5), [%r67+(8+i*4)]; ld.shared.f32 %f228, [%r64+156]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(766+i*-1), %f228, %f(227+i*2), %f(224+i*2); ld.shared.f32 %f(229+i*1), [%r67+(524+i*500)]; for.size.12 i in range(0, 23, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(231+i*14+j*2), %f213, %f(230+i*14+j*2), %f(764+i*-2+j*-1); ld.shared.f32 %f(232+i*14+j*2), [%r67+(1536+i*1024+j*-508)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(235+i*14+j*2), %f218, %f(234+i*14+j*2), %f(231+i*14+j*2); ld.shared.f32 %f(236+i*14+j*2), [%r67+(1540+i*1024+j*-508)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(239+i*14+j*2), %f223, %f(238+i*14+j*2), %f(235+i*14+j*2); ld.shared.f32 %f(240+i*14+j*2), [%r67+(1544+i*1024+j*-508)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(764+i*-2+j*-1), %f228, %f(242+i*14+j*1), %f(239+i*14+j*2); ld.shared.f32 %f(243+i*14+j*1), [%r67+(1548+i*1024+j*500)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(553+i*2), %f213, %f(552+i*2), %f(718+i*-1); ld.shared.f32 %f(554+i*2), [%r67+(25088+i*-508)]; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(557+i*4+j*2), %f(218+j*5), %f(556+i*4+j*2), %f(553+i*4+j*2); ld.shared.f32 %f(558+i*4+j*2), [%r67+(25092+i*4+j*-508)]; fma.rn.f32 %f718, %f228, %f564, %f561; ld.shared.f32 %f565, [%r67+25100]; fma.rn.f32 %f717, %f228, %f565, %f563; add.s32 %r89, %r89, 1; setp.ne.s32 %p13, %r89, 32; @%p13 bra $L__BB0_27; add.s32 %r85, %r85, 1; setp.lt.u32 %p14, %r85, 4; @%p14 bra $L__BB0_1; mul.hi.s32 %r83, %r1, 1321528399; shr.u32 %r82, %r83, 31; shr.s32 %r81, %r83, 2; add.s32 %r80, %r81, %r82; ld.param.u64 %rd26, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd25, %rd26; ld.param.u64 %rd24, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd23, %rd24; mul.lo.s32 %r79, %r80, 13; sub.s32 %r78, %r1, %r79; mov.u32 %r77, %ctaid.x; mad.lo.s32 %r76, %r77, 13, %r78; mul.lo.s32 %r68, %r80, 50; mad.lo.s32 %r69, %r80, 8450, %r76; mul.wide.s32 %rd19, %r68, 4; add.s64 %rd20, %rd23, %rd19; ld.global.nc.f32 %f566, [%rd20]; add.f32 %f567, %f766, %f566; mov.f32 %f568, 0f00000000; max.f32 %f569, %f567, %f568; mul.wide.s32 %rd21, %r69, 4; add.s64 %rd22, %rd25, %rd21; for.size.4 i in range(0, 49, 1): st.global.f32 [%rd22+(0+i*676)], %f(569+i*3); ld.global.nc.f32 %f(570+i*3), [%rd20+(4+i*4)]; add.f32 %f(571+i*3), %f(765+i*-1), %f(570+i*3); max.f32 %f(572+i*3), %f(571+i*3), %f568; st.global.f32 [%rd22+33124], %f716; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[50]; __shared__ float pad_temp_shared[1664]; __shared__ float placeholder_shared[128000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 25; ++ff_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = ff_outer_inner_init * 2 + ff_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 4; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 7; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x / 13 < 128) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 260 + (int)threadIdx.x < 1664) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 260 + (int)threadIdx.x; int index2 = rc_outer_outer * 21632 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3380 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 493; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 260 + (int)threadIdx.x < 128000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 260 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 260 + (int)threadIdx.x >> 7) * 512 + rc_outer_outer * 128 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 260 + (int)threadIdx.x & 127); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 25; ++ff_outer_inner) { for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index5 = ff_outer_inner * 2 + ff_inner; int index6 = rc_outer_inner * 52 + rc_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 6400 + ff_outer_inner * 256 + ff_inner * 128 + rc_outer_inner * 4 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } for (int ax1_inner = 0; ax1_inner < 50; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 8450 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 50 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpmkfui6w9
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<206>; .reg .b32 %r<29>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1352]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[320]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; shl.b32 %r11, %r1, 2; mov.u32 %r12, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r12, %r11; and.b32 %r13, %r1, 1; mov.u32 %r14, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r14, %r11; mul.hi.s32 %r15, %r1, 1626496491; shr.u32 %r16, %r15, 31; shr.s32 %r17, %r15, 6; add.s32 %r5, %r17, %r16; mul.lo.s32 %r18, %r5, 169; sub.s32 %r19, %r1, %r18; for.size.2 i in range(0, 2, 1): shl.b32 %r(20+i*1), %r(19+i*-14), (2+i*1); add.s32 %r(4+i*2), %r(12+i*2), %r(20+i*1); mov.u32 %r7, %ctaid.x; shl.b32 %r22, %r1, 8; and.b32 %r23, %r22, -512; mad.lo.s32 %r24, %r7, 20480, %r23; or.b32 %r25, %r24, %r13; mul.wide.s32 %rd13, %r25, 4; add.s64 %rd22, %rd12, %rd13; cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r1, 4; add.s64 %rd21, %rd14, %rd15; mov.u32 %r28, 0; mov.f32 %f186, 0f00000000; for.size.1 i in range(0, 19, 1): mov.f32 %f(187+i*1), %f186; $L__BB0_1: bar.sync 0; ld.global.nc.f32 %f61, [%rd21]; st.shared.f32 [%r2], %f61; setp.gt.s32 %p1, %r1, 79; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f62, [%rd22]; st.shared.f32 [%r3], %f62; $L__BB0_3: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(63+i*1), [%r(6+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(65+i*2), %f64, %f(63+i*3), %f(205+i*-1); ld.shared.f32 %f(66+i*2), [%r6+(16+i*16)]; for.size.2 i in range(0, 17, 1): fma.rn.f32 %f(69+i*2), %f64, %f(68+i*2), %f(203+i*-1); ld.shared.f32 %f(70+i*2), [%r6+(48+i*16)]; fma.rn.f32 %f103, %f64, %f102, %f186; ld.shared.f32 %f104, [%r6+4]; ld.shared.f32 %f105, [%r4+676]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(205+i*-1), %f105, %f(104+i*2), %f(65+i*2); ld.shared.f32 %f(106+i*1), [%r6+(20+i*16)]; for.size.2 i in range(0, 17, 1): fma.rn.f32 %f(203+i*-1), %f105, %f(107+i*1), %f(69+i*2); ld.shared.f32 %f(108+i*1), [%r6+(52+i*16)]; fma.rn.f32 %f186, %f105, %f124, %f103; for.size.1 i in range(0, 2, 1): add.s64 %rd(22+i*-1), %rd(22+i*-1), (8+i*1344); add.s32 %r28, %r28, 1; setp.ne.s32 %p2, %r28, 256; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd16, %rd8; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(26+i*1), %r7, (40+i*6720), %r(5+i*-4); mul.wide.s32 %rd17, %r26, 4; add.s64 %rd18, %rd16, %rd17; ld.global.nc.f32 %f125, [%rd18]; add.f32 %f126, %f205, %f125; mov.f32 %f127, 0f00000000; max.f32 %f128, %f126, %f127; mul.wide.s32 %rd19, %r27, 4; add.s64 %rd20, %rd1, %rd19; for.size.4 i in range(0, 19, 1): st.global.f32 [%rd20+(0+i*1352)], %f(128+i*3); ld.global.nc.f32 %f(129+i*3), [%rd18+(8+i*8)]; add.f32 %f(130+i*3), %f(204+i*-1), %f(129+i*3); max.f32 %f(131+i*3), %f(130+i*3), %f127; st.global.f32 [%rd20+25688], %f185; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[20]; __shared__ float pad_temp_shared[338]; __shared__ float placeholder_shared[80]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[19] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 256; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 338 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; if ((int)threadIdx.x < 80) { int index2 = (int)blockIdx.x * 20480 + ((int)threadIdx.x >> 1) * 512 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index3 = (int)threadIdx.x / 169 * 2; compute[0] = compute[0] + pad_temp_shared[index0 % 169] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 4]); compute[2] = compute[2] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 8]); compute[3] = compute[3] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 12]); compute[4] = compute[4] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 16]); compute[5] = compute[5] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 20]); compute[6] = compute[6] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 24]); compute[7] = compute[7] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 28]); compute[8] = compute[8] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 32]); compute[9] = compute[9] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 36]); compute[10] = compute[10] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 40]); compute[11] = compute[11] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 44]); compute[12] = compute[12] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 48]); compute[13] = compute[13] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 52]); compute[14] = compute[14] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 56]); compute[15] = compute[15] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 60]); compute[16] = compute[16] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 64]); compute[17] = compute[17] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 68]); compute[18] = compute[18] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 72]); compute[19] = compute[19] + (pad_temp_shared[index0 % 169] * placeholder_shared[index3 + 76]); int index4 = (int)threadIdx.x % 169 + 169; compute[0] = compute[0] + (pad_temp_shared[index4] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index3 + 5]); compute[2] = compute[2] + (pad_temp_shared[index4] * placeholder_shared[index3 + 9]); compute[3] = compute[3] + (pad_temp_shared[index4] * placeholder_shared[index3 + 13]); compute[4] = compute[4] + (pad_temp_shared[index4] * placeholder_shared[index3 + 17]); compute[5] = compute[5] + (pad_temp_shared[index4] * placeholder_shared[index3 + 21]); compute[6] = compute[6] + (pad_temp_shared[index4] * placeholder_shared[index3 + 25]); compute[7] = compute[7] + (pad_temp_shared[index4] * placeholder_shared[index3 + 29]); compute[8] = compute[8] + (pad_temp_shared[index4] * placeholder_shared[index3 + 33]); compute[9] = compute[9] + (pad_temp_shared[index4] * placeholder_shared[index3 + 37]); compute[10] = compute[10] + (pad_temp_shared[index4] * placeholder_shared[index3 + 41]); compute[11] = compute[11] + (pad_temp_shared[index4] * placeholder_shared[index3 + 45]); compute[12] = compute[12] + (pad_temp_shared[index4] * placeholder_shared[index3 + 49]); compute[13] = compute[13] + (pad_temp_shared[index4] * placeholder_shared[index3 + 53]); compute[14] = compute[14] + (pad_temp_shared[index4] * placeholder_shared[index3 + 57]); compute[15] = compute[15] + (pad_temp_shared[index4] * placeholder_shared[index3 + 61]); compute[16] = compute[16] + (pad_temp_shared[index4] * placeholder_shared[index3 + 65]); compute[17] = compute[17] + (pad_temp_shared[index4] * placeholder_shared[index3 + 69]); compute[18] = compute[18] + (pad_temp_shared[index4] * placeholder_shared[index3 + 73]); compute[19] = compute[19] + (pad_temp_shared[index4] * placeholder_shared[index3 + 77]); } for (int vthread_s = 0; vthread_s < 20; ++vthread_s) { int index5 = (int)blockIdx.x * 6760 + vthread_s * 338 + (int)threadIdx.x; int index6 = vthread_s; int index7 = (int)blockIdx.x * 40 + vthread_s * 2 + (int)threadIdx.x / 169; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpw_4ajuzf
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<231>; .reg .b32 %r<50>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r20, %r1, 2; mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r21, %r20; shr.u32 %r22, %r1, 4; and.b32 %r23, %r1, 15; bfi.b32 %r3, %r22, %r23, 9, 23; mov.u32 %r24, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r24, %r20; for.size.5 i in range(0, 2, 1): add.s32 %r(25+i*4), %r1, (6500+i*6500); shr.u32 %r(26+i*4), %r(25+i*4), 4; add.s32 %r(27+i*4), %r1, (4+i*4); and.b32 %r(28+i*4), %r(27+i*4), 15; bfi.b32 %r(5+i*1), %r(26+i*4), %r(28+i*4), 9, 23; mul.hi.s32 %r33, %r1, 1321528399; shr.u32 %r34, %r33, 31; shr.s32 %r35, %r33, 2; add.s32 %r8, %r35, %r34; mul.lo.s32 %r36, %r8, 13; sub.s32 %r7, %r1, %r36; for.size.2 i in range(0, 2, 1): shl.b32 %r(37+i*1), %r(8+i*-1), (6+i*-4); add.s32 %r(9+i*30), %r(24+i*-3), %r(37+i*1); add.s32 %r10, %r39, 312; mov.u32 %r47, 0; mov.f32 %f205, 0f00000000; for.size.1 i in range(0, 25, 1): mov.f32 %f(206+i*1), %f205; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 2703; @%p1 bra $L__BB0_3; mad.lo.s32 %r40, %r47, 2704, %r1; mul.wide.s32 %rd9, %r40, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f105, [%rd10]; st.shared.f32 [%r2], %f105; $L__BB0_3: setp.gt.s32 %p2, %r1, 2999; shl.b32 %r12, %r47, 4; for.size.5 i in range(0, 2, 1): add.s32 %r(41+i*1), %r(3+i*2), %r12; mul.wide.s32 %rd(11+i*2), %r(41+i*1), 4; add.s64 %rd(12+i*2), %rd1, %rd(11+i*2); ld.global.nc.f32 %f(106+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r4+(0+i*26000)], %f(106+i*1); @%p2 bra $L__BB0_5; add.s32 %r43, %r6, %r12; mul.wide.s32 %rd15, %r43, 4; add.s64 %rd16, %rd1, %rd15; ld.global.nc.f32 %f108, [%rd16]; st.shared.f32 [%r4+52000], %f108; $L__BB0_5: bar.sync 0; mov.u32 %r49, 32000; mov.u32 %r48, %r10; $L__BB0_6: add.s32 %r45, %r9, %r49; ld.shared.f32 %f109, [%r45+-32000]; ld.shared.f32 %f110, [%r48+-312]; fma.rn.f32 %f230, %f110, %f109, %f230; ld.shared.f32 %f111, [%r45]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(229+i*-2), %f(110+i*2), %f111, %f(229+i*-2); ld.shared.f32 %f(112+i*1), [%r48+(-260+i*52)]; fma.rn.f32 %f(228+i*-2), %f(112+i*1), %f109, %f(228+i*-2); for.size.3 i in range(0, 10, 1): fma.rn.f32 %f(225+i*-2), %f(113+i*1), %f111, %f(225+i*-2); ld.shared.f32 %f(114+i*1), [%r48+(-156+i*52)]; fma.rn.f32 %f(224+i*-2), %f(114+i*1), %f109, %f(224+i*-2); fma.rn.f32 %f205, %f123, %f111, %f205; for.size.1 i in range(0, 2, 1): add.s32 %r(48+i*1), %r(48+i*1), (676+i*-672); setp.ne.s32 %p3, %r49, 32064; @%p3 bra $L__BB0_6; add.s32 %r47, %r47, 1; setp.lt.u32 %p4, %r47, 32; @%p4 bra $L__BB0_1; mul.wide.s32 %rd17, %r8, 4; add.s64 %rd18, %rd4, %rd17; mad.lo.s32 %r46, %r8, 169, %r7; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(124+i*1), [%rd18+(2000+i*-2000)]; add.f32 %f126, %f230, %f125; mov.f32 %f127, 0f00000000; max.f32 %f128, %f126, %f127; mul.wide.s32 %rd19, %r46, 4; add.s64 %rd20, %rd3, %rd19; for.size.4 i in range(0, 12, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd20+(0+i*52+j*338000)], %f(128+i*4+j*2); add.f32 %f(129+i*4+j*2), %f(229+i*-2+j*-1), %f(124+j*1); max.f32 %f(130+i*4+j*2), %f(129+i*4+j*2), %f127; st.global.f32 [%rd20+624], %f176; add.f32 %f177, %f205, %f124; max.f32 %f178, %f177, %f127; st.global.f32 [%rd20+338624], %f178; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[26]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; compute[0] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[19] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[20] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[21] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[22] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[23] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[24] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[25] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 2704) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } int index0 = (int)threadIdx.x; int index2 = ((int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 15); placeholder_shared[index0] = placeholder1[index2]; int index3 = ((int)threadIdx.x + 6500 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x + 4 & 15); placeholder_shared[index0 + 6500] = placeholder1[index3]; if ((int)threadIdx.x < 3000) { int index4 = ((int)threadIdx.x + 13000 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x + 8 & 15); placeholder_shared[index0 + 13000] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { int index5 = rc_outer_inner * 169 + (int)threadIdx.x % 13; int index6 = (int)threadIdx.x / 13 * 16 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[13] = compute[13] + (pad_temp_shared[index5] * placeholder_shared[index6 + 8000]); compute[1] = compute[1] + (pad_temp_shared[index5 + 13] * placeholder_shared[index6]); compute[14] = compute[14] + (pad_temp_shared[index5 + 13] * placeholder_shared[index6 + 8000]); compute[2] = compute[2] + (pad_temp_shared[index5 + 26] * placeholder_shared[index6]); compute[15] = compute[15] + (pad_temp_shared[index5 + 26] * placeholder_shared[index6 + 8000]); compute[3] = compute[3] + (pad_temp_shared[index5 + 39] * placeholder_shared[index6]); compute[16] = compute[16] + (pad_temp_shared[index5 + 39] * placeholder_shared[index6 + 8000]); compute[4] = compute[4] + (pad_temp_shared[index5 + 52] * placeholder_shared[index6]); compute[17] = compute[17] + (pad_temp_shared[index5 + 52] * placeholder_shared[index6 + 8000]); compute[5] = compute[5] + (pad_temp_shared[index5 + 65] * placeholder_shared[index6]); compute[18] = compute[18] + (pad_temp_shared[index5 + 65] * placeholder_shared[index6 + 8000]); compute[6] = compute[6] + (pad_temp_shared[index5 + 78] * placeholder_shared[index6]); compute[19] = compute[19] + (pad_temp_shared[index5 + 78] * placeholder_shared[index6 + 8000]); compute[7] = compute[7] + (pad_temp_shared[index5 + 91] * placeholder_shared[index6]); compute[20] = compute[20] + (pad_temp_shared[index5 + 91] * placeholder_shared[index6 + 8000]); compute[8] = compute[8] + (pad_temp_shared[index5 + 104] * placeholder_shared[index6]); compute[21] = compute[21] + (pad_temp_shared[index5 + 104] * placeholder_shared[index6 + 8000]); compute[9] = compute[9] + (pad_temp_shared[index5 + 117] * placeholder_shared[index6]); compute[22] = compute[22] + (pad_temp_shared[index5 + 117] * placeholder_shared[index6 + 8000]); compute[10] = compute[10] + (pad_temp_shared[index5 + 130] * placeholder_shared[index6]); compute[23] = compute[23] + (pad_temp_shared[index5 + 130] * placeholder_shared[index6 + 8000]); compute[11] = compute[11] + (pad_temp_shared[index5 + 143] * placeholder_shared[index6]); compute[24] = compute[24] + (pad_temp_shared[index5 + 143] * placeholder_shared[index6 + 8000]); compute[12] = compute[12] + (pad_temp_shared[index5 + 156] * placeholder_shared[index6]); compute[25] = compute[25] + (pad_temp_shared[index5 + 156] * placeholder_shared[index6 + 8000]); } } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index0 = (int)threadIdx.x; int index7 = (int)threadIdx.x / 13 * 169 + ax2_inner * 13 + (int)threadIdx.x % 13; int index8 = ax2_inner; T_relu[index7] = max(compute[index8] + placeholder2[index0 / 13], 0.000000e+00f); int index9 = (int)threadIdx.x / 13 + 500; T_relu[index7 + 84500] = max(compute[index8 + 13] + placeholder2[index9], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp3reldchc
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<206>; .reg .b32 %r<45>; .reg .b64 %rd<19>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r18, %r1, 2; mov.u32 %r19, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r19, %r18; for.size.2 i in range(0, 2, 1): shl.b32 %r(20+i*2), %r1, (6+i*-5); and.b32 %r(21+i*2), %r(20+i*2), (-512+i*526); or.b32 %r3, %r23, %r21; shl.b32 %r24, %r1, 3; mov.u32 %r25, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r25, %r24; or.b32 %r26, %r22, 1; shr.u32 %r27, %r26, 4; and.b32 %r28, %r26, 15; bfi.b32 %r5, %r27, %r28, 9, 23; mul.hi.s32 %r29, %r1, 1626496491; shr.u32 %r30, %r29, 31; shr.s32 %r31, %r29, 6; add.s32 %r7, %r31, %r30; mul.lo.s32 %r32, %r7, 169; sub.s32 %r6, %r1, %r32; shl.b32 %r33, %r6, 2; add.s32 %r8, %r19, %r33; mad.lo.s32 %r9, %r7, 1280, %r25; mov.u32 %r42, 0; mov.f32 %f186, 0f00000000; for.size.1 i in range(0, 19, 1): mov.f32 %f(187+i*1), %f186; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 2703; @%p1 bra $L__BB0_3; mad.lo.s32 %r34, %r42, 2704, %r1; mul.wide.s32 %rd9, %r34, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f81, [%rd10]; st.shared.f32 [%r2], %f81; $L__BB0_3: setp.gt.s32 %p2, %r1, 7999; @%p2 bra $L__BB0_5; shl.b32 %r35, %r42, 4; for.size.5 i in range(0, 2, 1): add.s32 %r(36+i*1), %r(3+i*2), %r35; mul.wide.s32 %rd(11+i*2), %r(36+i*1), 4; add.s64 %rd(12+i*2), %rd1, %rd(11+i*2); ld.global.nc.f32 %f(82+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r4+(0+i*4)], %f(82+i*1); $L__BB0_5: bar.sync 0; mov.u32 %r43, 640; mov.u32 %r44, %r8; $L__BB0_6: add.s32 %r39, %r9, %r43; ld.shared.f32 %f84, [%r39+-640]; for.size.20 i in range(0, 2, 1): ld.shared.f32 %f(85+i*10), [%r(44+i*-5)+0]; fma.rn.f32 %f(205+i*-10), %f85, %f(84+i*11), %f(205+i*-10); ld.shared.f32 %f(86+i*10), [%r39+(-576+i*640)]; fma.rn.f32 %f(204+i*-10), %f85, %f(86+i*10), %f(204+i*-10); ld.shared.f32 %f(87+i*10), [%r39+(-512+i*640)]; fma.rn.f32 %f(203+i*-10), %f85, %f(87+i*10), %f(203+i*-10); ld.shared.f32 %f(88+i*10), [%r39+(-448+i*640)]; fma.rn.f32 %f(202+i*-10), %f85, %f(88+i*10), %f(202+i*-10); ld.shared.f32 %f(89+i*10), [%r39+(-384+i*640)]; fma.rn.f32 %f(201+i*-10), %f85, %f(89+i*10), %f(201+i*-10); ld.shared.f32 %f(90+i*10), [%r39+(-320+i*640)]; fma.rn.f32 %f(200+i*-10), %f85, %f(90+i*10), %f(200+i*-10); ld.shared.f32 %f(91+i*10), [%r39+(-256+i*640)]; fma.rn.f32 %f(199+i*-10), %f85, %f(91+i*10), %f(199+i*-10); ld.shared.f32 %f(92+i*10), [%r39+(-192+i*640)]; fma.rn.f32 %f(198+i*-10), %f85, %f(92+i*10), %f(198+i*-10); ld.shared.f32 %f(93+i*10), [%r39+(-128+i*640)]; fma.rn.f32 %f(197+i*-10), %f85, %f(93+i*10), %f(197+i*-10); ld.shared.f32 %f(94+i*10), [%r39+(-64+i*640)]; fma.rn.f32 %f(196+i*-10), %f85, %f(94+i*10), %f(196+i*-10); for.size.1 i in range(0, 2, 1): add.s32 %r(44+i*-1), %r(44+i*-1), (676+i*-672); setp.ne.s32 %p3, %r43, 704; @%p3 bra $L__BB0_6; add.s32 %r42, %r42, 1; setp.lt.u32 %p4, %r42, 32; @%p4 bra $L__BB0_1; mul.lo.s32 %r40, %r7, 20; mad.lo.s32 %r41, %r7, 3380, %r6; mul.wide.s32 %rd15, %r40, 4; add.s64 %rd16, %rd4, %rd15; ld.global.nc.f32 %f105, [%rd16]; add.f32 %f106, %f205, %f105; mov.f32 %f107, 0f00000000; max.f32 %f108, %f106, %f107; mul.wide.s32 %rd17, %r41, 4; add.s64 %rd18, %rd3, %rd17; for.size.4 i in range(0, 19, 1): st.global.f32 [%rd18+(0+i*676)], %f(108+i*3); ld.global.nc.f32 %f(109+i*3), [%rd16+(4+i*4)]; add.f32 %f(110+i*3), %f(204+i*-1), %f(109+i*3); max.f32 %f(111+i*3), %f(110+i*3), %f107; st.global.f32 [%rd18+12844], %f165; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[20]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[19] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 2704) { int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 8000) { int index0 = (int)threadIdx.x; int index2 = ((int)threadIdx.x >> 3) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x & 7) * 2; placeholder_shared[index0 * 2] = placeholder1[index2]; } if ((int)threadIdx.x < 8000) { int index3 = (int)threadIdx.x * 2 + 1; int index4 = ((int)threadIdx.x * 2 + 1 >> 4) * 512 + rc_outer_outer * 16 + ((int)threadIdx.x * 2 + 1 & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { int index5 = rc_outer_inner * 169 + (int)threadIdx.x % 169; int index6 = (int)threadIdx.x / 169 * 320 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index6 + 16]); compute[2] = compute[2] + (pad_temp_shared[index5] * placeholder_shared[index6 + 32]); compute[3] = compute[3] + (pad_temp_shared[index5] * placeholder_shared[index6 + 48]); compute[4] = compute[4] + (pad_temp_shared[index5] * placeholder_shared[index6 + 64]); compute[5] = compute[5] + (pad_temp_shared[index5] * placeholder_shared[index6 + 80]); compute[6] = compute[6] + (pad_temp_shared[index5] * placeholder_shared[index6 + 96]); compute[7] = compute[7] + (pad_temp_shared[index5] * placeholder_shared[index6 + 112]); compute[8] = compute[8] + (pad_temp_shared[index5] * placeholder_shared[index6 + 128]); compute[9] = compute[9] + (pad_temp_shared[index5] * placeholder_shared[index6 + 144]); compute[10] = compute[10] + (pad_temp_shared[index5] * placeholder_shared[index6 + 160]); compute[11] = compute[11] + (pad_temp_shared[index5] * placeholder_shared[index6 + 176]); compute[12] = compute[12] + (pad_temp_shared[index5] * placeholder_shared[index6 + 192]); compute[13] = compute[13] + (pad_temp_shared[index5] * placeholder_shared[index6 + 208]); compute[14] = compute[14] + (pad_temp_shared[index5] * placeholder_shared[index6 + 224]); compute[15] = compute[15] + (pad_temp_shared[index5] * placeholder_shared[index6 + 240]); compute[16] = compute[16] + (pad_temp_shared[index5] * placeholder_shared[index6 + 256]); compute[17] = compute[17] + (pad_temp_shared[index5] * placeholder_shared[index6 + 272]); compute[18] = compute[18] + (pad_temp_shared[index5] * placeholder_shared[index6 + 288]); compute[19] = compute[19] + (pad_temp_shared[index5] * placeholder_shared[index6 + 304]); } } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { int index7 = (int)threadIdx.x / 169 * 3380 + ax1_inner * 169 + (int)threadIdx.x % 169; int index8 = ax1_inner; int index9 = (int)threadIdx.x / 169 * 20 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpmttlxsua
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[1352]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<11>; .reg .f32 %f<631>; .reg .b32 %r<121>; .reg .b64 %rd<43>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[12800]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd11, [default_function_kernel0_param_1]; ld.param.u64 %rd12, [default_function_kernel0_param_2]; ld.param.u64 %rd13, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd14, %rd10; cvta.to.global.u64 %rd1, %rd12; add.u64 %rd2, %SPL, 0; mov.u64 %rd16, 0; mov.u32 %r113, 0; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(52+i*104)], %r113; mov.f32 %f5, 0f00000000; for.size.1 i in range(0, 13, 1): st.local.v2.f32 [%rd2+(0+i*104)], {%f5, %f5}; for.size.4 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd2+(56+i*4+j*104)], %r113; for.size.1 j in range(0, 12, 1): st.local.u32 [%rd2+(8+i*4+j*104)], %r113; st.local.v2.f32 [%rd2+1256], {%f5, %f5}; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd2+(64+i*-48+j*104)], %r113; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd2+(68+i*-48+j*104)], %r113; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1216+i*104)], %r113; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(24+i*104)], %r113; for.size.1 i in range(0, 11, 1): st.local.v2.f32 [%rd2+(72+i*104)], {%f5, %f5}; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd2+(1220+i*104)], %r113; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(28+i*104)], %r113; for.size.1 i in range(0, 3, 1): st.local.u32 [%rd2+(1072+i*104)], %r113; for.size.1 i in range(0, 13, 1): st.local.v2.f32 [%rd2+(80+i*104)], {%f5, %f5}; for.size.1 i in range(0, 10, 1): st.local.v2.f32 [%rd2+(32+i*104)], {%f5, %f5}; for.size.1 i in range(0, 3, 1): st.local.u32 [%rd2+(1076+i*104)], %r113; for.size.1 i in range(0, 8, 1): st.local.u32 [%rd2+(192+i*104)], %r113; st.local.u64 [%rd2+88], %rd16; for.size.1 i in range(0, 8, 1): st.local.u32 [%rd2+(196+i*104)], %r113; for.size.1 i in range(0, 4, 1): st.local.v2.f32 [%rd2+(1024+i*104)], {%f5, %f5}; for.size.1 i in range(0, 13, 1): st.local.v2.f32 [%rd2+(40+i*104)], {%f5, %f5}; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.u32 [%rd2+(96+i*-48+j*104)], %r113; for.size.1 i in range(0, 13, 1): st.local.u32 [%rd2+(100+i*104)], %r113; cvta.to.global.u64 %rd3, %rd11; cvta.to.global.u64 %rd4, %rd13; mov.u32 %r1, %ctaid.x; mul.lo.s32 %r2, %r1, 51200; mov.u32 %r3, %tid.x; shl.b32 %r4, %r3, 6; add.s32 %r5, %r3, 150; shl.b32 %r29, %r3, 2; mov.u32 %r30, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r31, %r30, %r29; add.s32 %r6, %r31, 400; add.s64 %rd5, %rd14, 400; $L__BB0_1: bar.sync 0; mad.lo.s32 %r33, %r113, 5408, %r3; mul.wide.s32 %rd17, %r33, 4; add.s64 %rd42, %rd5, %rd17; mov.u32 %r116, 112; mov.u32 %r114, %r6; mov.u32 %r115, %r5; bra.uni $L__BB0_2; $L__BB0_11: for.size.1 i in range(0, 2, 1): add.s32 %r(115+i*-1), %r(115+i*-1), (200+i*600); add.s64 %rd42, %rd42, 800; $L__BB0_2: add.s32 %r34, %r115, -150; setp.gt.s32 %p1, %r34, 5407; add.s64 %rd8, %rd42, -400; @%p1 bra $L__BB0_4; ld.global.nc.f32 %f6, [%rd8]; add.s32 %r109, %r114, -400; st.shared.f32 [%r109], %f6; $L__BB0_4: add.s32 %r116, %r116, -4; setp.eq.s32 %p2, %r116, 0; @%p2 bra $L__BB0_12; add.s32 %r35, %r115, -100; setp.gt.s32 %p3, %r35, 5407; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f7, [%rd8+200]; add.s32 %r110, %r114, -400; st.shared.f32 [%r110+200], %f7; $L__BB0_7: add.s32 %r36, %r115, -50; setp.gt.s32 %p4, %r36, 5407; @%p4 bra $L__BB0_9; ld.global.nc.f32 %f8, [%rd8+400]; add.s32 %r111, %r114, -400; st.shared.f32 [%r111+400], %f8; $L__BB0_9: setp.gt.s32 %p5, %r115, 5407; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f9, [%rd8+600]; add.s32 %r112, %r114, -400; st.shared.f32 [%r112+600], %f9; bra.uni $L__BB0_11; $L__BB0_12: shl.b32 %r38, %r113, 5; add.s32 %r16, %r38, %r2; mov.u32 %r117, 0; $L__BB0_13: mad.lo.s32 %r39, %r117, 50, %r3; shl.b32 %r40, %r39, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(41+i*1), %r(40+i*-1), (-512+i*543); for.size.1 i in range(0, 2, 1): add.s32 %r(43+i*1), %r(16+i*27), %r(42+i*-1); mul.wide.s32 %rd18, %r44, 4; add.s64 %rd19, %rd3, %rd18; ld.global.nc.f32 %f10, [%rd19]; shl.b32 %r45, %r39, 2; mov.u32 %r46, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r47, %r46, %r45; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r47+(0+i*200)], %f(10+i*1); add.s32 %r(48+i*6), %r39, (50+i*50); shl.b32 %r(49+i*6), %r(48+i*6), 4; and.b32 %r(50+i*6), %r(49+i*6), -512; and.b32 %r(51+i*6), %r(48+i*6), 31; add.s32 %r(52+i*6), %r16, %r(51+i*6); add.s32 %r(53+i*6), %r(52+i*6), %r(50+i*6); mul.wide.s32 %rd(20+i*2), %r(53+i*6), 4; add.s64 %rd(21+i*2), %rd3, %rd(20+i*2); ld.global.nc.f32 %f(11+i*1), [%rd(21+i*2)+0]; st.shared.f32 [%r47+1400], %f17; add.s32 %r117, %r117, 8; setp.ne.s32 %p6, %r117, 64; @%p6 bra $L__BB0_13; bar.sync 0; mov.u32 %r90, 0; mov.u32 %r118, %r90; $L__BB0_15: mul.lo.s32 %r20, %r118, 338; for.size.2 i in range(0, 2, 1): shl.b32 %r(92+i*2), %r(118+i*-25), (1+i*1); add.s32 %r(93+i*3), %r(92+i*-46), %r(4+i*90); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): ld.shared.f32 %f(1+i*2+j*1), [%r96+(0+i*128+j*4)]; mov.u32 %r119, %r90; $L__BB0_16: mad.lo.s32 %r97, %r119, 13, %r20; mul.wide.s32 %rd34, %r119, 4; add.s64 %rd35, %rd2, %rd34; shl.b32 %r98, %r97, 2; add.s32 %r100, %r30, %r98; for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(18+i*3), [%r100+(0+i*4)]; ld.local.f32 %f(19+i*3), [%rd35+(0+i*104)]; fma.rn.f32 %f(20+i*3), %f(18+i*3), %f1, %f(19+i*3); for.size.2 i in range(0, 13, 1): ld.local.f32 %f(57+i*2), [%rd35+(52+i*104)]; fma.rn.f32 %f(58+i*2), %f(18+i*3), %f3, %f(57+i*2); for.size.1 i in range(0, 12, 1): ld.shared.f32 %f(83+i*1), [%r100+(680+i*4)]; ld.shared.f32 %f95, [%r100+676]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(96+i*1), %f(95+i*-12), %f2, %f(20+i*3); st.local.f32 [%rd35+(0+i*104)], %f(96+i*1); for.size.2 i in range(0, 11, 1): fma.rn.f32 %f(98+i*1), %f(84+i*1), %f2, %f(26+i*3); st.local.f32 [%rd35+(208+i*104)], %f(98+i*1); for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(109+i*1), %f(95+i*-12), %f4, %f(58+i*2); st.local.f32 [%rd35+(52+i*104)], %f(109+i*1); for.size.2 i in range(0, 11, 1): fma.rn.f32 %f(111+i*1), %f(84+i*1), %f4, %f(62+i*2); st.local.f32 [%rd35+(260+i*104)], %f(111+i*1); add.s32 %r119, %r119, 1; setp.ne.s32 %p7, %r119, 13; @%p7 bra $L__BB0_16; add.s32 %r118, %r118, 1; setp.lt.u32 %p8, %r118, 16; @%p8 bra $L__BB0_15; add.s32 %r113, %r113, 1; setp.lt.u32 %p9, %r113, 16; @%p9 bra $L__BB0_1; shl.b32 %r102, %r3, 1; mad.lo.s32 %r25, %r1, 100, %r102; mul.lo.s32 %r103, %r3, 338; mad.lo.s32 %r26, %r1, 16900, %r103; mov.u32 %r120, 0; $L__BB0_20: add.s32 %r105, %r25, %r120; mul.wide.s32 %rd36, %r105, 4; add.s64 %rd37, %rd4, %rd36; mad.lo.s32 %r106, %r120, 169, %r26; mul.lo.s32 %r107, %r120, 13; mul.wide.s32 %rd38, %r107, 4; add.s64 %rd39, %rd2, %rd38; ld.local.f32 %f122, [%rd39]; ld.global.nc.f32 %f123, [%rd37]; add.f32 %f124, %f122, %f123; mov.f32 %f125, 0f00000000; max.f32 %f126, %f124, %f125; mul.wide.s32 %rd40, %r106, 4; add.s64 %rd41, %rd1, %rd40; for.size.4 i in range(0, 12, 1): st.global.f32 [%rd41+(0+i*4)], %f(126+i*3); ld.local.f32 %f(127+i*3), [%rd39+(104+i*104)]; add.f32 %f(128+i*3), %f(127+i*3), %f123; max.f32 %f(129+i*3), %f(128+i*3), %f125; for.size.5 i in range(0, 12, 1): for.size.4 j in range(0, 13, 1): st.global.f32 [%rd41+(48+i*52+j*4)], %f(162+i*39+j*3); ld.local.f32 %f(163+i*39+j*3), [%rd39+(4+i*4+j*104)]; add.f32 %f(164+i*39+j*3), %f(163+i*39+j*3), %f123; max.f32 %f(165+i*39+j*3), %f(164+i*39+j*3), %f125; st.global.f32 [%rd41+672], %f630; add.s32 %r108, %r120, 1; mov.u32 %r120, 1; setp.lt.u32 %p10, %r108, 2; @%p10 bra $L__BB0_20; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[338]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[3200]; for (int yy_outer_inner_init = 0; yy_outer_inner_init < 13; ++yy_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = ff_inner_init * 13 + yy_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 26] = 0.000000e+00f; compute[index0 + 52] = 0.000000e+00f; compute[index0 + 78] = 0.000000e+00f; compute[index0 + 104] = 0.000000e+00f; compute[index0 + 130] = 0.000000e+00f; compute[index0 + 156] = 0.000000e+00f; compute[index0 + 182] = 0.000000e+00f; compute[index0 + 208] = 0.000000e+00f; compute[index0 + 234] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 286] = 0.000000e+00f; compute[index0 + 312] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 109; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 50 + (int)threadIdx.x < 5408) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 50 + (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 50 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 64; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 50 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 51200 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 50 + (int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 50 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int yy_outer_inner = 0; yy_outer_inner < 13; ++yy_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index5 = ff_inner * 13 + yy_outer_inner; int index6 = rc_outer_inner * 338 + rc_inner * 169 + yy_outer_inner * 13; int index7 = (int)threadIdx.x * 64 + ff_inner * 32 + rc_outer_inner * 2 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 26] = compute[index5 + 26] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 52] = compute[index5 + 52] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 78] = compute[index5 + 78] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 104] = compute[index5 + 104] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 130] = compute[index5 + 130] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 156] = compute[index5 + 156] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 182] = compute[index5 + 182] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 208] = compute[index5 + 208] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); compute[index5 + 234] = compute[index5 + 234] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 260] = compute[index5 + 260] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7]); compute[index5 + 286] = compute[index5 + 286] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7]); compute[index5 + 312] = compute[index5 + 312] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7]); } } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index8 = (int)blockIdx.x * 16900 + (int)threadIdx.x * 338 + ax1_inner * 169 + ax2_inner * 13; int index9 = ax1_inner * 13 + ax2_inner; int index10 = (int)blockIdx.x * 100 + (int)threadIdx.x * 2 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 1] = max(compute[index9 + 26] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 2] = max(compute[index9 + 52] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 3] = max(compute[index9 + 78] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 4] = max(compute[index9 + 104] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 5] = max(compute[index9 + 130] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 6] = max(compute[index9 + 156] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 7] = max(compute[index9 + 182] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 8] = max(compute[index9 + 208] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 9] = max(compute[index9 + 234] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 10] = max(compute[index9 + 260] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 11] = max(compute[index9 + 286] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 12] = max(compute[index9 + 312] + placeholder2[index10], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp3zk0d8v9
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<236>; .reg .b32 %r<76>; .reg .b64 %rd<31>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[208]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[8000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; mul.hi.s32 %r26, %r1, 1321528399; shr.u32 %r27, %r26, 31; shr.s32 %r28, %r26, 2; add.s32 %r29, %r28, %r27; mov.u32 %r30, %ctaid.x; mul.hi.s32 %r31, %r30, 1321528399; shr.u32 %r32, %r31, 31; shr.s32 %r33, %r31, 2; add.s32 %r5, %r33, %r32; mul.lo.s32 %r34, %r5, 13; sub.s32 %r35, %r30, %r34; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(2+i*34), %r(35+i*-6), 13; sub.s32 %r37, %r1, %r36; mad.lo.s32 %r38, %r29, 169, %r37; add.s32 %r3, %r38, %r2; shl.b32 %r39, %r1, 2; mov.u32 %r40, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r40, %r39; mul.lo.s32 %r6, %r5, 256000; shl.b32 %r41, %r1, 7; for.size.1 i in range(0, 2, 1): and.b32 %r(7+i*1), %r(41+i*-40), (-512+i*515); mov.u32 %r42, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r9, %r42, %r39; add.s32 %r43, %r1, 250; shl.b32 %r44, %r43, 7; for.size.1 i in range(0, 2, 1): and.b32 %r(10+i*1), %r(44+i*-1), (-512+i*515); for.size.2 i in range(0, 6, 1): add.s32 %r(45+i*1), %r41, (64000+i*32000); and.b32 %r(12+i*1), %r(45+i*1), -512; shl.b32 %r51, %r1, 5; add.s32 %r18, %r42, %r51; mov.u32 %r73, 0; mov.f32 %f210, 0f00000000; for.size.1 i in range(0, 25, 1): mov.f32 %f(211+i*1), %f210; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 51; @%p1 bra $L__BB0_3; mad.lo.s32 %r52, %r73, 676, %r3; mul.wide.s32 %rd9, %r52, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f105, [%rd10]; st.shared.f32 [%r4], %f105; $L__BB0_3: shl.b32 %r55, %r73, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(56+i*1), %r(55+i*1), %r(6+i*2); for.size.6 i in range(0, 2, 1): add.s32 %r(58+i*2), %r(57+i*2), %r(7+i*3); mul.wide.s32 %rd(11+i*2), %r(58+i*2), 4; add.s64 %rd(12+i*2), %rd1, %rd(11+i*2); ld.global.nc.f32 %f(106+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r9+(0+i*1000)], %f(106+i*1); add.s32 %r(59+i*2), %r(56+i*1), %r(11+i*1); for.size.5 i in range(0, 2, 1): mul.wide.s32 %rd(15+i*2), %r(61+i*1), 4; add.s64 %rd(16+i*2), %rd1, %rd(15+i*2); ld.global.nc.f32 %f(108+i*1), [%rd(16+i*2)+0]; st.shared.f32 [%r9+(2000+i*1000)], %f(108+i*1); add.s32 %r(62+i*1), %r(59+i*-2), %r(13+i*1); for.size.5 i in range(0, 2, 1): mul.wide.s32 %rd(19+i*2), %r(63+i*1), 4; add.s64 %rd(20+i*2), %rd1, %rd(19+i*2); ld.global.nc.f32 %f(110+i*1), [%rd(20+i*2)+0]; st.shared.f32 [%r9+(4000+i*1000)], %f(110+i*1); add.s32 %r(64+i*1), %r(59+i*-2), %r(15+i*1); mul.wide.s32 %rd23, %r65, 4; add.s64 %rd24, %rd1, %rd23; ld.global.nc.f32 %f112, [%rd24]; st.shared.f32 [%r9+6000], %f112; add.s32 %r66, %r59, %r17; mul.wide.s32 %rd25, %r66, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f113, [%rd26]; st.shared.f32 [%r9+7000], %f113; bar.sync 0; mov.u32 %r75, 16; mov.u32 %r74, %r40; $L__BB0_4: ld.shared.f32 %f114, [%r74]; add.s32 %r67, %r18, %r75; ld.shared.f32 %f115, [%r67+-16]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(235+i*-1), %f(114+i*2), %f115, %f(235+i*-1); ld.shared.f32 %f(116+i*1), [%r74+(4+i*4)]; for.size.2 i in range(0, 10, 1): fma.rn.f32 %f(233+i*-1), %f(117+i*1), %f115, %f(233+i*-1); ld.shared.f32 %f(118+i*1), [%r74+(12+i*4)]; fma.rn.f32 %f223, %f127, %f115, %f223; ld.shared.f32 %f128, [%r67]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(222+i*-1), %f(114+i*2), %f128, %f(222+i*-1); for.size.1 i in range(0, 11, 1): fma.rn.f32 %f(220+i*-1), %f(117+i*1), %f128, %f(220+i*-1); for.size.1 i in range(0, 2, 1): add.s32 %r(74+i*1), %r(74+i*1), (52+i*-48); setp.ne.s32 %p2, %r75, 32; @%p2 bra $L__BB0_4; add.s32 %r73, %r73, 1; setp.lt.u32 %p3, %r73, 128; @%p3 bra $L__BB0_1; shl.b32 %r68, %r1, 1; mad.lo.s32 %r69, %r5, 500, %r68; mul.lo.s32 %r70, %r1, 338; mad.lo.s32 %r71, %r5, 84500, %r70; add.s32 %r72, %r71, %r2; mul.wide.s32 %rd27, %r69, 4; add.s64 %rd28, %rd4, %rd27; ld.global.nc.f32 %f129, [%rd28]; add.f32 %f130, %f235, %f129; mov.f32 %f131, 0f00000000; max.f32 %f132, %f130, %f131; mul.wide.s32 %rd29, %r72, 4; add.s64 %rd30, %rd3, %rd29; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd30+(0+i*4)], %f(132+i*2); add.f32 %f(133+i*2), %f(234+i*-1), %f129; max.f32 %f(134+i*2), %f(133+i*2), %f131; st.global.f32 [%rd30+48], %f156; ld.global.nc.f32 %f157, [%rd28+4]; for.size.3 i in range(0, 13, 1): add.f32 %f(158+i*2), %f(222+i*-1), %f157; max.f32 %f(159+i*2), %f(158+i*2), %f131; st.global.f32 [%rd30+(676+i*4)], %f(159+i*2); ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[26]; __shared__ float pad_temp_shared[52]; __shared__ float placeholder_shared[2000]; for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 13; ++xx_inner_init) { int index0 = ff_inner_init * 13 + xx_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 52) { int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 676 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 8; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 250 + (int)threadIdx.x; int index4 = (int)blockIdx.x / 13 * 256000 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 250 + (int)threadIdx.x >> 2) * 512 + rc_outer_outer * 4 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 250 + (int)threadIdx.x & 3); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { for (int xx_inner = 0; xx_inner < 13; ++xx_inner) { int index5 = ff_inner * 13 + xx_inner; int index6 = rc_outer_inner * 13 + xx_inner; int index7 = (int)threadIdx.x * 8 + ff_inner * 4 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x / 13 * 84500 + (int)threadIdx.x * 338 + ax1_inner * 169 + (int)blockIdx.x % 13 * 13 + ax3_inner; int index9 = ax1_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x / 13 * 500 + (int)threadIdx.x * 2 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpxeiqirna
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[67600]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<12>; .reg .f32 %f<709>; .reg .b32 %r<135>; .reg .b64 %rd<39>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[43264]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[51200]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd8, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd8; cvta.to.global.u64 %rd4, %rd9; add.u64 %rd5, %SPL, 0; mov.u32 %r36, 0; mov.u32 %r124, %r36; $L__BB0_1: mov.u32 %r125, %r36; $L__BB0_2: mad.lo.s32 %r38, %r125, 169, %r124; mul.wide.s32 %rd11, %r38, 4; add.s64 %rd12, %rd5, %rd11; mov.u32 %r39, 0; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 10, 1): st.local.u32 [%rd12+(0+i*52+j*6760)], %r39; add.s32 %r125, %r125, 1; setp.ne.s32 %p1, %r125, 10; @%p1 bra $L__BB0_2; add.s32 %r124, %r124, 1; setp.lt.u32 %p2, %r124, 13; @%p2 bra $L__BB0_1; mov.u32 %r5, %tid.x; mov.u32 %r6, %ctaid.x; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(7+i*1), %r(6+i*-1), (102400+i*-101760); mov.u32 %r40, 0; mov.u32 %r126, %r40; $L__BB0_5: bar.sync 0; mad.lo.s32 %r10, %r126, 10816, %r5; mov.u32 %r127, %r40; $L__BB0_6: shl.b32 %r42, %r127, 1; add.s32 %r43, %r10, %r42; mul.wide.s32 %rd13, %r43, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f24, [%rd14]; add.s32 %r44, %r42, %r5; shl.b32 %r45, %r44, 2; mov.u32 %r46, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r47, %r46, %r45; for.size.2 i in range(0, 15, 1): st.shared.f32 [%r47+(0+i*8)], %f(24+i*1); ld.global.nc.f32 %f(25+i*1), [%rd14+(8+i*8)]; st.shared.f32 [%r47+120], %f39; add.s32 %r127, %r127, 16; setp.ne.s32 %p3, %r127, 5408; @%p3 bra $L__BB0_6; shl.b32 %r49, %r126, 6; add.s32 %r13, %r49, %r7; mov.u32 %r128, 0; $L__BB0_8: shl.b32 %r50, %r128, 1; add.s32 %r51, %r50, %r5; shl.b32 %r52, %r51, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(53+i*1), %r(52+i*-1), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(55+i*1), %r(13+i*42), %r(54+i*-1); mul.wide.s32 %rd15, %r56, 4; add.s64 %rd16, %rd2, %rd15; ld.global.nc.f32 %f40, [%rd16]; shl.b32 %r57, %r51, 2; mov.u32 %r58, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r59, %r58, %r57; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r59+(0+i*8)], %f(40+i*1); add.s32 %r(60+i*6), %r51, (2+i*2); shl.b32 %r(61+i*6), %r(60+i*6), 3; and.b32 %r(62+i*6), %r(61+i*6), -512; and.b32 %r(63+i*6), %r(60+i*6), 63; add.s32 %r(64+i*6), %r13, %r(63+i*6); add.s32 %r(65+i*6), %r(64+i*6), %r(62+i*6); mul.wide.s32 %rd(17+i*2), %r(65+i*6), 4; add.s64 %rd(18+i*2), %rd2, %rd(17+i*2); ld.global.nc.f32 %f(41+i*1), [%rd(18+i*2)+0]; st.shared.f32 [%r59+56], %f47; add.s32 %r128, %r128, 8; setp.ne.s32 %p4, %r128, 6400; @%p4 bra $L__BB0_8; bar.sync 0; mov.u32 %r102, 0; mov.u32 %r129, %r102; $L__BB0_10: mul.lo.s32 %r17, %r129, 676; shl.b32 %r104, %r129, 2; add.s32 %r18, %r104, %r8; mov.u32 %r130, %r102; $L__BB0_11: add.s32 %r20, %r130, %r17; mov.u32 %r105, 0; mov.u32 %r131, %r105; $L__BB0_12: mad.lo.s32 %r107, %r131, 169, %r20; add.s32 %r22, %r18, %r131; shl.b32 %r108, %r107, 2; add.s32 %r110, %r46, %r108; for.size.1 i in range(0, 13, 1): ld.shared.f32 %f(1+i*1), [%r110+(0+i*52)]; mov.u32 %r132, %r105; $L__BB0_13: mad.lo.s32 %r111, %r132, 169, %r130; for.size.2 i in range(0, 2, 1): shl.b32 %r(112+i*2), %r(132+i*-19), (6+i*-4); add.s32 %r(113+i*3), %r(22+i*36), %r(112+i*2); for.size.1 i in range(0, 9, 1): ld.shared.f32 %f(48+i*1), [%r116+(5120+i*5120)]; mul.wide.s32 %rd31, %r111, 4; add.s64 %rd32, %rd5, %rd31; ld.shared.f32 %f57, [%r116]; for.size.8 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): ld.local.f32 %f(58+i*20+j*2), [%rd32+(0+i*52+j*6760)]; fma.rn.f32 %f(59+i*20+j*2), %f(1+j*1), %f(57+j*-9), %f(58+i*20+j*2); st.local.f32 [%rd32+(0+i*52+j*6760)], %f(59+i*20+j*2); for.size.3 j in range(0, 8, 1): ld.local.f32 %f(62+i*20+j*2), [%rd32+(13520+i*52+j*6760)]; fma.rn.f32 %f(63+i*20+j*2), %f(1+j*1), %f(49+j*1), %f(62+i*20+j*2); st.local.f32 [%rd32+(13520+i*52+j*6760)], %f(63+i*20+j*2); add.s32 %r132, %r132, 1; setp.ne.s32 %p5, %r132, 10; @%p5 bra $L__BB0_13; add.s32 %r131, %r131, 1; setp.lt.u32 %p6, %r131, 4; @%p6 bra $L__BB0_12; add.s32 %r130, %r130, 1; setp.lt.u32 %p7, %r130, 13; @%p7 bra $L__BB0_11; add.s32 %r129, %r129, 1; setp.lt.u32 %p8, %r129, 16; @%p8 bra $L__BB0_10; add.s32 %r126, %r126, 1; setp.lt.u32 %p9, %r126, 8; @%p9 bra $L__BB0_5; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(118+i*1), %r5, (10+i*1680); mad.lo.s32 %r(29+i*1), %r6, (200+i*33600), %r(118+i*1); mov.u32 %r117, 0; mov.u32 %r133, %r117; $L__BB0_19: mul.lo.s32 %r32, %r133, 169; add.s32 %r121, %r29, %r133; mul.wide.s32 %rd33, %r121, 4; add.s64 %rd34, %rd4, %rd33; for.size.1 i in range(0, 10, 1): ld.global.nc.f32 %f(14+i*1), [%rd34+(0+i*80)]; mov.u32 %r134, %r117; $L__BB0_20: mad.lo.s32 %r122, %r134, 13, %r32; add.s32 %r123, %r122, %r30; mul.wide.s32 %rd35, %r122, 4; add.s64 %rd36, %rd5, %rd35; ld.local.f32 %f318, [%rd36]; add.f32 %f319, %f318, %f14; mov.f32 %f320, 0f00000000; max.f32 %f321, %f319, %f320; mul.wide.s32 %rd37, %r123, 4; add.s64 %rd38, %rd3, %rd37; for.size.4 i in range(0, 9, 1): st.global.f32 [%rd38+(0+i*13520)], %f(321+i*3); ld.local.f32 %f(322+i*3), [%rd36+(6760+i*6760)]; add.f32 %f(323+i*3), %f(322+i*3), %f(15+i*1); max.f32 %f(324+i*3), %f(323+i*3), %f320; for.size.10 i in range(0, 12, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd38+(121680+i*4+j*-121676)], %f(348+i*30+j*3); ld.local.f32 %f(349+i*30+j*3), [%rd36+(4+i*4+j*6760)]; add.f32 %f(350+i*30+j*3), %f(349+i*30+j*3), %f(14+j*1); max.f32 %f(351+i*30+j*3), %f(350+i*30+j*3), %f320; for.size.4 j in range(0, 8, 1): st.global.f32 [%rd38+(13524+i*4+j*13520)], %f(354+i*30+j*3); ld.local.f32 %f(355+i*30+j*3), [%rd36+(13524+i*4+j*6760)]; add.f32 %f(356+i*30+j*3), %f(355+i*30+j*3), %f(16+j*1); max.f32 %f(357+i*30+j*3), %f(356+i*30+j*3), %f320; st.global.f32 [%rd38+121728], %f708; add.s32 %r134, %r134, 1; setp.ne.s32 %p10, %r134, 13; @%p10 bra $L__BB0_20; add.s32 %r133, %r133, 1; setp.lt.u32 %p11, %r133, 10; @%p11 bra $L__BB0_19; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16900]; __shared__ float pad_temp_shared[10816]; __shared__ float placeholder_shared[12800]; for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 10; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_inner_init * 169 + yy_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 1690] = 0.000000e+00f; compute[index0 + 3380] = 0.000000e+00f; compute[index0 + 5070] = 0.000000e+00f; compute[index0 + 6760] = 0.000000e+00f; compute[index0 + 8450] = 0.000000e+00f; compute[index0 + 10140] = 0.000000e+00f; compute[index0 + 11830] = 0.000000e+00f; compute[index0 + 13520] = 0.000000e+00f; compute[index0 + 15210] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 5408; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2 + (int)threadIdx.x; int index2 = rc_outer_outer * 10816 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 6400; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x; int index4 = (int)blockIdx.x * 102400 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int ff_inner = 0; ff_inner < 10; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = ff_inner * 169 + yy_inner * 13 + xx_outer_inner; int index6 = rc_outer_inner * 676 + rc_inner * 169 + yy_inner * 13 + xx_outer_inner; int index7 = (int)threadIdx.x * 640 + ff_inner * 64 + rc_outer_inner * 4 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 1690] = compute[index5 + 1690] + (pad_temp_shared[index6] * placeholder_shared[index7 + 1280]); compute[index5 + 3380] = compute[index5 + 3380] + (pad_temp_shared[index6] * placeholder_shared[index7 + 2560]); compute[index5 + 5070] = compute[index5 + 5070] + (pad_temp_shared[index6] * placeholder_shared[index7 + 3840]); compute[index5 + 6760] = compute[index5 + 6760] + (pad_temp_shared[index6] * placeholder_shared[index7 + 5120]); compute[index5 + 8450] = compute[index5 + 8450] + (pad_temp_shared[index6] * placeholder_shared[index7 + 6400]); compute[index5 + 10140] = compute[index5 + 10140] + (pad_temp_shared[index6] * placeholder_shared[index7 + 7680]); compute[index5 + 11830] = compute[index5 + 11830] + (pad_temp_shared[index6] * placeholder_shared[index7 + 8960]); compute[index5 + 13520] = compute[index5 + 13520] + (pad_temp_shared[index6] * placeholder_shared[index7 + 10240]); compute[index5 + 15210] = compute[index5 + 15210] + (pad_temp_shared[index6] * placeholder_shared[index7 + 11520]); } } } } } } for (int ax1_inner = 0; ax1_inner < 10; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)blockIdx.x * 33800 + (int)threadIdx.x * 1690 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index9 = ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index10 = (int)blockIdx.x * 200 + (int)threadIdx.x * 10 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 3380] = max(compute[index9 + 1690] + placeholder2[index10 + 20], 0.000000e+00f); T_relu[index8 + 6760] = max(compute[index9 + 3380] + placeholder2[index10 + 40], 0.000000e+00f); T_relu[index8 + 10140] = max(compute[index9 + 5070] + placeholder2[index10 + 60], 0.000000e+00f); T_relu[index8 + 13520] = max(compute[index9 + 6760] + placeholder2[index10 + 80], 0.000000e+00f); T_relu[index8 + 16900] = max(compute[index9 + 8450] + placeholder2[index10 + 100], 0.000000e+00f); T_relu[index8 + 20280] = max(compute[index9 + 10140] + placeholder2[index10 + 120], 0.000000e+00f); T_relu[index8 + 23660] = max(compute[index9 + 11830] + placeholder2[index10 + 140], 0.000000e+00f); T_relu[index8 + 27040] = max(compute[index9 + 13520] + placeholder2[index10 + 160], 0.000000e+00f); T_relu[index8 + 30420] = max(compute[index9 + 15210] + placeholder2[index10 + 180], 0.000000e+00f); } } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpva7qenui
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<12>; .reg .f32 %f<1015>; .reg .b32 %r<152>; .reg .b64 %rd<41>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[3328]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256000]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd3, %rd6; mov.u32 %r1, %tid.x; shl.b32 %r39, %r1, 1; mov.u32 %r2, %ctaid.x; mul.hi.s32 %r40, %r1, 1321528399; shr.u32 %r41, %r40, 31; shr.s32 %r42, %r40, 2; add.s32 %r4, %r42, %r41; mul.lo.s32 %r43, %r4, 13; sub.s32 %r3, %r1, %r43; mul.lo.s32 %r5, %r4, 6400; mul.hi.s32 %r44, %r39, 1321528399; shr.u32 %r45, %r44, 31; shr.s32 %r46, %r44, 2; add.s32 %r47, %r46, %r45; and.b32 %r48, %r47, 63; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(6+i*43), %r(48+i*-1), (169+i*-156); sub.s32 %r50, %r39, %r49; mul.lo.s32 %r7, %r50, 13; mad.lo.s32 %r51, %r48, 13, %r50; shl.b32 %r52, %r51, 2; mov.u32 %r53, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r8, %r53, %r52; or.b32 %r54, %r39, 1; mul.hi.s32 %r55, %r54, 1321528399; shr.u32 %r56, %r55, 31; shr.s32 %r57, %r55, 2; add.s32 %r58, %r57, %r56; and.b32 %r59, %r58, 63; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(9+i*51), %r(59+i*-1), (169+i*-156); sub.s32 %r61, %r54, %r60; for.size.6 i in range(0, 2, 1): mul.lo.s32 %r(10+i*2), %r(61+i*4), (13+i*156); mad.lo.s32 %r(62+i*4), %r(59+i*6), 13, %r(61+i*-11); shl.b32 %r(63+i*4), %r(62+i*4), 2; add.s32 %r(11+i*2), %r53, %r(63+i*4); add.s32 %r(64+i*4), %r(47+i*11), 20; and.b32 %r(65+i*4), %r(64+i*4), 63; for.size.7 i in range(0, 2, 1): for.size.6 j in range(0, 2, 1): mul.lo.s32 %r(14+i*4+j*2), %r(69+i*8+j*4), 169; mad.lo.s32 %r(70+i*8+j*4), %r(69+i*8+j*4), 13, %r(61+j*-11); shl.b32 %r(71+i*8+j*4), %r(70+i*8+j*4), 2; add.s32 %r(15+i*4+j*2), %r53, %r(71+i*8+j*4); add.s32 %r(72+i*8+j*4), %r(47+j*11), (40+j*20); and.b32 %r(73+i*8+j*4), %r(72+i*8+j*4), 63; mul.lo.s32 %r22, %r85, 169; mad.lo.s32 %r86, %r85, 13, %r61; shl.b32 %r87, %r86, 2; add.s32 %r23, %r53, %r87; mov.u32 %r149, 0; mov.f32 %f815, 0f00000000; for.size.1 i in range(0, 99, 1): mov.f32 %f(816+i*1), %f815; $L__BB0_1: bar.sync 0; mad.lo.s32 %r25, %r149, 10816, %r2; setp.gt.s32 %p1, %r1, 415; @%p1 bra $L__BB0_3; for.size.1 i in range(0, 2, 1): add.s32 %r(88+i*1), %r(25+i*63), %r(7+i*-1); mul.wide.s32 %rd9, %r89, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f401, [%rd10]; st.shared.f32 [%r8], %f401; for.size.1 i in range(0, 2, 1): add.s32 %r(90+i*1), %r(25+i*65), %r(10+i*-1); mul.wide.s32 %rd11, %r91, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f402, [%rd12]; st.shared.f32 [%r11], %f402; $L__BB0_3: setp.gt.s32 %p2, %r1, 285; @%p2 bra $L__BB0_5; for.size.1 i in range(0, 2, 1): add.s32 %r(92+i*1), %r(25+i*67), %r(7+i*5); mul.wide.s32 %rd13, %r93, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f403, [%rd14]; st.shared.f32 [%r13], %f403; for.size.1 i in range(0, 2, 1): add.s32 %r(94+i*1), %r(25+i*69), %r(10+i*4); mul.wide.s32 %rd15, %r95, 4; add.s64 %rd16, %rd1, %rd15; ld.global.nc.f32 %f404, [%rd16]; st.shared.f32 [%r15], %f404; $L__BB0_5: setp.gt.s32 %p3, %r1, 155; @%p3 bra $L__BB0_7; for.size.1 i in range(0, 2, 1): add.s32 %r(96+i*1), %r(25+i*71), %r(7+i*9); mul.wide.s32 %rd17, %r97, 4; add.s64 %rd18, %rd1, %rd17; ld.global.nc.f32 %f405, [%rd18]; st.shared.f32 [%r17], %f405; for.size.1 i in range(0, 2, 1): add.s32 %r(98+i*1), %r(25+i*73), %r(10+i*8); mul.wide.s32 %rd19, %r99, 4; add.s64 %rd20, %rd1, %rd19; ld.global.nc.f32 %f406, [%rd20]; st.shared.f32 [%r19], %f406; $L__BB0_7: setp.gt.s32 %p4, %r1, 25; @%p4 bra $L__BB0_9; for.size.1 i in range(0, 2, 1): add.s32 %r(100+i*1), %r(25+i*75), %r(7+i*13); mul.wide.s32 %rd21, %r101, 4; add.s64 %rd22, %rd1, %rd21; ld.global.nc.f32 %f407, [%rd22]; st.shared.f32 [%r21], %f407; for.size.1 i in range(0, 2, 1): add.s32 %r(102+i*1), %r(25+i*77), %r(10+i*12); mul.wide.s32 %rd23, %r103, 4; add.s64 %rd24, %rd1, %rd23; ld.global.nc.f32 %f408, [%rd24]; st.shared.f32 [%r23], %f408; $L__BB0_9: mov.u32 %r150, 0; bra.uni $L__BB0_10; $L__BB0_19: add.s32 %r150, %r150, 4; $L__BB0_10: mad.lo.s32 %r28, %r150, 130, %r1; setp.gt.s32 %p5, %r28, 63999; shl.b32 %r105, %r28, 2; mov.u32 %r106, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r29, %r106, %r105; @%p5 bra $L__BB0_12; for.size.1 i in range(0, 2, 1): shl.b32 %r(139+i*-32), %r(149+i*-121), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(108+i*1), %r(107+i*-79), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(110+i*1), %r(109+i*1), %r(139+i*-31); mul.wide.s32 %rd25, %r111, 4; add.s64 %rd26, %rd3, %rd25; ld.global.nc.f32 %f409, [%rd26]; st.shared.f32 [%r29], %f409; $L__BB0_12: add.s32 %r30, %r150, 1; setp.eq.s32 %p6, %r30, 493; @%p6 bra $L__BB0_20; mad.lo.s32 %r31, %r30, 130, %r1; setp.gt.s32 %p7, %r31, 63999; @%p7 bra $L__BB0_15; for.size.1 i in range(0, 2, 1): shl.b32 %r(142+i*-30), %r(149+i*-118), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(113+i*1), %r(112+i*-81), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(115+i*1), %r(114+i*1), %r(142+i*-29); mul.wide.s32 %rd27, %r116, 4; add.s64 %rd28, %rd3, %rd27; ld.global.nc.f32 %f410, [%rd28]; st.shared.f32 [%r29+520], %f410; $L__BB0_15: add.s32 %r32, %r28, 260; setp.gt.s32 %p8, %r32, 63999; @%p8 bra $L__BB0_17; for.size.1 i in range(0, 2, 1): shl.b32 %r(141+i*-24), %r(149+i*-117), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(118+i*1), %r(117+i*-85), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(120+i*1), %r(119+i*1), %r(141+i*-23); mul.wide.s32 %rd29, %r121, 4; add.s64 %rd30, %rd3, %rd29; ld.global.nc.f32 %f411, [%rd30]; st.shared.f32 [%r29+1040], %f411; $L__BB0_17: add.s32 %r33, %r28, 390; setp.gt.s32 %p9, %r33, 63999; @%p9 bra $L__BB0_19; for.size.1 i in range(0, 2, 1): shl.b32 %r(140+i*-18), %r(149+i*-116), (6+i*-3); for.size.1 i in range(0, 2, 1): and.b32 %r(123+i*1), %r(122+i*-89), (-512+i*575); for.size.1 i in range(0, 2, 1): add.s32 %r(125+i*1), %r(124+i*1), %r(140+i*-17); mul.wide.s32 %rd31, %r126, 4; add.s64 %rd32, %rd3, %rd31; ld.global.nc.f32 %f412, [%rd32]; st.shared.f32 [%r29+1560], %f412; bra.uni $L__BB0_19; $L__BB0_20: bar.sync 0; mov.u32 %r151, 0; $L__BB0_21: mov.u32 %r144, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r143, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r128, %r151, 13, %r3; shl.b32 %r129, %r128, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(131+i*1), %r(143+i*8), %r(129+i*-124); shl.b32 %r133, %r132, 2; add.s32 %r135, %r144, %r133; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(413+i*1), [%r(135+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(914+i*-1), %f414, %f(413+i*2), %f(914+i*-1); ld.shared.f32 %f(415+i*1), [%r135+(256+i*256)]; for.size.2 i in range(0, 97, 1): fma.rn.f32 %f(912+i*-1), %f414, %f(416+i*1), %f(912+i*-1); ld.shared.f32 %f(417+i*1), [%r135+(768+i*256)]; fma.rn.f32 %f815, %f414, %f513, %f815; add.s32 %r151, %r151, 1; setp.ne.s32 %p10, %r151, 64; @%p10 bra $L__BB0_21; add.s32 %r149, %r149, 1; setp.lt.u32 %p11, %r149, 8; @%p11 bra $L__BB0_1; ld.param.u64 %rd40, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd39, %rd40; ld.param.u64 %rd38, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd37, %rd38; mul.hi.s32 %r148, %r1, 1321528399; shr.u32 %r147, %r148, 31; shr.s32 %r146, %r148, 2; add.s32 %r145, %r146, %r147; mul.lo.s32 %r136, %r145, 100; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(137+i*1), %r(145+i*-142), (16900+i*-16887), %r(2+i*135); mul.wide.s32 %rd33, %r136, 4; add.s64 %rd34, %rd37, %rd33; ld.global.nc.f32 %f514, [%rd34]; add.f32 %f515, %f914, %f514; mov.f32 %f516, 0f00000000; max.f32 %f517, %f515, %f516; mul.wide.s32 %rd35, %r138, 4; add.s64 %rd36, %rd39, %rd35; for.size.4 i in range(0, 99, 1): st.global.f32 [%rd36+(0+i*676)], %f(517+i*3); ld.global.nc.f32 %f(518+i*3), [%rd34+(4+i*4)]; add.f32 %f(519+i*3), %f(913+i*-1), %f(518+i*3); max.f32 %f(520+i*3), %f(519+i*3), %f516; st.global.f32 [%rd36+66924], %f814; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[100]; __shared__ float pad_temp_shared[832]; __shared__ float placeholder_shared[64000]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 25; ++ff_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 4; ++ff_inner_init) { int index0 = ff_outer_inner_init * 4 + ff_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 4; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { for (int ax0_ax1_fused_ax2_fused_ax3_fused_inner_s = 0; ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 2; ++ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 260 + (int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s < 832) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 130 + (int)threadIdx.x < 416) { int index1 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) / 13 & 63) * 13 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) % 13; int index2 = rc_outer_outer * 10816 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) / 13 & 63) * 169 + ((int)threadIdx.x * 2 + ax0_ax1_fused_ax2_fused_ax3_fused_inner_s) % 13 * 13 + (int)blockIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 493; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 130 + (int)threadIdx.x < 64000) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 130 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 130 + (int)threadIdx.x >> 6) * 512 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 130 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 25; ++ff_outer_inner) { for (int ff_inner = 0; ff_inner < 4; ++ff_inner) { int index5 = ff_outer_inner * 4 + ff_inner; int index6 = rc_outer_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x / 13 * 6400 + ff_outer_inner * 256 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } for (int ax1_inner = 0; ax1_inner < 100; ++ax1_inner) { int index8 = (int)threadIdx.x / 13 * 16900 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x; int index9 = ax1_inner; int index10 = (int)threadIdx.x / 13 * 100 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmprgiz6es1
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<61>; .reg .b32 %r<50>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[5120]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd4, [default_function_kernel0_param_2]; ld.param.u64 %rd5, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd7; mov.u32 %r20, %ctaid.x; mul.lo.s32 %r21, %r20, 20480; mov.u32 %r1, %tid.x; shl.b32 %r22, %r1, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(23+i*1), %r(22+i*-21), (-512+i*543); or.b32 %r25, %r24, %r21; add.s32 %r2, %r25, %r23; shl.b32 %r26, %r1, 2; mov.u32 %r27, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r27, %r26; mul.hi.s32 %r28, %r1, 1626496491; shr.u32 %r29, %r28, 31; shr.s32 %r30, %r28, 6; add.s32 %r5, %r30, %r29; mul.lo.s32 %r31, %r5, 169; sub.s32 %r4, %r1, %r31; mov.u32 %r32, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r6, %r32, %r26; for.size.3 i in range(0, 2, 1): shl.b32 %r(33+i*2), %r(4+i*1), (2+i*6); add.s32 %r(34+i*2), %r(32+i*-5), %r(33+i*2); add.s32 %r(7+i*1), %r(34+i*2), (2704+i*-2576); mov.u32 %r46, 0; mov.f32 %f59, 0f00000000; mov.f32 %f60, %f59; $L__BB0_1: bar.sync 0; mad.lo.s32 %r37, %r46, 5408, %r1; mul.wide.s32 %rd8, %r37, 4; add.s64 %rd3, %rd1, %rd8; setp.gt.s32 %p1, %r1, 5407; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f9, [%rd3]; st.shared.f32 [%r6], %f9; $L__BB0_3: setp.gt.s32 %p2, %r1, 2027; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f10, [%rd3+13520]; st.shared.f32 [%r6+13520], %f10; $L__BB0_5: setp.gt.s32 %p3, %r1, 1279; @%p3 bra $L__BB0_7; shl.b32 %r38, %r46, 5; add.s32 %r39, %r2, %r38; mul.wide.s32 %rd9, %r39, 4; add.s64 %rd10, %rd2, %rd9; ld.global.nc.f32 %f11, [%rd10]; st.shared.f32 [%r3], %f11; $L__BB0_7: bar.sync 0; mov.u32 %r49, 2704; mov.u32 %r47, %r8; mov.u32 %r48, %r7; $L__BB0_8: for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(12+i*5), [%r47+(-128+i*4)]; ld.shared.f32 %f(13+i*5), [%r48+(-2704+i*676)]; fma.rn.f32 %f(14+i*5), %f(13+i*5), %f(12+i*5), %f(60+i*-46); ld.shared.f32 %f(15+i*5), [%r47+(0+i*4)]; fma.rn.f32 %f(16+i*5), %f(13+i*5), %f(15+i*5), %f(59+i*-43); for.size.5 i in range(0, 5, 1): ld.shared.f32 %f(22+i*5), [%r47+(-120+i*4)]; ld.shared.f32 %f(23+i*5), [%r48+(-1352+i*676)]; fma.rn.f32 %f(24+i*5), %f(23+i*5), %f(22+i*5), %f(19+i*5); ld.shared.f32 %f(25+i*5), [%r47+(8+i*4)]; fma.rn.f32 %f(26+i*5), %f(23+i*5), %f(25+i*5), %f(21+i*5); ld.shared.f32 %f47, [%r47+-100]; ld.shared.f32 %f48, [%r48+2028]; fma.rn.f32 %f60, %f48, %f47, %f44; ld.shared.f32 %f49, [%r47+28]; fma.rn.f32 %f59, %f48, %f49, %f46; for.size.1 i in range(0, 2, 1): add.s32 %r(48+i*-1), %r(48+i*-1), (5408+i*-5376); add.s32 %r49, %r49, 5408; setp.ne.s32 %p4, %r49, 24336; @%p4 bra $L__BB0_8; add.s32 %r46, %r46, 1; setp.lt.u32 %p5, %r46, 16; @%p5 bra $L__BB0_1; shl.b32 %r42, %r5, 1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(43+i*1), %r20, (40+i*6720), %r(42+i*-38); mad.lo.s32 %r45, %r5, 338, %r44; cvta.to.global.u64 %rd11, %rd5; mul.wide.s32 %rd12, %r43, 4; add.s64 %rd13, %rd11, %rd12; ld.global.nc.f32 %f50, [%rd13]; add.f32 %f51, %f60, %f50; mov.f32 %f52, 0f00000000; max.f32 %f53, %f51, %f52; cvta.to.global.u64 %rd14, %rd4; mul.wide.s32 %rd15, %r45, 4; add.s64 %rd16, %rd14, %rd15; st.global.f32 [%rd16], %f53; ld.global.nc.f32 %f54, [%rd13+4]; add.f32 %f55, %f59, %f54; max.f32 %f56, %f55, %f52; st.global.f32 [%rd16+676], %f56; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[1280]; for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 2; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 20 + (int)threadIdx.x / 169 < 32) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 260 + (int)threadIdx.x / 13 < 416) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3380 + (int)threadIdx.x < 5408) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3380 + (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3380 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } } } if ((int)threadIdx.x < 1280) { int index3 = (int)threadIdx.x; int index4 = (int)blockIdx.x * 20480 + ((int)threadIdx.x >> 5) * 512 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index5 = ff_inner; int index6 = rc_outer_inner * 338 + rc_inner * 169 + (int)threadIdx.x % 169; int index7 = (int)threadIdx.x / 169 * 64 + ff_inner * 32 + rc_outer_inner * 2 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index8 = (int)blockIdx.x * 6760 + (int)threadIdx.x / 169 * 338 + ax1_inner * 169 + (int)threadIdx.x % 169; int index9 = ax1_inner; int index10 = (int)blockIdx.x * 40 + (int)threadIdx.x / 169 * 2 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmp3ywg6rxa
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<209>; .reg .b32 %r<40>; .reg .b64 %rd<30>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[104]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[320]; ld.param.u64 %rd12, [default_function_kernel0_param_0]; ld.param.u64 %rd13, [default_function_kernel0_param_1]; ld.param.u64 %rd14, [default_function_kernel0_param_2]; ld.param.u64 %rd11, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd15, %rd13; cvta.to.global.u64 %rd1, %rd14; mov.u32 %r13, %ctaid.x; mul.hi.s32 %r14, %r13, 1321528399; shr.u32 %r15, %r14, 31; shr.s32 %r16, %r14, 2; add.s32 %r4, %r16, %r15; mul.lo.s32 %r17, %r4, 13; sub.s32 %r1, %r13, %r17; mov.u32 %r2, %tid.x; shl.b32 %r18, %r2, 2; mov.u32 %r19, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r19, %r18; and.b32 %r20, %r2, 1; mov.u32 %r21, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r21, %r18; mul.hi.s32 %r22, %r2, 1321528399; shr.u32 %r23, %r22, 31; shr.s32 %r24, %r22, 2; add.s32 %r8, %r24, %r23; mul.lo.s32 %r25, %r8, 13; sub.s32 %r6, %r2, %r25; shl.b32 %r26, %r6, 2; add.s32 %r7, %r19, %r26; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(9+i*18), %r(8+i*-6), (160+i*-147), %r(21+i*-20); cvta.to.global.u64 %rd16, %rd12; mul.wide.s32 %rd17, %r27, 4; add.s64 %rd29, %rd16, %rd17; mul.lo.s32 %r28, %r4, 20480; add.s32 %r29, %r2, %r28; mul.wide.s32 %rd18, %r29, 4; add.s64 %rd19, %rd15, %rd18; add.s64 %rd28, %rd19, 79872; shl.b32 %r30, %r2, 8; and.b32 %r31, %r30, -512; add.s32 %r32, %r28, %r31; or.b32 %r33, %r32, %r20; mul.wide.s32 %rd20, %r33, 4; add.s64 %rd21, %rd15, %rd20; add.s64 %rd27, %rd21, 26624; mov.u32 %r39, 0; mov.f32 %f189, 0f00000000; for.size.1 i in range(0, 19, 1): mov.f32 %f(190+i*1), %f189; $L__BB0_1: bar.sync 0; ld.global.nc.f32 %f61, [%rd29]; for.size.2 i in range(0, 2, 1): st.shared.f32 [%r(3+i*2)+0], %f(61+i*1); ld.global.nc.f32 %f(62+i*1), [%rd27+(-26624+i*26624)]; st.shared.f32 [%r5+104], %f63; ld.global.nc.f32 %f64, [%rd27+26624]; st.shared.f32 [%r5+208], %f64; setp.gt.s32 %p1, %r2, 1; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f65, [%rd28]; st.shared.f32 [%r5+312], %f65; $L__BB0_3: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(66+i*1), [%r(9+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(68+i*2), %f67, %f(66+i*3), %f(208+i*-1); ld.shared.f32 %f(69+i*2), [%r9+(8+i*8)]; for.size.2 i in range(0, 17, 1): fma.rn.f32 %f(72+i*2), %f67, %f(71+i*2), %f(206+i*-1); ld.shared.f32 %f(73+i*2), [%r9+(24+i*8)]; fma.rn.f32 %f106, %f67, %f105, %f189; ld.shared.f32 %f107, [%r9+4]; for.size.12 i in range(0, 2, 1): ld.shared.f32 %f(108+i*6), [%r(7+i*2)+52]; fma.rn.f32 %f(208+i*-6), %f108, %f(107+i*7), %f(68+i*12); ld.shared.f32 %f(109+i*6), [%r9+(12+i*48)]; fma.rn.f32 %f(207+i*-6), %f108, %f(109+i*6), %f(70+i*12); ld.shared.f32 %f(110+i*6), [%r9+(20+i*48)]; fma.rn.f32 %f(206+i*-6), %f108, %f(110+i*6), %f(72+i*12); ld.shared.f32 %f(111+i*6), [%r9+(28+i*48)]; fma.rn.f32 %f(205+i*-6), %f108, %f(111+i*6), %f(74+i*12); ld.shared.f32 %f(112+i*6), [%r9+(36+i*48)]; fma.rn.f32 %f(204+i*-6), %f108, %f(112+i*6), %f(76+i*12); ld.shared.f32 %f(113+i*6), [%r9+(44+i*48)]; fma.rn.f32 %f(203+i*-6), %f108, %f(113+i*6), %f(78+i*12); for.size.2 i in range(0, 8, 1): ld.shared.f32 %f(120+i*1), [%r9+(100+i*8)]; fma.rn.f32 %f(196+i*-1), %f108, %f(120+i*1), %f(92+i*2); for.size.1 i in range(0, 2, 1): add.s64 %rd(29+i*-1), %rd(29+i*-1), (1352+i*-1344); add.s64 %rd27, %rd27, 8; add.s32 %r39, %r39, 1; setp.ne.s32 %p2, %r39, 256; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd22, %rd11; mul.lo.s32 %r34, %r8, 20; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(35+i*1), %r(4+i*4), (40+i*3340), %r(34+i*-33); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(37+i*1), %r(4+i*2), (6760+i*-6747), %r(36+i*1); mul.wide.s32 %rd23, %r35, 4; add.s64 %rd24, %rd22, %rd23; ld.global.nc.f32 %f128, [%rd24]; add.f32 %f129, %f208, %f128; mov.f32 %f130, 0f00000000; max.f32 %f131, %f129, %f130; mul.wide.s32 %rd25, %r38, 4; add.s64 %rd26, %rd1, %rd25; for.size.4 i in range(0, 19, 1): st.global.f32 [%rd26+(0+i*676)], %f(131+i*3); ld.global.nc.f32 %f(132+i*3), [%rd24+(4+i*4)]; add.f32 %f(133+i*3), %f(207+i*-1), %f(132+i*3); max.f32 %f(134+i*3), %f(133+i*3), %f130; st.global.f32 [%rd26+12844], %f188; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[20]; __shared__ float pad_temp_shared[26]; __shared__ float placeholder_shared[80]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[19] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 256; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 338 + (int)threadIdx.x * 13 + (int)blockIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; int index2 = (int)blockIdx.x / 13 * 20480 + ((int)threadIdx.x >> 1) * 512 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index0] = placeholder1[index2]; placeholder_shared[index0 + 26] = placeholder1[index2 + 6656]; placeholder_shared[index0 + 52] = placeholder1[index2 + 13312]; if ((int)threadIdx.x < 2) { int index3 = (int)blockIdx.x / 13 * 20480 + rc_outer_outer * 2 + (int)threadIdx.x + 19968; placeholder_shared[index0 + 78] = placeholder1[index3]; } __syncthreads(); int index4 = (int)threadIdx.x / 13 * 40; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index4]; compute[1] = compute[1] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 2]); compute[2] = compute[2] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 4]); compute[3] = compute[3] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 6]); compute[4] = compute[4] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 8]); compute[5] = compute[5] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 10]); compute[6] = compute[6] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 12]); compute[7] = compute[7] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 14]); compute[8] = compute[8] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 16]); compute[9] = compute[9] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 18]); compute[10] = compute[10] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 20]); compute[11] = compute[11] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 22]); compute[12] = compute[12] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 24]); compute[13] = compute[13] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 26]); compute[14] = compute[14] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 28]); compute[15] = compute[15] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 30]); compute[16] = compute[16] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 32]); compute[17] = compute[17] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 34]); compute[18] = compute[18] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 36]); compute[19] = compute[19] + (pad_temp_shared[index0 % 13] * placeholder_shared[index4 + 38]); int index5 = (int)threadIdx.x % 13 + 13; compute[0] = compute[0] + (pad_temp_shared[index5] * placeholder_shared[index4 + 1]); compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index4 + 3]); compute[2] = compute[2] + (pad_temp_shared[index5] * placeholder_shared[index4 + 5]); compute[3] = compute[3] + (pad_temp_shared[index5] * placeholder_shared[index4 + 7]); compute[4] = compute[4] + (pad_temp_shared[index5] * placeholder_shared[index4 + 9]); compute[5] = compute[5] + (pad_temp_shared[index5] * placeholder_shared[index4 + 11]); compute[6] = compute[6] + (pad_temp_shared[index5] * placeholder_shared[index4 + 13]); compute[7] = compute[7] + (pad_temp_shared[index5] * placeholder_shared[index4 + 15]); compute[8] = compute[8] + (pad_temp_shared[index5] * placeholder_shared[index4 + 17]); compute[9] = compute[9] + (pad_temp_shared[index5] * placeholder_shared[index4 + 19]); compute[10] = compute[10] + (pad_temp_shared[index5] * placeholder_shared[index4 + 21]); compute[11] = compute[11] + (pad_temp_shared[index5] * placeholder_shared[index4 + 23]); compute[12] = compute[12] + (pad_temp_shared[index5] * placeholder_shared[index4 + 25]); compute[13] = compute[13] + (pad_temp_shared[index5] * placeholder_shared[index4 + 27]); compute[14] = compute[14] + (pad_temp_shared[index5] * placeholder_shared[index4 + 29]); compute[15] = compute[15] + (pad_temp_shared[index5] * placeholder_shared[index4 + 31]); compute[16] = compute[16] + (pad_temp_shared[index5] * placeholder_shared[index4 + 33]); compute[17] = compute[17] + (pad_temp_shared[index5] * placeholder_shared[index4 + 35]); compute[18] = compute[18] + (pad_temp_shared[index5] * placeholder_shared[index4 + 37]); compute[19] = compute[19] + (pad_temp_shared[index5] * placeholder_shared[index4 + 39]); } for (int ax1_inner = 0; ax1_inner < 20; ++ax1_inner) { int index6 = (int)blockIdx.x / 13 * 6760 + (int)threadIdx.x / 13 * 3380 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x % 13; int index7 = ax1_inner; int index8 = (int)blockIdx.x / 13 * 40 + (int)threadIdx.x / 13 * 20 + ax1_inner; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmppibzvq7i
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<7>; .reg .f32 %f<1882>; .reg .b32 %r<111>; .reg .b64 %rd<49>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[10816]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64000]; ld.param.u64 %rd2, [default_function_kernel0_param_0]; ld.param.u64 %rd3, [default_function_kernel0_param_1]; mov.u32 %r108, 0; mov.f32 %f1713, 0f00000000; mov.u32 %r9, %tid.x; cvta.to.global.u64 %rd6, %rd2; shl.b32 %r12, %r9, 2; mov.u32 %r13, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r14, %r13, %r12; cvta.to.global.u64 %rd8, %rd3; for.size.1 i in range(0, 168, 1): mov.f32 %f(1714+i*1), %f1713; $L__BB0_1: setp.gt.s32 %p1, %r9, 2703; bar.sync 0; mad.lo.s32 %r10, %r108, 2704, %r9; mul.wide.s32 %rd7, %r10, 4; add.s64 %rd1, %rd6, %rd7; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f846, [%rd1]; st.shared.f32 [%r14], %f846; $L__BB0_3: setp.gt.s32 %p2, %r9, 1703; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f847, [%rd1+4000]; st.shared.f32 [%r14+4000], %f847; $L__BB0_5: setp.gt.s32 %p3, %r9, 703; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f848, [%rd1+8000]; st.shared.f32 [%r14+8000], %f848; $L__BB0_7: shl.b32 %r100, %r9, 2; and.b32 %r99, %r9, 15; bfi.b32 %r28, %r108, %r99, 4, 28; shl.b32 %r29, %r9, 5; and.b32 %r30, %r29, -512; add.s32 %r31, %r28, %r30; mul.wide.s32 %rd9, %r31, 4; add.s64 %rd10, %rd8, %rd9; ld.global.nc.f32 %f849, [%rd10]; mov.u32 %r33, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r34, %r33, %r100; mov.u32 %r109, 0; st.shared.f32 [%r34], %f849; add.s32 %r35, %r9, 1000; and.b32 %r36, %r35, 15; bfi.b32 %r37, %r108, %r36, 4, 28; shl.b32 %r38, %r35, 5; for.size.7 i in range(0, 2, 1): and.b32 %r(39+i*3), %r(38+i*3), -512; add.s32 %r(40+i*3), %r(37+i*-9), %r(39+i*3); mul.wide.s32 %rd(11+i*2), %r(40+i*3), 4; add.s64 %rd(12+i*2), %rd8, %rd(11+i*2); ld.global.nc.f32 %f(850+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r34+(4000+i*4000)], %f(850+i*1); add.s32 %r(41+i*3), %r29, (64000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(45+i*3), %r(44+i*3), -512; add.s32 %r(46+i*3), %r(37+i*-9), %r(45+i*3); mul.wide.s32 %rd(15+i*2), %r(46+i*3), 4; add.s64 %rd(16+i*2), %rd8, %rd(15+i*2); ld.global.nc.f32 %f(852+i*1), [%rd(16+i*2)+0]; st.shared.f32 [%r34+(12000+i*4000)], %f(852+i*1); add.s32 %r(47+i*3), %r29, (128000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(51+i*3), %r(50+i*3), -512; add.s32 %r(52+i*3), %r(37+i*-9), %r(51+i*3); mul.wide.s32 %rd(19+i*2), %r(52+i*3), 4; add.s64 %rd(20+i*2), %rd8, %rd(19+i*2); ld.global.nc.f32 %f(854+i*1), [%rd(20+i*2)+0]; st.shared.f32 [%r34+(20000+i*4000)], %f(854+i*1); add.s32 %r(53+i*3), %r29, (192000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(57+i*3), %r(56+i*3), -512; add.s32 %r(58+i*3), %r(37+i*-9), %r(57+i*3); mul.wide.s32 %rd(23+i*2), %r(58+i*3), 4; add.s64 %rd(24+i*2), %rd8, %rd(23+i*2); ld.global.nc.f32 %f(856+i*1), [%rd(24+i*2)+0]; st.shared.f32 [%r34+(28000+i*4000)], %f(856+i*1); add.s32 %r(59+i*3), %r29, (256000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(63+i*3), %r(62+i*3), -512; add.s32 %r(64+i*3), %r(37+i*-9), %r(63+i*3); mul.wide.s32 %rd(27+i*2), %r(64+i*3), 4; add.s64 %rd(28+i*2), %rd8, %rd(27+i*2); ld.global.nc.f32 %f(858+i*1), [%rd(28+i*2)+0]; st.shared.f32 [%r34+(36000+i*4000)], %f(858+i*1); add.s32 %r(65+i*3), %r29, (320000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(69+i*3), %r(68+i*3), -512; add.s32 %r(70+i*3), %r(37+i*-9), %r(69+i*3); mul.wide.s32 %rd(31+i*2), %r(70+i*3), 4; add.s64 %rd(32+i*2), %rd8, %rd(31+i*2); ld.global.nc.f32 %f(860+i*1), [%rd(32+i*2)+0]; st.shared.f32 [%r34+(44000+i*4000)], %f(860+i*1); add.s32 %r(71+i*3), %r29, (384000+i*32000); for.size.7 i in range(0, 2, 1): and.b32 %r(75+i*3), %r(74+i*3), -512; add.s32 %r(76+i*3), %r(37+i*-9), %r(75+i*3); mul.wide.s32 %rd(35+i*2), %r(76+i*3), 4; add.s64 %rd(36+i*2), %rd8, %rd(35+i*2); ld.global.nc.f32 %f(862+i*1), [%rd(36+i*2)+0]; st.shared.f32 [%r34+(52000+i*4000)], %f(862+i*1); add.s32 %r(77+i*3), %r29, (448000+i*32000); and.b32 %r81, %r80, -512; add.s32 %r82, %r37, %r81; mul.wide.s32 %rd39, %r82, 4; add.s64 %rd40, %rd8, %rd39; ld.global.nc.f32 %f864, [%rd40]; st.shared.f32 [%r34+60000], %f864; bar.sync 0; $L__BB0_8: mov.u32 %r110, 0; $L__BB0_9: mov.u32 %r107, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r106, _ZZ24default_function_kernel0E18placeholder_shared; for.size.1 i in range(0, 2, 1): shl.b32 %r(105+i*-1), %r(9+i*100), (4+i*-3); add.s32 %r103, %r104, %r105; mul.lo.s32 %r102, %r109, 338; mad.lo.s32 %r88, %r110, 169, %r102; for.size.2 i in range(0, 2, 1): add.s32 %r(89+i*3), %r(103+i*3), %r(110+i*-20); shl.b32 %r(90+i*3), %r(89+i*-1), 2; add.s32 %r95, %r107, %r93; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(865+i*1), [%r(95+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1881+i*-1), %f(865+i*2), %f866, %f(1881+i*-1); ld.shared.f32 %f(867+i*1), [%r95+(4+i*4)]; for.size.2 i in range(0, 166, 1): fma.rn.f32 %f(1879+i*-1), %f(868+i*1), %f866, %f(1879+i*-1); ld.shared.f32 %f(869+i*1), [%r95+(12+i*4)]; fma.rn.f32 %f1713, %f1034, %f866, %f1713; add.s32 %r96, %r110, 1; mov.u32 %r110, 1; setp.lt.u32 %p4, %r96, 2; @%p4 bra $L__BB0_9; add.s32 %r109, %r109, 1; setp.lt.u32 %p5, %r109, 8; @%p5 bra $L__BB0_8; add.s32 %r108, %r108, 1; setp.lt.u32 %p6, %r108, 32; @%p6 bra $L__BB0_1; ld.param.u64 %rd48, [default_function_kernel0_param_2]; ld.param.u64 %rd47, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd41, %rd47; mul.wide.s32 %rd42, %r9, 4; add.s64 %rd43, %rd41, %rd42; mul.lo.s32 %r98, %r9, 169; ld.global.nc.f32 %f1035, [%rd43]; add.f32 %f1036, %f1881, %f1035; mov.f32 %f1037, 0f00000000; max.f32 %f1038, %f1036, %f1037; cvta.to.global.u64 %rd44, %rd48; mul.wide.s32 %rd45, %r98, 4; add.s64 %rd46, %rd44, %rd45; for.size.3 i in range(0, 168, 1): st.global.f32 [%rd46+(0+i*4)], %f(1038+i*2); add.f32 %f(1039+i*2), %f(1880+i*-1), %f1035; max.f32 %f(1040+i*2), %f(1039+i*2), %f1037; st.global.f32 [%rd46+672], %f1374; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[169]; __shared__ float pad_temp_shared[2704]; __shared__ float placeholder_shared[16000]; for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 13; ++xx_inner_init) { int index0 = yy_inner_init * 13 + xx_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 3; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1000 + (int)threadIdx.x < 2704) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1000 + (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1000 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 16; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 1000 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 1000 + (int)threadIdx.x >> 4) * 512 + rc_outer_outer * 16 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 1000 + (int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 2; ++rc_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { for (int xx_inner = 0; xx_inner < 13; ++xx_inner) { int index5 = yy_inner * 13 + xx_inner; int index6 = rc_outer_inner * 338 + rc_inner * 169 + yy_inner * 13 + xx_inner; int index7 = (int)threadIdx.x * 16 + rc_outer_inner * 2 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = (int)threadIdx.x * 169 + ax2_inner * 13 + ax3_inner; int index9 = ax2_inner * 13 + ax3_inner; int index10 = (int)threadIdx.x; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } }
[677e98c19c9fb48e26671b8368491e3e,1,512,13,13,1000,512,1,1,1,1000,1,1,1,1000,13,13]
tmpvgpgd4m9
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<12>; .reg .f32 %f<142>; .reg .b32 %r<57>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[26624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[204800]; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd8, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd10; cvta.to.global.u64 %rd2, %rd11; mov.u32 %r1, %tid.x; shl.b32 %r2, %r1, 1; mov.u32 %r3, %ctaid.x; mul.hi.s32 %r24, %r3, 1321528399; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 2; add.s32 %r27, %r26, %r25; mul.lo.s32 %r28, %r27, 13; sub.s32 %r4, %r3, %r28; mad.lo.s32 %r5, %r1, 26, %r4; mov.u32 %r52, 0; $L__BB0_1: mad.lo.s32 %r7, %r52, 200, %r2; setp.gt.s32 %p1, %r7, 6655; mad.lo.s32 %r29, %r52, 2600, %r5; mul.wide.s32 %rd12, %r29, 4; add.s64 %rd3, %rd1, %rd12; shl.b32 %r30, %r7, 2; mov.u32 %r55, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r8, %r55, %r30; @%p1 bra $L__BB0_4; mad.lo.s32 %r32, %r52, 100, %r1; setp.gt.s32 %p2, %r32, 3327; @%p2 bra $L__BB0_4; ld.global.nc.f32 %f27, [%rd3]; st.shared.f32 [%r8], %f27; $L__BB0_4: setp.gt.s32 %p3, %r7, 6654; @%p3 bra $L__BB0_7; mad.lo.s32 %r33, %r52, 100, %r1; setp.gt.s32 %p4, %r33, 3327; @%p4 bra $L__BB0_7; ld.global.nc.f32 %f28, [%rd3+52]; st.shared.f32 [%r8+4], %f28; $L__BB0_7: for.size.1 i in range(0, 2, 1): add.s32 %r(9+i*1), %r(52+i*-45), (1+i*199); setp.gt.s32 %p5, %r10, 6655; @%p5 bra $L__BB0_10; mad.lo.s32 %r34, %r9, 100, %r1; setp.gt.s32 %p6, %r34, 3327; @%p6 bra $L__BB0_10; ld.global.nc.f32 %f29, [%rd3+10400]; st.shared.f32 [%r8+800], %f29; $L__BB0_10: setp.gt.s32 %p7, %r10, 6654; @%p7 bra $L__BB0_13; mad.lo.s32 %r35, %r9, 100, %r1; setp.gt.s32 %p8, %r35, 3327; @%p8 bra $L__BB0_13; ld.global.nc.f32 %f30, [%rd3+10452]; st.shared.f32 [%r8+804], %f30; $L__BB0_13: add.s32 %r52, %r52, 2; setp.ne.s32 %p9, %r52, 34; @%p9 bra $L__BB0_1; cvta.to.global.u64 %rd13, %rd8; cvta.to.global.u64 %rd4, %rd9; mad.lo.s32 %r40, %r27, 51200, %r1; mul.wide.s32 %rd14, %r40, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd20, %rd15, 3200; shl.b32 %r41, %r1, 2; mov.u32 %r42, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r43, %r42, %r41; add.s32 %r53, %r43, 3200; mov.u32 %r54, 0; $L__BB0_15: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(31+i*1), [%rd20+(-3200+i*400)]; st.shared.f32 [%r53+(-3200+i*400)], %f(31+i*1); add.s64 %rd20, %rd20, 6400; for.size.1 i in range(0, 2, 1): add.s32 %r(53+i*1), %r(53+i*1), (6400+i*-6384); setp.ne.s32 %p10, %r54, 512; @%p10 bra $L__BB0_15; bar.sync 0; shl.b32 %r46, %r1, 11; add.s32 %r18, %r42, %r46; mov.f32 %f129, 0f00000000; mov.u32 %r56, 4; for.size.1 i in range(0, 12, 1): mov.f32 %f(130+i*1), %f129; $L__BB0_17: add.s32 %r48, %r18, %r56; for.size.27 i in range(0, 2, 1): ld.shared.f32 %f(60+i*27), [%r48+(-4+i*4)]; ld.shared.f32 %f(61+i*27), [%r55+(0+i*52)]; fma.rn.f32 %f(62+i*67), %f(61+i*27), %f(60+i*27), %f(129+i*-67); ld.shared.f32 %f(63+i*26), [%r55+(4+i*52)]; fma.rn.f32 %f(64+i*66), %f(60+i*27), %f(63+i*26), %f(130+i*-66); ld.shared.f32 %f(65+i*25), [%r55+(8+i*52)]; fma.rn.f32 %f(66+i*65), %f(65+i*25), %f(60+i*27), %f(131+i*-65); ld.shared.f32 %f(67+i*24), [%r55+(12+i*52)]; fma.rn.f32 %f(68+i*64), %f(60+i*27), %f(67+i*24), %f(132+i*-64); ld.shared.f32 %f(69+i*23), [%r55+(16+i*52)]; fma.rn.f32 %f(70+i*63), %f(69+i*23), %f(60+i*27), %f(133+i*-63); ld.shared.f32 %f(71+i*22), [%r55+(20+i*52)]; fma.rn.f32 %f(72+i*62), %f(60+i*27), %f(71+i*22), %f(134+i*-62); ld.shared.f32 %f(73+i*21), [%r55+(24+i*52)]; fma.rn.f32 %f(74+i*61), %f(73+i*21), %f(60+i*27), %f(135+i*-61); ld.shared.f32 %f(75+i*20), [%r55+(28+i*52)]; fma.rn.f32 %f(76+i*60), %f(60+i*27), %f(75+i*20), %f(136+i*-60); ld.shared.f32 %f(77+i*19), [%r55+(32+i*52)]; fma.rn.f32 %f(78+i*59), %f(77+i*19), %f(60+i*27), %f(137+i*-59); ld.shared.f32 %f(79+i*18), [%r55+(36+i*52)]; fma.rn.f32 %f(80+i*58), %f(60+i*27), %f(79+i*18), %f(138+i*-58); ld.shared.f32 %f(81+i*17), [%r55+(40+i*52)]; fma.rn.f32 %f(82+i*57), %f(81+i*17), %f(60+i*27), %f(139+i*-57); ld.shared.f32 %f(83+i*16), [%r55+(44+i*52)]; fma.rn.f32 %f(84+i*56), %f(60+i*27), %f(83+i*16), %f(140+i*-56); ld.shared.f32 %f(85+i*15), [%r55+(48+i*52)]; fma.rn.f32 %f(86+i*55), %f(85+i*15), %f(60+i*27), %f(141+i*-55); for.size.1 i in range(0, 2, 1): add.s32 %r(55+i*1), %r(55+i*1), (104+i*-96); setp.ne.s32 %p11, %r56, 2052; @%p11 bra $L__BB0_17; mad.lo.s32 %r49, %r27, 100, %r1; mul.wide.s32 %rd16, %r49, 4; add.s64 %rd17, %rd4, %rd16; ld.global.nc.f32 %f101, [%rd17]; add.f32 %f102, %f129, %f101; mov.f32 %f103, 0f00000000; max.f32 %f104, %f102, %f103; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(50+i*1), %r(1+i*26), (169+i*16731), %r(4+i*46); mul.wide.s32 %rd18, %r51, 4; add.s64 %rd19, %rd2, %rd18; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd19+(0+i*52)], %f(104+i*2); add.f32 %f(105+i*2), %f(130+i*1), %f101; max.f32 %f(106+i*2), %f(105+i*2), %f103; st.global.f32 [%rd19+624], %f128; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[13]; __shared__ float pad_temp_shared[6656]; __shared__ float placeholder_shared[51200]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 34; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x * 2 < 6656) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 3328) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x * 2; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2600 + (int)threadIdx.x * 26 + (int)blockIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; } } if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x * 2 < 6655) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 100 + (int)threadIdx.x < 3328) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 200 + (int)threadIdx.x * 2; int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2600 + (int)threadIdx.x * 26 + (int)blockIdx.x % 13; pad_temp_shared[index0 + 1] = placeholder[index1 + 13]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 512; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; int index3 = (int)blockIdx.x / 13 * 51200 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 100 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 512; ++rc_outer_inner) { int index4 = rc_outer_inner * 13; int index5 = (int)threadIdx.x * 512 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4 + 1] * placeholder_shared[index5]); compute[2] = compute[2] + (pad_temp_shared[index4 + 2] * placeholder_shared[index5]); compute[3] = compute[3] + (pad_temp_shared[index4 + 3] * placeholder_shared[index5]); compute[4] = compute[4] + (pad_temp_shared[index4 + 4] * placeholder_shared[index5]); compute[5] = compute[5] + (pad_temp_shared[index4 + 5] * placeholder_shared[index5]); compute[6] = compute[6] + (pad_temp_shared[index4 + 6] * placeholder_shared[index5]); compute[7] = compute[7] + (pad_temp_shared[index4 + 7] * placeholder_shared[index5]); compute[8] = compute[8] + (pad_temp_shared[index4 + 8] * placeholder_shared[index5]); compute[9] = compute[9] + (pad_temp_shared[index4 + 9] * placeholder_shared[index5]); compute[10] = compute[10] + (pad_temp_shared[index4 + 10] * placeholder_shared[index5]); compute[11] = compute[11] + (pad_temp_shared[index4 + 11] * placeholder_shared[index5]); compute[12] = compute[12] + (pad_temp_shared[index4 + 12] * placeholder_shared[index5]); } int index6 = (int)blockIdx.x / 13 * 16900 + (int)threadIdx.x * 169 + (int)blockIdx.x % 13; int index7 = (int)blockIdx.x / 13 * 100 + (int)threadIdx.x; T_relu[index6] = max(compute[0] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 13] = max(compute[1] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 26] = max(compute[2] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 39] = max(compute[3] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 52] = max(compute[4] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 65] = max(compute[5] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 78] = max(compute[6] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 91] = max(compute[7] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 104] = max(compute[8] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 117] = max(compute[9] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 130] = max(compute[10] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 143] = max(compute[11] + placeholder2[index7], 0.000000e+00f); T_relu[index6 + 156] = max(compute[12] + placeholder2[index7], 0.000000e+00f); }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpk4kvw4mk
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[16224]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<1074>; .reg .b32 %r<167>; .reg .b64 %rd<41>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[86528]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[3072]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd4, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.u32 %r29, 0; mov.u32 %r158, %r29; $L__BB0_1: mul.lo.s32 %r30, %r158, 507; mul.wide.s32 %rd9, %r30, 4; add.s64 %rd10, %rd1, %rd9; for.size.1 i in range(0, 507, 1): st.local.u32 [%rd10+(0+i*4)], %r29; add.s32 %r158, %r158, 1; setp.ne.s32 %p1, %r158, 8; @%p1 bra $L__BB0_1; mov.u32 %r32, 0; cvta.to.global.u64 %rd11, %rd4; cvta.to.global.u64 %rd28, %rd5; mov.u32 %r159, %r32; $L__BB0_3: bar.sync 0; mul.lo.s32 %r4, %r159, 2704; mov.u32 %r160, %r32; $L__BB0_4: mov.u32 %r34, %tid.x; shl.b32 %r35, %r160, 4; add.s32 %r36, %r35, %r34; mul.hi.s32 %r37, %r36, 1626496491; shr.u32 %r38, %r37, 31; shr.s32 %r39, %r37, 10; add.s32 %r40, %r39, %r38; mul.lo.s32 %r41, %r40, 2704; sub.s32 %r42, %r36, %r41; add.s32 %r43, %r42, %r4; mad.lo.s32 %r44, %r40, 43264, %r43; mul.wide.s32 %rd12, %r44, 4; add.s64 %rd13, %rd11, %rd12; ld.global.nc.f32 %f170, [%rd13]; shl.b32 %r45, %r36, 2; mov.u32 %r46, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r47, %r46, %r45; for.size.13 i in range(0, 7, 1): st.shared.f32 [%r47+(0+i*64)], %f(170+i*1); add.s32 %r(48+i*9), %r36, (16+i*16); mul.hi.s32 %r(49+i*9), %r(48+i*9), 1626496491; shr.u32 %r(50+i*9), %r(49+i*9), 31; shr.s32 %r(51+i*9), %r(49+i*9), 10; add.s32 %r(52+i*9), %r(51+i*9), %r(50+i*9); mul.lo.s32 %r(53+i*9), %r(52+i*9), 2704; sub.s32 %r(54+i*9), %r(48+i*9), %r(53+i*9); add.s32 %r(55+i*9), %r(54+i*9), %r4; mad.lo.s32 %r(56+i*9), %r(52+i*9), 43264, %r(55+i*9); mul.wide.s32 %rd(14+i*2), %r(56+i*9), 4; add.s64 %rd(15+i*2), %rd11, %rd(14+i*2); ld.global.nc.f32 %f(171+i*1), [%rd(15+i*2)+0]; st.shared.f32 [%r47+448], %f177; add.s32 %r160, %r160, 8; setp.ne.s32 %p2, %r160, 1352; @%p2 bra $L__BB0_4; mov.u32 %r143, %tid.x; shl.b32 %r112, %r159, 4; add.s32 %r114, %r112, %r143; mul.wide.s32 %rd29, %r114, 4; add.s64 %rd30, %rd28, %rd29; ld.global.nc.f32 %f178, [%rd30]; shl.b32 %r115, %r143, 2; mov.u32 %r116, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r117, %r116, %r115; mov.u32 %r161, 0; for.size.2 i in range(0, 47, 1): st.shared.f32 [%r117+(0+i*64)], %f(178+i*1); ld.global.nc.f32 %f(179+i*1), [%rd30+(1024+i*1024)]; st.shared.f32 [%r117+3008], %f225; bar.sync 0; $L__BB0_6: mov.u32 %r162, 0; $L__BB0_7: mov.u32 %r148, %tid.x; mov.u32 %r163, 0; $L__BB0_8: mul.lo.s32 %r157, %r161, 1352; mad.lo.s32 %r156, %r162, 169, %r157; mov.u32 %r164, 0; mov.u32 %r150, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r123, %r163, 2704, %r156; shl.b32 %r124, %r123, 2; add.s32 %r126, %r150, %r124; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(1+i*1), [%r126+(0+i*624)]; for.size.1 i in range(0, 155, 1): ld.shared.f32 %f(3+i*1), [%r126+(4+i*4)]; for.size.1 i in range(0, 12, 1): ld.shared.f32 %f(158+i*1), [%r126+(628+i*4)]; $L__BB0_9: shl.b32 %r155, %r161, 3; mov.u32 %r154, %tid.x; mad.lo.s32 %r153, %r154, 48, %r155; add.s32 %r152, %r153, %r162; mov.u32 %r141, _ZZ24default_function_kernel0E18placeholder_shared; mul.lo.s32 %r140, %r163, 507; mad.lo.s32 %r127, %r164, 169, %r140; for.size.2 i in range(0, 2, 1): shl.b32 %r(128+i*2), %r(164+i*-35), (4+i*-2); add.s32 %r(129+i*3), %r(152+i*-11), %r(128+i*2); mul.wide.s32 %rd31, %r127, 4; add.s64 %rd32, %rd1, %rd31; ld.shared.f32 %f226, [%r132]; for.size.3 i in range(0, 2, 1): ld.local.f32 %f(227+i*2), [%rd32+(0+i*4)]; fma.rn.f32 %f(228+i*2), %f(1+i*2), %f226, %f(227+i*2); st.local.f32 [%rd32+(0+i*4)], %f(228+i*2); for.size.3 i in range(0, 154, 1): ld.local.f32 %f(231+i*2), [%rd32+(8+i*4)]; fma.rn.f32 %f(232+i*2), %f(4+i*1), %f226, %f(231+i*2); st.local.f32 [%rd32+(8+i*4)], %f(232+i*2); for.size.3 i in range(0, 2, 1): ld.local.f32 %f(539+i*2), [%rd32+(624+i*4)]; fma.rn.f32 %f(540+i*2), %f(2+i*156), %f226, %f(539+i*2); st.local.f32 [%rd32+(624+i*4)], %f(540+i*2); for.size.3 i in range(0, 11, 1): ld.local.f32 %f(543+i*2), [%rd32+(632+i*4)]; fma.rn.f32 %f(544+i*2), %f(159+i*1), %f226, %f(543+i*2); st.local.f32 [%rd32+(632+i*4)], %f(544+i*2); add.s32 %r164, %r164, 1; setp.ne.s32 %p3, %r164, 3; @%p3 bra $L__BB0_9; add.s32 %r163, %r163, 1; setp.lt.u32 %p4, %r163, 8; @%p4 bra $L__BB0_8; add.s32 %r162, %r162, 1; setp.lt.u32 %p5, %r162, 8; @%p5 bra $L__BB0_7; add.s32 %r161, %r161, 1; setp.lt.u32 %p6, %r161, 2; @%p6 bra $L__BB0_6; add.s32 %r159, %r159, 1; setp.lt.u32 %p7, %r159, 16; @%p7 bra $L__BB0_3; ld.param.u64 %rd40, [default_function_kernel0_param_3]; ld.param.u64 %rd39, [default_function_kernel0_param_2]; mov.u32 %r142, %tid.x; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(21+i*1), %r142, (3+i*504); cvta.to.global.u64 %rd2, %rd39; cvta.to.global.u64 %rd3, %rd40; mov.u32 %r133, 0; mov.u32 %r165, %r133; $L__BB0_15: mul.lo.s32 %r24, %r165, 507; mad.lo.s32 %r25, %r165, 8112, %r22; mov.u32 %r166, %r133; $L__BB0_16: mul.lo.s32 %r136, %r166, 169; for.size.1 i in range(0, 2, 1): add.s32 %r(137+i*1), %r(136+i*30), %r(24+i*-3); mul.wide.s32 %rd33, %r138, 4; add.s64 %rd34, %rd3, %rd33; add.s32 %r139, %r25, %r136; mul.wide.s32 %rd35, %r137, 4; add.s64 %rd36, %rd1, %rd35; ld.local.f32 %f565, [%rd36]; ld.global.nc.f32 %f566, [%rd34]; add.f32 %f567, %f565, %f566; mov.f32 %f568, 0f00000000; max.f32 %f569, %f567, %f568; mul.wide.s32 %rd37, %r139, 4; add.s64 %rd38, %rd2, %rd37; for.size.4 i in range(0, 168, 1): st.global.f32 [%rd38+(0+i*4)], %f(569+i*3); ld.local.f32 %f(570+i*3), [%rd36+(4+i*4)]; add.f32 %f(571+i*3), %f(570+i*3), %f566; max.f32 %f(572+i*3), %f(571+i*3), %f568; st.global.f32 [%rd38+672], %f1073; add.s32 %r166, %r166, 1; setp.ne.s32 %p8, %r166, 3; @%p8 bra $L__BB0_16; add.s32 %r165, %r165, 1; setp.lt.u32 %p9, %r165, 8; @%p9 bra $L__BB0_15; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4056]; __shared__ float pad_temp_shared[21632]; __shared__ float placeholder_shared[768]; for (int nn_inner_init = 0; nn_inner_init < 8; ++nn_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 3; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 13; ++xx_inner_init) { int index0 = nn_inner_init * 507 + ff_inner_init * 169 + yy_inner_init * 13 + xx_inner_init; compute[index0] = 0.000000e+00f; } } } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 1352; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) / 2704 * 43264 + rc_outer_outer * 2704 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) % 2704; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 48; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 16 + (int)threadIdx.x; int index4 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 256 + rc_outer_outer * 16 + (int)threadIdx.x; placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int nn_inner = 0; nn_inner < 8; ++nn_inner) { for (int ff_inner = 0; ff_inner < 3; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { for (int xx_inner = 0; xx_inner < 13; ++xx_inner) { int index5 = nn_inner * 507 + ff_inner * 169 + yy_inner * 13 + xx_inner; int index6 = nn_inner * 2704 + rc_outer_inner * 1352 + rc_inner * 169 + yy_inner * 13 + xx_inner; int index7 = (int)threadIdx.x * 48 + ff_inner * 16 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } } } for (int ax0_inner = 0; ax0_inner < 8; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { int index8 = ax0_inner * 8112 + (int)threadIdx.x * 507 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index9 = ax0_inner * 507 + ax1_inner * 169 + ax2_inner * 13 + ax3_inner; int index10 = (int)threadIdx.x * 3 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpj6bt8l19
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<11>; .reg .f32 %f<301>; .reg .b32 %r<136>; .reg .b64 %rd<31>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd6; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(36+i*4), %r1, (1626496491+i*-304968092); shr.u32 %r(37+i*4), %r(36+i*4), 31; shr.s32 %r(38+i*4), %r(36+i*4), (6+i*-4); add.s32 %r(39+i*-37), %r(38+i*4), %r(37+i*4); mov.u32 %r3, %ctaid.x; mul.lo.s32 %r43, %r39, 169; sub.s32 %r4, %r1, %r43; mad.lo.s32 %r5, %r3, 173056, %r4; shr.u32 %r44, %r1, 7; and.b32 %r45, %r1, 127; prmt.b32 %r6, %r44, %r45, 8452; shl.b32 %r46, %r1, 2; mov.u32 %r47, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r47, %r46; for.size.5 i in range(0, 5, 1): add.s32 %r(48+i*4), %r1, (1014+i*1014); shr.u32 %r(49+i*4), %r(48+i*4), 7; add.s32 %r(50+i*4), %r1, (118+i*-10); and.b32 %r(51+i*4), %r(50+i*4), 127; prmt.b32 %r(8+i*1), %r(49+i*4), %r(51+i*4), 8452; shl.b32 %r68, %r1, 1; add.s32 %r69, %r68, 12168; and.b32 %r70, %r69, -256; add.s32 %r71, %r1, %r70; add.s32 %r13, %r71, 68; mul.hi.s32 %r72, %r1, 542165497; shr.u32 %r73, %r72, 31; shr.s32 %r74, %r72, 6; add.s32 %r14, %r74, %r73; mad.lo.s32 %r15, %r14, 43264, %r4; mul.lo.s32 %r75, %r14, 507; sub.s32 %r76, %r1, %r75; mul.hi.s32 %r77, %r76, 1626496491; shr.u32 %r78, %r77, 31; shr.s32 %r79, %r77, 6; add.s32 %r16, %r79, %r78; shl.b32 %r17, %r16, 9; add.s32 %r18, %r39, 6; mov.u32 %r80, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r19, %r80, %r46; mov.u32 %r130, 0; mov.f32 %f269, 0f00000000; for.size.1 i in range(0, 31, 1): mov.f32 %f(270+i*1), %f269; $L__BB0_1: mov.u32 %r131, 0; bar.sync 0; mad.lo.s32 %r22, %r130, 21632, %r5; mov.u32 %r132, %r19; mov.u32 %r133, %r2; mov.u32 %r134, %r18; $L__BB0_2: mov.u32 %r23, %r131; add.s32 %r82, %r134, -6; setp.gt.s32 %p1, %r82, 511; @%p1 bra $L__BB0_6; setp.gt.s32 %p2, %r133, 6655; @%p2 bra $L__BB0_6; add.s32 %r83, %r1, %r23; setp.gt.s32 %p3, %r83, 86527; @%p3 bra $L__BB0_6; shr.s32 %r85, %r82, 7; and.b32 %r86, %r82, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(87+i*1), %r(85+i*1), (43264+i*-43095), %r(22+i*65); mul.wide.s32 %rd9, %r88, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f129, [%rd10]; st.shared.f32 [%r132], %f129; $L__BB0_6: setp.gt.s32 %p4, %r134, 511; @%p4 bra $L__BB0_10; add.s32 %r89, %r133, 78; setp.gt.s32 %p5, %r89, 6655; @%p5 bra $L__BB0_10; add.s32 %r90, %r1, %r23; add.s32 %r91, %r90, 1014; setp.gt.s32 %p6, %r91, 86527; @%p6 bra $L__BB0_10; shr.s32 %r92, %r134, 7; and.b32 %r93, %r134, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(94+i*1), %r(92+i*1), (43264+i*-43095), %r(22+i*72); mul.wide.s32 %rd11, %r95, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f130, [%rd12]; st.shared.f32 [%r132+4056], %f130; $L__BB0_10: for.size.1 i in range(0, 2, 1): add.s32 %r(134+i*-1), %r(134+i*-1), (12+i*144); for.size.1 i in range(0, 2, 1): add.s32 %r(132+i*-1), %r(132+i*-109), (8112+i*-6084); setp.ne.s32 %p7, %r23, 85176; @%p7 bra $L__BB0_2; setp.gt.s32 %p8, %r1, 59; shl.b32 %r31, %r130, 7; for.size.5 i in range(0, 2, 1): add.s32 %r(96+i*1), %r(6+i*2), %r31; mul.wide.s32 %rd(13+i*2), %r(96+i*1), 4; add.s64 %rd(14+i*2), %rd2, %rd(13+i*2); ld.global.nc.f32 %f(131+i*1), [%rd(14+i*2)+0]; st.shared.f32 [%r7+(0+i*4056)], %f(131+i*1); for.size.5 i in range(0, 4, 1): add.s32 %r(98+i*1), %r(9+i*1), %r31; mul.wide.s32 %rd(17+i*2), %r(98+i*1), 4; add.s64 %rd(18+i*2), %rd2, %rd(17+i*2); ld.global.nc.f32 %f(133+i*1), [%rd(18+i*2)+0]; st.shared.f32 [%r7+(8112+i*4056)], %f(133+i*1); @%p8 bra $L__BB0_13; add.s32 %r102, %r13, %r31; mul.wide.s32 %rd25, %r102, 4; add.s64 %rd26, %rd2, %rd25; ld.global.nc.f32 %f137, [%rd26]; st.shared.f32 [%r7+24336], %f137; $L__BB0_13: bar.sync 0; mov.u32 %r135, 0; $L__BB0_14: mov.u32 %r117, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r116, _ZZ24default_function_kernel0E18placeholder_shared; mad.lo.s32 %r104, %r135, 169, %r15; for.size.2 i in range(0, 2, 1): add.s32 %r(105+i*3), %r(135+i*-19), %r(17+i*89); shl.b32 %r(106+i*3), %r(105+i*-1), 2; add.s32 %r111, %r117, %r109; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(138+i*1), [%r(111+i*-3)+0]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(300+i*-8), %f138, %f(139+i*1), %f(300+i*-8); ld.shared.f32 %f(140+i*1), [%r108+(6144+i*6144)]; for.size.6 i in range(0, 3, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(276+i*-1+j*23), %f138, %f(142+i*4+j*1), %f(276+i*-1+j*23); ld.shared.f32 %f(143+i*4+j*1), [%r108+(512+i*512+j*6144)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(291+i*-1+j*-8), %f138, %f(144+i*4+j*1), %f(291+i*-1+j*-8); ld.shared.f32 %f(145+i*4+j*1), [%r108+(12800+i*512+j*6144)]; fma.rn.f32 %f273, %f138, %f154, %f273; ld.shared.f32 %f155, [%r111+86528]; for.size.2 i in range(0, 4, 1): for.size.1 j in range(0, 4, 1): fma.rn.f32 %f(296+i*-1+j*-8), %f155, %f(139+i*4+j*1), %f(296+i*-1+j*-8); add.s32 %r135, %r135, 1; setp.ne.s32 %p9, %r135, 128; @%p9 bra $L__BB0_14; add.s32 %r130, %r130, 1; setp.lt.u32 %p10, %r130, 2; @%p10 bra $L__BB0_1; mov.u32 %r128, %ctaid.x; mul.hi.s32 %r127, %r1, 542165497; shr.u32 %r126, %r127, 31; shr.s32 %r125, %r127, 6; add.s32 %r124, %r125, %r126; mul.lo.s32 %r123, %r124, 507; sub.s32 %r122, %r1, %r123; mul.hi.s32 %r121, %r122, 1626496491; shr.u32 %r120, %r121, 31; shr.s32 %r119, %r121, 6; add.s32 %r118, %r119, %r120; shl.b32 %r112, %r118, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(113+i*1), %r(128+i*-4), (32448+i*-16224), %r(4+i*109); mad.lo.s32 %r115, %r118, 676, %r114; mul.wide.s32 %rd27, %r112, 4; add.s64 %rd28, %rd4, %rd27; ld.global.nc.f32 %f156, [%rd28]; add.f32 %f157, %f300, %f156; mov.f32 %f158, 0f00000000; max.f32 %f159, %f157, %f158; mul.wide.s32 %rd29, %r115, 4; add.s64 %rd30, %rd3, %rd29; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd30+(0+i*8112)], %f(159+i*3); ld.global.nc.f32 %f(160+i*3), [%rd28+(48+i*48)]; add.f32 %f(161+i*3), %f(292+i*-8), %f(160+i*3); max.f32 %f(162+i*3), %f(161+i*3), %f158; for.size.10 i in range(0, 3, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd30+(24336+i*676+j*-23660)], %f(168+i*12+j*3); ld.global.nc.f32 %f(169+i*12+j*3), [%rd28+(4+i*4+j*48)]; add.f32 %f(170+i*12+j*3), %f(299+i*-1+j*-8), %f(169+i*12+j*3); max.f32 %f(171+i*12+j*3), %f(170+i*12+j*3), %f158; for.size.4 j in range(0, 2, 1): st.global.f32 [%rd30+(8788+i*676+j*8112)], %f(174+i*12+j*3); ld.global.nc.f32 %f(175+i*12+j*3), [%rd28+(100+i*4+j*48)]; add.f32 %f(176+i*12+j*3), %f(283+i*-1+j*-8), %f(175+i*12+j*3); max.f32 %f(177+i*12+j*3), %f(176+i*12+j*3), %f158; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd30+(26364+i*6084)], %f(204+i*2); add.f32 %f(205+i*2), %f(296+i*-8), %f(156+i*4); max.f32 %f(206+i*2), %f(205+i*2), %f158; for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(40560+i*676+j*8112)], %f(208+i*8+j*2); add.f32 %f(209+i*8+j*2), %f(280+i*-1+j*-8), %f(163+i*12+j*3); max.f32 %f(210+i*8+j*2), %f(209+i*8+j*2), %f158; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(56784+i*676+j*-23660)], %f(212+i*8+j*2); add.f32 %f(213+i*8+j*2), %f(295+i*-1+j*-8), %f(169+i*12+j*3); max.f32 %f(214+i*8+j*2), %f(213+i*8+j*2), %f158; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd30+(42588+i*8112)], %f(232+i*2); add.f32 %f(233+i*2), %f(277+i*-8), %f(199+i*3); max.f32 %f(234+i*2), %f(233+i*2), %f158; st.global.f32 [%rd30+58812], %f236; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[32]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[6144]; for (int nn_inner_init = 0; nn_inner_init < 2; ++nn_inner_init) { int index0 = nn_inner_init * 4; compute[index0] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 16] = 0.000000e+00f; compute[index0 + 24] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 17] = 0.000000e+00f; compute[index0 + 25] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 10] = 0.000000e+00f; compute[index0 + 18] = 0.000000e+00f; compute[index0 + 26] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 11] = 0.000000e+00f; compute[index0 + 19] = 0.000000e+00f; compute[index0 + 27] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 86; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 169 < 512) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 78 + (int)threadIdx.x / 13 < 6656) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x < 86528) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x; int index2 = (int)blockIdx.x * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 169 >> 7) * 43264 + rc_outer_outer * 21632 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 169 & 127) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1] = placeholder[index2]; } } } } int index3 = (int)threadIdx.x; int index4 = ((int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x & 127); placeholder_shared[index3] = placeholder1[index4]; int index5 = ((int)threadIdx.x + 1014 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 118 & 127); placeholder_shared[index3 + 1014] = placeholder1[index5]; int index6 = ((int)threadIdx.x + 2028 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 108 & 127); placeholder_shared[index3 + 2028] = placeholder1[index6]; int index7 = ((int)threadIdx.x + 3042 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 98 & 127); placeholder_shared[index3 + 3042] = placeholder1[index7]; int index8 = ((int)threadIdx.x + 4056 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 88 & 127); placeholder_shared[index3 + 4056] = placeholder1[index8]; int index9 = ((int)threadIdx.x + 5070 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 78 & 127); placeholder_shared[index3 + 5070] = placeholder1[index9]; if ((int)threadIdx.x < 60) { int index10 = ((int)threadIdx.x + 6084 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 68); placeholder_shared[index3 + 6084] = placeholder1[index10]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { for (int nn_inner = 0; nn_inner < 2; ++nn_inner) { int index11 = nn_inner * 4; int index12 = (int)threadIdx.x / 507 * 43264 + nn_inner * 21632 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index13 = (int)threadIdx.x % 507 / 169 * 512 + rc_outer_inner; compute[index11] = compute[index11] + pad_temp_shared[index12] * placeholder_shared[index13]; compute[index11 + 8] = compute[index11 + 8] + (pad_temp_shared[index12] * placeholder_shared[index13 + 1536]); compute[index11 + 16] = compute[index11 + 16] + (pad_temp_shared[index12] * placeholder_shared[index13 + 3072]); compute[index11 + 24] = compute[index11 + 24] + (pad_temp_shared[index12] * placeholder_shared[index13 + 4608]); compute[index11 + 1] = compute[index11 + 1] + (pad_temp_shared[index12] * placeholder_shared[index13 + 128]); compute[index11 + 9] = compute[index11 + 9] + (pad_temp_shared[index12] * placeholder_shared[index13 + 1664]); compute[index11 + 17] = compute[index11 + 17] + (pad_temp_shared[index12] * placeholder_shared[index13 + 3200]); compute[index11 + 25] = compute[index11 + 25] + (pad_temp_shared[index12] * placeholder_shared[index13 + 4736]); compute[index11 + 2] = compute[index11 + 2] + (pad_temp_shared[index12] * placeholder_shared[index13 + 256]); compute[index11 + 10] = compute[index11 + 10] + (pad_temp_shared[index12] * placeholder_shared[index13 + 1792]); compute[index11 + 18] = compute[index11 + 18] + (pad_temp_shared[index12] * placeholder_shared[index13 + 3328]); compute[index11 + 26] = compute[index11 + 26] + (pad_temp_shared[index12] * placeholder_shared[index13 + 4864]); compute[index11 + 3] = compute[index11 + 3] + (pad_temp_shared[index12] * placeholder_shared[index13 + 384]); compute[index11 + 11] = compute[index11 + 11] + (pad_temp_shared[index12] * placeholder_shared[index13 + 1920]); compute[index11 + 19] = compute[index11 + 19] + (pad_temp_shared[index12] * placeholder_shared[index13 + 3456]); compute[index11 + 27] = compute[index11 + 27] + (pad_temp_shared[index12] * placeholder_shared[index13 + 4992]); } } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index14 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 507 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 507 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index15 = ax0_inner * 4 + ax1_inner; int index16 = (int)threadIdx.x % 507 / 169 * 4 + ax1_inner; T_relu[index14] = max(compute[index15] + placeholder2[index16], 0.000000e+00f); T_relu[index14 + 2028] = max(compute[index15 + 8] + placeholder2[index16 + 12], 0.000000e+00f); T_relu[index14 + 4056] = max(compute[index15 + 16] + placeholder2[index16 + 24], 0.000000e+00f); T_relu[index14 + 6084] = max(compute[index15 + 24] + placeholder2[index16 + 36], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmphm8px6pz
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[1248]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<8>; .reg .f32 %f<367>; .reg .b32 %r<140>; .reg .b64 %rd<44>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[6656]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[3072]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; ld.param.u64 %rd13, [default_function_kernel0_param_2]; ld.param.u64 %rd14, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd11; cvta.to.global.u64 %rd2, %rd13; add.u64 %rd42, %SPL, 0; mov.f32 %f10, 0f00000000; for.size.1 i in range(0, 78, 1): st.local.v4.f32 [%rd42+(0+i*16)], {%f10, %f10, %f10, %f10}; cvta.to.global.u64 %rd4, %rd12; cvta.to.global.u64 %rd5, %rd14; mov.u32 %r1, %ctaid.x; mov.u32 %r2, %tid.x; mul.lo.s32 %r3, %r2, 48; shl.b32 %r29, %r2, 2; mov.u32 %r30, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r30, %r29; add.s32 %r5, %r2, 112; mov.u32 %r31, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r6, %r31, %r29; mov.u32 %r131, 0; $L__BB0_1: bar.sync 0; mad.lo.s32 %r8, %r131, 2704, %r1; mov.u32 %r134, 256; mov.u32 %r132, %r2; mov.u32 %r133, %r5; $L__BB0_2: mul.hi.s32 %r33, %r132, 1321528399; shr.u32 %r34, %r33, 31; shr.s32 %r35, %r33, 6; add.s32 %r36, %r35, %r34; mul.lo.s32 %r37, %r36, 208; sub.s32 %r38, %r132, %r37; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(39+i*1), %r(36+i*2), (43264+i*-43251), %r(8+i*31); mul.wide.s32 %rd16, %r40, 4; add.s64 %rd17, %rd1, %rd16; ld.global.nc.f32 %f11, [%rd17]; add.s32 %r41, %r6, %r134; for.size.13 i in range(0, 4, 1): st.shared.f32 [%r41+(-256+i*64)], %f(11+i*1); add.s32 %r(42+i*9), %r132, (16+i*16); mul.hi.s32 %r(43+i*9), %r(42+i*9), 1321528399; shr.u32 %r(44+i*9), %r(43+i*9), 31; shr.s32 %r(45+i*9), %r(43+i*9), 6; add.s32 %r(46+i*9), %r(45+i*9), %r(44+i*9); mul.lo.s32 %r(47+i*9), %r(46+i*9), 208; sub.s32 %r(48+i*9), %r(42+i*9), %r(47+i*9); mad.lo.s32 %r(49+i*9), %r(46+i*9), 43264, %r8; mad.lo.s32 %r(50+i*9), %r(48+i*9), 13, %r(49+i*9); mul.wide.s32 %rd(18+i*2), %r(50+i*9), 4; add.s64 %rd(19+i*2), %rd1, %rd(18+i*2); ld.global.nc.f32 %f(12+i*1), [%rd(19+i*2)+0]; for.size.13 i in range(0, 2, 1): st.shared.f32 [%r41+(0+i*64)], %f(15+i*1); add.s32 %r(78+i*9), %r133, (-32+i*16); mul.hi.s32 %r(79+i*9), %r(78+i*9), 1321528399; shr.u32 %r(80+i*9), %r(79+i*9), 31; shr.s32 %r(81+i*9), %r(79+i*9), 6; add.s32 %r(82+i*9), %r(81+i*9), %r(80+i*9); mul.lo.s32 %r(83+i*9), %r(82+i*9), 208; sub.s32 %r(84+i*9), %r(78+i*9), %r(83+i*9); mad.lo.s32 %r(85+i*9), %r(82+i*9), 43264, %r8; mad.lo.s32 %r(86+i*9), %r(84+i*9), 13, %r(85+i*9); mul.wide.s32 %rd(26+i*2), %r(86+i*9), 4; add.s64 %rd(27+i*2), %rd1, %rd(26+i*2); ld.global.nc.f32 %f(16+i*1), [%rd(27+i*2)+0]; st.shared.f32 [%r41+128], %f17; mul.hi.s32 %r96, %r133, 1321528399; shr.u32 %r97, %r96, 31; shr.s32 %r98, %r96, 6; add.s32 %r99, %r98, %r97; mul.lo.s32 %r100, %r99, 208; sub.s32 %r101, %r133, %r100; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(102+i*1), %r(99+i*2), (43264+i*-43251), %r(8+i*94); mul.wide.s32 %rd30, %r103, 4; add.s64 %rd31, %rd1, %rd30; ld.global.nc.f32 %f18, [%rd31]; st.shared.f32 [%r41+192], %f18; for.size.1 i in range(0, 2, 1): add.s32 %r(133+i*-1), %r(133+i*-1), 128; add.s32 %r134, %r134, 512; setp.ne.s32 %p1, %r134, 6912; @%p1 bra $L__BB0_2; shl.b32 %r105, %r131, 4; add.s32 %r106, %r105, %r2; mul.wide.s32 %rd32, %r106, 4; add.s64 %rd33, %rd4, %rd32; ld.global.nc.f32 %f19, [%rd33]; mov.u32 %r104, 0; for.size.2 i in range(0, 47, 1): st.shared.f32 [%r4+(0+i*64)], %f(19+i*1); ld.global.nc.f32 %f(20+i*1), [%rd33+(1024+i*1024)]; st.shared.f32 [%r4+3008], %f66; bar.sync 0; mov.u32 %r135, %r104; $L__BB0_4: mul.lo.s32 %r16, %r135, 13; add.s32 %r108, %r3, %r135; shl.b32 %r109, %r108, 2; add.s32 %r111, %r30, %r109; for.size.1 i in range(0, 3, 1): ld.shared.f32 %f(1+i*1), [%r111+(0+i*64)]; mov.u32 %r136, %r104; $L__BB0_5: mad.lo.s32 %r112, %r136, 208, %r16; mul.lo.s32 %r113, %r136, 39; mul.wide.s32 %rd34, %r113, 4; add.s64 %rd35, %rd42, %rd34; shl.b32 %r114, %r112, 2; add.s32 %r116, %r31, %r114; for.size.4 i in range(0, 13, 1): ld.shared.f32 %f(67+i*3), [%r116+(0+i*4)]; ld.local.f32 %f(68+i*3), [%rd35+(0+i*4)]; fma.rn.f32 %f(69+i*3), %f(67+i*3), %f1, %f(68+i*3); st.local.f32 [%rd35+(0+i*4)], %f(69+i*3); for.size.4 i in range(0, 2, 1): for.size.3 j in range(0, 13, 1): ld.local.f32 %f(106+i*26+j*2), [%rd35+(52+i*52+j*4)]; fma.rn.f32 %f(107+i*26+j*2), %f(67+j*3), %f(2+j*1), %f(106+i*26+j*2); st.local.f32 [%rd35+(52+i*52+j*4)], %f(107+i*26+j*2); add.s32 %r136, %r136, 1; setp.ne.s32 %p2, %r136, 8; @%p2 bra $L__BB0_5; add.s32 %r135, %r135, 1; setp.lt.u32 %p3, %r135, 8; @%p3 bra $L__BB0_4; mov.u32 %r117, 0; mov.u32 %r137, %r117; $L__BB0_8: mul.lo.s32 %r21, %r137, 13; add.s32 %r119, %r3, %r137; shl.b32 %r120, %r119, 2; add.s32 %r122, %r30, %r120; for.size.1 i in range(0, 3, 1): ld.shared.f32 %f(4+i*1), [%r122+(32+i*64)]; mov.u32 %r138, %r117; $L__BB0_9: mad.lo.s32 %r123, %r138, 208, %r21; mul.lo.s32 %r124, %r138, 39; mul.wide.s32 %rd36, %r124, 4; add.s64 %rd37, %rd42, %rd36; shl.b32 %r125, %r123, 2; add.s32 %r127, %r31, %r125; for.size.4 i in range(0, 13, 1): ld.shared.f32 %f(158+i*3), [%r127+(416+i*4)]; ld.local.f32 %f(159+i*3), [%rd37+(0+i*4)]; fma.rn.f32 %f(160+i*3), %f(158+i*3), %f4, %f(159+i*3); st.local.f32 [%rd37+(0+i*4)], %f(160+i*3); for.size.4 i in range(0, 2, 1): for.size.3 j in range(0, 13, 1): ld.local.f32 %f(197+i*26+j*2), [%rd37+(52+i*52+j*4)]; fma.rn.f32 %f(198+i*26+j*2), %f(158+j*3), %f(5+j*1), %f(197+i*26+j*2); st.local.f32 [%rd37+(52+i*52+j*4)], %f(198+i*26+j*2); add.s32 %r138, %r138, 1; setp.ne.s32 %p4, %r138, 8; @%p4 bra $L__BB0_9; add.s32 %r137, %r137, 1; setp.lt.u32 %p5, %r137, 8; @%p5 bra $L__BB0_8; add.s32 %r131, %r131, 1; setp.lt.u32 %p6, %r131, 16; @%p6 bra $L__BB0_1; mul.lo.s32 %r129, %r2, 3; mul.wide.s32 %rd38, %r129, 4; add.s64 %rd39, %rd5, %rd38; for.size.1 i in range(0, 3, 1): ld.global.nc.f32 %f(7+i*1), [%rd39+(0+i*4)]; mad.lo.s32 %r130, %r2, 507, %r1; mul.wide.s32 %rd40, %r130, 4; add.s64 %rd41, %rd2, %rd40; add.s64 %rd43, %rd41, 988; mov.u32 %r139, 0; $L__BB0_13: ld.local.f32 %f249, [%rd42]; add.f32 %f250, %f249, %f7; mov.f32 %f251, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(252+i*3), %f(250+i*4), %f251; st.global.f32 [%rd43+(-988+i*52)], %f(252+i*3); ld.local.f32 %f(253+i*3), [%rd42+(4+i*4)]; add.f32 %f(254+i*3), %f(253+i*3), %f7; for.size.4 i in range(0, 10, 1): max.f32 %f(258+i*3), %f(257+i*3), %f251; st.global.f32 [%rd43+(-884+i*52)], %f(258+i*3); ld.local.f32 %f(259+i*3), [%rd42+(12+i*4)]; add.f32 %f(260+i*3), %f(259+i*3), %f7; for.size.5 i in range(0, 2, 1): for.size.4 j in range(0, 13, 1): max.f32 %f(288+i*39+j*3), %f(287+i*39+j*3), %f251; st.global.f32 [%rd43+(-364+i*676+j*52)], %f(288+i*39+j*3); ld.local.f32 %f(289+i*39+j*3), [%rd42+(52+i*52+j*4)]; add.f32 %f(290+i*39+j*3), %f(289+i*39+j*3), %f(8+j*1); max.f32 %f366, %f365, %f251; st.global.f32 [%rd43+988], %f366; for.size.1 i in range(0, 2, 1): add.s64 %rd(43+i*-1), %rd(43+i*-1), (32448+i*-32292); add.s32 %r139, %r139, 1; setp.ne.s32 %p7, %r139, 8; @%p7 bra $L__BB0_13; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[312]; __shared__ float pad_temp_shared[1664]; __shared__ float placeholder_shared[768]; for (int nn_inner_init = 0; nn_inner_init < 8; ++nn_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 3; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = nn_inner_init * 39 + ff_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 104; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) / 208 * 43264 + rc_outer_outer * 2704 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) % 208 * 13 + (int)blockIdx.x; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 48; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 16 + (int)threadIdx.x; int index4 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 256 + rc_outer_outer * 16 + (int)threadIdx.x; placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int nn_inner = 0; nn_inner < 8; ++nn_inner) { for (int ff_inner = 0; ff_inner < 3; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = nn_inner * 39 + ff_inner * 13 + yy_inner; int index6 = nn_inner * 208 + rc_outer_inner * 104 + rc_inner * 13 + yy_inner; int index7 = (int)threadIdx.x * 48 + ff_inner * 16 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } } for (int ax0_inner = 0; ax0_inner < 8; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index8 = ax0_inner * 8112 + (int)threadIdx.x * 507 + ax1_inner * 169 + ax2_inner * 13 + (int)blockIdx.x; int index9 = ax0_inner * 39 + ax1_inner * 13 + ax2_inner; int index10 = (int)threadIdx.x * 3 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpxgmw0axt
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<184>; .reg .b32 %r<54>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r15, %r1, 2; mov.u32 %r16, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r16, %r15; mov.u32 %r3, %ctaid.x; shl.b32 %r17, %r1, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(18+i*1), %r(17+i*-16), (-256+i*287); bfi.b32 %r20, %r3, %r19, 12, 20; add.s32 %r4, %r20, %r18; mov.u32 %r21, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r21, %r15; mul.hi.s32 %r22, %r1, 1626496491; shr.u32 %r23, %r22, 31; for.size.2 i in range(0, 2, 1): shr.s32 %r(24+i*1), %r22, (10+i*-4); add.s32 %r(6+i*20), %r(24+i*1), %r23; mul.lo.s32 %r27, %r26, 169; sub.s32 %r28, %r1, %r27; mad.lo.s32 %r7, %r6, 21632, %r28; mul.lo.s32 %r29, %r6, 2704; sub.s32 %r8, %r1, %r29; mul.hi.s32 %r30, %r8, 1626496491; shr.u32 %r31, %r30, 31; shr.s32 %r32, %r30, 6; add.s32 %r9, %r32, %r31; shl.b32 %r10, %r9, 5; mov.u32 %r52, 0; mov.f32 %f180, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(181+i*1), %f180; $L__BB0_1: bar.sync 0; mad.lo.s32 %r33, %r52, 5408, %r1; mul.wide.s32 %rd9, %r33, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 8, 1): ld.global.nc.f32 %f(17+i*1), [%rd10+(0+i*173056)]; st.shared.f32 [%r2+(0+i*21632)], %f(17+i*1); setp.gt.s32 %p1, %r1, 511; @%p1 bra $L__BB0_3; shl.b32 %r34, %r52, 5; add.s32 %r35, %r4, %r34; mul.wide.s32 %rd11, %r35, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f25, [%rd12]; st.shared.f32 [%r5], %f25; $L__BB0_3: bar.sync 0; mov.u32 %r53, 0; $L__BB0_4: mad.lo.s32 %r38, %r53, 2704, %r7; for.size.2 i in range(0, 2, 1): shl.b32 %r(39+i*3), %r(38+i*15), (2+i*2); add.s32 %r(41+i*2), %r(16+i*26), %r(39+i*-29); shl.b32 %r44, %r43, 2; add.s32 %r46, %r21, %r44; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(26+i*1), [%r(46+i*-5)+0]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(28+i*2), %f(27+i*2), %f26, %f(183+i*-1); ld.shared.f32 %f(29+i*2), [%r41+(21632+i*21632)]; for.size.9 i in range(0, 2, 1): fma.rn.f32 %f(34+i*9), %f(33+i*9), %f(26+i*9), %f(180+i*-146); ld.shared.f32 %f(35+i*9), [%r46+(4+i*4)]; ld.shared.f32 %f(36+i*9), [%r41+(676+i*676)]; fma.rn.f32 %f(37+i*9), %f(36+i*9), %f(35+i*9), %f(28+i*9); ld.shared.f32 %f(38+i*9), [%r41+(22308+i*676)]; fma.rn.f32 %f(39+i*9), %f(38+i*9), %f(35+i*9), %f(30+i*9); ld.shared.f32 %f(40+i*9), [%r41+(43940+i*676)]; fma.rn.f32 %f(41+i*9), %f(40+i*9), %f(35+i*9), %f(32+i*9); ld.shared.f32 %f(42+i*9), [%r41+(65572+i*676)]; for.size.9 i in range(0, 12, 1): fma.rn.f32 %f(52+i*9), %f(51+i*9), %f(44+i*9), %f(43+i*9); ld.shared.f32 %f(53+i*9), [%r46+(12+i*4)]; ld.shared.f32 %f(54+i*9), [%r41+(2028+i*676)]; fma.rn.f32 %f(55+i*9), %f(54+i*9), %f(53+i*9), %f(46+i*9); ld.shared.f32 %f(56+i*9), [%r41+(23660+i*676)]; fma.rn.f32 %f(57+i*9), %f(56+i*9), %f(53+i*9), %f(48+i*9); ld.shared.f32 %f(58+i*9), [%r41+(45292+i*676)]; fma.rn.f32 %f(59+i*9), %f(58+i*9), %f(53+i*9), %f(50+i*9); ld.shared.f32 %f(60+i*9), [%r41+(66924+i*676)]; fma.rn.f32 %f160, %f159, %f152, %f151; ld.shared.f32 %f161, [%r46+60]; for.size.2 i in range(0, 4, 1): ld.shared.f32 %f(162+i*1), [%r41+(10140+i*21632)]; fma.rn.f32 %f(183+i*-1), %f(162+i*1), %f161, %f(154+i*2); add.s32 %r47, %r53, 1; mov.u32 %r53, 1; setp.lt.u32 %p2, %r47, 2; @%p2 bra $L__BB0_4; add.s32 %r52, %r52, 1; setp.lt.u32 %p3, %r52, 8; @%p3 bra $L__BB0_1; shl.b32 %r48, %r3, 4; add.s32 %r49, %r9, %r48; mul.wide.s32 %rd13, %r49, 4; add.s64 %rd14, %rd4, %rd13; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(50+i*1), %r(3+i*3), (2704+i*29744), %r(8+i*42); ld.global.nc.f32 %f166, [%rd14]; add.f32 %f167, %f183, %f166; mov.f32 %f168, 0f00000000; max.f32 %f169, %f167, %f168; mul.wide.s32 %rd15, %r51, 4; add.s64 %rd16, %rd3, %rd15; for.size.3 i in range(0, 3, 1): st.global.f32 [%rd16+(0+i*32448)], %f(169+i*2); add.f32 %f(170+i*2), %f(182+i*-1), %f166; max.f32 %f(171+i*2), %f(170+i*2), %f168; st.global.f32 [%rd16+97344], %f175; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[512]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 5408 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 5408] = placeholder[index1 + 43264]; pad_temp_shared[index0 + 10816] = placeholder[index1 + 86528]; pad_temp_shared[index0 + 16224] = placeholder[index1 + 129792]; pad_temp_shared[index0 + 21632] = placeholder[index1 + 173056]; pad_temp_shared[index0 + 27040] = placeholder[index1 + 216320]; pad_temp_shared[index0 + 32448] = placeholder[index1 + 259584]; pad_temp_shared[index0 + 37856] = placeholder[index1 + 302848]; if ((int)threadIdx.x < 512) { int index2 = (int)blockIdx.x * 4096 + ((int)threadIdx.x >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { int index3 = (int)threadIdx.x / 2704 * 21632 + rc_outer_inner * 2704 + (int)threadIdx.x % 169; int index4 = (int)threadIdx.x % 2704 / 169 * 32 + rc_outer_inner * 16; compute[0] = compute[0] + pad_temp_shared[index3] * placeholder_shared[index4]; compute[1] = compute[1] + (pad_temp_shared[index3 + 5408] * placeholder_shared[index4]); compute[2] = compute[2] + (pad_temp_shared[index3 + 10816] * placeholder_shared[index4]); compute[3] = compute[3] + (pad_temp_shared[index3 + 16224] * placeholder_shared[index4]); compute[0] = compute[0] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 1]); compute[1] = compute[1] + (pad_temp_shared[index3 + 5577] * placeholder_shared[index4 + 1]); compute[2] = compute[2] + (pad_temp_shared[index3 + 10985] * placeholder_shared[index4 + 1]); compute[3] = compute[3] + (pad_temp_shared[index3 + 16393] * placeholder_shared[index4 + 1]); compute[0] = compute[0] + (pad_temp_shared[index3 + 338] * placeholder_shared[index4 + 2]); compute[1] = compute[1] + (pad_temp_shared[index3 + 5746] * placeholder_shared[index4 + 2]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11154] * placeholder_shared[index4 + 2]); compute[3] = compute[3] + (pad_temp_shared[index3 + 16562] * placeholder_shared[index4 + 2]); compute[0] = compute[0] + (pad_temp_shared[index3 + 507] * placeholder_shared[index4 + 3]); compute[1] = compute[1] + (pad_temp_shared[index3 + 5915] * placeholder_shared[index4 + 3]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11323] * placeholder_shared[index4 + 3]); compute[3] = compute[3] + (pad_temp_shared[index3 + 16731] * placeholder_shared[index4 + 3]); compute[0] = compute[0] + (pad_temp_shared[index3 + 676] * placeholder_shared[index4 + 4]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6084] * placeholder_shared[index4 + 4]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11492] * placeholder_shared[index4 + 4]); compute[3] = compute[3] + (pad_temp_shared[index3 + 16900] * placeholder_shared[index4 + 4]); compute[0] = compute[0] + (pad_temp_shared[index3 + 845] * placeholder_shared[index4 + 5]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6253] * placeholder_shared[index4 + 5]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11661] * placeholder_shared[index4 + 5]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17069] * placeholder_shared[index4 + 5]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1014] * placeholder_shared[index4 + 6]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6422] * placeholder_shared[index4 + 6]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11830] * placeholder_shared[index4 + 6]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17238] * placeholder_shared[index4 + 6]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1183] * placeholder_shared[index4 + 7]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6591] * placeholder_shared[index4 + 7]); compute[2] = compute[2] + (pad_temp_shared[index3 + 11999] * placeholder_shared[index4 + 7]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17407] * placeholder_shared[index4 + 7]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1352] * placeholder_shared[index4 + 8]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6760] * placeholder_shared[index4 + 8]); compute[2] = compute[2] + (pad_temp_shared[index3 + 12168] * placeholder_shared[index4 + 8]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17576] * placeholder_shared[index4 + 8]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1521] * placeholder_shared[index4 + 9]); compute[1] = compute[1] + (pad_temp_shared[index3 + 6929] * placeholder_shared[index4 + 9]); compute[2] = compute[2] + (pad_temp_shared[index3 + 12337] * placeholder_shared[index4 + 9]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17745] * placeholder_shared[index4 + 9]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1690] * placeholder_shared[index4 + 10]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7098] * placeholder_shared[index4 + 10]); compute[2] = compute[2] + (pad_temp_shared[index3 + 12506] * placeholder_shared[index4 + 10]); compute[3] = compute[3] + (pad_temp_shared[index3 + 17914] * placeholder_shared[index4 + 10]); compute[0] = compute[0] + (pad_temp_shared[index3 + 1859] * placeholder_shared[index4 + 11]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7267] * placeholder_shared[index4 + 11]); compute[2] = compute[2] + (pad_temp_shared[index3 + 12675] * placeholder_shared[index4 + 11]); compute[3] = compute[3] + (pad_temp_shared[index3 + 18083] * placeholder_shared[index4 + 11]); compute[0] = compute[0] + (pad_temp_shared[index3 + 2028] * placeholder_shared[index4 + 12]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7436] * placeholder_shared[index4 + 12]); compute[2] = compute[2] + (pad_temp_shared[index3 + 12844] * placeholder_shared[index4 + 12]); compute[3] = compute[3] + (pad_temp_shared[index3 + 18252] * placeholder_shared[index4 + 12]); compute[0] = compute[0] + (pad_temp_shared[index3 + 2197] * placeholder_shared[index4 + 13]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7605] * placeholder_shared[index4 + 13]); compute[2] = compute[2] + (pad_temp_shared[index3 + 13013] * placeholder_shared[index4 + 13]); compute[3] = compute[3] + (pad_temp_shared[index3 + 18421] * placeholder_shared[index4 + 13]); compute[0] = compute[0] + (pad_temp_shared[index3 + 2366] * placeholder_shared[index4 + 14]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7774] * placeholder_shared[index4 + 14]); compute[2] = compute[2] + (pad_temp_shared[index3 + 13182] * placeholder_shared[index4 + 14]); compute[3] = compute[3] + (pad_temp_shared[index3 + 18590] * placeholder_shared[index4 + 14]); compute[0] = compute[0] + (pad_temp_shared[index3 + 2535] * placeholder_shared[index4 + 15]); compute[1] = compute[1] + (pad_temp_shared[index3 + 7943] * placeholder_shared[index4 + 15]); compute[2] = compute[2] + (pad_temp_shared[index3 + 13351] * placeholder_shared[index4 + 15]); compute[3] = compute[3] + (pad_temp_shared[index3 + 18759] * placeholder_shared[index4 + 15]); } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { int index5 = (int)threadIdx.x / 2704 * 32448 + ax0_inner * 8112 + (int)blockIdx.x * 2704 + (int)threadIdx.x % 2704; int index6 = ax0_inner; int index7 = (int)blockIdx.x * 16 + (int)threadIdx.x % 2704 / 169; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpiixvrh1b
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<70>; .reg .b32 %r<184>; .reg .b64 %rd<53>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1536]; ld.param.u64 %rd25, [default_function_kernel0_param_0]; ld.param.u64 %rd26, [default_function_kernel0_param_1]; ld.param.u64 %rd27, [default_function_kernel0_param_2]; ld.param.u64 %rd24, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd28, %rd26; cvta.to.global.u64 %rd51, %rd25; cvta.to.global.u64 %rd2, %rd27; mov.u32 %r1, %tid.x; shl.b32 %r13, %r1, 4; mul.hi.s32 %r14, %r13, 1626496491; shr.u32 %r15, %r14, 31; shr.s32 %r16, %r14, 9; add.s32 %r17, %r16, %r15; mul.lo.s32 %r18, %r17, 1352; sub.s32 %r19, %r13, %r18; shl.b32 %r20, %r1, 6; mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r21, %r20; for.size.7 i in range(0, 15, 1): or.b32 %r(22+i*7), %r13, (1+i*1); mul.hi.s32 %r(23+i*7), %r(22+i*7), 1626496491; shr.u32 %r(24+i*7), %r(23+i*7), 31; shr.s32 %r(25+i*7), %r(23+i*7), 9; add.s32 %r(26+i*7), %r(25+i*7), %r(24+i*7); mul.lo.s32 %r(27+i*7), %r(26+i*7), 1352; sub.s32 %r(28+i*7), %r(22+i*7), %r(27+i*7); shr.u32 %r127, %r1, 3; and.b32 %r128, %r1, 7; shl.b32 %r129, %r1, 2; mov.u32 %r130, _ZZ24default_function_kernel0E18placeholder_shared; for.size.4 i in range(0, 2, 1): add.s32 %r(3+i*1), %r(130+i*3), %r(129+i*3); mul.hi.s32 %r(131+i*3), %r1, (542165497+i*1084330994); shr.u32 %r(132+i*3), %r(131+i*3), 31; shr.s32 %r(133+i*3), %r(131+i*3), (10+i*-4); add.s32 %r137, %r136, %r135; mul.lo.s32 %r138, %r137, 169; sub.s32 %r139, %r1, %r138; mad.lo.s32 %r140, %r4, 2704, %r139; shl.b32 %r141, %r140, 2; add.s32 %r5, %r21, %r141; mul.lo.s32 %r142, %r4, 8112; sub.s32 %r6, %r1, %r142; mul.hi.s32 %r143, %r6, 1626496491; shr.u32 %r144, %r143, 31; shr.s32 %r145, %r143, 6; add.s32 %r7, %r145, %r144; shl.b32 %r146, %r7, 5; add.s32 %r8, %r130, %r146; prmt.b32 %r147, %r127, %r128, 8452; mul.wide.s32 %rd29, %r147, 4; add.s64 %rd52, %rd28, %rd29; mov.u32 %r9, %ctaid.x; mul.lo.s32 %r148, %r9, 173056; for.size.3 i in range(0, 15, 1): add.s32 %r(149+i*2), %r(126+i*-7), %r148; mad.lo.s32 %r(150+i*2), %r(124+i*-7), 43264, %r(149+i*2); mul.wide.s32 %rd(4+i*1), %r(150+i*2), 4; add.s32 %r179, %r19, %r148; mad.lo.s32 %r180, %r17, 43264, %r179; mul.wide.s32 %rd19, %r180, 4; mov.u32 %r183, 0; mov.f32 %f68, 0f00000000; mov.f32 %f69, %f68; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 337; @%p1 bra $L__BB0_3; for.size.3 i in range(0, 16, 1): add.s64 %rd(30+i*1), %rd51, %rd(19+i*-1); ld.global.nc.f32 %f(7+i*1), [%rd(30+i*1)+0]; st.shared.f32 [%r2+(0+i*4)], %f(7+i*1); $L__BB0_3: setp.gt.s32 %p2, %r1, 383; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f23, [%rd52]; st.shared.f32 [%r3], %f23; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(24+i*1), [%r(8+i*-3)+0]; for.size.5 i in range(0, 2, 1): fma.rn.f32 %f(26+i*5), %f(25+i*5), %f(24+i*5), %f(69+i*-43); ld.shared.f32 %f(27+i*5), [%r5+(5408+i*676)]; fma.rn.f32 %f(28+i*5), %f(27+i*5), %f(24+i*5), %f(68+i*-40); ld.shared.f32 %f(29+i*5), [%r8+(4+i*4)]; ld.shared.f32 %f(30+i*5), [%r5+(676+i*676)]; for.size.5 i in range(0, 5, 1): fma.rn.f32 %f(36+i*5), %f(35+i*5), %f(34+i*5), %f(31+i*5); ld.shared.f32 %f(37+i*5), [%r5+(6760+i*676)]; fma.rn.f32 %f(38+i*5), %f(37+i*5), %f(34+i*5), %f(33+i*5); ld.shared.f32 %f(39+i*5), [%r8+(12+i*4)]; ld.shared.f32 %f(40+i*5), [%r5+(2028+i*676)]; fma.rn.f32 %f69, %f60, %f59, %f56; ld.shared.f32 %f61, [%r5+10140]; fma.rn.f32 %f68, %f61, %f59, %f58; for.size.1 i in range(0, 2, 1): add.s64 %rd(52+i*-1), %rd(52+i*-1), (32+i*5376); add.s32 %r183, %r183, 1; setp.ne.s32 %p3, %r183, 32; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd46, %rd24; mul.wide.s32 %rd47, %r7, 4; add.s64 %rd48, %rd46, %rd47; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(181+i*1), %r(9+i*-5), (32448+i*-16224), %r(6+i*175); ld.global.nc.f32 %f62, [%rd48]; add.f32 %f63, %f69, %f62; mov.f32 %f64, 0f00000000; max.f32 %f65, %f63, %f64; mul.wide.s32 %rd49, %r182, 4; add.s64 %rd50, %rd2, %rd49; st.global.f32 [%rd50], %f65; add.f32 %f66, %f68, %f62; max.f32 %f67, %f66, %f64; st.global.f32 [%rd50+32448], %f67; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[384]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index1 = (int)blockIdx.x * 173056 + (int)threadIdx.x * 16 / 1352 * 43264 + rc_outer_outer * 1352 + (int)threadIdx.x * 16 % 1352; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index2 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 1) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 1) % 1352; pad_temp_shared[index0 + 1] = placeholder[index2]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index3 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 2) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 2) % 1352; pad_temp_shared[index0 + 2] = placeholder[index3]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index4 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 3) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 3) % 1352; pad_temp_shared[index0 + 3] = placeholder[index4]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index5 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 4) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 4) % 1352; pad_temp_shared[index0 + 4] = placeholder[index5]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index6 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 5) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 5) % 1352; pad_temp_shared[index0 + 5] = placeholder[index6]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index7 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 6) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 6) % 1352; pad_temp_shared[index0 + 6] = placeholder[index7]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index8 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 7) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 7) % 1352; pad_temp_shared[index0 + 7] = placeholder[index8]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index9 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 8) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 8) % 1352; pad_temp_shared[index0 + 8] = placeholder[index9]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index10 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 9) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 9) % 1352; pad_temp_shared[index0 + 9] = placeholder[index10]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index11 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 10) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 10) % 1352; pad_temp_shared[index0 + 10] = placeholder[index11]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index12 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 11) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 11) % 1352; pad_temp_shared[index0 + 11] = placeholder[index12]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index13 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 12) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 12) % 1352; pad_temp_shared[index0 + 12] = placeholder[index13]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index14 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 13) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 13) % 1352; pad_temp_shared[index0 + 13] = placeholder[index14]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index15 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 14) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 14) % 1352; pad_temp_shared[index0 + 14] = placeholder[index15]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index16 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 15) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 15) % 1352; pad_temp_shared[index0 + 15] = placeholder[index16]; } if ((int)threadIdx.x < 384) { int index17 = (int)threadIdx.x; int index18 = ((int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index17] = placeholder1[index18]; } __syncthreads(); int index19 = (int)threadIdx.x / 8112 * 2704 + (int)threadIdx.x % 169; int index20 = (int)threadIdx.x % 8112 / 169 * 8; compute[0] = compute[0] + pad_temp_shared[index19] * placeholder_shared[index20]; compute[1] = compute[1] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20]); compute[0] = compute[0] + (pad_temp_shared[index19 + 169] * placeholder_shared[index20 + 1]); compute[1] = compute[1] + (pad_temp_shared[index19 + 1521] * placeholder_shared[index20 + 1]); compute[0] = compute[0] + (pad_temp_shared[index19 + 338] * placeholder_shared[index20 + 2]); compute[1] = compute[1] + (pad_temp_shared[index19 + 1690] * placeholder_shared[index20 + 2]); compute[0] = compute[0] + (pad_temp_shared[index19 + 507] * placeholder_shared[index20 + 3]); compute[1] = compute[1] + (pad_temp_shared[index19 + 1859] * placeholder_shared[index20 + 3]); compute[0] = compute[0] + (pad_temp_shared[index19 + 676] * placeholder_shared[index20 + 4]); compute[1] = compute[1] + (pad_temp_shared[index19 + 2028] * placeholder_shared[index20 + 4]); compute[0] = compute[0] + (pad_temp_shared[index19 + 845] * placeholder_shared[index20 + 5]); compute[1] = compute[1] + (pad_temp_shared[index19 + 2197] * placeholder_shared[index20 + 5]); compute[0] = compute[0] + (pad_temp_shared[index19 + 1014] * placeholder_shared[index20 + 6]); compute[1] = compute[1] + (pad_temp_shared[index19 + 2366] * placeholder_shared[index20 + 6]); compute[0] = compute[0] + (pad_temp_shared[index19 + 1183] * placeholder_shared[index20 + 7]); compute[1] = compute[1] + (pad_temp_shared[index19 + 2535] * placeholder_shared[index20 + 7]); } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { int index21 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 8112 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 8112; int index22 = ax0_inner; int index23 = (int)threadIdx.x % 8112 / 169; T_relu[index21] = max(compute[index22] + placeholder2[index23], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp4ulgvf68
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<168>; .reg .b32 %r<101>; .reg .b64 %rd<42>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1664]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256]; ld.param.u64 %rd20, [default_function_kernel0_param_0]; ld.param.u64 %rd21, [default_function_kernel0_param_1]; ld.param.u64 %rd22, [default_function_kernel0_param_2]; ld.param.u64 %rd23, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd39, %rd20; cvta.to.global.u64 %rd2, %rd22; cvta.to.global.u64 %rd3, %rd23; mov.u32 %r15, %ctaid.x; mul.hi.s32 %r16, %r15, -770891565; add.s32 %r17, %r16, %r15; shr.u32 %r18, %r17, 31; shr.s32 %r19, %r17, 6; add.s32 %r1, %r19, %r18; mov.u32 %r2, %tid.x; for.size.7 i in range(0, 2, 1): mul.hi.s32 %r(20+i*6), %r(2+i*13), 1321528399; shr.u32 %r(21+i*6), %r(20+i*6), 31; shr.s32 %r(22+i*6), %r(20+i*6), 2; add.s32 %r(23+i*6), %r(22+i*6), %r(21+i*6); mul.lo.s32 %r(24+i*6), %r(23+i*6), 13; sub.s32 %r(25+i*-21), %r(2+i*13), %r(24+i*6); mul.lo.s32 %r(3+i*28), %r(25+i*-24), (13+i*65); sub.s32 %r32, %r15, %r31; mul.hi.s32 %r33, %r32, 1321528399; shr.u32 %r34, %r33, 31; shr.s32 %r35, %r33, 2; add.s32 %r5, %r35, %r34; shl.b32 %r36, %r5, 11; shr.s32 %r37, %r20, 3; add.s32 %r6, %r37, %r21; mad.lo.s32 %r38, %r6, 104, %r25; mul.lo.s32 %r39, %r6, 26; sub.s32 %r40, %r2, %r39; mul.hi.s32 %r41, %r40, 1321528399; shr.u32 %r42, %r41, 31; shr.s32 %r43, %r41, 2; add.s32 %r7, %r43, %r42; shr.s32 %r44, %r23, 3; shl.b32 %r45, %r2, 2; mov.u32 %r46, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r8, %r46, %r45; for.size.3 i in range(0, 2, 1): add.s32 %r(47+i*3), %r23, (4+i*4); shr.s32 %r(48+i*3), %r(47+i*3), 3; add.s32 %r(49+i*3), %r(2+i*21), (52+i*-40); for.size.2 i in range(0, 3, 1): shr.s32 %r(53+i*2), %r(52+i*2), 3; add.s32 %r(54+i*2), %r23, (16+i*4); shr.s32 %r59, %r58, 3; and.b32 %r60, %r58, 7; add.s32 %r61, %r23, 28; shr.s32 %r62, %r61, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(63+i*1), %r(61+i*-59), 7; mov.u32 %r65, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r9, %r65, %r45; and.b32 %r66, %r49, 7; for.size.2 i in range(0, 2, 1): shl.b32 %r(67+i*1), %r(38+i*-31), (2+i*4); add.s32 %r(10+i*1), %r(46+i*19), %r(67+i*1); shl.b32 %r69, %r49, 5; and.b32 %r70, %r69, -256; add.s32 %r71, %r36, %r70; or.b32 %r72, %r71, %r66; cvta.to.global.u64 %rd24, %rd21; mul.wide.s32 %rd25, %r72, 4; add.s64 %rd41, %rd24, %rd25; shl.b32 %r73, %r2, 5; and.b32 %r74, %r73, -256; add.s32 %r75, %r36, %r74; or.b32 %r76, %r75, %r64; mul.wide.s32 %rd26, %r76, 4; add.s64 %rd40, %rd24, %rd26; mad.lo.s32 %r77, %r63, 169, %r4; add.s32 %r78, %r77, %r3; mul.lo.s32 %r79, %r1, 173056; add.s32 %r80, %r78, %r79; mad.lo.s32 %r81, %r62, 43264, %r80; mul.wide.s32 %rd6, %r81, 4; mad.lo.s32 %r82, %r60, 169, %r4; for.size.1 i in range(0, 2, 1): add.s32 %r(83+i*1), %r(82+i*1), %r(3+i*76); for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): mad.lo.s32 %r(85+i*2+j*1), %r(59+i*-4+j*-2), 43264, %r(84+j*-4); mul.wide.s32 %rd(7+i*2+j*1), %r(85+i*2+j*1), 4; for.size.2 i in range(0, 2, 1): mad.lo.s32 %r(89+i*1), %r(51+i*-3), 43264, %r(84+i*-4); mul.wide.s32 %rd(11+i*1), %r(89+i*1), 4; mad.lo.s32 %r91, %r44, 43264, %r84; mul.wide.s32 %rd13, %r91, 4; mov.u32 %r100, 0; mov.f32 %f160, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(161+i*1), %f160; $L__BB0_1: bar.sync 0; for.size.3 i in range(0, 8, 1): add.s64 %rd(27+i*1), %rd39, %rd(13+i*-1); ld.global.nc.f32 %f(25+i*1), [%rd(27+i*1)+0]; st.shared.f32 [%r8+(0+i*208)], %f(25+i*1); setp.gt.s32 %p1, %r2, 63; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f33, [%rd40]; st.shared.f32 [%r9], %f33; $L__BB0_3: setp.gt.s32 %p2, %r2, 11; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f34, [%rd41]; st.shared.f32 [%r9+208], %f34; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(35+i*1), [%r(11+i*-1)+0]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(37+i*14), %f(36+i*14), %f(35+i*14), %f(167+i*-130); ld.shared.f32 %f(38+i*14), [%r11+(128+i*4)]; fma.rn.f32 %f(39+i*14), %f(36+i*14), %f(38+i*14), %f(165+i*-126); ld.shared.f32 %f(40+i*14), [%r10+(832+i*52)]; fma.rn.f32 %f(41+i*14), %f(40+i*14), %f(35+i*14), %f(163+i*-122); fma.rn.f32 %f(42+i*14), %f(40+i*14), %f(38+i*14), %f(161+i*-119); ld.shared.f32 %f(43+i*14), [%r11+(32+i*4)]; fma.rn.f32 %f(44+i*14), %f(36+i*14), %f(43+i*14), %f(166+i*-122); ld.shared.f32 %f(45+i*14), [%r11+(160+i*4)]; fma.rn.f32 %f(46+i*14), %f(36+i*14), %f(45+i*14), %f(164+i*-118); fma.rn.f32 %f(47+i*14), %f(40+i*14), %f(43+i*14), %f(162+i*-115); fma.rn.f32 %f(48+i*14), %f(40+i*14), %f(45+i*14), %f(160+i*-112); ld.shared.f32 %f(49+i*14), [%r11+(4+i*4)]; ld.shared.f32 %f(50+i*14), [%r10+(52+i*52)]; for.size.14 i in range(0, 5, 1): fma.rn.f32 %f(65+i*14), %f(64+i*14), %f(63+i*14), %f(51+i*14); ld.shared.f32 %f(66+i*14), [%r11+(136+i*4)]; fma.rn.f32 %f(67+i*14), %f(64+i*14), %f(66+i*14), %f(53+i*14); ld.shared.f32 %f(68+i*14), [%r10+(936+i*52)]; fma.rn.f32 %f(69+i*14), %f(68+i*14), %f(63+i*14), %f(55+i*14); fma.rn.f32 %f(70+i*14), %f(68+i*14), %f(66+i*14), %f(56+i*14); ld.shared.f32 %f(71+i*14), [%r11+(40+i*4)]; fma.rn.f32 %f(72+i*14), %f(64+i*14), %f(71+i*14), %f(58+i*14); ld.shared.f32 %f(73+i*14), [%r11+(168+i*4)]; fma.rn.f32 %f(74+i*14), %f(64+i*14), %f(73+i*14), %f(60+i*14); fma.rn.f32 %f(75+i*14), %f(68+i*14), %f(71+i*14), %f(61+i*14); fma.rn.f32 %f(76+i*14), %f(68+i*14), %f(73+i*14), %f(62+i*14); ld.shared.f32 %f(77+i*14), [%r11+(12+i*4)]; ld.shared.f32 %f(78+i*14), [%r10+(156+i*52)]; fma.rn.f32 %f167, %f134, %f133, %f121; ld.shared.f32 %f135, [%r11+156]; fma.rn.f32 %f165, %f134, %f135, %f123; ld.shared.f32 %f136, [%r10+1196]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(163+i*-2), %f136, %f(133+i*2), %f(125+i*1); for.size.2 i in range(0, 2, 1): ld.shared.f32 %f(137+i*1), [%r11+(60+i*128)]; fma.rn.f32 %f(166+i*-2), %f134, %f(137+i*1), %f(128+i*2); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(162+i*-2), %f136, %f(137+i*1), %f(131+i*1); for.size.1 i in range(0, 2, 1): add.s64 %rd(41+i*-1), %rd(41+i*-1), 32; add.s64 %rd39, %rd39, 5408; add.s32 %r100, %r100, 1; setp.ne.s32 %p3, %r100, 32; @%p3 bra $L__BB0_1; for.size.1 i in range(0, 2, 1): shl.b32 %r(92+i*1), %r(5+i*2), (3+i*-2); for.size.3 i in range(0, 2, 1): add.s32 %r(94+i*3), %r(92+i*4), %r(93+i*-90); mad.lo.s32 %r(95+i*3), %r(6+i*-1), (8112+i*-6760), %r(4+i*93); mad.lo.s32 %r(96+i*3), %r(1+i*6), (32448+i*-32110), %r(95+i*3); mul.wide.s32 %rd35, %r94, 4; add.s64 %rd36, %rd3, %rd35; ld.global.nc.f32 %f139, [%rd36]; add.f32 %f140, %f167, %f139; mov.f32 %f141, 0f00000000; max.f32 %f142, %f140, %f141; mul.wide.s32 %rd37, %r99, 4; add.s64 %rd38, %rd2, %rd37; st.global.f32 [%rd38], %f142; ld.global.nc.f32 %f143, [%rd36+16]; for.size.3 i in range(0, 2, 1): add.f32 %f(144+i*2), %f(165+i*-2), %f(143+i*-4); max.f32 %f(145+i*2), %f(144+i*2), %f141; st.global.f32 [%rd38+(2704+i*62192)], %f(145+i*2); for.size.4 i in range(0, 2, 1): add.f32 %f(148+i*3), %f(161+i*5), %f(143+i*7); max.f32 %f(149+i*3), %f(148+i*3), %f141; st.global.f32 [%rd38+(67600+i*-66924)], %f(149+i*3); ld.global.nc.f32 %f(150+i*3), [%rd36+(4+i*16)]; for.size.3 i in range(0, 2, 1): add.f32 %f(154+i*2), %f(164+i*-2), %f(153+i*-3); max.f32 %f(155+i*2), %f(154+i*2), %f141; st.global.f32 [%rd38+(3380+i*62192)], %f(155+i*2); add.f32 %f158, %f160, %f153; max.f32 %f159, %f158, %f141; st.global.f32 [%rd38+68276], %f159; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[416]; __shared__ float placeholder_shared[64]; for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = ff_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 8; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 52 + (int)threadIdx.x; int index2 = (int)blockIdx.x / 78 * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4 + (int)threadIdx.x / 13 >> 3) * 43264 + rc_outer_outer * 1352 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4 + (int)threadIdx.x / 13 & 7) * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 2; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x < 64) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x; int index4 = (int)blockIdx.x % 78 / 13 * 2048 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 52 + (int)threadIdx.x & 7); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index5 = ff_inner; int index6 = (int)threadIdx.x / 26 * 104 + rc_outer_inner * 52 + rc_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x % 26 / 13 * 16 + ff_inner * 8 + rc_outer_inner * 4 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 2] = compute[index5 + 2] + (pad_temp_shared[index6] * placeholder_shared[index7 + 32]); compute[index5 + 4] = compute[index5 + 4] + (pad_temp_shared[index6 + 208] * placeholder_shared[index7]); compute[index5 + 6] = compute[index5 + 6] + (pad_temp_shared[index6 + 208] * placeholder_shared[index7 + 32]); } } } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index8 = (int)blockIdx.x / 78 * 32448 + (int)threadIdx.x / 26 * 8112 + (int)blockIdx.x % 78 / 13 * 1352 + (int)threadIdx.x % 26 / 13 * 338 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x % 13; int index9 = ax1_inner; int index10 = (int)blockIdx.x % 78 / 13 * 8 + (int)threadIdx.x % 26 / 13 * 2 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 676] = max(compute[index9 + 2] + placeholder2[index10 + 4], 0.000000e+00f); T_relu[index8 + 16224] = max(compute[index9 + 4] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 16900] = max(compute[index9 + 6] + placeholder2[index10 + 4], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp9s2bam8r
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<149>; .reg .b32 %r<117>; .reg .b64 %rd<31>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[86528]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[768]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r29, %r1, 2; mov.u32 %r30, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r30, %r29; add.s32 %r31, %r1, 4056; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(32+i*4), %r(31+i*-30), 1626496491; shr.u32 %r(33+i*4), %r(32+i*4), 31; shr.s32 %r(34+i*4), %r(32+i*4), (10+i*-4); add.s32 %r(35+i*4), %r(34+i*4), %r(33+i*4); mul.lo.s32 %r40, %r39, 169; sub.s32 %r3, %r1, %r40; add.s32 %r41, %r40, 1352; for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(42+i*6), %r(35+i*12), 43264, %r3; add.s32 %r(4+i*1), %r(42+i*6), %r41; add.s32 %r(43+i*6), %r1, (6760+i*2704); mul.hi.s32 %r(44+i*6), %r(43+i*6), 1626496491; shr.u32 %r(45+i*6), %r(44+i*6), 31; shr.s32 %r(46+i*6), %r(44+i*6), 10; add.s32 %r(47+i*6), %r(46+i*6), %r(45+i*6); for.size.7 i in range(0, 4, 1): mad.lo.s32 %r(54+i*6), %r(53+i*6), 43264, %r3; add.s32 %r(6+i*1), %r(54+i*6), %r41; add.s32 %r(55+i*6), %r1, (12168+i*2704); mul.hi.s32 %r(56+i*6), %r(55+i*6), 1626496491; shr.u32 %r(57+i*6), %r(56+i*6), 31; shr.s32 %r(58+i*6), %r(56+i*6), 10; add.s32 %r(59+i*6), %r(58+i*6), %r(57+i*6); mad.lo.s32 %r78, %r77, 43264, %r3; add.s32 %r10, %r78, %r41; mov.u32 %r11, %ctaid.x; mul.lo.s32 %r79, %r11, 3072; shl.b32 %r80, %r1, 4; for.size.1 i in range(0, 2, 1): and.b32 %r(81+i*1), %r(80+i*-79), (-256+i*271); or.b32 %r83, %r82, %r79; add.s32 %r12, %r83, %r81; mov.u32 %r84, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r13, %r84, %r29; shr.s32 %r85, %r36, 7; add.s32 %r14, %r85, %r37; mul.lo.s32 %r86, %r14, 338; sub.s32 %r87, %r1, %r86; mul.hi.s32 %r88, %r87, 1626496491; shr.u32 %r89, %r88, 31; shr.s32 %r90, %r88, 6; add.s32 %r15, %r90, %r89; shl.b32 %r91, %r15, 7; add.s32 %r92, %r84, %r91; add.s32 %r16, %r92, 320; mad.lo.s32 %r93, %r14, 2704, %r3; shl.b32 %r94, %r93, 2; add.s32 %r95, %r30, %r94; add.s32 %r17, %r95, 43264; mov.u32 %r113, 0; mov.f32 %f137, 0f00000000; for.size.1 i in range(0, 11, 1): mov.f32 %f(138+i*1), %f137; $L__BB0_1: bar.sync 0; mul.lo.s32 %r96, %r113, 2704; add.s32 %r97, %r96, %r1; mul.wide.s32 %rd9, %r97, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(49+i*1), [%rd10+(0+i*5408)]; st.shared.f32 [%r2+(0+i*5408)], %f(49+i*1); for.size.7 i in range(0, 7, 1): ld.global.nc.f32 %f(51+i*2), [%rd10+(173056+i*173056)]; st.shared.f32 [%r2+(10816+i*10816)], %f(51+i*2); add.s32 %r(98+i*1), %r(4+i*1), %r96; mul.wide.s32 %rd(11+i*2), %r(98+i*1), 4; add.s64 %rd(12+i*2), %rd2, %rd(11+i*2); ld.global.nc.f32 %f(52+i*2), [%rd(12+i*2)+0]; st.shared.f32 [%r2+(16224+i*10816)], %f(52+i*2); setp.gt.s32 %p1, %r1, 191; @%p1 bra $L__BB0_3; shl.b32 %r105, %r113, 4; add.s32 %r106, %r12, %r105; mul.wide.s32 %rd25, %r106, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f65, [%rd26]; st.shared.f32 [%r13], %f65; $L__BB0_3: bar.sync 0; mov.u32 %r116, 320; mov.u32 %r114, %r17; mov.u32 %r115, %r16; $L__BB0_4: for.size.20 i in range(0, 2, 1): ld.shared.f32 %f(66+i*20), [%r115+(-320+i*4)]; ld.shared.f32 %f(67+i*20), [%r114+(-43264+i*676)]; fma.rn.f32 %f(68+i*80), %f(67+i*20), %f(66+i*20), %f(148+i*-80); ld.shared.f32 %f(69+i*19), [%r115+(-64+i*4)]; fma.rn.f32 %f(70+i*77), %f(67+i*20), %f(69+i*19), %f(147+i*-77); ld.shared.f32 %f(71+i*18), [%r115+(192+i*4)]; fma.rn.f32 %f(72+i*74), %f(67+i*20), %f(71+i*18), %f(146+i*-74); ld.shared.f32 %f(73+i*17), [%r114+(0+i*676)]; fma.rn.f32 %f(74+i*71), %f(73+i*17), %f(66+i*20), %f(145+i*-71); fma.rn.f32 %f(75+i*69), %f(73+i*17), %f(69+i*19), %f(144+i*-69); fma.rn.f32 %f(76+i*67), %f(73+i*17), %f(71+i*18), %f(143+i*-67); ld.shared.f32 %f(77+i*14), [%r115+(-256+i*4)]; fma.rn.f32 %f(78+i*64), %f(67+i*20), %f(77+i*14), %f(142+i*-64); ld.shared.f32 %f(79+i*13), [%r115+(0+i*4)]; fma.rn.f32 %f(80+i*61), %f(67+i*20), %f(79+i*13), %f(141+i*-61); ld.shared.f32 %f(81+i*12), [%r115+(256+i*4)]; fma.rn.f32 %f(82+i*58), %f(67+i*20), %f(81+i*12), %f(140+i*-58); fma.rn.f32 %f(83+i*56), %f(73+i*17), %f(77+i*14), %f(139+i*-56); fma.rn.f32 %f(84+i*54), %f(73+i*17), %f(79+i*13), %f(138+i*-54); fma.rn.f32 %f(85+i*52), %f(73+i*17), %f(81+i*12), %f(137+i*-52); for.size.1 i in range(0, 2, 1): add.s32 %r(115+i*-1), %r(115+i*-1), (8+i*1344); add.s32 %r116, %r116, 8; setp.ne.s32 %p2, %r116, 384; @%p2 bra $L__BB0_4; add.s32 %r113, %r113, 1; setp.lt.u32 %p3, %r113, 16; @%p3 bra $L__BB0_1; shl.b32 %r108, %r15, 1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(109+i*1), %r11, (12+i*2016), %r(108+i*-105); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(111+i*1), %r(14+i*1), (8112+i*-7774), %r(110+i*1); mul.wide.s32 %rd27, %r109, 4; add.s64 %rd28, %rd4, %rd27; ld.global.nc.f32 %f94, [%rd28]; add.f32 %f95, %f148, %f94; mov.f32 %f96, 0f00000000; max.f32 %f97, %f95, %f96; mul.wide.s32 %rd29, %r112, 4; add.s64 %rd30, %rd3, %rd29; for.size.4 i in range(0, 2, 1): st.global.f32 [%rd30+(0+i*2704)], %f(97+i*3); ld.global.nc.f32 %f(98+i*3), [%rd28+(16+i*16)]; add.f32 %f(99+i*3), %f(147+i*-1), %f(98+i*3); max.f32 %f(100+i*3), %f(99+i*3), %f96; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd30+(5408+i*124384)], %f(103+i*2); add.f32 %f(104+i*2), %f(145+i*-1), %f(94+i*4); max.f32 %f(105+i*2), %f(104+i*2), %f96; st.global.f32 [%rd30+132496], %f107; for.size.4 i in range(0, 2, 1): add.f32 %f(108+i*3), %f(143+i*-1), %f(101+i*9); max.f32 %f(109+i*3), %f(108+i*3), %f96; st.global.f32 [%rd30+(135200+i*-134524)], %f(109+i*3); ld.global.nc.f32 %f(110+i*3), [%rd28+(4+i*16)]; add.f32 %f114, %f141, %f113; max.f32 %f115, %f114, %f96; st.global.f32 [%rd30+3380], %f115; ld.global.nc.f32 %f116, [%rd28+36]; for.size.3 i in range(0, 2, 1): add.f32 %f(117+i*2), %f(140+i*-1), %f(116+i*-6); max.f32 %f(118+i*2), %f(117+i*2), %f96; st.global.f32 [%rd30+(6084+i*124384)], %f(118+i*2); for.size.3 i in range(0, 2, 1): add.f32 %f(121+i*2), %f(138+i*-1), %f(113+i*3); max.f32 %f(122+i*2), %f(121+i*2), %f96; st.global.f32 [%rd30+(133172+i*2704)], %f(122+i*2); ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[12]; __shared__ float pad_temp_shared[21632]; __shared__ float placeholder_shared[192]; compute[0] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[11] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 1352] = placeholder[index1 + 1352]; pad_temp_shared[index0 + 2704] = placeholder[index1 + 43264]; int index2 = ((int)threadIdx.x + 4056) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 4056] = placeholder[index2]; pad_temp_shared[index0 + 5408] = placeholder[index1 + 86528]; int index3 = ((int)threadIdx.x + 6760) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 6760] = placeholder[index3]; pad_temp_shared[index0 + 8112] = placeholder[index1 + 129792]; int index4 = ((int)threadIdx.x + 9464) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 9464] = placeholder[index4]; pad_temp_shared[index0 + 10816] = placeholder[index1 + 173056]; int index5 = ((int)threadIdx.x + 12168) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 12168] = placeholder[index5]; pad_temp_shared[index0 + 13520] = placeholder[index1 + 216320]; int index6 = ((int)threadIdx.x + 14872) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 14872] = placeholder[index6]; pad_temp_shared[index0 + 16224] = placeholder[index1 + 259584]; int index7 = ((int)threadIdx.x + 17576) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 17576] = placeholder[index7]; pad_temp_shared[index0 + 18928] = placeholder[index1 + 302848]; int index8 = ((int)threadIdx.x + 20280) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 20280] = placeholder[index8]; if ((int)threadIdx.x < 192) { int index9 = (int)blockIdx.x * 3072 + ((int)threadIdx.x >> 4) * 256 + rc_outer_outer * 16 + ((int)threadIdx.x & 15); placeholder_shared[index0] = placeholder1[index9]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { int index10 = (int)threadIdx.x / 338 * 2704 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index11 = (int)threadIdx.x % 338 / 169 * 32 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index10] * placeholder_shared[index11]; compute[2] = compute[2] + (pad_temp_shared[index10] * placeholder_shared[index11 + 64]); compute[4] = compute[4] + (pad_temp_shared[index10] * placeholder_shared[index11 + 128]); compute[6] = compute[6] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11]); compute[8] = compute[8] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11 + 64]); compute[10] = compute[10] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11 + 128]); compute[1] = compute[1] + (pad_temp_shared[index10] * placeholder_shared[index11 + 16]); compute[3] = compute[3] + (pad_temp_shared[index10] * placeholder_shared[index11 + 80]); compute[5] = compute[5] + (pad_temp_shared[index10] * placeholder_shared[index11 + 144]); compute[7] = compute[7] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11 + 16]); compute[9] = compute[9] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11 + 80]); compute[11] = compute[11] + (pad_temp_shared[index10 + 10816] * placeholder_shared[index11 + 144]); } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index12 = (int)threadIdx.x / 338 * 8112 + (int)blockIdx.x * 2028 + (int)threadIdx.x % 338 / 169 * 338 + ax1_inner * 169 + (int)threadIdx.x % 169; int index13 = ax1_inner; int index14 = (int)blockIdx.x * 12 + (int)threadIdx.x % 338 / 169 * 2 + ax1_inner; T_relu[index12] = max(compute[index13] + placeholder2[index14], 0.000000e+00f); T_relu[index12 + 676] = max(compute[index13 + 2] + placeholder2[index14 + 4], 0.000000e+00f); T_relu[index12 + 1352] = max(compute[index13 + 4] + placeholder2[index14 + 8], 0.000000e+00f); T_relu[index12 + 32448] = max(compute[index13 + 6] + placeholder2[index14], 0.000000e+00f); T_relu[index12 + 33124] = max(compute[index13 + 8] + placeholder2[index14 + 4], 0.000000e+00f); T_relu[index12 + 33800] = max(compute[index13 + 10] + placeholder2[index14 + 8], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpq3q5noff
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<40>; .reg .b32 %r<38>; .reg .b64 %rd<24>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[64]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r13, %ctaid.x; mul.hi.s32 %r14, %r13, 1626496491; shr.u32 %r15, %r14, 31; shr.s32 %r16, %r14, 6; add.s32 %r4, %r16, %r15; mul.lo.s32 %r17, %r4, 169; sub.s32 %r1, %r13, %r17; mov.u32 %r2, %tid.x; shl.b32 %r18, %r2, 6; mov.u32 %r19, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r19, %r18; and.b32 %r20, %r2, 1; shl.b32 %r21, %r2, 2; mov.u32 %r22, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r22, %r21; shr.s32 %r6, %r2, 3; shl.b32 %r23, %r6, 4; add.s32 %r7, %r19, %r23; and.b32 %r8, %r2, 7; shl.b32 %r24, %r2, 3; and.b32 %r25, %r24, 56; add.s32 %r9, %r22, %r25; for.size.1 i in range(0, 2, 1): shl.b32 %r(26+i*1), %r(4+i*-2), (11+i*-4); and.b32 %r28, %r27, -256; add.s32 %r29, %r26, %r28; or.b32 %r30, %r29, %r20; mul.wide.s32 %rd13, %r30, 4; add.s64 %rd23, %rd12, %rd13; mad.lo.s32 %r31, %r2, 346112, %r1; cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r31, 4; add.s64 %rd16, %rd14, %rd15; add.s64 %rd22, %rd16, 692224; mov.u32 %r37, 0; mov.f32 %f38, 0f00000000; mov.f32 %f39, %f38; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r2, 0; @%p1 bra $L__BB0_3; for.size.3 i in range(0, 8, 1): for.size.2 j in range(0, 2, 1): ld.global.nc.f32 %f(7+i*2+j*1), [%rd22+(-692224+i*173056+j*676)]; st.shared.f32 [%r3+(0+i*8+j*4)], %f(7+i*2+j*1); $L__BB0_3: setp.gt.s32 %p2, %r2, 15; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f23, [%rd23]; st.shared.f32 [%r5], %f23; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(24+i*1), [%r(9+i*-2)+0]; fma.rn.f32 %f26, %f25, %f24, %f39; ld.shared.f32 %f27, [%r7+8]; fma.rn.f32 %f28, %f27, %f24, %f38; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(29+i*1), [%r(9+i*-2)+4]; fma.rn.f32 %f39, %f30, %f29, %f26; ld.shared.f32 %f31, [%r7+12]; fma.rn.f32 %f38, %f31, %f29, %f28; for.size.1 i in range(0, 2, 1): add.s64 %rd(23+i*-1), %rd(23+i*-1), (8+i*1344); add.s32 %r37, %r37, 1; setp.ne.s32 %p3, %r37, 128; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd8; shl.b32 %r32, %r4, 3; or.b32 %r33, %r32, %r8; mul.wide.s32 %rd18, %r33, 4; add.s64 %rd19, %rd17, %rd18; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(34+i*1), %r(6+i*-2), (16224+i*-14872), %r(1+i*33); mad.lo.s32 %r36, %r8, 169, %r35; ld.global.nc.f32 %f32, [%rd19]; add.f32 %f33, %f39, %f32; mov.f32 %f34, 0f00000000; max.f32 %f35, %f33, %f34; mul.wide.s32 %rd20, %r36, 4; add.s64 %rd21, %rd1, %rd20; st.global.f32 [%rd21], %f35; add.f32 %f36, %f38, %f32; max.f32 %f37, %f36, %f34; st.global.f32 [%rd21+32448], %f37; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[2]; __shared__ float pad_temp_shared[16]; __shared__ float placeholder_shared[16]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 1] = placeholder[index1 + 169]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 2] = placeholder[index1 + 43264]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 3] = placeholder[index1 + 43433]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 4] = placeholder[index1 + 86528]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 5] = placeholder[index1 + 86697]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 6] = placeholder[index1 + 129792]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 7] = placeholder[index1 + 129961]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 8] = placeholder[index1 + 173056]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 9] = placeholder[index1 + 173225]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 10] = placeholder[index1 + 216320]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 11] = placeholder[index1 + 216489]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 12] = placeholder[index1 + 259584]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 13] = placeholder[index1 + 259753]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 14] = placeholder[index1 + 302848]; } if ((int)threadIdx.x < 1) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 346112 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 15] = placeholder[index1 + 303017]; } if ((int)threadIdx.x < 16) { int index2 = (int)threadIdx.x; int index3 = (int)blockIdx.x / 169 * 2048 + ((int)threadIdx.x >> 1) * 256 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); int index4 = ((int)threadIdx.x >> 3) * 4; int index5 = ((int)threadIdx.x & 7) * 2; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4 + 2] * placeholder_shared[index5]); compute[0] = compute[0] + (pad_temp_shared[index4 + 1] * placeholder_shared[index5 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4 + 3] * placeholder_shared[index5 + 1]); } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { int index6 = ((int)threadIdx.x >> 3) * 16224 + ax0_inner * 8112 + (int)blockIdx.x / 169 * 1352 + ((int)threadIdx.x & 7) * 169 + (int)blockIdx.x % 169; int index7 = ax0_inner; int index8 = (int)blockIdx.x / 169 * 8 + ((int)threadIdx.x & 7); T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp9f1exv9v
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<149>; .reg .b32 %r<51>; .reg .b64 %rd<17>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[692224]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[49152]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd3, [default_function_kernel0_param_1]; ld.param.u64 %rd4, [default_function_kernel0_param_2]; ld.param.u64 %rd5, [default_function_kernel0_param_3]; mov.u32 %r15, %ctaid.x; mov.u32 %r1, %tid.x; mad.lo.s32 %r16, %r15, 173056, %r1; cvta.to.global.u64 %rd7, %rd6; mul.wide.s32 %rd8, %r16, 4; add.s64 %rd1, %rd7, %rd8; ld.global.nc.f32 %f17, [%rd1]; shl.b32 %r17, %r1, 2; mov.u32 %r18, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r18, %r17; for.size.2 i in range(0, 41, 1): st.shared.f32 [%r2+(0+i*16224)], %f(17+i*1); ld.global.nc.f32 %f(18+i*1), [%rd1+(16224+i*16224)]; st.shared.f32 [%r2+665184], %f58; setp.gt.s32 %p1, %r1, 2703; @%p1 bra $L__BB0_2; ld.global.nc.f32 %f59, [%rd1+681408]; st.shared.f32 [%r2+681408], %f59; $L__BB0_2: shl.b32 %r19, %r1, 1; cvta.to.global.u64 %rd9, %rd3; mul.wide.s32 %rd10, %r19, 4; add.s64 %rd2, %rd9, %rd10; ld.global.nc.f32 %f60, [%rd2]; shl.b32 %r20, %r1, 3; mov.u32 %r21, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r21, %r20; st.shared.f32 [%r3], %f60; ld.global.nc.f32 %f61, [%rd2+4]; st.shared.f32 [%r3+4], %f61; setp.gt.s32 %p2, %r1, 2087; @%p2 bra $L__BB0_4; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(62+i*1), [%rd2+(32448+i*4)]; st.shared.f32 [%r3+(32448+i*4)], %f(62+i*1); $L__BB0_4: bar.sync 0; mul.hi.s32 %r23, %r1, 542165497; shr.u32 %r24, %r23, 31; shr.s32 %r25, %r23, 8; add.s32 %r6, %r25, %r24; mul.lo.s32 %r26, %r6, 2028; sub.s32 %r27, %r1, %r26; mul.hi.s32 %r28, %r27, 1626496491; shr.u32 %r29, %r28, 31; shr.s32 %r30, %r28, 6; add.s32 %r4, %r30, %r29; shl.b32 %r31, %r4, 12; add.s32 %r33, %r21, %r31; add.s32 %r49, %r33, 2048; mov.u32 %r50, 2048; mul.hi.s32 %r34, %r1, 1626496491; shr.u32 %r35, %r34, 31; shr.s32 %r36, %r34, 6; add.s32 %r37, %r36, %r35; mul.lo.s32 %r38, %r37, 169; sub.s32 %r7, %r1, %r38; mad.lo.s32 %r39, %r6, 86528, %r7; shl.b32 %r40, %r39, 2; add.s32 %r42, %r18, %r40; add.s32 %r48, %r42, 173056; mov.f32 %f141, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(142+i*1), %f141; $L__BB0_5: for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(72+i*14), [%r49+(-2048+i*4)]; ld.shared.f32 %f(73+i*14), [%r48+(-173056+i*676)]; fma.rn.f32 %f(74+i*14), %f(73+i*14), %f(72+i*14), %f(148+i*-74); ld.shared.f32 %f(75+i*14), [%r49+(-1024+i*4)]; fma.rn.f32 %f(76+i*14), %f(73+i*14), %f(75+i*14), %f(147+i*-71); ld.shared.f32 %f(77+i*14), [%r49+(0+i*4)]; fma.rn.f32 %f(78+i*14), %f(73+i*14), %f(77+i*14), %f(146+i*-68); ld.shared.f32 %f(79+i*14), [%r49+(1024+i*4)]; fma.rn.f32 %f(80+i*14), %f(73+i*14), %f(79+i*14), %f(145+i*-65); ld.shared.f32 %f(81+i*14), [%r48+(0+i*676)]; fma.rn.f32 %f(82+i*14), %f(81+i*14), %f(72+i*14), %f(144+i*-62); fma.rn.f32 %f(83+i*14), %f(81+i*14), %f(75+i*14), %f(143+i*-60); fma.rn.f32 %f(84+i*14), %f(81+i*14), %f(77+i*14), %f(142+i*-58); fma.rn.f32 %f(85+i*14), %f(81+i*14), %f(79+i*14), %f(141+i*-56); for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(100+i*14), [%r49+(-2040+i*4)]; ld.shared.f32 %f(101+i*14), [%r48+(-171704+i*676)]; fma.rn.f32 %f(102+i*46), %f(101+i*14), %f(100+i*14), %f(88+i*14); ld.shared.f32 %f(103+i*13), [%r49+(-1016+i*4)]; fma.rn.f32 %f(104+i*43), %f(101+i*14), %f(103+i*13), %f(90+i*14); ld.shared.f32 %f(105+i*12), [%r49+(8+i*4)]; fma.rn.f32 %f(106+i*40), %f(101+i*14), %f(105+i*12), %f(92+i*14); ld.shared.f32 %f(107+i*11), [%r49+(1032+i*4)]; fma.rn.f32 %f(108+i*37), %f(101+i*14), %f(107+i*11), %f(94+i*14); ld.shared.f32 %f(109+i*10), [%r48+(1352+i*676)]; fma.rn.f32 %f(110+i*34), %f(109+i*10), %f(100+i*14), %f(96+i*14); fma.rn.f32 %f(111+i*32), %f(109+i*10), %f(103+i*13), %f(97+i*14); fma.rn.f32 %f(112+i*30), %f(109+i*10), %f(105+i*12), %f(98+i*14); fma.rn.f32 %f(113+i*28), %f(109+i*10), %f(107+i*11), %f(99+i*14); for.size.1 i in range(0, 2, 1): add.s32 %r(49+i*-1), %r(49+i*-1), (16+i*2688); add.s32 %r50, %r50, 16; setp.ne.s32 %p3, %r50, 3072; @%p3 bra $L__BB0_5; shl.b32 %r43, %r4, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(45+i*1), %r(15+i*-9), (32448+i*-16224), %r(7+i*38); mad.lo.s32 %r47, %r4, 676, %r46; cvta.to.global.u64 %rd11, %rd5; mul.wide.s32 %rd12, %r43, 4; add.s64 %rd13, %rd11, %rd12; ld.global.nc.f32 %f120, [%rd13]; add.f32 %f121, %f148, %f120; mov.f32 %f122, 0f00000000; max.f32 %f123, %f121, %f122; cvta.to.global.u64 %rd14, %rd4; mul.wide.s32 %rd15, %r47, 4; add.s64 %rd16, %rd14, %rd15; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd16+(0+i*676)], %f(123+i*3); ld.global.nc.f32 %f(124+i*3), [%rd13+(4+i*4)]; add.f32 %f(125+i*3), %f(147+i*-1), %f(124+i*3); max.f32 %f(126+i*3), %f(125+i*3), %f122; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd16+(2028+i*30420)], %f(132+i*2); add.f32 %f(133+i*2), %f(144+i*-1), %f(120+i*4); max.f32 %f(134+i*2), %f(133+i*2), %f122; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd16+(33124+i*676)], %f(136+i*2); add.f32 %f(137+i*2), %f(142+i*-1), %f(127+i*3); max.f32 %f(138+i*2), %f(137+i*2), %f122; st.global.f32 [%rd16+34476], %f140; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[173056]; __shared__ float placeholder_shared[12288]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x * 173056 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 4056] = placeholder[index1 + 4056]; pad_temp_shared[index0 + 8112] = placeholder[index1 + 8112]; pad_temp_shared[index0 + 12168] = placeholder[index1 + 12168]; pad_temp_shared[index0 + 16224] = placeholder[index1 + 16224]; pad_temp_shared[index0 + 20280] = placeholder[index1 + 20280]; pad_temp_shared[index0 + 24336] = placeholder[index1 + 24336]; pad_temp_shared[index0 + 28392] = placeholder[index1 + 28392]; pad_temp_shared[index0 + 32448] = placeholder[index1 + 32448]; pad_temp_shared[index0 + 36504] = placeholder[index1 + 36504]; pad_temp_shared[index0 + 40560] = placeholder[index1 + 40560]; pad_temp_shared[index0 + 44616] = placeholder[index1 + 44616]; pad_temp_shared[index0 + 48672] = placeholder[index1 + 48672]; pad_temp_shared[index0 + 52728] = placeholder[index1 + 52728]; pad_temp_shared[index0 + 56784] = placeholder[index1 + 56784]; pad_temp_shared[index0 + 60840] = placeholder[index1 + 60840]; pad_temp_shared[index0 + 64896] = placeholder[index1 + 64896]; pad_temp_shared[index0 + 68952] = placeholder[index1 + 68952]; pad_temp_shared[index0 + 73008] = placeholder[index1 + 73008]; pad_temp_shared[index0 + 77064] = placeholder[index1 + 77064]; pad_temp_shared[index0 + 81120] = placeholder[index1 + 81120]; pad_temp_shared[index0 + 85176] = placeholder[index1 + 85176]; pad_temp_shared[index0 + 89232] = placeholder[index1 + 89232]; pad_temp_shared[index0 + 93288] = placeholder[index1 + 93288]; pad_temp_shared[index0 + 97344] = placeholder[index1 + 97344]; pad_temp_shared[index0 + 101400] = placeholder[index1 + 101400]; pad_temp_shared[index0 + 105456] = placeholder[index1 + 105456]; pad_temp_shared[index0 + 109512] = placeholder[index1 + 109512]; pad_temp_shared[index0 + 113568] = placeholder[index1 + 113568]; pad_temp_shared[index0 + 117624] = placeholder[index1 + 117624]; pad_temp_shared[index0 + 121680] = placeholder[index1 + 121680]; pad_temp_shared[index0 + 125736] = placeholder[index1 + 125736]; pad_temp_shared[index0 + 129792] = placeholder[index1 + 129792]; pad_temp_shared[index0 + 133848] = placeholder[index1 + 133848]; pad_temp_shared[index0 + 137904] = placeholder[index1 + 137904]; pad_temp_shared[index0 + 141960] = placeholder[index1 + 141960]; pad_temp_shared[index0 + 146016] = placeholder[index1 + 146016]; pad_temp_shared[index0 + 150072] = placeholder[index1 + 150072]; pad_temp_shared[index0 + 154128] = placeholder[index1 + 154128]; pad_temp_shared[index0 + 158184] = placeholder[index1 + 158184]; pad_temp_shared[index0 + 162240] = placeholder[index1 + 162240]; pad_temp_shared[index0 + 166296] = placeholder[index1 + 166296]; if ((int)threadIdx.x < 2704) { pad_temp_shared[index0 + 170352] = placeholder[index1 + 170352]; } placeholder_shared[index0 * 2] = placeholder1[index0 * 2]; int index2 = (int)threadIdx.x * 2 + 1; placeholder_shared[index2] = placeholder1[index2]; if ((int)threadIdx.x < 2088) { int index3 = (int)threadIdx.x * 2 + 8112; placeholder_shared[index3] = placeholder1[index3]; } if ((int)threadIdx.x < 2088) { int index4 = (int)threadIdx.x * 2 + 8113; placeholder_shared[index4] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index5 = (int)threadIdx.x / 2028 * 86528 + rc_outer_inner * 338 + (int)threadIdx.x % 169; int index6 = (int)threadIdx.x % 2028 / 169 * 1024 + rc_outer_inner * 2; compute[0] = compute[0] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index6 + 256]); compute[2] = compute[2] + (pad_temp_shared[index5] * placeholder_shared[index6 + 512]); compute[3] = compute[3] + (pad_temp_shared[index5] * placeholder_shared[index6 + 768]); compute[4] = compute[4] + (pad_temp_shared[index5 + 43264] * placeholder_shared[index6]); compute[5] = compute[5] + (pad_temp_shared[index5 + 43264] * placeholder_shared[index6 + 256]); compute[6] = compute[6] + (pad_temp_shared[index5 + 43264] * placeholder_shared[index6 + 512]); compute[7] = compute[7] + (pad_temp_shared[index5 + 43264] * placeholder_shared[index6 + 768]); compute[0] = compute[0] + (pad_temp_shared[index5 + 169] * placeholder_shared[index6 + 1]); compute[1] = compute[1] + (pad_temp_shared[index5 + 169] * placeholder_shared[index6 + 257]); compute[2] = compute[2] + (pad_temp_shared[index5 + 169] * placeholder_shared[index6 + 513]); compute[3] = compute[3] + (pad_temp_shared[index5 + 169] * placeholder_shared[index6 + 769]); compute[4] = compute[4] + (pad_temp_shared[index5 + 43433] * placeholder_shared[index6 + 1]); compute[5] = compute[5] + (pad_temp_shared[index5 + 43433] * placeholder_shared[index6 + 257]); compute[6] = compute[6] + (pad_temp_shared[index5 + 43433] * placeholder_shared[index6 + 513]); compute[7] = compute[7] + (pad_temp_shared[index5 + 43433] * placeholder_shared[index6 + 769]); } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index7 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 2028 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 2028 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index8 = ax0_inner * 4 + ax1_inner; int index9 = (int)threadIdx.x % 2028 / 169 * 4 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp6_n4ctcg
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<209>; .reg .b32 %r<30>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[64]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[96]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r10, %ctaid.x; mul.hi.s32 %r11, %r10, 1626496491; shr.u32 %r12, %r11, 31; shr.s32 %r13, %r11, 6; add.s32 %r4, %r13, %r12; mul.lo.s32 %r14, %r4, 169; sub.s32 %r1, %r10, %r14; mov.u32 %r2, %tid.x; shl.b32 %r15, %r2, 5; mov.u32 %r16, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r16, %r15; and.b32 %r17, %r2, 1; shl.b32 %r18, %r2, 2; mov.u32 %r19, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r19, %r18; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(6+i*14), %r2, (24+i*173032), %r(19+i*-18); cvta.to.global.u64 %rd13, %rd9; mul.wide.s32 %rd14, %r20, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd24, %rd15, 346112; shl.b32 %r21, %r2, 7; and.b32 %r22, %r21, -256; mad.lo.s32 %r23, %r4, 3072, %r22; or.b32 %r24, %r23, %r17; mul.wide.s32 %rd16, %r24, 4; add.s64 %rd17, %rd12, %rd16; add.s64 %rd23, %rd17, 6144; mov.u32 %r29, 0; mov.f32 %f185, 0f00000000; for.size.1 i in range(0, 23, 1): mov.f32 %f(186+i*1), %f185; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r2, 1; @%p1 bra $L__BB0_3; for.size.3 i in range(0, 4, 1): for.size.2 j in range(0, 2, 1): ld.global.nc.f32 %f(73+i*2+j*1), [%rd24+(-346112+i*173056+j*676)]; st.shared.f32 [%r3+(0+i*8+j*4)], %f(73+i*2+j*1); $L__BB0_3: for.size.2 i in range(0, 6, 1): ld.global.nc.f32 %f(81+i*1), [%rd23+(-6144+i*2048)]; st.shared.f32 [%r5+(0+i*16)], %f(81+i*1); bar.sync 0; for.size.35 i in range(0, 2, 1): ld.shared.f32 %f(87+i*35), [%r6+(0+i*4)]; ld.shared.f32 %f(88+i*35), [_ZZ24default_function_kernel0E15pad_temp_shared+(0+i*4)]; fma.rn.f32 %f(89+i*119), %f(88+i*35), %f(87+i*35), %f(208+i*-119); ld.shared.f32 %f(90+i*34), [_ZZ24default_function_kernel0E15pad_temp_shared+(8+i*4)]; fma.rn.f32 %f(91+i*116), %f(87+i*35), %f(90+i*34), %f(207+i*-116); ld.shared.f32 %f(92+i*33), [_ZZ24default_function_kernel0E15pad_temp_shared+(16+i*4)]; fma.rn.f32 %f(93+i*113), %f(92+i*33), %f(87+i*35), %f(206+i*-113); ld.shared.f32 %f(94+i*32), [_ZZ24default_function_kernel0E15pad_temp_shared+(24+i*4)]; fma.rn.f32 %f(95+i*110), %f(87+i*35), %f(94+i*32), %f(205+i*-110); ld.shared.f32 %f(96+i*31), [%r6+(8+i*4)]; fma.rn.f32 %f(97+i*107), %f(88+i*35), %f(96+i*31), %f(204+i*-107); fma.rn.f32 %f(98+i*105), %f(96+i*31), %f(90+i*34), %f(203+i*-105); fma.rn.f32 %f(99+i*103), %f(92+i*33), %f(96+i*31), %f(202+i*-103); fma.rn.f32 %f(100+i*101), %f(96+i*31), %f(94+i*32), %f(201+i*-101); ld.shared.f32 %f(101+i*27), [%r6+(16+i*4)]; fma.rn.f32 %f(102+i*98), %f(88+i*35), %f(101+i*27), %f(200+i*-98); fma.rn.f32 %f(103+i*96), %f(101+i*27), %f(90+i*34), %f(199+i*-96); fma.rn.f32 %f(104+i*94), %f(92+i*33), %f(101+i*27), %f(198+i*-94); fma.rn.f32 %f(105+i*92), %f(101+i*27), %f(94+i*32), %f(197+i*-92); ld.shared.f32 %f(106+i*23), [_ZZ24default_function_kernel0E15pad_temp_shared+(32+i*4)]; fma.rn.f32 %f(107+i*89), %f(106+i*23), %f(87+i*35), %f(196+i*-89); ld.shared.f32 %f(108+i*22), [_ZZ24default_function_kernel0E15pad_temp_shared+(40+i*4)]; fma.rn.f32 %f(109+i*86), %f(87+i*35), %f(108+i*22), %f(195+i*-86); ld.shared.f32 %f(110+i*21), [_ZZ24default_function_kernel0E15pad_temp_shared+(48+i*4)]; fma.rn.f32 %f(111+i*83), %f(110+i*21), %f(87+i*35), %f(194+i*-83); ld.shared.f32 %f(112+i*20), [_ZZ24default_function_kernel0E15pad_temp_shared+(56+i*4)]; fma.rn.f32 %f(113+i*80), %f(87+i*35), %f(112+i*20), %f(193+i*-80); fma.rn.f32 %f(114+i*78), %f(106+i*23), %f(96+i*31), %f(192+i*-78); fma.rn.f32 %f(115+i*76), %f(96+i*31), %f(108+i*22), %f(191+i*-76); fma.rn.f32 %f(116+i*74), %f(110+i*21), %f(96+i*31), %f(190+i*-74); fma.rn.f32 %f(117+i*72), %f(96+i*31), %f(112+i*20), %f(189+i*-72); fma.rn.f32 %f(118+i*70), %f(106+i*23), %f(101+i*27), %f(188+i*-70); fma.rn.f32 %f(119+i*68), %f(101+i*27), %f(108+i*22), %f(187+i*-68); fma.rn.f32 %f(120+i*66), %f(110+i*21), %f(101+i*27), %f(186+i*-66); fma.rn.f32 %f(121+i*64), %f(101+i*27), %f(112+i*20), %f(185+i*-64); for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (1352+i*-1344); add.s32 %r29, %r29, 1; setp.ne.s32 %p2, %r29, 128; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd18, %rd8; mul.lo.s32 %r25, %r2, 3; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(26+i*1), %r(4+i*-2), (12+i*495), %r(25+i*-24); mad.lo.s32 %r28, %r4, 2028, %r27; mul.wide.s32 %rd19, %r26, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f133, [%rd20]; add.f32 %f134, %f208, %f133; mov.f32 %f135, 0f00000000; max.f32 %f136, %f134, %f135; mul.wide.s32 %rd21, %r28, 4; add.s64 %rd22, %rd1, %rd21; for.size.4 i in range(0, 2, 1): st.global.f32 [%rd22+(0+i*676)], %f(136+i*3); ld.global.nc.f32 %f(137+i*3), [%rd20+(4+i*4)]; add.f32 %f(138+i*3), %f(204+i*-4), %f(137+i*3); max.f32 %f(139+i*3), %f(138+i*3), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(1352+i*31096)], %f(142+i*2); add.f32 %f(143+i*2), %f(207+i*-4), %f(133+i*4); max.f32 %f(144+i*2), %f(143+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(33124+i*676)], %f(146+i*2); add.f32 %f(147+i*2), %f(199+i*7), %f(140+i*-7); max.f32 %f(148+i*2), %f(147+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(64896+i*676)], %f(150+i*2); add.f32 %f(151+i*2), %f(202+i*-4), %f(137+i*3); max.f32 %f(152+i*2), %f(151+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(66248+i*31096)], %f(154+i*2); add.f32 %f(155+i*2), %f(205+i*-4), %f(133+i*4); max.f32 %f(156+i*2), %f(155+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(98020+i*676)], %f(158+i*2); add.f32 %f(159+i*2), %f(197+i*-1), %f(140+i*-7); max.f32 %f(160+i*2), %f(159+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(129792+i*676)], %f(162+i*2); add.f32 %f(163+i*2), %f(192+i*-4), %f(137+i*3); max.f32 %f(164+i*2), %f(163+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(131144+i*31096)], %f(166+i*2); add.f32 %f(167+i*2), %f(195+i*-4), %f(133+i*4); max.f32 %f(168+i*2), %f(167+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(162916+i*676)], %f(170+i*2); add.f32 %f(171+i*2), %f(187+i*7), %f(140+i*-7); max.f32 %f(172+i*2), %f(171+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(194688+i*676)], %f(174+i*2); add.f32 %f(175+i*2), %f(190+i*-4), %f(137+i*3); max.f32 %f(176+i*2), %f(175+i*2), %f135; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd22+(196040+i*31096)], %f(178+i*2); add.f32 %f(179+i*2), %f(193+i*-4), %f(133+i*4); max.f32 %f(180+i*2), %f(179+i*2), %f135; st.global.f32 [%rd22+227812], %f182; add.f32 %f183, %f185, %f140; max.f32 %f184, %f183, %f135; st.global.f32 [%rd22+228488], %f184; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[24]; __shared__ float pad_temp_shared[16]; __shared__ float placeholder_shared[24]; compute[0] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[21] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[19] = 0.000000e+00f; compute[22] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[20] = 0.000000e+00f; compute[23] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 1] = placeholder[index1 + 169]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 2] = placeholder[index1 + 43264]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 3] = placeholder[index1 + 43433]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 4] = placeholder[index1 + 86528]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 5] = placeholder[index1 + 86697]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 6] = placeholder[index1 + 129792]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 8; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 338 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 7] = placeholder[index1 + 129961]; } int index2 = (int)threadIdx.x; int index3 = (int)blockIdx.x / 169 * 3072 + ((int)threadIdx.x >> 1) * 256 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index2] = placeholder1[index3]; placeholder_shared[index2 + 4] = placeholder1[index3 + 512]; placeholder_shared[index2 + 8] = placeholder1[index3 + 1024]; placeholder_shared[index2 + 12] = placeholder1[index3 + 1536]; placeholder_shared[index2 + 16] = placeholder1[index3 + 2048]; placeholder_shared[index2 + 20] = placeholder1[index3 + 2560]; __syncthreads(); compute[0] = compute[0] + pad_temp_shared[0] * placeholder_shared[index2 * 6]; compute[3] = compute[3] + pad_temp_shared[2] * placeholder_shared[index2 * 6]; compute[6] = compute[6] + pad_temp_shared[4] * placeholder_shared[index2 * 6]; compute[9] = compute[9] + pad_temp_shared[6] * placeholder_shared[index2 * 6]; int index4 = (int)threadIdx.x * 6 + 2; compute[1] = compute[1] + pad_temp_shared[0] * placeholder_shared[index4]; compute[4] = compute[4] + pad_temp_shared[2] * placeholder_shared[index4]; compute[7] = compute[7] + pad_temp_shared[4] * placeholder_shared[index4]; compute[10] = compute[10] + pad_temp_shared[6] * placeholder_shared[index4]; int index5 = (int)threadIdx.x * 6 + 4; compute[2] = compute[2] + pad_temp_shared[0] * placeholder_shared[index5]; compute[5] = compute[5] + pad_temp_shared[2] * placeholder_shared[index5]; compute[8] = compute[8] + pad_temp_shared[4] * placeholder_shared[index5]; compute[11] = compute[11] + pad_temp_shared[6] * placeholder_shared[index5]; compute[12] = compute[12] + pad_temp_shared[8] * placeholder_shared[index2 * 6]; compute[15] = compute[15] + pad_temp_shared[10] * placeholder_shared[index2 * 6]; compute[18] = compute[18] + pad_temp_shared[12] * placeholder_shared[index2 * 6]; compute[21] = compute[21] + pad_temp_shared[14] * placeholder_shared[index2 * 6]; compute[13] = compute[13] + pad_temp_shared[8] * placeholder_shared[index4]; compute[16] = compute[16] + pad_temp_shared[10] * placeholder_shared[index4]; compute[19] = compute[19] + pad_temp_shared[12] * placeholder_shared[index4]; compute[22] = compute[22] + pad_temp_shared[14] * placeholder_shared[index4]; compute[14] = compute[14] + pad_temp_shared[8] * placeholder_shared[index5]; compute[17] = compute[17] + pad_temp_shared[10] * placeholder_shared[index5]; compute[20] = compute[20] + pad_temp_shared[12] * placeholder_shared[index5]; compute[23] = compute[23] + pad_temp_shared[14] * placeholder_shared[index5]; int index6 = (int)threadIdx.x * 6 + 1; compute[0] = compute[0] + pad_temp_shared[1] * placeholder_shared[index6]; compute[3] = compute[3] + pad_temp_shared[3] * placeholder_shared[index6]; compute[6] = compute[6] + pad_temp_shared[5] * placeholder_shared[index6]; compute[9] = compute[9] + pad_temp_shared[7] * placeholder_shared[index6]; int index7 = (int)threadIdx.x * 6 + 3; compute[1] = compute[1] + pad_temp_shared[1] * placeholder_shared[index7]; compute[4] = compute[4] + pad_temp_shared[3] * placeholder_shared[index7]; compute[7] = compute[7] + pad_temp_shared[5] * placeholder_shared[index7]; compute[10] = compute[10] + pad_temp_shared[7] * placeholder_shared[index7]; int index8 = (int)threadIdx.x * 6 + 5; compute[2] = compute[2] + pad_temp_shared[1] * placeholder_shared[index8]; compute[5] = compute[5] + pad_temp_shared[3] * placeholder_shared[index8]; compute[8] = compute[8] + pad_temp_shared[5] * placeholder_shared[index8]; compute[11] = compute[11] + pad_temp_shared[7] * placeholder_shared[index8]; compute[12] = compute[12] + pad_temp_shared[9] * placeholder_shared[index6]; compute[15] = compute[15] + pad_temp_shared[11] * placeholder_shared[index6]; compute[18] = compute[18] + pad_temp_shared[13] * placeholder_shared[index6]; compute[21] = compute[21] + pad_temp_shared[15] * placeholder_shared[index6]; compute[13] = compute[13] + pad_temp_shared[9] * placeholder_shared[index7]; compute[16] = compute[16] + pad_temp_shared[11] * placeholder_shared[index7]; compute[19] = compute[19] + pad_temp_shared[13] * placeholder_shared[index7]; compute[22] = compute[22] + pad_temp_shared[15] * placeholder_shared[index7]; compute[14] = compute[14] + pad_temp_shared[9] * placeholder_shared[index8]; compute[17] = compute[17] + pad_temp_shared[11] * placeholder_shared[index8]; compute[20] = compute[20] + pad_temp_shared[13] * placeholder_shared[index8]; compute[23] = compute[23] + pad_temp_shared[15] * placeholder_shared[index8]; } for (int ax0_inner = 0; ax0_inner < 8; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { int index9 = ax0_inner * 8112 + (int)blockIdx.x / 169 * 2028 + (int)threadIdx.x * 507 + ax1_inner * 169 + (int)blockIdx.x % 169; int index10 = ax0_inner * 3 + ax1_inner; int index11 = (int)blockIdx.x / 169 * 12 + (int)threadIdx.x * 3 + ax1_inner; T_relu[index9] = max(compute[index10] + placeholder2[index11], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpnse0pdqv
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<448>; .reg .b32 %r<207>; .reg .b64 %rd<43>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1664]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[3072]; ld.param.u64 %rd1, [default_function_kernel0_param_0]; ld.param.u64 %rd2, [default_function_kernel0_param_1]; mov.u32 %r38, %ctaid.x; mul.hi.s32 %r39, %r38, 1321528399; shr.u32 %r40, %r39, 31; shr.s32 %r41, %r39, 2; add.s32 %r1, %r41, %r40; mul.lo.s32 %r42, %r1, 13; sub.s32 %r2, %r38, %r42; mov.u32 %r43, %tid.x; for.size.9 i in range(0, 12, 1): mul.hi.s32 %r(44+i*7), %r(43+i*7), 1321528399; shr.u32 %r(45+i*7), %r(44+i*7), 31; shr.s32 %r(46+i*7), %r(44+i*7), 6; add.s32 %r(47+i*7), %r(46+i*7), %r(45+i*7); mul.lo.s32 %r(3+i*2), %r(47+i*7), 43264; mul.lo.s32 %r(48+i*7), %r(47+i*7), 208; sub.s32 %r(49+i*7), %r(43+i*7), %r(48+i*7); mul.lo.s32 %r(4+i*2), %r(49+i*7), 13; add.s32 %r(50+i*7), %r43, (32+i*32); mul.hi.s32 %r128, %r127, 1321528399; shr.u32 %r129, %r128, 31; shr.s32 %r130, %r128, 6; add.s32 %r131, %r130, %r129; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(27+i*105), %r131, (43264+i*-43056); sub.s32 %r133, %r127, %r132; mul.lo.s32 %r28, %r133, 13; mov.u32 %r204, 0; mov.f32 %f409, 0f00000000; cvta.to.global.u64 %rd5, %rd1; cvta.to.global.u64 %rd32, %rd2; for.size.1 i in range(0, 38, 1): mov.f32 %f(410+i*1), %f409; $L__BB0_1: mov.u32 %r205, 0; bar.sync 0; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(135+i*1), %r(1+i*203), (86528+i*-83824), %r(2+i*133); for.size.1 i in range(0, 2, 1): add.s32 %r(137+i*1), %r(136+i*1), %r(3+i*1); mul.wide.s32 %rd6, %r138, 4; add.s64 %rd7, %rd5, %rd6; ld.global.nc.f32 %f196, [%rd7]; shl.b32 %r140, %r43, 2; mov.u32 %r141, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r142, %r141, %r140; for.size.6 i in range(0, 12, 1): st.shared.f32 [%r142+(0+i*128)], %f(196+i*1); add.s32 %r(143+i*2), %r136, %r(5+i*2); add.s32 %r(144+i*2), %r(143+i*2), %r(6+i*2); mul.wide.s32 %rd(8+i*2), %r(144+i*2), 4; add.s64 %rd(9+i*2), %rd5, %rd(8+i*2); ld.global.nc.f32 %f(197+i*1), [%rd(9+i*2)+0]; st.shared.f32 [%r142+1536], %f208; shr.u32 %r167, %r43, 4; and.b32 %r168, %r43, 15; prmt.b32 %r169, %r167, %r168, 8452; shl.b32 %r170, %r204, 4; add.s32 %r171, %r169, %r170; mul.wide.s32 %rd33, %r171, 4; add.s64 %rd34, %rd32, %rd33; ld.global.nc.f32 %f209, [%rd34]; mov.u32 %r172, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r173, %r172, %r140; for.size.2 i in range(0, 23, 1): st.shared.f32 [%r173+(0+i*128)], %f(209+i*1); ld.global.nc.f32 %f(210+i*1), [%rd34+(2048+i*2048)]; st.shared.f32 [%r173+2944], %f232; bar.sync 0; $L__BB0_2: mov.u32 %r206, 0; $L__BB0_3: mov.u32 %r200, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r199, _ZZ24default_function_kernel0E18placeholder_shared; shl.b32 %r198, %r205, 3; and.b32 %r197, %r43, 15; mad.lo.s32 %r196, %r197, 48, %r198; shr.s32 %r195, %r43, 4; mul.lo.s32 %r194, %r195, 208; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(193+i*-13), %r(205+i*1), (104+i*-91), %r(194+i*-1); for.size.2 i in range(0, 2, 1): add.s32 %r(181+i*3), %r(196+i*3), %r(206+i*-24); shl.b32 %r(182+i*3), %r(181+i*-1), 2; add.s32 %r187, %r200, %r185; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(233+i*1), [%r(187+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(447+i*-1), %f(233+i*2), %f234, %f(447+i*-1); ld.shared.f32 %f(235+i*1), [%r187+(4+i*4)]; for.size.2 i in range(0, 10, 1): fma.rn.f32 %f(445+i*-1), %f(236+i*1), %f234, %f(445+i*-1); ld.shared.f32 %f(237+i*1), [%r187+(12+i*4)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(435+i*-13), %f246, %f(234+i*13), %f(435+i*-13); ld.shared.f32 %f(247+i*1), [%r184+(64+i*64)]; fma.rn.f32 %f(434+i*-13), %f233, %f(247+i*1), %f(434+i*-13); fma.rn.f32 %f(433+i*-13), %f235, %f(247+i*1), %f(433+i*-13); fma.rn.f32 %f(432+i*-13), %f236, %f(247+i*1), %f(432+i*-13); fma.rn.f32 %f(431+i*-13), %f237, %f(247+i*1), %f(431+i*-13); fma.rn.f32 %f(430+i*-13), %f238, %f(247+i*1), %f(430+i*-13); fma.rn.f32 %f(429+i*-13), %f239, %f(247+i*1), %f(429+i*-13); fma.rn.f32 %f(428+i*-13), %f240, %f(247+i*1), %f(428+i*-13); fma.rn.f32 %f(427+i*-13), %f241, %f(247+i*1), %f(427+i*-13); fma.rn.f32 %f(426+i*-13), %f242, %f(247+i*1), %f(426+i*-13); fma.rn.f32 %f(425+i*-13), %f243, %f(247+i*1), %f(425+i*-13); fma.rn.f32 %f(424+i*-13), %f244, %f(247+i*1), %f(424+i*-13); fma.rn.f32 %f(423+i*-13), %f245, %f(247+i*1), %f(423+i*-13); fma.rn.f32 %f409, %f246, %f248, %f409; add.s32 %r206, %r206, 1; setp.ne.s32 %p1, %r206, 8; @%p1 bra $L__BB0_3; add.s32 %r205, %r205, 1; setp.lt.u32 %p2, %r205, 2; @%p2 bra $L__BB0_2; add.s32 %r204, %r204, 1; setp.lt.u32 %p3, %r204, 16; @%p3 bra $L__BB0_1; ld.param.u64 %rd42, [default_function_kernel0_param_2]; ld.param.u64 %rd41, [default_function_kernel0_param_3]; and.b32 %r201, %r43, 15; mul.lo.s32 %r190, %r201, 3; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(191+i*1), %r(43+i*-42), (507+i*15717), %r(2+i*189); cvta.to.global.u64 %rd35, %rd41; mul.wide.u32 %rd36, %r190, 4; add.s64 %rd37, %rd35, %rd36; ld.global.nc.f32 %f249, [%rd37]; add.f32 %f250, %f447, %f249; mov.f32 %f251, 0f00000000; max.f32 %f252, %f250, %f251; cvta.to.global.u64 %rd38, %rd42; mul.wide.s32 %rd39, %r192, 4; add.s64 %rd40, %rd38, %rd39; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd40+(0+i*52)], %f(252+i*2); add.f32 %f(253+i*2), %f(446+i*-1), %f249; max.f32 %f(254+i*2), %f(253+i*2), %f251; for.size.40 i in range(0, 2, 1): st.global.f32 [%rd40+(624+i*676)], %f(276+i*27); ld.global.nc.f32 %f(277+i*27), [%rd37+(4+i*4)]; add.f32 %f(278+i*27), %f(434+i*-13), %f(277+i*27); max.f32 %f(279+i*27), %f(278+i*27), %f251; st.global.f32 [%rd40+(676+i*676)], %f(279+i*27); add.f32 %f(280+i*27), %f(433+i*-13), %f(277+i*27); max.f32 %f(281+i*27), %f(280+i*27), %f251; st.global.f32 [%rd40+(728+i*676)], %f(281+i*27); add.f32 %f(282+i*27), %f(432+i*-13), %f(277+i*27); max.f32 %f(283+i*27), %f(282+i*27), %f251; st.global.f32 [%rd40+(780+i*676)], %f(283+i*27); add.f32 %f(284+i*27), %f(431+i*-13), %f(277+i*27); max.f32 %f(285+i*27), %f(284+i*27), %f251; st.global.f32 [%rd40+(832+i*676)], %f(285+i*27); add.f32 %f(286+i*27), %f(430+i*-13), %f(277+i*27); max.f32 %f(287+i*27), %f(286+i*27), %f251; st.global.f32 [%rd40+(884+i*676)], %f(287+i*27); add.f32 %f(288+i*27), %f(429+i*-13), %f(277+i*27); max.f32 %f(289+i*27), %f(288+i*27), %f251; st.global.f32 [%rd40+(936+i*676)], %f(289+i*27); add.f32 %f(290+i*27), %f(428+i*-13), %f(277+i*27); max.f32 %f(291+i*27), %f(290+i*27), %f251; st.global.f32 [%rd40+(988+i*676)], %f(291+i*27); add.f32 %f(292+i*27), %f(427+i*-13), %f(277+i*27); max.f32 %f(293+i*27), %f(292+i*27), %f251; st.global.f32 [%rd40+(1040+i*676)], %f(293+i*27); add.f32 %f(294+i*27), %f(426+i*-13), %f(277+i*27); max.f32 %f(295+i*27), %f(294+i*27), %f251; st.global.f32 [%rd40+(1092+i*676)], %f(295+i*27); add.f32 %f(296+i*27), %f(425+i*-13), %f(277+i*27); max.f32 %f(297+i*27), %f(296+i*27), %f251; st.global.f32 [%rd40+(1144+i*676)], %f(297+i*27); add.f32 %f(298+i*27), %f(424+i*-13), %f(277+i*27); max.f32 %f(299+i*27), %f(298+i*27), %f251; st.global.f32 [%rd40+(1196+i*676)], %f(299+i*27); add.f32 %f(300+i*27), %f(423+i*-13), %f(277+i*27); max.f32 %f(301+i*27), %f(300+i*27), %f251; st.global.f32 [%rd40+(1248+i*676)], %f(301+i*27); add.f32 %f(302+i*27), %f(422+i*-13), %f(277+i*27); max.f32 %f(303+i*27), %f(302+i*27), %f251; st.global.f32 [%rd40+1976], %f330; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[39]; __shared__ float pad_temp_shared[416]; __shared__ float placeholder_shared[768]; for (int ff_inner_init = 0; ff_inner_init < 3; ++ff_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = ff_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 13; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 32 + (int)threadIdx.x; int index2 = (int)blockIdx.x / 13 * 86528 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 32 + (int)threadIdx.x) / 208 * 43264 + rc_outer_outer * 2704 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 32 + (int)threadIdx.x) % 208 * 13 + (int)blockIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 24; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 32 + (int)threadIdx.x; int index4 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 512 + ((int)threadIdx.x >> 4) * 256 + rc_outer_outer * 16 + ((int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int ff_inner = 0; ff_inner < 3; ++ff_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = ff_inner * 13 + yy_inner; int index6 = ((int)threadIdx.x >> 4) * 208 + rc_outer_inner * 104 + rc_inner * 13 + yy_inner; int index7 = ((int)threadIdx.x & 15) * 48 + ff_inner * 16 + rc_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index8 = (int)blockIdx.x / 13 * 16224 + (int)threadIdx.x * 507 + ax1_inner * 169 + ax2_inner * 13 + (int)blockIdx.x % 13; int index9 = ax1_inner * 13 + ax2_inner; int index10 = ((int)threadIdx.x & 15) * 3 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpdr0rfpix
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<229>; .reg .b32 %r<128>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r32, %r1, 2; mov.u32 %r33, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r33, %r32; add.s32 %r34, %r1, 8112; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(35+i*4), %r(34+i*-33), 1626496491; shr.u32 %r(36+i*4), %r(35+i*4), 31; shr.s32 %r(37+i*4), %r(35+i*4), (11+i*-5); add.s32 %r(38+i*-24), %r(37+i*4), %r(36+i*4); mul.lo.s32 %r42, %r14, 169; sub.s32 %r43, %r1, %r42; add.s32 %r44, %r42, 2704; for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(45+i*6), %r(38+i*12), 43264, %r43; add.s32 %r(3+i*1), %r(45+i*6), %r44; add.s32 %r(46+i*6), %r1, (13520+i*5408); mul.hi.s32 %r(47+i*6), %r(46+i*6), 1626496491; shr.u32 %r(48+i*6), %r(47+i*6), 31; shr.s32 %r(49+i*6), %r(47+i*6), 11; add.s32 %r(50+i*6), %r(49+i*6), %r(48+i*6); for.size.7 i in range(0, 4, 1): mad.lo.s32 %r(57+i*6), %r(56+i*6), 43264, %r43; add.s32 %r(5+i*1), %r(57+i*6), %r44; add.s32 %r(58+i*6), %r1, (24336+i*5408); mul.hi.s32 %r(59+i*6), %r(58+i*6), 1626496491; shr.u32 %r(60+i*6), %r(59+i*6), 31; shr.s32 %r(61+i*6), %r(59+i*6), 11; add.s32 %r(62+i*6), %r(61+i*6), %r(60+i*6); mad.lo.s32 %r81, %r80, 43264, %r43; add.s32 %r9, %r81, %r44; mov.u32 %r10, %ctaid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(82+i*1), %r(10+i*-9), (12+i*-8); and.b32 %r84, %r83, -256; add.s32 %r85, %r84, %r82; shl.b32 %r86, %r1, 1; and.b32 %r87, %r86, 30; or.b32 %r11, %r85, %r87; shl.b32 %r88, %r1, 3; mov.u32 %r89, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r12, %r89, %r88; or.b32 %r90, %r86, 1; and.b32 %r91, %r90, 31; bfi.b32 %r92, %r10, %r91, 12, 20; add.s32 %r13, %r92, %r84; shl.b32 %r93, %r14, 5; add.s32 %r15, %r43, 2704; or.b32 %r16, %r93, 16; for.size.3 i in range(0, 2, 1): shl.b32 %r(94+i*2), %r(14+i*29), (7+i*-5); add.s32 %r(95+i*2), %r(89+i*-56), %r(94+i*2); add.s32 %r(17+i*1), %r(95+i*2), (8+i*86520); mov.u32 %r123, 0; mov.f32 %f213, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(214+i*1), %f213; $L__BB0_1: bar.sync 0; mul.lo.s32 %r98, %r123, 5408; add.s32 %r99, %r98, %r1; mul.wide.s32 %rd9, %r99, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(49+i*1), [%rd10+(0+i*10816)]; st.shared.f32 [%r2+(0+i*10816)], %f(49+i*1); for.size.7 i in range(0, 7, 1): ld.global.nc.f32 %f(51+i*2), [%rd10+(173056+i*173056)]; st.shared.f32 [%r2+(21632+i*21632)], %f(51+i*2); add.s32 %r(100+i*1), %r(3+i*1), %r98; mul.wide.s32 %rd(11+i*2), %r(100+i*1), 4; add.s64 %rd(12+i*2), %rd2, %rd(11+i*2); ld.global.nc.f32 %f(52+i*2), [%rd(12+i*2)+0]; st.shared.f32 [%r2+(32448+i*21632)], %f(52+i*2); setp.gt.s32 %p1, %r1, 255; @%p1 bra $L__BB0_3; shl.b32 %r107, %r123, 5; for.size.5 i in range(0, 2, 1): add.s32 %r(108+i*1), %r(11+i*2), %r107; mul.wide.s32 %rd(25+i*2), %r(108+i*1), 4; add.s64 %rd(26+i*2), %rd1, %rd(25+i*2); ld.global.nc.f32 %f(65+i*1), [%rd(26+i*2)+0]; st.shared.f32 [%r12+(0+i*4)], %f(65+i*1); $L__BB0_3: bar.sync 0; mov.u32 %r126, 8; mov.u32 %r124, %r18; mov.u32 %r125, %r17; $L__BB0_4: for.size.17 i in range(0, 2, 1): ld.shared.f32 %f(67+i*17), [%r125+(-8+i*4)]; ld.shared.f32 %f(68+i*17), [%r124+(-86528+i*676)]; fma.rn.f32 %f(69+i*17), %f(68+i*17), %f(67+i*17), %f(220+i*-151); ld.shared.f32 %f(70+i*17), [%r124+(-64896+i*676)]; fma.rn.f32 %f(71+i*17), %f(70+i*17), %f(67+i*17), %f(219+i*-148); ld.shared.f32 %f(72+i*17), [%r124+(-43264+i*676)]; fma.rn.f32 %f(73+i*17), %f(72+i*17), %f(67+i*17), %f(218+i*-145); ld.shared.f32 %f(74+i*17), [%r124+(-21632+i*676)]; fma.rn.f32 %f(75+i*17), %f(74+i*17), %f(67+i*17), %f(217+i*-142); ld.shared.f32 %f(76+i*17), [%r124+(0+i*676)]; fma.rn.f32 %f(77+i*17), %f(76+i*17), %f(67+i*17), %f(216+i*-139); ld.shared.f32 %f(78+i*17), [%r124+(21632+i*676)]; fma.rn.f32 %f(79+i*17), %f(78+i*17), %f(67+i*17), %f(215+i*-136); ld.shared.f32 %f(80+i*17), [%r124+(43264+i*676)]; fma.rn.f32 %f(81+i*17), %f(80+i*17), %f(67+i*17), %f(214+i*-133); ld.shared.f32 %f(82+i*17), [%r124+(64896+i*676)]; fma.rn.f32 %f(83+i*17), %f(82+i*17), %f(67+i*17), %f(213+i*-130); for.size.17 i in range(0, 2, 1): ld.shared.f32 %f(101+i*17), [%r125+(0+i*4)]; ld.shared.f32 %f(102+i*17), [%r124+(-85176+i*676)]; fma.rn.f32 %f(103+i*117), %f(102+i*17), %f(101+i*17), %f(86+i*17); ld.shared.f32 %f(104+i*16), [%r124+(-63544+i*676)]; fma.rn.f32 %f(105+i*114), %f(104+i*16), %f(101+i*17), %f(88+i*17); ld.shared.f32 %f(106+i*15), [%r124+(-41912+i*676)]; fma.rn.f32 %f(107+i*111), %f(106+i*15), %f(101+i*17), %f(90+i*17); ld.shared.f32 %f(108+i*14), [%r124+(-20280+i*676)]; fma.rn.f32 %f(109+i*108), %f(108+i*14), %f(101+i*17), %f(92+i*17); ld.shared.f32 %f(110+i*13), [%r124+(1352+i*676)]; fma.rn.f32 %f(111+i*105), %f(110+i*13), %f(101+i*17), %f(94+i*17); ld.shared.f32 %f(112+i*12), [%r124+(22984+i*676)]; fma.rn.f32 %f(113+i*102), %f(112+i*12), %f(101+i*17), %f(96+i*17); ld.shared.f32 %f(114+i*11), [%r124+(44616+i*676)]; fma.rn.f32 %f(115+i*99), %f(114+i*11), %f(101+i*17), %f(98+i*17); ld.shared.f32 %f(116+i*10), [%r124+(66248+i*676)]; fma.rn.f32 %f(117+i*96), %f(116+i*10), %f(101+i*17), %f(100+i*17); for.size.1 i in range(0, 2, 1): add.s32 %r(125+i*-1), %r(125+i*-1), (16+i*2688); add.s32 %r126, %r126, 16; setp.ne.s32 %p2, %r126, 72; @%p2 bra $L__BB0_4; mov.u32 %r127, 0; $L__BB0_6: mad.lo.s32 %r112, %r127, 169, %r15; shl.b32 %r113, %r112, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(115+i*1), %r(33+i*-17), %r(113+i*14); shl.b32 %r117, %r116, 2; add.s32 %r119, %r89, %r117; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(127+i*1), [%r(119+i*-4)+0]; for.size.2 i in range(0, 7, 1): fma.rn.f32 %f(129+i*2), %f(128+i*2), %f127, %f(220+i*-1); ld.shared.f32 %f(130+i*2), [%r115+(21632+i*21632)]; for.size.17 i in range(0, 2, 1): fma.rn.f32 %f(143+i*17), %f(142+i*17), %f(127+i*17), %f(213+i*-70); ld.shared.f32 %f(144+i*17), [%r119+(4+i*4)]; ld.shared.f32 %f(145+i*17), [%r115+(676+i*676)]; fma.rn.f32 %f(146+i*17), %f(145+i*17), %f(144+i*17), %f(129+i*17); ld.shared.f32 %f(147+i*17), [%r115+(22308+i*676)]; fma.rn.f32 %f(148+i*17), %f(147+i*17), %f(144+i*17), %f(131+i*17); ld.shared.f32 %f(149+i*17), [%r115+(43940+i*676)]; fma.rn.f32 %f(150+i*17), %f(149+i*17), %f(144+i*17), %f(133+i*17); ld.shared.f32 %f(151+i*17), [%r115+(65572+i*676)]; fma.rn.f32 %f(152+i*17), %f(151+i*17), %f(144+i*17), %f(135+i*17); ld.shared.f32 %f(153+i*17), [%r115+(87204+i*676)]; fma.rn.f32 %f(154+i*17), %f(153+i*17), %f(144+i*17), %f(137+i*17); ld.shared.f32 %f(155+i*17), [%r115+(108836+i*676)]; fma.rn.f32 %f(156+i*17), %f(155+i*17), %f(144+i*17), %f(139+i*17); ld.shared.f32 %f(157+i*17), [%r115+(130468+i*676)]; fma.rn.f32 %f(158+i*17), %f(157+i*17), %f(144+i*17), %f(141+i*17); ld.shared.f32 %f(159+i*17), [%r115+(152100+i*676)]; fma.rn.f32 %f177, %f176, %f161, %f160; ld.shared.f32 %f178, [%r119+12]; for.size.2 i in range(0, 8, 1): ld.shared.f32 %f(179+i*1), [%r115+(2028+i*21632)]; fma.rn.f32 %f(220+i*-1), %f(179+i*1), %f178, %f(163+i*2); add.s32 %r127, %r127, 4; setp.ne.s32 %p3, %r127, 16; @%p3 bra $L__BB0_6; add.s32 %r123, %r123, 1; setp.lt.u32 %p4, %r123, 8; @%p4 bra $L__BB0_1; shl.b32 %r120, %r10, 4; add.s32 %r121, %r120, %r14; mul.wide.s32 %rd29, %r121, 4; add.s64 %rd30, %rd4, %rd29; mad.lo.s32 %r122, %r10, 2704, %r1; ld.global.nc.f32 %f187, [%rd30]; add.f32 %f188, %f220, %f187; mov.f32 %f189, 0f00000000; max.f32 %f190, %f188, %f189; mul.wide.s32 %rd31, %r122, 4; add.s64 %rd32, %rd3, %rd31; for.size.3 i in range(0, 7, 1): st.global.f32 [%rd32+(0+i*32448)], %f(190+i*2); add.f32 %f(191+i*2), %f(219+i*-1), %f187; max.f32 %f(192+i*2), %f(191+i*2), %f189; st.global.f32 [%rd32+227136], %f204; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[512]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 5408 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 2704] = placeholder[index1 + 2704]; pad_temp_shared[index0 + 5408] = placeholder[index1 + 43264]; int index2 = ((int)threadIdx.x + 8112) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 8112] = placeholder[index2]; pad_temp_shared[index0 + 10816] = placeholder[index1 + 86528]; int index3 = ((int)threadIdx.x + 13520) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 13520] = placeholder[index3]; pad_temp_shared[index0 + 16224] = placeholder[index1 + 129792]; int index4 = ((int)threadIdx.x + 18928) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 18928] = placeholder[index4]; pad_temp_shared[index0 + 21632] = placeholder[index1 + 173056]; int index5 = ((int)threadIdx.x + 24336) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 24336] = placeholder[index5]; pad_temp_shared[index0 + 27040] = placeholder[index1 + 216320]; int index6 = ((int)threadIdx.x + 29744) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 29744] = placeholder[index6]; pad_temp_shared[index0 + 32448] = placeholder[index1 + 259584]; int index7 = ((int)threadIdx.x + 35152) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 35152] = placeholder[index7]; pad_temp_shared[index0 + 37856] = placeholder[index1 + 302848]; int index8 = ((int)threadIdx.x + 40560) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 40560] = placeholder[index8]; if ((int)threadIdx.x < 256) { int index9 = (int)blockIdx.x * 4096 + ((int)threadIdx.x >> 4) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x & 15) * 2; placeholder_shared[index0 * 2] = placeholder1[index9]; } if ((int)threadIdx.x < 256) { int index10 = (int)threadIdx.x * 2 + 1; int index11 = (int)blockIdx.x * 4096 + ((int)threadIdx.x * 2 + 1 >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x * 2 + 1 & 31); placeholder_shared[index10] = placeholder1[index11]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 16; ++rc_inner) { int index12 = rc_outer_inner * 2704 + rc_inner * 169 + (int)threadIdx.x % 169; int index13 = (int)threadIdx.x / 169 * 32 + rc_outer_inner * 16 + rc_inner; compute[0] = compute[0] + pad_temp_shared[index12] * placeholder_shared[index13]; compute[1] = compute[1] + (pad_temp_shared[index12 + 5408] * placeholder_shared[index13]); compute[2] = compute[2] + (pad_temp_shared[index12 + 10816] * placeholder_shared[index13]); compute[3] = compute[3] + (pad_temp_shared[index12 + 16224] * placeholder_shared[index13]); compute[4] = compute[4] + (pad_temp_shared[index12 + 21632] * placeholder_shared[index13]); compute[5] = compute[5] + (pad_temp_shared[index12 + 27040] * placeholder_shared[index13]); compute[6] = compute[6] + (pad_temp_shared[index12 + 32448] * placeholder_shared[index13]); compute[7] = compute[7] + (pad_temp_shared[index12 + 37856] * placeholder_shared[index13]); } } } for (int ax0_inner = 0; ax0_inner < 8; ++ax0_inner) { int index14 = ax0_inner * 8112 + (int)blockIdx.x * 2704 + (int)threadIdx.x; int index15 = ax0_inner; int index16 = (int)blockIdx.x * 16 + (int)threadIdx.x / 169; T_relu[index14] = max(compute[index15] + placeholder2[index16], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmplhe3zwxk
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<124>; .reg .b32 %r<62>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[416]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[64]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r13, %ctaid.x; mul.hi.s32 %r14, %r13, -770891565; add.s32 %r15, %r14, %r13; shr.u32 %r16, %r15, 31; shr.s32 %r17, %r15, 6; add.s32 %r1, %r17, %r16; mov.u32 %r2, %tid.x; mul.hi.s32 %r18, %r2, 1321528399; shr.u32 %r19, %r18, 31; shr.s32 %r20, %r18, 3; add.s32 %r21, %r20, %r19; mul.lo.s32 %r22, %r21, 26; sub.s32 %r23, %r2, %r22; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(24+i*4), %r(23+i*-10), 1321528399; shr.u32 %r(25+i*4), %r(24+i*4), 31; shr.s32 %r(26+i*4), %r(24+i*4), 2; add.s32 %r(27+i*4), %r(26+i*4), %r(25+i*4); mul.lo.s32 %r32, %r31, 13; sub.s32 %r33, %r13, %r32; mul.lo.s32 %r3, %r33, 13; shr.s32 %r34, %r18, 2; add.s32 %r7, %r34, %r19; mul.lo.s32 %r35, %r7, 13; sub.s32 %r36, %r2, %r35; shl.b32 %r37, %r2, 2; mov.u32 %r38, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r38, %r37; mul.lo.s32 %r39, %r1, 78; sub.s32 %r40, %r13, %r39; mul.hi.s32 %r41, %r40, 1321528399; shr.u32 %r42, %r41, 31; shr.s32 %r43, %r41, 2; add.s32 %r5, %r43, %r42; shl.b32 %r44, %r5, 11; and.b32 %r45, %r2, 1; mov.u32 %r46, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r46, %r37; mad.lo.s32 %r8, %r7, 26, %r36; shl.b32 %r47, %r8, 2; for.size.2 i in range(0, 2, 1): add.s32 %r(9+i*40), %r(38+i*10), %r(47+i*-44); mad.lo.s32 %r(48+i*2), %r(27+i*-26), (169+i*172887), %r(36+i*13); mad.lo.s32 %r51, %r21, 43264, %r50; cvta.to.global.u64 %rd13, %rd9; mul.wide.s32 %rd14, %r51, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd24, %rd15, 347464; shl.b32 %r52, %r2, 7; and.b32 %r53, %r52, -256; add.s32 %r54, %r44, %r53; or.b32 %r55, %r54, %r45; mul.wide.s32 %rd16, %r55, 4; add.s64 %rd17, %rd12, %rd16; add.s64 %rd23, %rd17, 8; mov.u32 %r61, 0; mov.f32 %f116, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(117+i*1), %f116; $L__BB0_1: bar.sync 0; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(33+i*1), [%rd24+(-347464+i*346112)]; st.shared.f32 [%r4+(0+i*208)], %f(33+i*1); setp.gt.s32 %p1, %r2, 15; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f35, [%rd23+-8]; st.shared.f32 [%r6], %f35; $L__BB0_3: bar.sync 0; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(36+i*5), [_ZZ24default_function_kernel0E18placeholder_shared+(0+i*4)]; ld.shared.f32 %f(37+i*5), [%r9+(0+i*52)]; fma.rn.f32 %f(38+i*-29), %f(37+i*5), %f(36+i*5), %f(123+i*-85); ld.shared.f32 %f(39+i*4), [_ZZ24default_function_kernel0E18placeholder_shared+(32+i*4)]; fma.rn.f32 %f(40+i*-30), %f(37+i*5), %f(39+i*4), %f(122+i*-82); for.size.6 i in range(0, 3, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(44+i*6+j*2), [_ZZ24default_function_kernel0E18placeholder_shared+(8+i*8+j*32)]; fma.rn.f32 %f(45+i*6+j*2), %f37, %f(44+i*6+j*2), %f(121+i*-2+j*-1); for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(48+i*6+j*1), [_ZZ24default_function_kernel0E18placeholder_shared+(12+i*8+j*32)]; fma.rn.f32 %f(11+i*2+j*1), %f42, %f(48+i*6+j*1), %f(45+i*6+j*2); bar.sync 0; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(62+i*1), [%rd24+(-346112+i*346112)]; st.shared.f32 [%r4+(0+i*208)], %f(62+i*1); @%p1 bra $L__BB0_5; ld.global.nc.f32 %f64, [%rd23]; st.shared.f32 [%r6], %f64; $L__BB0_5: bar.sync 0; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(65+i*5), [_ZZ24default_function_kernel0E18placeholder_shared+(0+i*4)]; ld.shared.f32 %f(66+i*5), [%r9+(0+i*52)]; fma.rn.f32 %f(67+i*56), %f(66+i*5), %f(65+i*5), %f(9+i*58); ld.shared.f32 %f(68+i*4), [_ZZ24default_function_kernel0E18placeholder_shared+(32+i*4)]; fma.rn.f32 %f(69+i*53), %f(66+i*5), %f(68+i*4), %f(10+i*59); for.size.6 i in range(0, 3, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(73+i*6+j*2), [_ZZ24default_function_kernel0E18placeholder_shared+(8+i*8+j*32)]; fma.rn.f32 %f(74+i*6+j*2), %f66, %f(73+i*6+j*2), %f(11+i*2+j*1); for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(77+i*6+j*1), [_ZZ24default_function_kernel0E18placeholder_shared+(12+i*8+j*32)]; fma.rn.f32 %f(121+i*-2+j*-1), %f71, %f(77+i*6+j*1), %f(74+i*6+j*2); for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (2704+i*-2688); add.s32 %r61, %r61, 2; setp.ne.s32 %p3, %r61, 128; @%p3 bra $L__BB0_1; cvta.to.global.u64 %rd18, %rd8; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(56+i*1), %r(7+i*-6), (8086+i*24362), %r(8+i*48); add.s32 %r58, %r57, %r3; mad.lo.s32 %r59, %r5, 1352, %r58; shl.b32 %r60, %r5, 3; mul.wide.s32 %rd19, %r60, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f91, [%rd20]; add.f32 %f92, %f123, %f91; mov.f32 %f93, 0f00000000; max.f32 %f94, %f92, %f93; mul.wide.s32 %rd21, %r59, 4; add.s64 %rd22, %rd1, %rd21; for.size.5 i in range(0, 3, 1): for.size.4 j in range(0, 2, 1): st.global.f32 [%rd22+(0+i*676+j*2704)], %f(94+i*6+j*3); ld.global.nc.f32 %f(95+i*6+j*3), [%rd20+(16+i*4+j*-12)]; add.f32 %f(96+i*6+j*3), %f(122+i*-2+j*-1), %f(95+i*6+j*3); max.f32 %f(97+i*6+j*3), %f(96+i*6+j*3), %f93; st.global.f32 [%rd22+2028], %f112; ld.global.nc.f32 %f113, [%rd20+28]; add.f32 %f114, %f116, %f113; max.f32 %f115, %f114, %f93; st.global.f32 [%rd22+4732], %f115; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[104]; __shared__ float placeholder_shared[16]; compute[0] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x / 78 * 173056 + (int)threadIdx.x / 26 * 43264 + rc_outer_outer * 338 + (int)threadIdx.x % 26 / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 52] = placeholder[index1 + 86528]; if ((int)threadIdx.x < 16) { int index2 = (int)blockIdx.x % 78 / 13 * 2048 + ((int)threadIdx.x >> 1) * 256 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index3 = (int)threadIdx.x / 13 * 26 + (int)threadIdx.x % 13; compute[0] = compute[0] + pad_temp_shared[index3] * placeholder_shared[0]; compute[4] = compute[4] + pad_temp_shared[index3] * placeholder_shared[8]; compute[0] = compute[0] + (pad_temp_shared[index3 + 13] * placeholder_shared[1]); compute[4] = compute[4] + (pad_temp_shared[index3 + 13] * placeholder_shared[9]); compute[1] = compute[1] + pad_temp_shared[index3] * placeholder_shared[2]; compute[5] = compute[5] + pad_temp_shared[index3] * placeholder_shared[10]; compute[1] = compute[1] + (pad_temp_shared[index3 + 13] * placeholder_shared[3]); compute[5] = compute[5] + (pad_temp_shared[index3 + 13] * placeholder_shared[11]); compute[2] = compute[2] + pad_temp_shared[index3] * placeholder_shared[4]; compute[6] = compute[6] + pad_temp_shared[index3] * placeholder_shared[12]; compute[2] = compute[2] + (pad_temp_shared[index3 + 13] * placeholder_shared[5]); compute[6] = compute[6] + (pad_temp_shared[index3 + 13] * placeholder_shared[13]); compute[3] = compute[3] + pad_temp_shared[index3] * placeholder_shared[6]; compute[7] = compute[7] + pad_temp_shared[index3] * placeholder_shared[14]; compute[3] = compute[3] + (pad_temp_shared[index3 + 13] * placeholder_shared[7]); compute[7] = compute[7] + (pad_temp_shared[index3 + 13] * placeholder_shared[15]); } for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index4 = (int)blockIdx.x / 78 * 32448 + (int)threadIdx.x / 13 * 8112 + (int)blockIdx.x % 78 / 13 * 1352 + ax1_inner * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; int index5 = ax1_inner; int index6 = (int)blockIdx.x % 78 / 13 * 8 + ax1_inner; T_relu[index4] = max(compute[index5] + placeholder2[index6], 0.000000e+00f); T_relu[index4 + 676] = max(compute[index5 + 4] + placeholder2[index6 + 4], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpixmglmhz
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<112>; .reg .b32 %r<146>; .reg .b64 %rd<39>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[53248]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd8, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd7; cvta.to.global.u64 %rd2, %rd6; cvta.to.global.u64 %rd3, %rd8; cvta.to.global.u64 %rd4, %rd9; mov.u32 %r1, %tid.x; mul.hi.s32 %r34, %r1, 1321528399; shr.u32 %r35, %r34, 31; shr.s32 %r36, %r34, 2; add.s32 %r2, %r36, %r35; mul.lo.s32 %r37, %r2, 169; mov.u32 %r38, %ctaid.x; mul.lo.s32 %r39, %r2, 13; sub.s32 %r40, %r1, %r39; mad.lo.s32 %r3, %r38, 13, %r40; add.s32 %r4, %r3, %r37; shl.b32 %r41, %r1, 2; mov.u32 %r42, _ZZ24default_function_kernel0E15pad_temp_shared; for.size.10 i in range(0, 2, 1): add.s32 %r(5+i*1), %r(42+i*9), %r(41+i*9); add.s32 %r(43+i*9), %r1, (1248+i*1248); mul.hi.s32 %r(44+i*9), %r(43+i*9), 1321528399; shr.u32 %r(45+i*9), %r(44+i*9), 31; shr.s32 %r(46+i*9), %r(44+i*9), 9; add.s32 %r(47+i*9), %r(46+i*9), %r(45+i*9); add.s32 %r(48+i*9), %r2, (96+i*-32); and.b32 %r(49+i*9), %r(48+i*9), 127; mul.lo.s32 %r(50+i*9), %r(49+i*9), 169; mad.lo.s32 %r(51+i*9), %r(47+i*9), 43264, %r3; add.s32 %r7, %r60, %r59; add.s32 %r61, %r1, 3744; mul.hi.s32 %r62, %r61, 1321528399; shr.u32 %r63, %r62, 31; shr.s32 %r64, %r62, 9; add.s32 %r65, %r64, %r63; add.s32 %r66, %r37, 5408; for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(67+i*6), %r(65+i*7), 43264, %r3; add.s32 %r(8+i*1), %r(67+i*6), %r(66+i*-16); add.s32 %r(68+i*6), %r1, (6240+i*1248); mul.hi.s32 %r(69+i*6), %r(68+i*6), 1321528399; shr.u32 %r(70+i*6), %r(69+i*6), 31; shr.s32 %r(71+i*6), %r(69+i*6), 9; add.s32 %r(72+i*6), %r(71+i*6), %r(70+i*6); for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(79+i*6), %r(78+i*6), 43264, %r3; add.s32 %r(10+i*1), %r(79+i*6), %r(59+i*7); add.s32 %r(80+i*6), %r1, (8736+i*2496); mul.hi.s32 %r(81+i*6), %r(80+i*6), 1321528399; shr.u32 %r(82+i*6), %r(81+i*6), 31; shr.s32 %r(83+i*6), %r(81+i*6), 9; add.s32 %r(84+i*6), %r(83+i*6), %r(82+i*6); mad.lo.s32 %r91, %r90, 43264, %r3; add.s32 %r12, %r91, %r50; add.s32 %r92, %r1, 12480; mul.hi.s32 %r93, %r92, 1321528399; shr.u32 %r94, %r93, 31; shr.s32 %r95, %r93, 9; add.s32 %r96, %r95, %r94; mad.lo.s32 %r97, %r96, 43264, %r4; add.s32 %r13, %r97, 10816; shr.u32 %r98, %r1, 7; and.b32 %r99, %r1, 127; prmt.b32 %r14, %r98, %r99, 8452; mov.u32 %r100, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r15, %r100, %r41; for.size.4 i in range(0, 3, 1): shr.u32 %r(101+i*3), %r(43+i*9), 7; add.s32 %r(102+i*3), %r1, (96+i*-32); and.b32 %r(103+i*3), %r(102+i*3), 127; prmt.b32 %r(16+i*1), %r(101+i*3), %r(103+i*3), 8452; mul.hi.s32 %r110, %r1, -770891565; add.s32 %r111, %r110, %r1; shr.u32 %r112, %r111, 31; shr.s32 %r113, %r111, 8; add.s32 %r114, %r113, %r112; mul.lo.s32 %r115, %r114, 312; sub.s32 %r116, %r1, %r115; mul.hi.s32 %r117, %r116, 1321528399; shr.u32 %r118, %r117, 31; shr.s32 %r119, %r117, 2; add.s32 %r19, %r119, %r118; shl.b32 %r120, %r19, 10; add.s32 %r121, %r100, %r120; add.s32 %r20, %r121, 512; mad.lo.s32 %r122, %r114, 1664, %r40; shl.b32 %r123, %r122, 2; add.s32 %r124, %r42, %r123; add.s32 %r21, %r124, 26624; mov.u32 %r142, 0; mov.f32 %f108, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(109+i*1), %f108; $L__BB0_1: bar.sync 0; mul.lo.s32 %r23, %r142, 21632; for.size.5 i in range(0, 2, 1): add.s32 %r(125+i*1), %r(4+i*2), %r23; mul.wide.s32 %rd(10+i*2), %r(125+i*1), 4; add.s64 %rd(11+i*2), %rd2, %rd(10+i*2); ld.global.nc.f32 %f(17+i*1), [%rd(11+i*2)+0]; st.shared.f32 [%r5+(0+i*4992)], %f(17+i*1); for.size.5 i in range(0, 2, 1): add.s32 %r(127+i*1), %r(7+i*1), %r23; mul.wide.s32 %rd(14+i*2), %r(127+i*1), 4; add.s64 %rd(15+i*2), %rd2, %rd(14+i*2); ld.global.nc.f32 %f(19+i*1), [%rd(15+i*2)+0]; st.shared.f32 [%r5+(9984+i*4992)], %f(19+i*1); ld.global.nc.f32 %f21, [%rd11+519168]; for.size.5 i in range(0, 3, 1): st.shared.f32 [%r5+(19968+i*4992)], %f(21+i*1); add.s32 %r(129+i*1), %r(9+i*1), %r23; mul.wide.s32 %rd(18+i*2), %r(129+i*1), 4; add.s64 %rd(19+i*2), %rd2, %rd(18+i*2); ld.global.nc.f32 %f(22+i*1), [%rd(19+i*2)+0]; st.shared.f32 [%r5+34944], %f24; ld.global.nc.f32 %f25, [%rd11+1038336]; st.shared.f32 [%r5+39936], %f25; add.s32 %r132, %r12, %r23; mul.wide.s32 %rd24, %r132, 4; add.s64 %rd25, %rd2, %rd24; ld.global.nc.f32 %f26, [%rd25]; st.shared.f32 [%r5+44928], %f26; setp.gt.s32 %p1, %r1, 831; @%p1 bra $L__BB0_3; add.s32 %r133, %r13, %r23; mul.wide.s32 %rd26, %r133, 4; add.s64 %rd27, %rd2, %rd26; ld.global.nc.f32 %f27, [%rd27]; st.shared.f32 [%r5+49920], %f27; $L__BB0_3: setp.gt.s32 %p2, %r1, 1151; shl.b32 %r134, %r142, 7; for.size.5 i in range(0, 2, 1): add.s32 %r(135+i*1), %r(14+i*2), %r134; mul.wide.s32 %rd(28+i*1), %r(135+i*1), 4; add.s64 %rd(5+i*25), %rd1, %rd(28+i*1); ld.global.nc.f32 %f(28+i*1), [%rd(5+i*25)+0]; st.shared.f32 [%r15+(0+i*4992)], %f(28+i*1); for.size.5 i in range(0, 2, 1): add.s32 %r(137+i*1), %r(17+i*1), %r134; mul.wide.s32 %rd(31+i*2), %r(137+i*1), 4; add.s64 %rd(32+i*2), %rd1, %rd(31+i*2); ld.global.nc.f32 %f(30+i*1), [%rd(32+i*2)+0]; st.shared.f32 [%r15+(9984+i*4992)], %f(30+i*1); @%p2 bra $L__BB0_5; ld.global.nc.f32 %f32, [%rd5+39936]; st.shared.f32 [%r15+19968], %f32; $L__BB0_5: bar.sync 0; mov.u32 %r145, 512; mov.u32 %r143, %r21; mov.u32 %r144, %r20; $L__BB0_6: for.size.8 i in range(0, 2, 1): ld.shared.f32 %f(33+i*8), [%r144+(-512+i*4)]; ld.shared.f32 %f(34+i*8), [%r143+(-26624+i*52)]; fma.rn.f32 %f(35+i*8), %f(34+i*8), %f(33+i*8), %f(111+i*-76); ld.shared.f32 %f(36+i*8), [%r143+(0+i*52)]; fma.rn.f32 %f(37+i*8), %f(36+i*8), %f(33+i*8), %f(110+i*-73); ld.shared.f32 %f(38+i*8), [%r144+(0+i*4)]; fma.rn.f32 %f(39+i*8), %f(34+i*8), %f(38+i*8), %f(109+i*-70); fma.rn.f32 %f(40+i*8), %f(38+i*8), %f(36+i*8), %f(108+i*-68); for.size.8 i in range(0, 5, 1): ld.shared.f32 %f(49+i*8), [%r144+(-504+i*4)]; ld.shared.f32 %f(50+i*8), [%r143+(-26520+i*52)]; fma.rn.f32 %f(51+i*8), %f(50+i*8), %f(49+i*8), %f(43+i*8); ld.shared.f32 %f(52+i*8), [%r143+(104+i*52)]; fma.rn.f32 %f(53+i*8), %f(52+i*8), %f(49+i*8), %f(45+i*8); ld.shared.f32 %f(54+i*8), [%r144+(8+i*4)]; fma.rn.f32 %f(55+i*8), %f(50+i*8), %f(54+i*8), %f(47+i*8); fma.rn.f32 %f(56+i*8), %f(54+i*8), %f(52+i*8), %f(48+i*8); ld.shared.f32 %f89, [%r144+-484]; for.size.2 i in range(0, 2, 1): ld.shared.f32 %f(90+i*1), [%r143+(-26260+i*26624)]; fma.rn.f32 %f(111+i*-1), %f(90+i*1), %f89, %f(83+i*2); ld.shared.f32 %f92, [%r144+28]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(109+i*-1), %f(90+i*2), %f(92+i*-1), %f(87+i*1); for.size.1 i in range(0, 2, 1): add.s32 %r(144+i*-1), %r(144+i*-1), (32+i*384); add.s32 %r145, %r145, 32; setp.ne.s32 %p3, %r145, 1024; @%p3 bra $L__BB0_6; add.s32 %r142, %r142, 1; setp.lt.u32 %p4, %r142, 2; @%p4 bra $L__BB0_1; shl.b32 %r140, %r19, 1; mad.lo.s32 %r141, %r2, 338, %r3; mul.wide.s32 %rd35, %r140, 4; add.s64 %rd36, %rd4, %rd35; ld.global.nc.f32 %f93, [%rd36]; add.f32 %f94, %f111, %f93; mov.f32 %f95, 0f00000000; max.f32 %f96, %f94, %f95; mul.wide.s32 %rd37, %r141, 4; add.s64 %rd38, %rd3, %rd37; st.global.f32 [%rd38], %f96; add.f32 %f97, %f110, %f93; max.f32 %f98, %f97, %f95; st.global.f32 [%rd38+129792], %f98; ld.global.nc.f32 %f99, [%rd36+4]; for.size.3 i in range(0, 2, 1): add.f32 %f(100+i*2), %f(109+i*-1), %f99; max.f32 %f(101+i*2), %f(100+i*2), %f95; st.global.f32 [%rd38+(676+i*129792)], %f(101+i*2); ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[13312]; __shared__ float placeholder_shared[6144]; compute[0] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = rc_outer_outer * 21632 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; int index2 = ((int)threadIdx.x + 1248) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 96 & 127) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 1248] = placeholder[index2]; int index3 = ((int)threadIdx.x + 2496) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 64 & 127) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 2496] = placeholder[index3]; int index4 = ((int)threadIdx.x + 3744) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 32) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 3744] = placeholder[index4]; pad_temp_shared[index0 + 4992] = placeholder[index1 + 129792]; int index5 = ((int)threadIdx.x + 6240) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 96 & 127) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 6240] = placeholder[index5]; int index6 = ((int)threadIdx.x + 7488) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 64 & 127) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 7488] = placeholder[index6]; int index7 = ((int)threadIdx.x + 8736) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 32) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 8736] = placeholder[index7]; pad_temp_shared[index0 + 9984] = placeholder[index1 + 259584]; int index8 = ((int)threadIdx.x + 11232) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 96 & 127) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 11232] = placeholder[index8]; if ((int)threadIdx.x < 832) { int index9 = ((int)threadIdx.x + 12480) / 1664 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 13 + 64) * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0 + 12480] = placeholder[index9]; } int index10 = ((int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x & 127); placeholder_shared[index0] = placeholder1[index10]; int index11 = ((int)threadIdx.x + 1248 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 96 & 127); placeholder_shared[index0 + 1248] = placeholder1[index11]; int index12 = ((int)threadIdx.x + 2496 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 64 & 127); placeholder_shared[index0 + 2496] = placeholder1[index12]; int index13 = ((int)threadIdx.x + 3744 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 32 & 127); placeholder_shared[index0 + 3744] = placeholder1[index13]; if ((int)threadIdx.x < 1152) { placeholder_shared[index0 + 4992] = placeholder1[index10 + 9984]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index14 = (int)threadIdx.x / 312 * 1664 + rc_outer_inner * 13 + (int)threadIdx.x % 13; int index15 = (int)threadIdx.x % 312 / 13 * 256 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index14] * placeholder_shared[index15]; compute[2] = compute[2] + (pad_temp_shared[index14 + 6656] * placeholder_shared[index15]); compute[1] = compute[1] + (pad_temp_shared[index14] * placeholder_shared[index15 + 128]); compute[3] = compute[3] + (pad_temp_shared[index14 + 6656] * placeholder_shared[index15 + 128]); } } for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index16 = (int)threadIdx.x / 13 * 338 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index17 = ax1_inner; int index18 = (int)threadIdx.x % 312 / 13 * 2 + ax1_inner; T_relu[index16] = max(compute[index17] + placeholder2[index18], 0.000000e+00f); T_relu[index16 + 32448] = max(compute[index17 + 2] + placeholder2[index18], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpb1bucsz7
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<11>; .reg .f32 %f<289>; .reg .b32 %r<125>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[13312]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1536]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd6; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; mul.hi.s32 %r35, %r1, 1321528399; shr.u32 %r36, %r35, 31; shr.s32 %r37, %r35, 2; add.s32 %r38, %r37, %r36; mov.u32 %r39, %ctaid.x; mul.hi.s32 %r40, %r39, 1321528399; shr.u32 %r41, %r40, 31; shr.s32 %r42, %r40, 2; add.s32 %r5, %r42, %r41; mul.lo.s32 %r43, %r5, 13; sub.s32 %r44, %r39, %r43; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(2+i*43), %r(44+i*-6), 13; sub.s32 %r3, %r1, %r45; add.s32 %r4, %r2, %r3; mul.lo.s32 %r6, %r5, 3072; mul.hi.s32 %r46, %r1, -770891565; add.s32 %r47, %r46, %r1; shr.u32 %r48, %r47, 31; shr.s32 %r49, %r47, 6; add.s32 %r7, %r49, %r48; mad.lo.s32 %r8, %r7, 1664, %r3; mul.lo.s32 %r50, %r7, 78; sub.s32 %r51, %r1, %r50; mul.hi.s32 %r52, %r51, 1321528399; shr.u32 %r53, %r52, 31; shr.s32 %r54, %r52, 2; add.s32 %r9, %r54, %r53; for.size.1 i in range(0, 2, 1): shl.b32 %r(10+i*45), %r(9+i*-8), (5+i*-2); and.b32 %r11, %r55, -256; shl.b32 %r56, %r1, 2; mov.u32 %r57, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r13, %r57, %r56; for.size.4 i in range(0, 2, 1): add.s32 %r(58+i*2), %r1, (156+i*156); shl.b32 %r(59+i*2), %r(58+i*2), 3; and.b32 %r(14+i*2), %r(59+i*2), -256; and.b32 %r(15+i*2), %r(58+i*2), 31; mov.u32 %r62, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r63, %r62, %r56; for.size.1 i in range(0, 2, 1): add.s32 %r(18+i*1), %r(63+i*-25), (624+i*-612); mov.u32 %r119, 0; mov.f32 %f281, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(282+i*1), %f281; $L__BB0_1: bar.sync 0; mad.lo.s32 %r22, %r119, 5408, %r4; mov.u32 %r122, 12; mov.u32 %r120, %r1; mov.u32 %r121, %r19; mov.u32 %r123, %r18; $L__BB0_2: add.s32 %r65, %r121, -12; setp.gt.s32 %p1, %r65, 255; @%p1 bra $L__BB0_5; setp.gt.s32 %p2, %r120, 3327; @%p2 bra $L__BB0_5; shr.s32 %r67, %r65, 5; and.b32 %r68, %r65, 31; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(69+i*1), %r(67+i*1), (43264+i*-43095), %r(22+i*47); mul.wide.s32 %rd9, %r70, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f33, [%rd10]; st.shared.f32 [%r123+-624], %f33; $L__BB0_5: setp.gt.s32 %p3, %r121, 255; @%p3 bra $L__BB0_8; add.s32 %r71, %r120, 156; setp.gt.s32 %p4, %r71, 3327; @%p4 bra $L__BB0_8; shr.s32 %r72, %r121, 5; and.b32 %r73, %r121, 31; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(74+i*1), %r(72+i*1), (43264+i*-43095), %r(22+i*52); mul.wide.s32 %rd11, %r75, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f34, [%rd12]; st.shared.f32 [%r123], %f34; $L__BB0_8: for.size.1 i in range(0, 2, 1): add.s32 %r(123+i*-2), %r(123+i*-2), (1248+i*-1224); for.size.1 i in range(0, 2, 1): add.s32 %r(120+i*2), %r(120+i*2), (312+i*-288); setp.ne.s32 %p5, %r122, 276; @%p5 bra $L__BB0_2; setp.gt.s32 %p6, %r1, 383; shl.b32 %r76, %r119, 5; add.s32 %r31, %r76, %r6; @%p6 bra $L__BB0_11; and.b32 %r116, %r1, 31; for.size.1 i in range(0, 2, 1): add.s32 %r(77+i*1), %r(31+i*46), %r(116+i*-105); mul.wide.s32 %rd13, %r78, 4; add.s64 %rd14, %rd3, %rd13; ld.global.nc.f32 %f35, [%rd14]; st.shared.f32 [%r13], %f35; $L__BB0_11: setp.gt.s32 %p7, %r1, 227; @%p7 bra $L__BB0_13; for.size.1 i in range(0, 2, 1): add.s32 %r(79+i*1), %r(31+i*48), %r(15+i*-1); mul.wide.s32 %rd15, %r80, 4; add.s64 %rd16, %rd3, %rd15; ld.global.nc.f32 %f36, [%rd16]; st.shared.f32 [%r13+624], %f36; $L__BB0_13: setp.gt.s32 %p8, %r1, 71; @%p8 bra $L__BB0_15; for.size.1 i in range(0, 2, 1): add.s32 %r(81+i*1), %r(31+i*50), %r(17+i*-1); mul.wide.s32 %rd17, %r82, 4; add.s64 %rd18, %rd3, %rd17; ld.global.nc.f32 %f37, [%rd18]; st.shared.f32 [%r13+1248], %f37; $L__BB0_15: bar.sync 0; mov.u32 %r124, 0; $L__BB0_16: mov.u32 %r118, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r117, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r85, %r124, 208, %r8; for.size.2 i in range(0, 2, 1): shl.b32 %r(86+i*2), %r(124+i*-39), (4+i*-2); add.s32 %r(87+i*3), %r(86+i*31), %r(10+i*78); shl.b32 %r91, %r87, 2; add.s32 %r93, %r118, %r91; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(38+i*1), [%r(93+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(40+i*2), %f39, %f(38+i*3), %f(288+i*-4); ld.shared.f32 %f(41+i*2), [%r93+(768+i*-764)]; for.size.5 i in range(0, 14, 1): ld.shared.f32 %f(44+i*5), [%r90+(52+i*52)]; fma.rn.f32 %f(45+i*5), %f(44+i*5), %f(43+i*5), %f(40+i*5); ld.shared.f32 %f(46+i*5), [%r93+(772+i*4)]; fma.rn.f32 %f(47+i*5), %f(44+i*5), %f(46+i*5), %f(42+i*5); ld.shared.f32 %f(48+i*5), [%r93+(8+i*4)]; ld.shared.f32 %f114, [%r90+780]; fma.rn.f32 %f288, %f114, %f113, %f110; ld.shared.f32 %f115, [%r93+828]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(284+i*-166), %f(114+i*2), %f(115+i*-74), %f(112+i*171); ld.shared.f32 %f(116+i*3), [%r90+(1664+i*52)]; fma.rn.f32 %f(117+i*3), %f(116+i*3), %f(38+i*5), %f(287+i*-170); for.size.3 i in range(0, 13, 1): fma.rn.f32 %f(121+i*3), %f(119+i*3), %f(46+i*5), %f(118+i*3); ld.shared.f32 %f(122+i*3), [%r90+(1768+i*52)]; fma.rn.f32 %f(123+i*3), %f(122+i*3), %f(48+i*5), %f(120+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(160+i*123), %f(158+i*3), %f(111+i*4), %f(157+i*3); ld.shared.f32 %f(161+i*1), [%r90+(2444+i*884)]; fma.rn.f32 %f(287+i*-124), %f(161+i*1), %f(113+i*-75), %f(159+i*127); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(164+i*3), %f(162+i*3), %f(41+i*5), %f(282+i*-118); ld.shared.f32 %f(165+i*3), [%r90+(3380+i*52)]; fma.rn.f32 %f(166+i*3), %f(165+i*3), %f(43+i*5), %f(163+i*3); for.size.3 i in range(0, 12, 1): fma.rn.f32 %f(170+i*3), %f(168+i*3), %f(51+i*5), %f(167+i*3); ld.shared.f32 %f(171+i*3), [%r90+(3484+i*52)]; fma.rn.f32 %f(172+i*3), %f(171+i*3), %f(53+i*5), %f(169+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(206+i*76), %f(204+i*3), %f(111+i*4), %f(203+i*3); ld.shared.f32 %f(207+i*1), [%r90+(4108+i*884)]; fma.rn.f32 %f(286+i*-77), %f(207+i*1), %f(113+i*-75), %f(205+i*80); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(210+i*3), %f(208+i*3), %f(41+i*5), %f(281+i*-71); ld.shared.f32 %f(211+i*3), [%r90+(5044+i*52)]; fma.rn.f32 %f(212+i*3), %f(211+i*3), %f(43+i*5), %f(209+i*3); for.size.3 i in range(0, 12, 1): fma.rn.f32 %f(216+i*3), %f(214+i*3), %f(51+i*5), %f(213+i*3); ld.shared.f32 %f(217+i*3), [%r90+(5148+i*52)]; fma.rn.f32 %f(218+i*3), %f(217+i*3), %f(53+i*5), %f(215+i*3); fma.rn.f32 %f252, %f250, %f111, %f249; ld.shared.f32 %f253, [%r90+5772]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(285+i*-4), %f253, %f(113+i*2), %f(251+i*1); add.s32 %r94, %r124, 1; mov.u32 %r124, 1; setp.lt.u32 %p9, %r94, 2; @%p9 bra $L__BB0_16; add.s32 %r119, %r119, 1; setp.lt.u32 %p10, %r119, 8; @%p10 bra $L__BB0_1; mul.hi.s32 %r115, %r1, -770891565; add.s32 %r114, %r115, %r1; shr.u32 %r113, %r114, 31; shr.s32 %r112, %r114, 6; add.s32 %r111, %r112, %r113; mul.lo.s32 %r110, %r111, 78; sub.s32 %r109, %r1, %r110; mul.hi.s32 %r108, %r109, 1321528399; shr.u32 %r107, %r108, 31; shr.s32 %r106, %r108, 2; add.s32 %r105, %r106, %r107; mov.u32 %r104, %ctaid.x; mul.hi.s32 %r103, %r104, 1321528399; shr.u32 %r102, %r103, 31; shr.s32 %r101, %r103, 2; add.s32 %r100, %r101, %r102; mad.lo.s32 %r95, %r100, 12, %r105; mul.wide.s32 %rd19, %r95, 4; add.s64 %rd20, %rd4, %rd19; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(96+i*1), %r(111+i*-11), (32448+i*-30420), %r(3+i*93); add.s32 %r98, %r97, %r2; mad.lo.s32 %r99, %r105, 169, %r98; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(254+i*1), [%rd20+(24+i*-24)]; add.f32 %f256, %f288, %f255; mov.f32 %f257, 0f00000000; max.f32 %f258, %f256, %f257; mul.wide.s32 %rd21, %r99, 4; add.s64 %rd22, %rd2, %rd21; for.size.4 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd22+(0+i*32448+j*4056)], %f(258+i*4+j*2); add.f32 %f(259+i*4+j*2), %f(284+i*-1+j*3), %f(254+j*1); max.f32 %f(260+i*4+j*2), %f(259+i*4+j*2), %f257; st.global.f32 [%rd22+97344], %f270; add.f32 %f271, %f281, %f254; max.f32 %f272, %f271, %f257; st.global.f32 [%rd22+101400], %f272; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[3328]; __shared__ float placeholder_shared[384]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 4; ++nn_outer_inner_init) { int index0 = nn_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 22; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 13 < 256) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 156 + (int)threadIdx.x < 3328) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 156 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 13 >> 5) * 43264 + rc_outer_outer * 5408 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 13 & 31) * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 3; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 156 + (int)threadIdx.x < 384) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 156 + (int)threadIdx.x; int index4 = (int)blockIdx.x / 13 * 3072 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 156 + (int)threadIdx.x >> 5) * 256 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 156 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 4; ++nn_outer_inner) { for (int rc_inner = 0; rc_inner < 16; ++rc_inner) { int index5 = nn_outer_inner; int index6 = (int)threadIdx.x / 78 * 1664 + nn_outer_inner * 416 + rc_outer_inner * 208 + rc_inner * 13 + (int)threadIdx.x % 13; int index7 = (int)threadIdx.x % 78 / 13 * 32 + rc_outer_inner * 16 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 4] = compute[index5 + 4] + (pad_temp_shared[index6] * placeholder_shared[index7 + 192]); } } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { int index8 = (int)threadIdx.x / 78 * 32448 + ax0_inner * 8112 + (int)blockIdx.x / 13 * 2028 + (int)threadIdx.x % 78 / 13 * 169 + (int)blockIdx.x % 13 * 13 + (int)threadIdx.x % 13; int index9 = ax0_inner; int index10 = (int)blockIdx.x / 13 * 12 + (int)threadIdx.x % 78 / 13; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 1014] = max(compute[index9 + 4] + placeholder2[index10 + 6], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpqh3ylson
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<260>; .reg .b32 %r<152>; .reg .b64 %rd<45>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[512]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1536]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd6; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r25, %ctaid.x; mul.hi.s32 %r26, %r25, 1626496491; shr.u32 %r27, %r26, 31; for.size.2 i in range(0, 2, 1): shr.s32 %r(28+i*1), %r26, (8+i*-2); add.s32 %r(1+i*29), %r(28+i*1), %r27; mul.lo.s32 %r31, %r30, 169; sub.s32 %r2, %r25, %r31; mad.lo.s32 %r3, %r1, 173056, %r2; mul.lo.s32 %r32, %r1, 676; sub.s32 %r33, %r25, %r32; mul.hi.s32 %r34, %r33, 1626496491; shr.u32 %r35, %r34, 31; shr.s32 %r36, %r34, 6; add.s32 %r4, %r36, %r35; mul.lo.s32 %r5, %r4, 3072; mov.u32 %r6, %tid.x; mul.lo.s32 %r7, %r6, 192; add.s32 %r8, %r6, 14; shl.b32 %r37, %r6, 2; mov.u32 %r38, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r9, %r38, %r37; mov.u32 %r146, 0; mov.f32 %f212, 0f00000000; for.size.1 i in range(0, 23, 1): mov.f32 %f(213+i*1), %f212; $L__BB0_1: bar.sync 0; mad.lo.s32 %r11, %r146, 5408, %r3; mov.u32 %r149, 32; mov.u32 %r147, %r6; mov.u32 %r148, %r8; $L__BB0_2: shr.s32 %r40, %r147, 5; and.b32 %r41, %r147, 31; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(42+i*1), %r(40+i*1), (43264+i*-43095), %r(11+i*31); mul.wide.s32 %rd9, %r43, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f97, [%rd10]; add.s32 %r44, %r9, %r149; for.size.9 i in range(0, 4, 1): st.shared.f32 [%r44+(-32+i*8)], %f(97+i*1); add.s32 %r(45+i*5), %r147, (2+i*2); shr.s32 %r(46+i*5), %r(45+i*5), 5; and.b32 %r(47+i*5), %r(45+i*5), 31; mad.lo.s32 %r(48+i*5), %r(46+i*5), 43264, %r11; mad.lo.s32 %r(49+i*5), %r(47+i*5), 169, %r(48+i*5); mul.wide.s32 %rd(11+i*2), %r(49+i*5), 4; add.s64 %rd(12+i*2), %rd1, %rd(11+i*2); ld.global.nc.f32 %f(98+i*1), [%rd(12+i*2)+0]; for.size.9 i in range(0, 2, 1): st.shared.f32 [%r44+(0+i*8)], %f(101+i*1); add.s32 %r(65+i*5), %r148, (-4+i*2); shr.s32 %r(66+i*5), %r(65+i*5), 5; and.b32 %r(67+i*5), %r(65+i*5), 31; mad.lo.s32 %r(68+i*5), %r(66+i*5), 43264, %r11; mad.lo.s32 %r(69+i*5), %r(67+i*5), 169, %r(68+i*5); mul.wide.s32 %rd(19+i*2), %r(69+i*5), 4; add.s64 %rd(20+i*2), %rd1, %rd(19+i*2); ld.global.nc.f32 %f(102+i*1), [%rd(20+i*2)+0]; st.shared.f32 [%r44+16], %f103; shr.s32 %r75, %r148, 5; and.b32 %r76, %r148, 31; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(77+i*1), %r(75+i*1), (43264+i*-43095), %r(11+i*66); mul.wide.s32 %rd23, %r78, 4; add.s64 %rd24, %rd1, %rd23; ld.global.nc.f32 %f104, [%rd24]; st.shared.f32 [%r44+24], %f104; add.s32 %r149, %r149, 64; setp.ne.s32 %p1, %r149, 544; for.size.1 i in range(0, 2, 1): add.s32 %r(16+i*131), %r148, (16+i*-14); mov.u32 %r148, %r16; @%p1 bra $L__BB0_2; shl.b32 %r80, %r146, 5; add.s32 %r18, %r80, %r5; mov.u32 %r150, 0; $L__BB0_4: shl.b32 %r81, %r150, 1; add.s32 %r82, %r81, %r6; shl.b32 %r83, %r82, 3; for.size.1 i in range(0, 2, 1): and.b32 %r(84+i*1), %r(83+i*-1), (-256+i*287); for.size.1 i in range(0, 2, 1): add.s32 %r(86+i*1), %r(18+i*68), %r(85+i*-1); mul.wide.s32 %rd25, %r87, 4; add.s64 %rd26, %rd3, %rd25; ld.global.nc.f32 %f105, [%rd26]; shl.b32 %r88, %r82, 2; mov.u32 %r89, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r90, %r89, %r88; for.size.10 i in range(0, 7, 1): st.shared.f32 [%r90+(0+i*8)], %f(105+i*1); add.s32 %r(91+i*6), %r82, (2+i*2); shl.b32 %r(92+i*6), %r(91+i*6), 3; and.b32 %r(93+i*6), %r(92+i*6), -256; and.b32 %r(94+i*6), %r(91+i*6), 31; add.s32 %r(95+i*6), %r18, %r(94+i*6); add.s32 %r(96+i*6), %r(95+i*6), %r(93+i*6); mul.wide.s32 %rd(27+i*2), %r(96+i*6), 4; add.s64 %rd(28+i*2), %rd3, %rd(27+i*2); ld.global.nc.f32 %f(106+i*1), [%rd(28+i*2)+0]; st.shared.f32 [%r90+56], %f112; add.s32 %r150, %r150, 8; setp.ne.s32 %p2, %r150, 192; @%p2 bra $L__BB0_4; bar.sync 0; mov.u32 %r151, 0; $L__BB0_6: for.size.2 i in range(0, 2, 1): add.s32 %r(134+i*3), %r(151+i*-62), %r(7+i*128); shl.b32 %r(135+i*3), %r(134+i*17), 2; add.s32 %r140, %r38, %r138; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(113+i*1), [%r(140+i*-3)+0]; for.size.2 i in range(0, 5, 1): fma.rn.f32 %f(115+i*2), %f113, %f(114+i*2), %f(235+i*-1); ld.shared.f32 %f(116+i*2), [%r137+(128+i*128)]; for.size.7 i in range(0, 2, 1): fma.rn.f32 %f(125+i*7), %f(113+i*13), %f124, %f(230+i*-6); ld.shared.f32 %f(126+i*7), [%r140+(128+i*128)]; fma.rn.f32 %f(127+i*7), %f(126+i*7), %f114, %f(229+i*-6); fma.rn.f32 %f(128+i*7), %f(126+i*7), %f116, %f(228+i*-6); fma.rn.f32 %f(129+i*7), %f(126+i*7), %f118, %f(227+i*-6); fma.rn.f32 %f(130+i*7), %f(126+i*7), %f120, %f(226+i*-6); fma.rn.f32 %f(131+i*7), %f(126+i*7), %f122, %f(225+i*-6); fma.rn.f32 %f139, %f133, %f124, %f218; ld.shared.f32 %f140, [%r140+384]; for.size.1 i in range(0, 6, 1): fma.rn.f32 %f(141+i*1), %f140, %f(114+i*2), %f(217+i*-1); for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(147+i*1), [%r(140+i*-3)+4]; for.size.2 i in range(0, 5, 1): fma.rn.f32 %f(235+i*-1), %f147, %f(148+i*1), %f(115+i*2); ld.shared.f32 %f(149+i*1), [%r137+(132+i*128)]; for.size.7 i in range(0, 2, 1): fma.rn.f32 %f(230+i*-6), %f(147+i*7), %f153, %f(125+i*7); ld.shared.f32 %f(154+i*1), [%r140+(132+i*128)]; fma.rn.f32 %f(229+i*-6), %f(154+i*1), %f148, %f(127+i*7); fma.rn.f32 %f(228+i*-6), %f(154+i*1), %f149, %f(128+i*7); fma.rn.f32 %f(227+i*-6), %f(154+i*1), %f150, %f(129+i*7); fma.rn.f32 %f(226+i*-6), %f(154+i*1), %f151, %f(130+i*7); fma.rn.f32 %f(225+i*-6), %f(154+i*1), %f152, %f(131+i*7); fma.rn.f32 %f218, %f155, %f153, %f139; ld.shared.f32 %f156, [%r140+388]; for.size.1 i in range(0, 6, 1): fma.rn.f32 %f(217+i*-1), %f156, %f(148+i*1), %f(141+i*1); add.s32 %r151, %r151, 2; setp.ne.s32 %p3, %r151, 32; @%p3 bra $L__BB0_6; add.s32 %r146, %r146, 1; setp.lt.u32 %p4, %r146, 8; @%p4 bra $L__BB0_1; mul.lo.s32 %r141, %r6, 6; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(142+i*1), %r(4+i*2), (12+i*1002), %r(141+i*-139); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(144+i*1), %r(1+i*3), (32448+i*-30420), %r(143+i*1); mul.wide.s32 %rd41, %r142, 4; add.s64 %rd42, %rd4, %rd41; ld.global.nc.f32 %f157, [%rd42]; add.f32 %f158, %f235, %f157; mov.f32 %f159, 0f00000000; max.f32 %f160, %f158, %f159; mul.wide.s32 %rd43, %r145, 4; add.s64 %rd44, %rd2, %rd43; for.size.4 i in range(0, 5, 1): st.global.f32 [%rd44+(0+i*676)], %f(160+i*3); ld.global.nc.f32 %f(161+i*3), [%rd42+(4+i*4)]; add.f32 %f(162+i*3), %f(234+i*-1), %f(161+i*3); max.f32 %f(163+i*3), %f(162+i*3), %f159; for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd44+(3380+i*32448+j*29068)], %f(175+i*12+j*2); add.f32 %f(176+i*12+j*2), %f(229+i*-6+j*-1), %f(157+j*4); max.f32 %f(177+i*12+j*2), %f(176+i*12+j*2), %f159; for.size.3 j in range(0, 4, 1): st.global.f32 [%rd44+(33124+i*32448+j*676)], %f(179+i*12+j*2); add.f32 %f(180+i*12+j*2), %f(227+i*-6+j*-1), %f(164+j*3); max.f32 %f(181+i*12+j*2), %f(180+i*12+j*2), %f159; st.global.f32 [%rd44+100724], %f211; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[24]; __shared__ float pad_temp_shared[128]; __shared__ float placeholder_shared[384]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 4; ++nn_outer_inner_init) { int index0 = nn_outer_inner_init * 6; compute[index0] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 64; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2 + (int)threadIdx.x; int index2 = (int)blockIdx.x / 676 * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2 + (int)threadIdx.x >> 5) * 43264 + rc_outer_outer * 5408 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2 + (int)threadIdx.x & 31) * 169 + (int)blockIdx.x % 169; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 192; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x; int index4 = (int)blockIdx.x % 676 / 169 * 3072 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x >> 5) * 256 + rc_outer_outer * 32 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 2 + (int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 4; ++nn_outer_inner) { int index5 = nn_outer_inner * 6; int index6 = nn_outer_inner * 32 + rc_outer_inner; int index7 = (int)threadIdx.x * 192 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 1] = compute[index5 + 1] + (pad_temp_shared[index6] * placeholder_shared[index7 + 32]); compute[index5 + 2] = compute[index5 + 2] + (pad_temp_shared[index6] * placeholder_shared[index7 + 64]); compute[index5 + 3] = compute[index5 + 3] + (pad_temp_shared[index6] * placeholder_shared[index7 + 96]); compute[index5 + 4] = compute[index5 + 4] + (pad_temp_shared[index6] * placeholder_shared[index7 + 128]); compute[index5 + 5] = compute[index5 + 5] + (pad_temp_shared[index6] * placeholder_shared[index7 + 160]); } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 6; ++ax1_inner) { int index8 = (int)blockIdx.x / 676 * 32448 + ax0_inner * 8112 + (int)blockIdx.x % 676 / 169 * 2028 + (int)threadIdx.x * 1014 + ax1_inner * 169 + (int)blockIdx.x % 169; int index9 = ax0_inner * 6 + ax1_inner; int index10 = (int)blockIdx.x % 676 / 169 * 12 + (int)threadIdx.x * 6 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpup2bnw3u
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<17>; .reg .f32 %f<174>; .reg .b32 %r<110>; .reg .b64 %rd<29>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd6; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(36+i*3), %r1, (1626496491+i*-304968092); shr.u32 %r(37+i*3), %r(36+i*3), 31; shr.s32 %r(38+i*3), %r(36+i*3), (6+i*-4); add.s32 %r(2+i*1), %r(38+i*3), %r(37+i*3); mov.u32 %r4, %ctaid.x; mul.lo.s32 %r42, %r2, 169; sub.s32 %r5, %r1, %r42; mad.lo.s32 %r6, %r4, 173056, %r5; shr.u32 %r43, %r1, 7; and.b32 %r44, %r1, 127; prmt.b32 %r7, %r43, %r44, 8452; shl.b32 %r45, %r1, 2; mov.u32 %r46, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r8, %r46, %r45; for.size.5 i in range(0, 2, 1): add.s32 %r(47+i*4), %r1, (2028+i*2028); shr.u32 %r(48+i*4), %r(47+i*4), 7; add.s32 %r(49+i*4), %r1, (108+i*-20); and.b32 %r(50+i*4), %r(49+i*4), 127; prmt.b32 %r(9+i*1), %r(48+i*4), %r(50+i*4), 8452; shl.b32 %r55, %r1, 1; add.s32 %r56, %r55, 12168; and.b32 %r57, %r56, -256; add.s32 %r58, %r1, %r57; add.s32 %r11, %r58, 68; shl.b32 %r12, %r2, 9; for.size.1 i in range(0, 2, 1): add.s32 %r(13+i*1), %r2, (36+i*-24); mov.u32 %r59, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r60, %r59, %r45; add.s32 %r15, %r60, 16224; mov.u32 %r35, 0; mov.f32 %f158, 0f00000000; for.size.1 i in range(0, 15, 1): mov.f32 %f(159+i*1), %f158; mov.u32 %r103, %r35; $L__BB0_1: bar.sync 0; mad.lo.s32 %r18, %r103, 21632, %r6; mov.u32 %r104, %r1; mov.u32 %r105, %r15; mov.u32 %r106, %r35; mov.u32 %r107, %r3; mov.u32 %r108, %r13; bra.uni $L__BB0_2; $L__BB0_19: for.size.1 i in range(0, 2, 1): add.s32 %r(108+i*-1), %r(108+i*-1), (48+i*576); for.size.1 i in range(0, 2, 1): add.s32 %r(106+i*-1), %r(106+i*-1), (48+i*32400); add.s32 %r104, %r104, 8112; $L__BB0_2: add.s32 %r62, %r108, -36; setp.gt.s32 %p1, %r62, 511; @%p1 bra $L__BB0_6; setp.gt.s32 %p2, %r107, 6655; @%p2 bra $L__BB0_6; setp.gt.s32 %p3, %r104, 86527; @%p3 bra $L__BB0_6; shr.s32 %r64, %r62, 7; and.b32 %r65, %r62, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(66+i*1), %r(64+i*1), (43264+i*-43095), %r(18+i*48); mul.wide.s32 %rd9, %r67, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f65, [%rd10]; st.shared.f32 [%r105+-16224], %f65; $L__BB0_6: add.s32 %r24, %r14, %r106; setp.gt.s32 %p4, %r24, 511; @%p4 bra $L__BB0_10; add.s32 %r68, %r107, 156; setp.gt.s32 %p5, %r68, 6655; @%p5 bra $L__BB0_10; add.s32 %r69, %r104, 2028; setp.gt.s32 %p6, %r69, 86527; @%p6 bra $L__BB0_10; shr.s32 %r70, %r24, 7; and.b32 %r71, %r24, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(72+i*1), %r(70+i*1), (43264+i*-43095), %r(18+i*54); mul.wide.s32 %rd11, %r73, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f66, [%rd12]; st.shared.f32 [%r105+-8112], %f66; $L__BB0_10: add.s32 %r74, %r2, %r106; add.s32 %r25, %r74, 24; setp.gt.s32 %p7, %r25, 511; @%p7 bra $L__BB0_14; add.s32 %r75, %r107, 312; setp.gt.s32 %p8, %r75, 6655; @%p8 bra $L__BB0_14; add.s32 %r76, %r104, 4056; setp.gt.s32 %p9, %r76, 86527; @%p9 bra $L__BB0_14; shr.s32 %r77, %r25, 7; and.b32 %r78, %r25, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(79+i*1), %r(77+i*1), (43264+i*-43095), %r(18+i*61); mul.wide.s32 %rd13, %r80, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f67, [%rd14]; st.shared.f32 [%r105], %f67; $L__BB0_14: setp.eq.s32 %p10, %r106, 480; @%p10 bra $L__BB0_20; setp.gt.s32 %p11, %r108, 511; @%p11 bra $L__BB0_19; add.s32 %r81, %r107, 468; setp.gt.s32 %p12, %r81, 6655; @%p12 bra $L__BB0_19; add.s32 %r82, %r104, 6084; setp.gt.s32 %p13, %r82, 86527; @%p13 bra $L__BB0_19; shr.s32 %r83, %r108, 7; and.b32 %r84, %r108, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(85+i*1), %r(83+i*1), (43264+i*-43095), %r(18+i*67); mul.wide.s32 %rd15, %r86, 4; add.s64 %rd16, %rd1, %rd15; ld.global.nc.f32 %f68, [%rd16]; st.shared.f32 [%r105+8112], %f68; bra.uni $L__BB0_19; $L__BB0_20: setp.gt.s32 %p14, %r1, 59; shl.b32 %r31, %r103, 7; for.size.5 i in range(0, 2, 1): add.s32 %r(87+i*1), %r(7+i*2), %r31; mul.wide.s32 %rd(17+i*2), %r(87+i*1), 4; add.s64 %rd(18+i*2), %rd2, %rd(17+i*2); ld.global.nc.f32 %f(69+i*1), [%rd(18+i*2)+0]; st.shared.f32 [%r8+(0+i*8112)], %f(69+i*1); add.s32 %r89, %r10, %r31; mul.wide.s32 %rd21, %r89, 4; add.s64 %rd22, %rd2, %rd21; ld.global.nc.f32 %f71, [%rd22]; st.shared.f32 [%r8+16224], %f71; @%p14 bra $L__BB0_22; add.s32 %r90, %r11, %r31; mul.wide.s32 %rd23, %r90, 4; add.s64 %rd24, %rd2, %rd23; ld.global.nc.f32 %f72, [%rd24]; st.shared.f32 [%r8+24336], %f72; $L__BB0_22: bar.sync 0; mov.u32 %r109, 0; $L__BB0_23: mad.lo.s32 %r92, %r109, 169, %r5; shl.b32 %r93, %r92, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(95+i*1), %r(59+i*50), %r(93+i*-81); shl.b32 %r97, %r96, 2; add.s32 %r99, %r46, %r97; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(73+i*1), [%r(99+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(75+i*2), %f74, %f(73+i*3), %f(173+i*-1); ld.shared.f32 %f(76+i*2), [%r99+(512+i*512)]; fma.rn.f32 %f79, %f74, %f78, %f171; ld.shared.f32 %f80, [%r99+1536]; for.size.5 i in range(0, 2, 1): fma.rn.f32 %f(81+i*5), %f(74+i*8), %f80, %f(170+i*-4); ld.shared.f32 %f(82+i*5), [%r95+(86528+i*86528)]; fma.rn.f32 %f(83+i*5), %f(82+i*5), %f73, %f(169+i*-4); fma.rn.f32 %f(84+i*5), %f(82+i*5), %f76, %f(168+i*-4); fma.rn.f32 %f(85+i*5), %f(82+i*5), %f78, %f(167+i*-4); fma.rn.f32 %f91, %f87, %f80, %f162; ld.shared.f32 %f92, [%r95+259584]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(93+i*1), %f92, %f(73+i*3), %f(161+i*-1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(95+i*1), %f92, %f(78+i*2), %f(159+i*-1); ld.shared.f32 %f97, [%r99+4]; ld.shared.f32 %f98, [%r95+676]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(173+i*-1), %f98, %f(97+i*2), %f(75+i*2); ld.shared.f32 %f(99+i*1), [%r99+(516+i*512)]; fma.rn.f32 %f171, %f98, %f100, %f79; ld.shared.f32 %f101, [%r99+1540]; for.size.5 i in range(0, 2, 1): fma.rn.f32 %f(170+i*-4), %f(98+i*4), %f101, %f(81+i*5); ld.shared.f32 %f(102+i*1), [%r95+(87204+i*86528)]; fma.rn.f32 %f(169+i*-4), %f(102+i*1), %f97, %f(83+i*5); fma.rn.f32 %f(168+i*-4), %f(102+i*1), %f99, %f(84+i*5); fma.rn.f32 %f(167+i*-4), %f(102+i*1), %f100, %f(85+i*5); fma.rn.f32 %f162, %f103, %f101, %f91; ld.shared.f32 %f104, [%r95+260260]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(161+i*-1), %f104, %f(97+i*2), %f(93+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(159+i*-1), %f104, %f(100+i*1), %f(95+i*1); add.s32 %r109, %r109, 2; setp.ne.s32 %p15, %r109, 128; @%p15 bra $L__BB0_23; add.s32 %r103, %r103, 1; setp.lt.u32 %p16, %r103, 2; @%p16 bra $L__BB0_1; shl.b32 %r100, %r2, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(101+i*1), %r(4+i*-2), (32448+i*-31772), %r(5+i*96); mul.wide.s32 %rd25, %r100, 4; add.s64 %rd26, %rd4, %rd25; ld.global.nc.f32 %f105, [%rd26]; add.f32 %f106, %f173, %f105; mov.f32 %f107, 0f00000000; max.f32 %f108, %f106, %f107; mul.wide.s32 %rd27, %r102, 4; add.s64 %rd28, %rd3, %rd27; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd28+(0+i*676)], %f(108+i*3); ld.global.nc.f32 %f(109+i*3), [%rd26+(4+i*4)]; add.f32 %f(110+i*3), %f(172+i*-1), %f(109+i*3); max.f32 %f(111+i*3), %f(110+i*3), %f107; for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd28+(2028+i*32448+j*30420)], %f(117+i*8+j*2); add.f32 %f(118+i*8+j*2), %f(169+i*-4+j*-1), %f(105+j*4); max.f32 %f(119+i*8+j*2), %f(118+i*8+j*2), %f107; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd28+(33124+i*32448+j*676)], %f(121+i*8+j*2); add.f32 %f(122+i*8+j*2), %f(167+i*-4+j*-1), %f(112+j*3); max.f32 %f(123+i*8+j*2), %f(122+i*8+j*2), %f107; st.global.f32 [%rd28+99372], %f141; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[6144]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[15] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 43; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 169 < 512) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 156 + (int)threadIdx.x / 13 < 6656) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2028 + (int)threadIdx.x < 86528) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2028 + (int)threadIdx.x; int index1 = (int)blockIdx.x * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 169 >> 7) * 43264 + rc_outer_outer * 21632 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 12 + (int)threadIdx.x / 169 & 127) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } } } } int index2 = (int)threadIdx.x; int index3 = ((int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x & 127); placeholder_shared[index2] = placeholder1[index3]; int index4 = ((int)threadIdx.x + 2028 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 108 & 127); placeholder_shared[index2 + 2028] = placeholder1[index4]; int index5 = ((int)threadIdx.x + 4056 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 88 & 127); placeholder_shared[index2 + 4056] = placeholder1[index5]; if ((int)threadIdx.x < 60) { int index6 = ((int)threadIdx.x + 6084 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 68); placeholder_shared[index2 + 6084] = placeholder1[index6]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index7 = rc_outer_inner * 169 + (int)threadIdx.x % 169; int index8 = (int)threadIdx.x / 169 * 512 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index7] * placeholder_shared[index8]; compute[1] = compute[1] + (pad_temp_shared[index7] * placeholder_shared[index8 + 128]); compute[2] = compute[2] + (pad_temp_shared[index7] * placeholder_shared[index8 + 256]); compute[3] = compute[3] + (pad_temp_shared[index7] * placeholder_shared[index8 + 384]); compute[4] = compute[4] + (pad_temp_shared[index7 + 21632] * placeholder_shared[index8]); compute[5] = compute[5] + (pad_temp_shared[index7 + 21632] * placeholder_shared[index8 + 128]); compute[6] = compute[6] + (pad_temp_shared[index7 + 21632] * placeholder_shared[index8 + 256]); compute[7] = compute[7] + (pad_temp_shared[index7 + 21632] * placeholder_shared[index8 + 384]); compute[8] = compute[8] + (pad_temp_shared[index7 + 43264] * placeholder_shared[index8]); compute[9] = compute[9] + (pad_temp_shared[index7 + 43264] * placeholder_shared[index8 + 128]); compute[10] = compute[10] + (pad_temp_shared[index7 + 43264] * placeholder_shared[index8 + 256]); compute[11] = compute[11] + (pad_temp_shared[index7 + 43264] * placeholder_shared[index8 + 384]); compute[12] = compute[12] + (pad_temp_shared[index7 + 64896] * placeholder_shared[index8]); compute[13] = compute[13] + (pad_temp_shared[index7 + 64896] * placeholder_shared[index8 + 128]); compute[14] = compute[14] + (pad_temp_shared[index7 + 64896] * placeholder_shared[index8 + 256]); compute[15] = compute[15] + (pad_temp_shared[index7 + 64896] * placeholder_shared[index8 + 384]); } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index9 = (int)blockIdx.x * 32448 + ax0_inner * 8112 + (int)threadIdx.x / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index10 = ax0_inner * 4 + ax1_inner; int index11 = (int)threadIdx.x / 169 * 4 + ax1_inner; T_relu[index9] = max(compute[index10] + placeholder2[index11], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpc03_279g
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<108>; .reg .b32 %r<129>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %ctaid.x; mul.lo.s32 %r32, %r1, 173056; mov.u32 %r2, %tid.x; add.s32 %r3, %r32, %r2; shl.b32 %r33, %r2, 2; mov.u32 %r34, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r34, %r33; add.s32 %r35, %r2, 16224; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(36+i*4), %r(35+i*-33), 1626496491; shr.u32 %r(37+i*4), %r(36+i*4), 31; shr.s32 %r(38+i*4), %r(36+i*4), (13+i*-7); add.s32 %r(39+i*4), %r(38+i*4), %r(37+i*4); add.s32 %r44, %r43, 96; and.b32 %r45, %r44, 127; mul.lo.s32 %r46, %r43, 169; sub.s32 %r5, %r2, %r46; add.s32 %r47, %r5, %r32; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(48+i*-42), %r(39+i*6), (43264+i*-43095), %r(47+i*1); for.size.7 i in range(0, 2, 1): add.s32 %r(49+i*6), %r2, (24336+i*8112); mul.hi.s32 %r(50+i*6), %r(49+i*6), 1626496491; shr.u32 %r(51+i*6), %r(50+i*6), 31; shr.s32 %r(52+i*6), %r(50+i*6), 13; add.s32 %r(53+i*6), %r(52+i*6), %r(51+i*6); mad.lo.s32 %r(54+i*6), %r(53+i*6), 43264, %r3; add.s32 %r(7+i*1), %r(54+i*6), (2704+i*8112); add.s32 %r61, %r2, 40560; mul.hi.s32 %r62, %r61, 1626496491; shr.u32 %r63, %r62, 31; shr.s32 %r64, %r62, 13; add.s32 %r65, %r64, %r63; add.s32 %r66, %r43, 112; and.b32 %r67, %r66, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(68+i*-59), %r(65+i*2), (43264+i*-43095), %r(47+i*21); for.size.8 i in range(0, 2, 1): for.size.7 j in range(0, 2, 1): add.s32 %r(69+i*12+j*6), %r2, (48672+i*24336+j*8112); mul.hi.s32 %r(70+i*12+j*6), %r(69+i*12+j*6), 1626496491; shr.u32 %r(71+i*12+j*6), %r(70+i*12+j*6), 31; shr.s32 %r(72+i*12+j*6), %r(70+i*12+j*6), 13; add.s32 %r(73+i*12+j*6), %r(72+i*12+j*6), %r(71+i*12+j*6); mad.lo.s32 %r(74+i*12+j*6), %r(73+i*12+j*6), 43264, %r3; add.s32 %r(10+i*2+j*1), %r(74+i*12+j*6), (5408+i*2704+j*8112); shr.u32 %r93, %r2, 7; and.b32 %r94, %r2, 127; prmt.b32 %r14, %r93, %r94, 8452; mov.u32 %r95, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r15, %r95, %r33; mul.hi.s32 %r96, %r2, 542165497; shr.u32 %r97, %r96, 31; shr.s32 %r98, %r96, 9; add.s32 %r16, %r98, %r97; mul.lo.s32 %r99, %r16, 4056; sub.s32 %r100, %r2, %r99; mul.hi.s32 %r101, %r100, 1626496491; shr.u32 %r102, %r101, 31; shr.s32 %r103, %r101, 6; add.s32 %r17, %r103, %r102; shl.b32 %r104, %r17, 10; add.s32 %r105, %r95, %r104; add.s32 %r18, %r105, 512; mad.lo.s32 %r106, %r16, 43264, %r5; shl.b32 %r107, %r106, 2; add.s32 %r108, %r34, %r107; add.s32 %r19, %r108, 86528; mov.u32 %r125, 0; mov.f32 %f104, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(105+i*1), %f104; $L__BB0_1: bar.sync 0; mul.lo.s32 %r21, %r125, 21632; add.s32 %r109, %r3, %r21; mul.wide.s32 %rd9, %r109, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(17+i*1), [%rd10+(0+i*32448)]; st.shared.f32 [%r4+(0+i*32448)], %f(17+i*1); for.size.5 i in range(0, 6, 1): add.s32 %r(110+i*1), %r(6+i*1), %r21; mul.wide.s32 %rd(11+i*2), %r(110+i*1), 4; add.s64 %rd(12+i*2), %rd2, %rd(11+i*2); ld.global.nc.f32 %f(19+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r4+(64896+i*32448)], %f(19+i*1); ld.global.nc.f32 %f25, [%rd10+519168]; st.shared.f32 [%r4+259584], %f25; add.s32 %r116, %r12, %r21; mul.wide.s32 %rd23, %r116, 4; add.s64 %rd24, %rd2, %rd23; ld.global.nc.f32 %f26, [%rd24]; st.shared.f32 [%r4+292032], %f26; setp.gt.s32 %p1, %r2, 5407; @%p1 bra $L__BB0_3; add.s32 %r117, %r13, %r21; mul.wide.s32 %rd25, %r117, 4; add.s64 %rd26, %rd2, %rd25; ld.global.nc.f32 %f27, [%rd26]; st.shared.f32 [%r4+324480], %f27; $L__BB0_3: setp.gt.s32 %p2, %r2, 6143; @%p2 bra $L__BB0_5; shl.b32 %r118, %r125, 7; add.s32 %r119, %r14, %r118; mul.wide.s32 %rd27, %r119, 4; add.s64 %rd28, %rd1, %rd27; ld.global.nc.f32 %f28, [%rd28]; st.shared.f32 [%r15], %f28; $L__BB0_5: bar.sync 0; mov.u32 %r128, 512; mov.u32 %r126, %r19; mov.u32 %r127, %r18; $L__BB0_6: for.size.8 i in range(0, 2, 1): ld.shared.f32 %f(29+i*8), [%r127+(-512+i*4)]; ld.shared.f32 %f(30+i*8), [%r126+(-86528+i*676)]; fma.rn.f32 %f(31+i*8), %f(30+i*8), %f(29+i*8), %f(107+i*-76); ld.shared.f32 %f(32+i*8), [%r127+(0+i*4)]; fma.rn.f32 %f(33+i*8), %f(30+i*8), %f(32+i*8), %f(106+i*-73); ld.shared.f32 %f(34+i*8), [%r126+(0+i*676)]; fma.rn.f32 %f(35+i*8), %f(34+i*8), %f(29+i*8), %f(105+i*-70); fma.rn.f32 %f(36+i*8), %f(34+i*8), %f(32+i*8), %f(104+i*-68); for.size.8 i in range(0, 5, 1): ld.shared.f32 %f(45+i*8), [%r127+(-504+i*4)]; ld.shared.f32 %f(46+i*8), [%r126+(-85176+i*676)]; fma.rn.f32 %f(47+i*8), %f(46+i*8), %f(45+i*8), %f(39+i*8); ld.shared.f32 %f(48+i*8), [%r127+(8+i*4)]; fma.rn.f32 %f(49+i*8), %f(46+i*8), %f(48+i*8), %f(41+i*8); ld.shared.f32 %f(50+i*8), [%r126+(1352+i*676)]; fma.rn.f32 %f(51+i*8), %f(50+i*8), %f(45+i*8), %f(43+i*8); fma.rn.f32 %f(52+i*8), %f(50+i*8), %f(48+i*8), %f(44+i*8); ld.shared.f32 %f85, [%r127+-484]; ld.shared.f32 %f86, [%r126+-81796]; fma.rn.f32 %f107, %f86, %f85, %f79; ld.shared.f32 %f87, [%r127+28]; fma.rn.f32 %f106, %f86, %f87, %f81; ld.shared.f32 %f88, [%r126+4732]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(105+i*-1), %f88, %f(85+i*2), %f(83+i*1); for.size.1 i in range(0, 2, 1): add.s32 %r(127+i*-1), %r(127+i*-1), (32+i*5376); add.s32 %r128, %r128, 32; setp.ne.s32 %p3, %r128, 1024; @%p3 bra $L__BB0_6; add.s32 %r125, %r125, 1; setp.lt.u32 %p4, %r125, 2; @%p4 bra $L__BB0_1; shl.b32 %r121, %r17, 1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(122+i*1), %r(1+i*15), (32448+i*-16224), %r(5+i*117); mad.lo.s32 %r124, %r17, 338, %r123; mul.wide.s32 %rd29, %r121, 4; add.s64 %rd30, %rd4, %rd29; ld.global.nc.f32 %f89, [%rd30]; add.f32 %f90, %f107, %f89; mov.f32 %f91, 0f00000000; max.f32 %f92, %f90, %f91; mul.wide.s32 %rd31, %r124, 4; add.s64 %rd32, %rd3, %rd31; st.global.f32 [%rd32], %f92; ld.global.nc.f32 %f93, [%rd30+4]; for.size.3 i in range(0, 2, 1): add.f32 %f(94+i*2), %f(106+i*-1), %f(93+i*-4); max.f32 %f(95+i*2), %f(94+i*2), %f91; st.global.f32 [%rd32+(676+i*31772)], %f(95+i*2); add.f32 %f98, %f104, %f93; max.f32 %f99, %f98, %f91; st.global.f32 [%rd32+33124], %f99; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[6144]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x * 173056 + rc_outer_outer * 21632 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 8112] = placeholder[index1 + 8112]; int index2 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 16224) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 96 & 127) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 16224] = placeholder[index2]; int index3 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 24336) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 24336] = placeholder[index3]; int index4 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 32448) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 64) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 32448] = placeholder[index4]; int index5 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 40560) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 112 & 127) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 40560] = placeholder[index5]; int index6 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 48672) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 32) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 48672] = placeholder[index6]; int index7 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 56784) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 80) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 56784] = placeholder[index7]; pad_temp_shared[index0 + 64896] = placeholder[index1 + 129792]; int index8 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 73008) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 48) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 73008] = placeholder[index8]; if ((int)threadIdx.x < 5408) { int index9 = (int)blockIdx.x * 173056 + ((int)threadIdx.x + 81120) / 21632 * 43264 + rc_outer_outer * 21632 + ((int)threadIdx.x / 169 + 96) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0 + 81120] = placeholder[index9]; } if ((int)threadIdx.x < 6144) { int index10 = ((int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x & 127); placeholder_shared[index0] = placeholder1[index10]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index11 = (int)threadIdx.x / 4056 * 43264 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index12 = (int)threadIdx.x % 4056 / 169 * 256 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index11] * placeholder_shared[index12]; compute[1] = compute[1] + (pad_temp_shared[index11] * placeholder_shared[index12 + 128]); compute[2] = compute[2] + (pad_temp_shared[index11 + 21632] * placeholder_shared[index12]); compute[3] = compute[3] + (pad_temp_shared[index11 + 21632] * placeholder_shared[index12 + 128]); } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index13 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 4056 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 4056 / 169 * 338 + ax1_inner * 169 + (int)threadIdx.x % 169; int index14 = ax0_inner * 2 + ax1_inner; int index15 = (int)threadIdx.x % 4056 / 169 * 2 + ax1_inner; T_relu[index13] = max(compute[index14] + placeholder2[index15], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpy36qm1a6
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<99>; .reg .b32 %r<45>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1664]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[768]; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd11, [default_function_kernel0_param_1]; ld.param.u64 %rd12, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd13, %rd11; cvta.to.global.u64 %rd1, %rd12; mov.u32 %r1, %tid.x; mul.hi.s32 %r13, %r1, 1321528399; shr.u32 %r14, %r13, 31; shr.s32 %r15, %r13, 4; add.s32 %r16, %r15, %r14; mul.lo.s32 %r17, %r16, 52; sub.s32 %r18, %r1, %r17; shl.b32 %r19, %r1, 2; mov.u32 %r20, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r20, %r19; shr.u32 %r21, %r1, 2; and.b32 %r22, %r1, 3; mov.u32 %r23, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r23, %r19; mul.hi.s32 %r24, %r1, -770891565; add.s32 %r25, %r24, %r1; shr.u32 %r26, %r25, 31; for.size.2 i in range(0, 2, 1): shr.s32 %r(27+i*1), %r(25+i*-12), (9+i*-7); add.s32 %r(4+i*25), %r(27+i*1), %r(26+i*-12); mul.lo.s32 %r30, %r29, 13; sub.s32 %r31, %r1, %r30; mad.lo.s32 %r32, %r4, 208, %r31; shl.b32 %r33, %r32, 2; add.s32 %r5, %r20, %r33; mul.lo.s32 %r34, %r4, 624; sub.s32 %r6, %r1, %r34; mul.hi.s32 %r35, %r6, 1321528399; shr.u32 %r36, %r35, 31; shr.s32 %r37, %r35, 2; add.s32 %r7, %r37, %r36; shl.b32 %r38, %r7, 4; add.s32 %r8, %r23, %r38; prmt.b32 %r39, %r21, %r22, 8452; mul.wide.s32 %rd14, %r39, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd24, %rd15, 16; mov.u32 %r9, %ctaid.x; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(40+i*1), %r(18+i*-2), (13+i*43251), %r(9+i*31); cvta.to.global.u64 %rd16, %rd10; mul.wide.s32 %rd17, %r41, 4; add.s64 %rd23, %rd16, %rd17; mov.u32 %r44, 0; mov.f32 %f95, 0f00000000; for.size.1 i in range(0, 3, 1): mov.f32 %f(96+i*1), %f95; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 415; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f17, [%rd23]; st.shared.f32 [%r2], %f17; $L__BB0_3: setp.gt.s32 %p2, %r1, 191; add.s64 %rd6, %rd24, -16; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f18, [%rd6]; st.shared.f32 [%r3], %f18; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(19+i*1), [%r(8+i*-3)+0]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(21+i*2), %f(20+i*2), %f19, %f(98+i*-1); ld.shared.f32 %f(22+i*2), [%r5+(208+i*208)]; for.size.9 i in range(0, 2, 1): fma.rn.f32 %f(27+i*9), %f(26+i*9), %f(19+i*9), %f(95+i*-68); ld.shared.f32 %f(28+i*9), [%r8+(4+i*4)]; ld.shared.f32 %f(29+i*9), [%r5+(52+i*52)]; fma.rn.f32 %f(30+i*9), %f(29+i*9), %f(28+i*9), %f(21+i*9); ld.shared.f32 %f(31+i*9), [%r5+(260+i*52)]; fma.rn.f32 %f(32+i*9), %f(31+i*9), %f(28+i*9), %f(23+i*9); ld.shared.f32 %f(33+i*9), [%r5+(468+i*52)]; fma.rn.f32 %f(34+i*9), %f(33+i*9), %f(28+i*9), %f(25+i*9); ld.shared.f32 %f(35+i*9), [%r5+(676+i*52)]; fma.rn.f32 %f45, %f44, %f37, %f36; ld.shared.f32 %f46, [%r8+12]; for.size.2 i in range(0, 4, 1): ld.shared.f32 %f(47+i*1), [%r5+(156+i*208)]; fma.rn.f32 %f(5+i*1), %f(47+i*1), %f46, %f(39+i*2); bar.sync 0; @%p1 bra $L__BB0_7; ld.global.nc.f32 %f51, [%rd23+2704]; st.shared.f32 [%r2], %f51; $L__BB0_7: @%p2 bra $L__BB0_9; ld.global.nc.f32 %f52, [%rd6+16]; st.shared.f32 [%r3], %f52; $L__BB0_9: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(53+i*1), [%r(8+i*-3)+0]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(55+i*2), %f(54+i*2), %f53, %f(5+i*1); ld.shared.f32 %f(56+i*2), [%r5+(208+i*208)]; for.size.9 i in range(0, 2, 1): fma.rn.f32 %f(61+i*9), %f(60+i*9), %f(53+i*9), %f(8+i*53); ld.shared.f32 %f(62+i*9), [%r8+(4+i*4)]; ld.shared.f32 %f(63+i*9), [%r5+(52+i*52)]; fma.rn.f32 %f(64+i*9), %f(63+i*9), %f(62+i*9), %f(55+i*9); ld.shared.f32 %f(65+i*9), [%r5+(260+i*52)]; fma.rn.f32 %f(66+i*9), %f(65+i*9), %f(62+i*9), %f(57+i*9); ld.shared.f32 %f(67+i*9), [%r5+(468+i*52)]; fma.rn.f32 %f(68+i*9), %f(67+i*9), %f(62+i*9), %f(59+i*9); ld.shared.f32 %f(69+i*9), [%r5+(676+i*52)]; fma.rn.f32 %f79, %f78, %f71, %f70; ld.shared.f32 %f80, [%r8+12]; for.size.2 i in range(0, 4, 1): ld.shared.f32 %f(81+i*1), [%r5+(156+i*208)]; fma.rn.f32 %f(98+i*-1), %f(81+i*1), %f80, %f(73+i*2); for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (32+i*5376); add.s32 %r44, %r44, 2; setp.ne.s32 %p5, %r44, 64; @%p5 bra $L__BB0_1; cvta.to.global.u64 %rd18, %rd9; mul.wide.s32 %rd19, %r7, 4; add.s64 %rd20, %rd18, %rd19; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(42+i*1), %r(4+i*2), (32448+i*-32435), %r(9+i*33); ld.global.nc.f32 %f85, [%rd20]; add.f32 %f86, %f98, %f85; mov.f32 %f87, 0f00000000; max.f32 %f88, %f86, %f87; mul.wide.s32 %rd21, %r43, 4; add.s64 %rd22, %rd1, %rd21; for.size.3 i in range(0, 3, 1): st.global.f32 [%rd22+(0+i*32448)], %f(88+i*2); add.f32 %f(89+i*2), %f(97+i*-1), %f85; max.f32 %f(90+i*2), %f(89+i*2), %f87; st.global.f32 [%rd22+97344], %f94; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[4]; __shared__ float pad_temp_shared[416]; __shared__ float placeholder_shared[192]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 416) { int index0 = (int)threadIdx.x; int index1 = (int)threadIdx.x / 52 * 43264 + rc_outer_outer * 676 + (int)threadIdx.x % 52 * 13 + (int)blockIdx.x; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 192) { int index0 = (int)threadIdx.x; int index2 = ((int)threadIdx.x >> 2) * 256 + rc_outer_outer * 4 + ((int)threadIdx.x & 3); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index3 = (int)threadIdx.x / 624 * 208 + (int)threadIdx.x % 13; int index4 = (int)threadIdx.x % 624 / 13 * 4; compute[0] = compute[0] + pad_temp_shared[index3] * placeholder_shared[index4]; compute[1] = compute[1] + (pad_temp_shared[index3 + 52] * placeholder_shared[index4]); compute[2] = compute[2] + (pad_temp_shared[index3 + 104] * placeholder_shared[index4]); compute[3] = compute[3] + (pad_temp_shared[index3 + 156] * placeholder_shared[index4]); compute[0] = compute[0] + (pad_temp_shared[index3 + 13] * placeholder_shared[index4 + 1]); compute[1] = compute[1] + (pad_temp_shared[index3 + 65] * placeholder_shared[index4 + 1]); compute[2] = compute[2] + (pad_temp_shared[index3 + 117] * placeholder_shared[index4 + 1]); compute[3] = compute[3] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 1]); compute[0] = compute[0] + (pad_temp_shared[index3 + 26] * placeholder_shared[index4 + 2]); compute[1] = compute[1] + (pad_temp_shared[index3 + 78] * placeholder_shared[index4 + 2]); compute[2] = compute[2] + (pad_temp_shared[index3 + 130] * placeholder_shared[index4 + 2]); compute[3] = compute[3] + (pad_temp_shared[index3 + 182] * placeholder_shared[index4 + 2]); compute[0] = compute[0] + (pad_temp_shared[index3 + 39] * placeholder_shared[index4 + 3]); compute[1] = compute[1] + (pad_temp_shared[index3 + 91] * placeholder_shared[index4 + 3]); compute[2] = compute[2] + (pad_temp_shared[index3 + 143] * placeholder_shared[index4 + 3]); compute[3] = compute[3] + (pad_temp_shared[index3 + 195] * placeholder_shared[index4 + 3]); } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { int index5 = (int)threadIdx.x / 624 * 32448 + ax0_inner * 8112 + (int)threadIdx.x % 624 * 13 + (int)blockIdx.x; int index6 = ax0_inner; int index7 = (int)threadIdx.x % 624 / 13; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp7y_g4rq_
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<16>; .reg .f32 %f<110>; .reg .b32 %r<101>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[692224]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[49152]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; ld.param.u64 %rd6, [default_function_kernel0_param_2]; ld.param.u64 %rd7, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd9, %rd8; mov.u32 %r38, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(39+i*3), %r38, (1626496491+i*-304968092); shr.u32 %r(40+i*3), %r(39+i*3), 31; shr.s32 %r(41+i*3), %r(39+i*3), (6+i*-4); add.s32 %r(1+i*91), %r(41+i*3), %r(40+i*3); for.size.1 i in range(0, 2, 1): add.s32 %r(97+i*-1), %r(92+i*-54), (312+i*11856); add.s32 %r95, %r1, 72; mov.u32 %r45, %ctaid.x; mad.lo.s32 %r46, %r45, 173056, %r38; mul.wide.s32 %rd10, %r46, 4; add.s64 %rd11, %rd9, %rd10; add.s64 %rd20, %rd11, 32448; mov.u32 %r93, 8112; shl.b32 %r47, %r38, 2; mov.u32 %r48, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r49, %r48, %r47; add.s32 %r90, %r49, 32448; mov.u32 %r94, 48; mov.u32 %r91, 0; bra.uni $L__BB0_1; $L__BB0_18: for.size.1 i in range(0, 2, 1): add.s32 %r(97+i*-1), %r(97+i*-1), (1248+i*14976); for.size.1 i in range(0, 2, 1): add.s32 %r(95+i*-1), %r(95+i*-1), 96; for.size.1 i in range(0, 2, 1): add.s32 %r(93+i*-1), %r(93+i*-1), (16224+i*-14976); add.s32 %r91, %r91, 96; add.s64 %rd20, %rd20, 64896; add.s32 %r90, %r90, 64896; $L__BB0_1: add.s32 %r50, %r1, %r91; setp.gt.s32 %p1, %r50, 1023; @%p1 bra $L__BB0_5; setp.gt.s32 %p2, %r92, 13311; @%p2 bra $L__BB0_5; add.s32 %r51, %r96, -12168; setp.gt.s32 %p3, %r51, 173055; @%p3 bra $L__BB0_5; ld.global.nc.f32 %f17, [%rd20+-32448]; st.shared.f32 [%r90+-32448], %f17; $L__BB0_5: add.s32 %r53, %r50, 24; setp.gt.s32 %p4, %r53, 1023; @%p4 bra $L__BB0_9; setp.gt.s32 %p5, %r97, 13311; @%p5 bra $L__BB0_9; add.s32 %r54, %r96, -8112; setp.gt.s32 %p6, %r54, 173055; @%p6 bra $L__BB0_9; ld.global.nc.f32 %f18, [%rd20+-16224]; st.shared.f32 [%r90+-16224], %f18; $L__BB0_9: add.s32 %r55, %r1, %r94; setp.gt.s32 %p7, %r55, 1023; @%p7 bra $L__BB0_13; add.s32 %r56, %r92, 624; setp.gt.s32 %p8, %r56, 13311; @%p8 bra $L__BB0_13; add.s32 %r58, %r38, %r93; setp.gt.s32 %p9, %r58, 173055; @%p9 bra $L__BB0_13; ld.global.nc.f32 %f19, [%rd20]; st.shared.f32 [%r90], %f19; $L__BB0_13: setp.eq.s32 %p10, %r94, 1008; @%p10 bra $L__BB0_19; setp.gt.s32 %p11, %r95, 1023; @%p11 bra $L__BB0_18; add.s32 %r59, %r97, 624; setp.gt.s32 %p12, %r59, 13311; @%p12 bra $L__BB0_18; setp.gt.s32 %p13, %r96, 173055; @%p13 bra $L__BB0_18; ld.global.nc.f32 %f20, [%rd20+16224]; st.shared.f32 [%r90+16224], %f20; bra.uni $L__BB0_18; $L__BB0_19: cvta.to.global.u64 %rd12, %rd5; mul.wide.s32 %rd13, %r38, 4; add.s64 %rd4, %rd12, %rd13; ld.global.nc.f32 %f21, [%rd4]; mov.u32 %r62, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r23, %r62, %r47; for.size.2 i in range(0, 2, 1): st.shared.f32 [%r23+(0+i*16224)], %f(21+i*1); ld.global.nc.f32 %f(22+i*1), [%rd4+(16224+i*16224)]; st.shared.f32 [%r23+32448], %f23; setp.gt.s32 %p14, %r38, 119; @%p14 bra $L__BB0_21; ld.global.nc.f32 %f24, [%rd4+48672]; st.shared.f32 [%r23+48672], %f24; $L__BB0_21: bar.sync 0; mul.hi.s32 %r65, %r38, 542165497; shr.u32 %r66, %r65, 31; shr.s32 %r67, %r65, 8; add.s32 %r24, %r67, %r66; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(72+i*1), %r(1+i*23), (169+i*1859); sub.s32 %r(25+i*49), %r38, %r(72+i*1); mul.hi.s32 %r75, %r74, 1626496491; shr.u32 %r76, %r75, 31; shr.s32 %r77, %r75, 6; add.s32 %r26, %r77, %r76; shl.b32 %r78, %r26, 12; add.s32 %r80, %r62, %r78; add.s32 %r99, %r80, 2048; mov.u32 %r100, 2048; mad.lo.s32 %r81, %r24, 86528, %r25; shl.b32 %r82, %r81, 2; add.s32 %r84, %r48, %r82; add.s32 %r98, %r84, 173056; mov.f32 %f102, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(103+i*1), %f102; $L__BB0_22: for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(33+i*14), [%r99+(-2048+i*4)]; ld.shared.f32 %f(34+i*14), [%r98+(-173056+i*676)]; fma.rn.f32 %f(35+i*14), %f(34+i*14), %f(33+i*14), %f(109+i*-74); ld.shared.f32 %f(36+i*14), [%r99+(-1024+i*4)]; fma.rn.f32 %f(37+i*14), %f(34+i*14), %f(36+i*14), %f(108+i*-71); ld.shared.f32 %f(38+i*14), [%r99+(0+i*4)]; fma.rn.f32 %f(39+i*14), %f(34+i*14), %f(38+i*14), %f(107+i*-68); ld.shared.f32 %f(40+i*14), [%r99+(1024+i*4)]; fma.rn.f32 %f(41+i*14), %f(34+i*14), %f(40+i*14), %f(106+i*-65); ld.shared.f32 %f(42+i*14), [%r98+(0+i*676)]; fma.rn.f32 %f(43+i*14), %f(42+i*14), %f(33+i*14), %f(105+i*-62); fma.rn.f32 %f(44+i*14), %f(42+i*14), %f(36+i*14), %f(104+i*-60); fma.rn.f32 %f(45+i*14), %f(42+i*14), %f(38+i*14), %f(103+i*-58); fma.rn.f32 %f(46+i*14), %f(42+i*14), %f(40+i*14), %f(102+i*-56); for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(61+i*14), [%r99+(-2040+i*4)]; ld.shared.f32 %f(62+i*14), [%r98+(-171704+i*676)]; fma.rn.f32 %f(63+i*46), %f(62+i*14), %f(61+i*14), %f(49+i*14); ld.shared.f32 %f(64+i*13), [%r99+(-1016+i*4)]; fma.rn.f32 %f(65+i*43), %f(62+i*14), %f(64+i*13), %f(51+i*14); ld.shared.f32 %f(66+i*12), [%r99+(8+i*4)]; fma.rn.f32 %f(67+i*40), %f(62+i*14), %f(66+i*12), %f(53+i*14); ld.shared.f32 %f(68+i*11), [%r99+(1032+i*4)]; fma.rn.f32 %f(69+i*37), %f(62+i*14), %f(68+i*11), %f(55+i*14); ld.shared.f32 %f(70+i*10), [%r98+(1352+i*676)]; fma.rn.f32 %f(71+i*34), %f(70+i*10), %f(61+i*14), %f(57+i*14); fma.rn.f32 %f(72+i*32), %f(70+i*10), %f(64+i*13), %f(58+i*14); fma.rn.f32 %f(73+i*30), %f(70+i*10), %f(66+i*12), %f(59+i*14); fma.rn.f32 %f(74+i*28), %f(70+i*10), %f(68+i*11), %f(60+i*14); for.size.1 i in range(0, 2, 1): add.s32 %r(99+i*-1), %r(99+i*-1), (16+i*2688); add.s32 %r100, %r100, 16; setp.ne.s32 %p15, %r100, 3072; @%p15 bra $L__BB0_22; shl.b32 %r85, %r26, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(87+i*1), %r(45+i*-21), (32448+i*-16224), %r(25+i*62); mad.lo.s32 %r89, %r26, 676, %r88; cvta.to.global.u64 %rd14, %rd7; mul.wide.s32 %rd15, %r85, 4; add.s64 %rd16, %rd14, %rd15; ld.global.nc.f32 %f81, [%rd16]; add.f32 %f82, %f109, %f81; mov.f32 %f83, 0f00000000; max.f32 %f84, %f82, %f83; cvta.to.global.u64 %rd17, %rd6; mul.wide.s32 %rd18, %r89, 4; add.s64 %rd19, %rd17, %rd18; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd19+(0+i*676)], %f(84+i*3); ld.global.nc.f32 %f(85+i*3), [%rd16+(4+i*4)]; add.f32 %f(86+i*3), %f(108+i*-1), %f(85+i*3); max.f32 %f(87+i*3), %f(86+i*3), %f83; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd19+(2028+i*30420)], %f(93+i*2); add.f32 %f(94+i*2), %f(105+i*-1), %f(81+i*4); max.f32 %f(95+i*2), %f(94+i*2), %f83; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd19+(33124+i*676)], %f(97+i*2); add.f32 %f(98+i*2), %f(103+i*-1), %f(88+i*3); max.f32 %f(99+i*2), %f(98+i*2), %f83; st.global.f32 [%rd19+34476], %f101; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[173056]; __shared__ float placeholder_shared[12288]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 43; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 24 + (int)threadIdx.x / 169 < 1024) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 312 + (int)threadIdx.x / 13 < 13312) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4056 + (int)threadIdx.x < 173056) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4056 + (int)threadIdx.x; int index1 = (int)blockIdx.x * 173056 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4056 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } } } } int index2 = (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index2]; placeholder_shared[index2 + 4056] = placeholder1[index2 + 4056]; placeholder_shared[index2 + 8112] = placeholder1[index2 + 8112]; if ((int)threadIdx.x < 120) { placeholder_shared[index2 + 12168] = placeholder1[index2 + 12168]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index3 = (int)threadIdx.x / 2028 * 86528 + rc_outer_inner * 338 + (int)threadIdx.x % 169; int index4 = (int)threadIdx.x % 2028 / 169 * 1024 + rc_outer_inner * 2; compute[0] = compute[0] + pad_temp_shared[index3] * placeholder_shared[index4]; compute[1] = compute[1] + (pad_temp_shared[index3] * placeholder_shared[index4 + 256]); compute[2] = compute[2] + (pad_temp_shared[index3] * placeholder_shared[index4 + 512]); compute[3] = compute[3] + (pad_temp_shared[index3] * placeholder_shared[index4 + 768]); compute[4] = compute[4] + (pad_temp_shared[index3 + 43264] * placeholder_shared[index4]); compute[5] = compute[5] + (pad_temp_shared[index3 + 43264] * placeholder_shared[index4 + 256]); compute[6] = compute[6] + (pad_temp_shared[index3 + 43264] * placeholder_shared[index4 + 512]); compute[7] = compute[7] + (pad_temp_shared[index3 + 43264] * placeholder_shared[index4 + 768]); compute[0] = compute[0] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 1]); compute[1] = compute[1] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 257]); compute[2] = compute[2] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 513]); compute[3] = compute[3] + (pad_temp_shared[index3 + 169] * placeholder_shared[index4 + 769]); compute[4] = compute[4] + (pad_temp_shared[index3 + 43433] * placeholder_shared[index4 + 1]); compute[5] = compute[5] + (pad_temp_shared[index3 + 43433] * placeholder_shared[index4 + 257]); compute[6] = compute[6] + (pad_temp_shared[index3 + 43433] * placeholder_shared[index4 + 513]); compute[7] = compute[7] + (pad_temp_shared[index3 + 43433] * placeholder_shared[index4 + 769]); } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index5 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 2028 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 2028 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index6 = ax0_inner * 4 + ax1_inner; int index7 = (int)threadIdx.x % 2028 / 169 * 4 + ax1_inner; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpqq_d198b
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<8>; .reg .f32 %f<215>; .reg .b32 %r<68>; .reg .b64 %rd<26>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[106496]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[49152]; ld.param.u64 %rd11, [default_function_kernel0_param_0]; ld.param.u64 %rd12, [default_function_kernel0_param_1]; ld.param.u64 %rd13, [default_function_kernel0_param_2]; ld.param.u64 %rd14, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd15, %rd11; cvta.to.global.u64 %rd1, %rd12; cvta.to.global.u64 %rd2, %rd13; cvta.to.global.u64 %rd3, %rd14; mov.u32 %r62, %tid.x; mul.hi.s32 %r30, %r62, 1321528399; shr.u32 %r31, %r30, 31; shr.s32 %r32, %r30, 2; add.s32 %r2, %r32, %r31; mov.u32 %r33, %ctaid.x; for.size.1 i in range(0, 2, 1): mul.lo.s32 %r(34+i*1), %r(33+i*-31), 13; sub.s32 %r3, %r62, %r35; add.s32 %r4, %r3, %r34; shl.b32 %r36, %r62, 2; mov.u32 %r37, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r38, %r37, %r36; add.s32 %r59, %r38, 6656; mov.u32 %r60, 6656; mad.lo.s32 %r39, %r2, 169, %r3; add.s32 %r40, %r39, %r34; mul.wide.s32 %rd16, %r40, 4; add.s64 %rd17, %rd15, %rd16; add.s64 %rd24, %rd17, 86528; $L__BB0_1: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(49+i*1), [%rd24+(-86528+i*10816)]; st.shared.f32 [%r59+(-6656+i*832)], %f(49+i*1); add.s32 %r59, %r59, 13312; add.s64 %rd24, %rd24, 173056; add.s32 %r60, %r60, 13312; setp.ne.s32 %p1, %r60, 113152; @%p1 bra $L__BB0_1; add.s32 %r63, %r62, 624; mul.wide.s32 %rd18, %r62, 4; add.s64 %rd19, %rd1, %rd18; add.s64 %rd25, %rd19, 1664; mov.u32 %r64, 416; mov.u32 %r43, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r44, %r43, %r36; add.s32 %r61, %r44, 1664; $L__BB0_3: add.s64 %rd9, %rd25, -1664; setp.gt.s32 %p2, %r62, 12287; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f65, [%rd9]; add.s32 %r55, %r61, -1664; st.shared.f32 [%r55], %f65; $L__BB0_5: add.s32 %r45, %r63, -416; setp.gt.s32 %p3, %r45, 12287; @%p3 bra $L__BB0_7; ld.global.nc.f32 %f66, [%rd9+832]; add.s32 %r56, %r61, -1664; st.shared.f32 [%r56+832], %f66; $L__BB0_7: add.s32 %r46, %r63, -208; setp.gt.s32 %p4, %r46, 12287; @%p4 bra $L__BB0_9; ld.global.nc.f32 %f67, [%rd9+1664]; add.s32 %r57, %r61, -1664; st.shared.f32 [%r57+1664], %f67; $L__BB0_9: setp.gt.s32 %p5, %r63, 12287; @%p5 bra $L__BB0_11; ld.global.nc.f32 %f68, [%rd9+2496]; add.s32 %r58, %r61, -1664; st.shared.f32 [%r58+2496], %f68; $L__BB0_11: for.size.1 i in range(0, 3, 1): add.s32 %r(64+i*-1), %r(64+i*-1), 832; add.s64 %rd25, %rd25, 3328; add.s32 %r61, %r61, 3328; setp.ne.s32 %p6, %r64, 12896; @%p6 bra $L__BB0_3; bar.sync 0; mad.lo.s32 %r49, %r2, 3072, %r43; add.s32 %r67, %r49, 1028; mov.u32 %r66, 1028; shl.b32 %r50, %r3, 2; add.s32 %r52, %r37, %r50; add.s32 %r65, %r52, 53248; mov.f32 %f191, 0f00000000; for.size.1 i in range(0, 23, 1): mov.f32 %f(192+i*1), %f191; $L__BB0_13: for.size.9 i in range(0, 2, 1): ld.shared.f32 %f(93+i*9), [%r67+(-1028+i*4)]; ld.shared.f32 %f(94+i*9), [%r65+(-53248+i*52)]; fma.rn.f32 %f(95+i*119), %f(94+i*9), %f(93+i*9), %f(214+i*-119); ld.shared.f32 %f(96+i*8), [%r65+(0+i*52)]; fma.rn.f32 %f(97+i*105), %f(96+i*8), %f(93+i*9), %f(202+i*-105); ld.shared.f32 %f(98+i*7), [%r65+(-39936+i*52)]; fma.rn.f32 %f(99+i*112), %f(98+i*7), %f(93+i*9), %f(211+i*-112); ld.shared.f32 %f(100+i*6), [%r65+(13312+i*52)]; fma.rn.f32 %f(101+i*98), %f(100+i*6), %f(93+i*9), %f(199+i*-98); for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(107+i*5), [%r67+(-4+i*4)]; fma.rn.f32 %f(108+i*105), %f(94+i*9), %f(107+i*5), %f(213+i*-105); fma.rn.f32 %f(109+i*92), %f(96+i*8), %f(107+i*5), %f(201+i*-92); fma.rn.f32 %f(110+i*100), %f(98+i*7), %f(107+i*5), %f(210+i*-100); fma.rn.f32 %f(111+i*87), %f(100+i*6), %f(107+i*5), %f(198+i*-87); for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(113+i*5), [%r67+(1020+i*4)]; fma.rn.f32 %f(114+i*98), %f(94+i*9), %f(113+i*5), %f(212+i*-98); fma.rn.f32 %f(115+i*85), %f(96+i*8), %f(113+i*5), %f(200+i*-85); fma.rn.f32 %f(116+i*93), %f(98+i*7), %f(113+i*5), %f(209+i*-93); fma.rn.f32 %f(117+i*80), %f(100+i*6), %f(113+i*5), %f(197+i*-80); for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(119+i*4+j*2), [%r65+(-26624+i*13312+j*53248)]; fma.rn.f32 %f(120+i*4+j*2), %f(119+i*4+j*2), %f93, %f(208+i*-3+j*-12); for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(127+i*2+j*1), [%r65+(-26572+i*13312+j*53248)]; fma.rn.f32 %f(208+i*-3+j*-12), %f(127+i*2+j*1), %f102, %f(120+i*4+j*2); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): fma.rn.f32 %f(131+i*2+j*1), %f(119+i*4+j*2), %f107, %f(207+i*-3+j*-12); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): fma.rn.f32 %f(207+i*-3+j*-12), %f(127+i*2+j*1), %f112, %f(131+i*2+j*1); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): fma.rn.f32 %f(135+i*2+j*1), %f(119+i*4+j*2), %f113, %f(206+i*-3+j*-12); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 2, 1): fma.rn.f32 %f(206+i*-3+j*-12), %f(127+i*2+j*1), %f118, %f(135+i*2+j*1); for.size.1 i in range(0, 2, 1): add.s32 %r(67+i*-2), %r(67+i*-2), (8+i*96); add.s32 %r66, %r66, 8; setp.ne.s32 %p7, %r66, 2052; @%p7 bra $L__BB0_13; mul.lo.s32 %r53, %r2, 3; mad.lo.s32 %r54, %r2, 507, %r4; mul.wide.s32 %rd20, %r53, 4; add.s64 %rd21, %rd3, %rd20; ld.global.nc.f32 %f139, [%rd21]; add.f32 %f140, %f214, %f139; mov.f32 %f141, 0f00000000; max.f32 %f142, %f140, %f141; mul.wide.s32 %rd22, %r54, 4; add.s64 %rd23, %rd2, %rd22; for.size.7 i in range(0, 2, 1): st.global.f32 [%rd23+(0+i*676)], %f(142+i*5); add.f32 %f(143+i*5), %f(202+i*-1), %f(139+i*6); max.f32 %f(144+i*5), %f(143+i*5), %f141; st.global.f32 [%rd23+(129792+i*676)], %f(144+i*5); ld.global.nc.f32 %f(145+i*5), [%rd21+(4+i*4)]; add.f32 %f(146+i*5), %f(213+i*-1), %f(145+i*5); max.f32 %f(147+i*5), %f(146+i*5), %f141; for.size.12 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd23+(1352+i*32448+j*129792)], %f(152+i*12+j*2); add.f32 %f(153+i*12+j*2), %f(200+i*-3+j*11), %f(150+j*-11); max.f32 %f(154+i*12+j*2), %f(153+i*12+j*2), %f141; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd23+(32448+i*32448+j*129792)], %f(156+i*12+j*2); add.f32 %f(157+i*12+j*2), %f(199+i*-3+j*11), %f(139+j*6); max.f32 %f(158+i*12+j*2), %f(157+i*12+j*2), %f141; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd23+(33124+i*32448+j*129792)], %f(160+i*12+j*2); add.f32 %f(161+i*12+j*2), %f(198+i*-3+j*11), %f(145+j*5); max.f32 %f(162+i*12+j*2), %f(161+i*12+j*2), %f141; st.global.f32 [%rd23+98696], %f188; add.f32 %f189, %f191, %f150; max.f32 %f190, %f189, %f141; st.global.f32 [%rd23+228488], %f190; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[24]; __shared__ float pad_temp_shared[26624]; __shared__ float placeholder_shared[12288]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 2; ++nn_outer_inner_init) { int index0 = nn_outer_inner_init * 6; compute[index0] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 15] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 13] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 16] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 14] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 17] = 0.000000e+00f; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 128; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 208 + (int)threadIdx.x; int index2 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 2704 + (int)threadIdx.x / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 60; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 208 + (int)threadIdx.x < 12288) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 208 + (int)threadIdx.x; placeholder_shared[index3] = placeholder1[index3]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 2; ++nn_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 3; ++ff_outer_inner) { int index4 = nn_outer_inner * 6 + ff_outer_inner; int index5 = nn_outer_inner * 6656 + rc_outer_inner * 26 + (int)threadIdx.x % 13; int index6 = (int)threadIdx.x / 13 * 768 + ff_outer_inner * 256 + rc_outer_inner * 2; compute[index4] = compute[index4] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[index4 + 12] = compute[index4 + 12] + (pad_temp_shared[index5 + 13312] * placeholder_shared[index6]); compute[index4 + 3] = compute[index4 + 3] + (pad_temp_shared[index5 + 3328] * placeholder_shared[index6]); compute[index4 + 15] = compute[index4 + 15] + (pad_temp_shared[index5 + 16640] * placeholder_shared[index6]); compute[index4] = compute[index4] + (pad_temp_shared[index5 + 13] * placeholder_shared[index6 + 1]); compute[index4 + 12] = compute[index4 + 12] + (pad_temp_shared[index5 + 13325] * placeholder_shared[index6 + 1]); compute[index4 + 3] = compute[index4 + 3] + (pad_temp_shared[index5 + 3341] * placeholder_shared[index6 + 1]); compute[index4 + 15] = compute[index4 + 15] + (pad_temp_shared[index5 + 16653] * placeholder_shared[index6 + 1]); } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { int index7 = ax0_inner * 8112 + (int)threadIdx.x / 13 * 507 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index8 = ax0_inner * 3 + ax1_inner; int index9 = (int)threadIdx.x / 13 * 3 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); T_relu[index7 + 32448] = max(compute[index8 + 12] + placeholder2[index9], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpiebnbzgy
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<255>; .reg .b32 %r<215>; .reg .b64 %rd<47>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1536]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %ctaid.x; mul.lo.s32 %r35, %r1, 173056; mov.u32 %r2, %tid.x; shl.b32 %r36, %r2, 4; mul.hi.s32 %r37, %r36, 1626496491; shr.u32 %r38, %r37, 31; shr.s32 %r39, %r37, 9; add.s32 %r40, %r39, %r38; mul.lo.s32 %r41, %r40, 1352; sub.s32 %r42, %r36, %r41; add.s32 %r43, %r42, %r35; mad.lo.s32 %r3, %r40, 43264, %r43; shl.b32 %r44, %r2, 6; mov.u32 %r45, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r45, %r44; for.size.9 i in range(0, 15, 1): or.b32 %r(46+i*8), %r36, (1+i*1); mul.hi.s32 %r(47+i*8), %r(46+i*8), 1626496491; shr.u32 %r(48+i*8), %r(47+i*8), 31; shr.s32 %r(49+i*8), %r(47+i*8), 9; add.s32 %r(50+i*8), %r(49+i*8), %r(48+i*8); mul.lo.s32 %r(51+i*8), %r(50+i*8), 1352; sub.s32 %r(52+i*8), %r(46+i*8), %r(51+i*8); add.s32 %r(53+i*8), %r(52+i*8), %r35; mad.lo.s32 %r(5+i*1), %r(50+i*8), 43264, %r(53+i*8); shr.u32 %r166, %r2, 3; and.b32 %r167, %r2, 7; prmt.b32 %r20, %r166, %r167, 8452; shl.b32 %r168, %r2, 2; mov.u32 %r169, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r21, %r169, %r168; mul.hi.s32 %r170, %r2, -770891565; add.s32 %r171, %r170, %r2; shr.u32 %r172, %r171, 31; shr.s32 %r173, %r171, 9; add.s32 %r22, %r173, %r172; mul.hi.s32 %r174, %r2, 1321528399; shr.u32 %r175, %r174, 31; shr.s32 %r176, %r174, 2; add.s32 %r177, %r176, %r175; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(178+i*2), %r(177+i*-155), (13+i*611); sub.s32 %r(179+i*-156), %r2, %r(178+i*2); mul.hi.s32 %r181, %r23, 1321528399; shr.u32 %r182, %r181, 31; shr.s32 %r183, %r181, 2; add.s32 %r24, %r183, %r182; shl.b32 %r184, %r24, 5; add.s32 %r25, %r169, %r184; mul.lo.s32 %r185, %r22, 2704; mad.lo.s32 %r186, %r179, 13, %r185; shl.b32 %r187, %r186, 2; add.s32 %r26, %r45, %r187; mov.u32 %r212, 0; mov.f32 %f229, 0f00000000; for.size.1 i in range(0, 25, 1): mov.f32 %f(230+i*1), %f229; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r2, 337; @%p1 bra $L__BB0_3; mul.lo.s32 %r188, %r212, 1352; for.size.5 i in range(0, 2, 1): add.s32 %r(189+i*1), %r(3+i*2), %r188; mul.wide.s32 %rd(9+i*2), %r(189+i*1), 4; add.s64 %rd(10+i*2), %rd2, %rd(9+i*2); ld.global.nc.f32 %f(105+i*1), [%rd(10+i*2)+0]; st.shared.f32 [%r4+(0+i*4)], %f(105+i*1); for.size.5 i in range(0, 14, 1): add.s32 %r(191+i*1), %r(6+i*1), %r188; mul.wide.s32 %rd(13+i*2), %r(191+i*1), 4; add.s64 %rd(14+i*2), %rd2, %rd(13+i*2); ld.global.nc.f32 %f(107+i*1), [%rd(14+i*2)+0]; st.shared.f32 [%r4+(8+i*4)], %f(107+i*1); $L__BB0_3: setp.gt.s32 %p2, %r2, 383; @%p2 bra $L__BB0_5; shl.b32 %r205, %r212, 3; add.s32 %r206, %r20, %r205; mul.wide.s32 %rd41, %r206, 4; add.s64 %rd42, %rd1, %rd41; ld.global.nc.f32 %f121, [%rd42]; st.shared.f32 [%r21], %f121; $L__BB0_5: bar.sync 0; mov.u32 %r213, 5408; mov.u32 %r214, %r25; $L__BB0_6: ld.shared.f32 %f122, [%r214]; add.s32 %r208, %r26, %r213; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 13, 1): ld.shared.f32 %f(123+i*13+j*1), [%r208+(-5408+i*5408+j*4)]; fma.rn.f32 %f(254+i*-13+j*-1), %f(123+i*13+j*1), %f122, %f(254+i*-13+j*-1); for.size.1 i in range(0, 2, 1): add.s32 %r(214+i*-1), %r(214+i*-1), (4+i*672); setp.ne.s32 %p3, %r213, 10816; @%p3 bra $L__BB0_6; add.s32 %r212, %r212, 1; setp.lt.u32 %p4, %r212, 32; @%p4 bra $L__BB0_1; mul.wide.s32 %rd43, %r24, 4; add.s64 %rd44, %rd4, %rd43; mul.lo.s32 %r209, %r1, 32448; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(210+i*1), %r(22+i*1), (16224+i*-16211), %r(209+i*1); ld.global.nc.f32 %f149, [%rd44]; add.f32 %f150, %f254, %f149; mov.f32 %f151, 0f00000000; max.f32 %f152, %f150, %f151; mul.wide.s32 %rd45, %r211, 4; add.s64 %rd46, %rd3, %rd45; for.size.3 i in range(0, 13, 1): st.global.f32 [%rd46+(0+i*4)], %f(152+i*2); add.f32 %f(153+i*2), %f(253+i*-1), %f149; max.f32 %f(154+i*2), %f(153+i*2), %f151; for.size.3 i in range(0, 12, 1): st.global.f32 [%rd46+(32448+i*4)], %f(178+i*2); add.f32 %f(179+i*2), %f(240+i*-1), %f149; max.f32 %f(180+i*2), %f(179+i*2), %f151; st.global.f32 [%rd46+32496], %f202; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[26]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[384]; compute[0] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[20] = 0.000000e+00f; compute[22] = 0.000000e+00f; compute[24] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[19] = 0.000000e+00f; compute[21] = 0.000000e+00f; compute[23] = 0.000000e+00f; compute[25] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index1 = (int)blockIdx.x * 173056 + (int)threadIdx.x * 16 / 1352 * 43264 + rc_outer_outer * 1352 + (int)threadIdx.x * 16 % 1352; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index2 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 1) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 1) % 1352; pad_temp_shared[index0 + 1] = placeholder[index2]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index3 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 2) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 2) % 1352; pad_temp_shared[index0 + 2] = placeholder[index3]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index4 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 3) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 3) % 1352; pad_temp_shared[index0 + 3] = placeholder[index4]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index5 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 4) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 4) % 1352; pad_temp_shared[index0 + 4] = placeholder[index5]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index6 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 5) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 5) % 1352; pad_temp_shared[index0 + 5] = placeholder[index6]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index7 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 6) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 6) % 1352; pad_temp_shared[index0 + 6] = placeholder[index7]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index8 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 7) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 7) % 1352; pad_temp_shared[index0 + 7] = placeholder[index8]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index9 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 8) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 8) % 1352; pad_temp_shared[index0 + 8] = placeholder[index9]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index10 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 9) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 9) % 1352; pad_temp_shared[index0 + 9] = placeholder[index10]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index11 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 10) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 10) % 1352; pad_temp_shared[index0 + 10] = placeholder[index11]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index12 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 11) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 11) % 1352; pad_temp_shared[index0 + 11] = placeholder[index12]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index13 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 12) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 12) % 1352; pad_temp_shared[index0 + 12] = placeholder[index13]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index14 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 13) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 13) % 1352; pad_temp_shared[index0 + 13] = placeholder[index14]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index15 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 14) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 14) % 1352; pad_temp_shared[index0 + 14] = placeholder[index15]; } if ((int)threadIdx.x < 338) { int index0 = (int)threadIdx.x * 16; int index16 = (int)blockIdx.x * 173056 + ((int)threadIdx.x * 16 + 15) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 15) % 1352; pad_temp_shared[index0 + 15] = placeholder[index16]; } if ((int)threadIdx.x < 384) { int index17 = (int)threadIdx.x; int index18 = ((int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index17] = placeholder1[index18]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { int index19 = (int)threadIdx.x / 624 * 2704 + rc_outer_inner * 169 + (int)threadIdx.x % 13 * 13; int index20 = (int)threadIdx.x % 624 / 13 * 8 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index19] * placeholder_shared[index20]; compute[2] = compute[2] + (pad_temp_shared[index19 + 1] * placeholder_shared[index20]); compute[4] = compute[4] + (pad_temp_shared[index19 + 2] * placeholder_shared[index20]); compute[6] = compute[6] + (pad_temp_shared[index19 + 3] * placeholder_shared[index20]); compute[8] = compute[8] + (pad_temp_shared[index19 + 4] * placeholder_shared[index20]); compute[10] = compute[10] + (pad_temp_shared[index19 + 5] * placeholder_shared[index20]); compute[12] = compute[12] + (pad_temp_shared[index19 + 6] * placeholder_shared[index20]); compute[14] = compute[14] + (pad_temp_shared[index19 + 7] * placeholder_shared[index20]); compute[16] = compute[16] + (pad_temp_shared[index19 + 8] * placeholder_shared[index20]); compute[18] = compute[18] + (pad_temp_shared[index19 + 9] * placeholder_shared[index20]); compute[20] = compute[20] + (pad_temp_shared[index19 + 10] * placeholder_shared[index20]); compute[22] = compute[22] + (pad_temp_shared[index19 + 11] * placeholder_shared[index20]); compute[24] = compute[24] + (pad_temp_shared[index19 + 12] * placeholder_shared[index20]); compute[1] = compute[1] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20]); compute[3] = compute[3] + (pad_temp_shared[index19 + 1353] * placeholder_shared[index20]); compute[5] = compute[5] + (pad_temp_shared[index19 + 1354] * placeholder_shared[index20]); compute[7] = compute[7] + (pad_temp_shared[index19 + 1355] * placeholder_shared[index20]); compute[9] = compute[9] + (pad_temp_shared[index19 + 1356] * placeholder_shared[index20]); compute[11] = compute[11] + (pad_temp_shared[index19 + 1357] * placeholder_shared[index20]); compute[13] = compute[13] + (pad_temp_shared[index19 + 1358] * placeholder_shared[index20]); compute[15] = compute[15] + (pad_temp_shared[index19 + 1359] * placeholder_shared[index20]); compute[17] = compute[17] + (pad_temp_shared[index19 + 1360] * placeholder_shared[index20]); compute[19] = compute[19] + (pad_temp_shared[index19 + 1361] * placeholder_shared[index20]); compute[21] = compute[21] + (pad_temp_shared[index19 + 1362] * placeholder_shared[index20]); compute[23] = compute[23] + (pad_temp_shared[index19 + 1363] * placeholder_shared[index20]); compute[25] = compute[25] + (pad_temp_shared[index19 + 1364] * placeholder_shared[index20]); } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { int index21 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 624 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 624 * 13; int index22 = ax0_inner; int index23 = (int)threadIdx.x % 624 / 13; T_relu[index21] = max(compute[index22] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 1] = max(compute[index22 + 2] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 2] = max(compute[index22 + 4] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 3] = max(compute[index22 + 6] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 4] = max(compute[index22 + 8] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 5] = max(compute[index22 + 10] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 6] = max(compute[index22 + 12] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 7] = max(compute[index22 + 14] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 8] = max(compute[index22 + 16] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 9] = max(compute[index22 + 18] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 10] = max(compute[index22 + 20] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 11] = max(compute[index22 + 22] + placeholder2[index23], 0.000000e+00f); T_relu[index21 + 12] = max(compute[index22 + 24] + placeholder2[index23], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpn_6g9i7l
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<162>; .reg .b32 %r<54>; .reg .b64 %rd<24>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1664]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[512]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r14, %ctaid.x; mul.hi.s32 %r15, %r14, -770891565; add.s32 %r16, %r15, %r14; shr.u32 %r17, %r16, 31; shr.s32 %r18, %r16, 5; add.s32 %r1, %r18, %r17; mul.hi.s32 %r19, %r14, 1321528399; shr.u32 %r20, %r19, 31; shr.s32 %r21, %r19, 2; add.s32 %r22, %r21, %r20; mul.lo.s32 %r23, %r22, 13; sub.s32 %r2, %r14, %r23; mov.u32 %r3, %tid.x; shl.b32 %r24, %r3, 2; mov.u32 %r25, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r25, %r24; mul.lo.s32 %r26, %r1, 39; sub.s32 %r27, %r14, %r26; mul.hi.s32 %r28, %r27, 1321528399; shr.u32 %r29, %r28, 31; shr.s32 %r30, %r28, 2; add.s32 %r5, %r30, %r29; shl.b32 %r31, %r5, 12; and.b32 %r32, %r3, 7; mov.u32 %r33, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r33, %r24; mul.hi.s32 %r34, %r3, 1321528399; shr.u32 %r35, %r34, 31; shr.s32 %r36, %r34, 2; add.s32 %r9, %r36, %r35; mul.lo.s32 %r37, %r9, 13; sub.s32 %r7, %r3, %r37; for.size.2 i in range(0, 2, 1): shl.b32 %r(38+i*1), %r(7+i*2), (2+i*4); add.s32 %r(8+i*2), %r(25+i*8), %r(38+i*1); shl.b32 %r40, %r3, 5; and.b32 %r41, %r40, -256; add.s32 %r42, %r31, %r41; or.b32 %r43, %r42, %r32; mul.wide.s32 %rd13, %r43, 4; add.s64 %rd23, %rd12, %rd13; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(44+i*1), %r(1+i*2), (173056+i*-173043), %r(2+i*42); cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r45, 4; add.s64 %rd16, %rd14, %rd15; add.s64 %rd22, %rd16, 346112; mov.u32 %r53, 0; mov.f32 %f154, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(155+i*1), %f154; $L__BB0_1: bar.sync 0; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(25+i*1), [%rd22+(-346112+i*173056)]; st.shared.f32 [%r4+(0+i*416)], %f(25+i*1); ld.global.nc.f32 %f29, [%rd23]; st.shared.f32 [%r6], %f29; setp.gt.s32 %p1, %r3, 23; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f30, [%rd23+13312]; st.shared.f32 [%r6+416], %f30; $L__BB0_3: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(31+i*1), [%r(10+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(33+i*2), %f32, %f(31+i*3), %f(161+i*-1); ld.shared.f32 %f(34+i*2), [%r10+(32+i*-28)]; for.size.5 i in range(0, 6, 1): ld.shared.f32 %f(37+i*5), [%r8+(52+i*52)]; fma.rn.f32 %f(38+i*5), %f(37+i*5), %f(36+i*5), %f(33+i*5); ld.shared.f32 %f(39+i*5), [%r10+(36+i*4)]; fma.rn.f32 %f(40+i*5), %f(37+i*5), %f(39+i*5), %f(35+i*5); ld.shared.f32 %f(41+i*5), [%r10+(8+i*4)]; ld.shared.f32 %f67, [%r8+364]; fma.rn.f32 %f161, %f67, %f66, %f63; ld.shared.f32 %f68, [%r10+60]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(160+i*-89), %f(67+i*2), %f(68+i*-34), %f(65+i*93); ld.shared.f32 %f(69+i*3), [%r8+(416+i*52)]; fma.rn.f32 %f(70+i*3), %f(69+i*3), %f(31+i*5), %f(159+i*-89); for.size.3 i in range(0, 5, 1): fma.rn.f32 %f(74+i*3), %f(72+i*3), %f(39+i*5), %f(71+i*3); ld.shared.f32 %f(75+i*3), [%r8+(520+i*52)]; fma.rn.f32 %f(76+i*3), %f(75+i*3), %f(41+i*5), %f(73+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(89+i*69), %f(87+i*3), %f(64+i*4), %f(86+i*3); ld.shared.f32 %f(90+i*1), [%r8+(780+i*52)]; fma.rn.f32 %f(159+i*-67), %f(90+i*1), %f(66+i*-35), %f(88+i*69); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(93+i*3), %f(91+i*3), %f(34+i*5), %f(156+i*-63); ld.shared.f32 %f(94+i*3), [%r8+(884+i*52)]; fma.rn.f32 %f(95+i*3), %f(94+i*3), %f(36+i*5), %f(92+i*3); for.size.3 i in range(0, 4, 1): fma.rn.f32 %f(99+i*3), %f(97+i*3), %f(44+i*5), %f(96+i*3); ld.shared.f32 %f(100+i*3), [%r8+(988+i*52)]; fma.rn.f32 %f(101+i*3), %f(100+i*3), %f(46+i*5), %f(98+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(111+i*45), %f(109+i*3), %f(64+i*4), %f(108+i*3); ld.shared.f32 %f(112+i*1), [%r8+(1196+i*52)]; fma.rn.f32 %f(157+i*-43), %f(112+i*1), %f(66+i*-35), %f(110+i*45); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(115+i*3), %f(113+i*3), %f(34+i*5), %f(154+i*-39); ld.shared.f32 %f(116+i*3), [%r8+(1300+i*52)]; fma.rn.f32 %f(117+i*3), %f(116+i*3), %f(36+i*5), %f(114+i*3); for.size.3 i in range(0, 4, 1): fma.rn.f32 %f(121+i*3), %f(119+i*3), %f(44+i*5), %f(118+i*3); ld.shared.f32 %f(122+i*3), [%r8+(1404+i*52)]; fma.rn.f32 %f(123+i*3), %f(122+i*3), %f(46+i*5), %f(120+i*3); fma.rn.f32 %f133, %f131, %f64, %f130; ld.shared.f32 %f134, [%r8+1612]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(155+i*-1), %f134, %f(66+i*2), %f(132+i*1); for.size.1 i in range(0, 2, 1): add.s64 %rd(23+i*-1), %rd(23+i*-1), (32+i*5376); add.s32 %r53, %r53, 1; setp.ne.s32 %p2, %r53, 32; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd8; for.size.1 i in range(0, 2, 1): shl.b32 %r(46+i*1), %r(9+i*-4), (1+i*3); add.s32 %r48, %r47, %r46; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(49+i*1), %r(1+i*8), (32448+i*-32110), %r(2+i*47); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(51+i*1), %r(7+i*-2), (13+i*2691), %r(50+i*1); mul.wide.s32 %rd18, %r48, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f135, [%rd19]; add.f32 %f136, %f161, %f135; mov.f32 %f137, 0f00000000; max.f32 %f138, %f136, %f137; mul.wide.s32 %rd20, %r52, 4; add.s64 %rd21, %rd1, %rd20; st.global.f32 [%rd21], %f138; ld.global.nc.f32 %f139, [%rd19+4]; for.size.4 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): add.f32 %f(140+i*4+j*2), %f(160+i*-2+j*-1), %f(139+j*-4); max.f32 %f(141+i*4+j*2), %f(140+i*4+j*2), %f137; st.global.f32 [%rd21+(676+i*32448+j*31772)], %f(141+i*4+j*2); add.f32 %f152, %f154, %f139; max.f32 %f153, %f152, %f137; st.global.f32 [%rd21+98020], %f153; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[416]; __shared__ float placeholder_shared[128]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x / 39 * 173056 + rc_outer_outer * 1352 + (int)threadIdx.x * 13 + (int)blockIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 104] = placeholder[index1 + 43264]; pad_temp_shared[index0 + 208] = placeholder[index1 + 86528]; pad_temp_shared[index0 + 312] = placeholder[index1 + 129792]; int index2 = (int)blockIdx.x % 39 / 13 * 4096 + ((int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index0] = placeholder1[index2]; if ((int)threadIdx.x < 24) { placeholder_shared[index0 + 104] = placeholder1[index2 + 3328]; } __syncthreads(); int index3 = (int)threadIdx.x / 13 * 16; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 8]); int index4 = (int)threadIdx.x % 13 + 13; compute[0] = compute[0] + (pad_temp_shared[index4] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index3 + 9]); int index5 = (int)threadIdx.x % 13 + 26; compute[0] = compute[0] + (pad_temp_shared[index5] * placeholder_shared[index3 + 2]); compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index3 + 10]); int index6 = (int)threadIdx.x % 13 + 39; compute[0] = compute[0] + (pad_temp_shared[index6] * placeholder_shared[index3 + 3]); compute[1] = compute[1] + (pad_temp_shared[index6] * placeholder_shared[index3 + 11]); int index7 = (int)threadIdx.x % 13 + 52; compute[0] = compute[0] + (pad_temp_shared[index7] * placeholder_shared[index3 + 4]); compute[1] = compute[1] + (pad_temp_shared[index7] * placeholder_shared[index3 + 12]); int index8 = (int)threadIdx.x % 13 + 65; compute[0] = compute[0] + (pad_temp_shared[index8] * placeholder_shared[index3 + 5]); compute[1] = compute[1] + (pad_temp_shared[index8] * placeholder_shared[index3 + 13]); int index9 = (int)threadIdx.x % 13 + 78; compute[0] = compute[0] + (pad_temp_shared[index9] * placeholder_shared[index3 + 6]); compute[1] = compute[1] + (pad_temp_shared[index9] * placeholder_shared[index3 + 14]); int index10 = (int)threadIdx.x % 13 + 91; compute[0] = compute[0] + (pad_temp_shared[index10] * placeholder_shared[index3 + 7]); compute[1] = compute[1] + (pad_temp_shared[index10] * placeholder_shared[index3 + 15]); int index11 = (int)threadIdx.x % 13 + 104; compute[2] = compute[2] + pad_temp_shared[index11] * placeholder_shared[index3]; compute[3] = compute[3] + (pad_temp_shared[index11] * placeholder_shared[index3 + 8]); int index12 = (int)threadIdx.x % 13 + 117; compute[2] = compute[2] + (pad_temp_shared[index12] * placeholder_shared[index3 + 1]); compute[3] = compute[3] + (pad_temp_shared[index12] * placeholder_shared[index3 + 9]); int index13 = (int)threadIdx.x % 13 + 130; compute[2] = compute[2] + (pad_temp_shared[index13] * placeholder_shared[index3 + 2]); compute[3] = compute[3] + (pad_temp_shared[index13] * placeholder_shared[index3 + 10]); int index14 = (int)threadIdx.x % 13 + 143; compute[2] = compute[2] + (pad_temp_shared[index14] * placeholder_shared[index3 + 3]); compute[3] = compute[3] + (pad_temp_shared[index14] * placeholder_shared[index3 + 11]); int index15 = (int)threadIdx.x % 13 + 156; compute[2] = compute[2] + (pad_temp_shared[index15] * placeholder_shared[index3 + 4]); compute[3] = compute[3] + (pad_temp_shared[index15] * placeholder_shared[index3 + 12]); int index16 = (int)threadIdx.x % 13 + 169; compute[2] = compute[2] + (pad_temp_shared[index16] * placeholder_shared[index3 + 5]); compute[3] = compute[3] + (pad_temp_shared[index16] * placeholder_shared[index3 + 13]); int index17 = (int)threadIdx.x % 13 + 182; compute[2] = compute[2] + (pad_temp_shared[index17] * placeholder_shared[index3 + 6]); compute[3] = compute[3] + (pad_temp_shared[index17] * placeholder_shared[index3 + 14]); int index18 = (int)threadIdx.x % 13 + 195; compute[2] = compute[2] + (pad_temp_shared[index18] * placeholder_shared[index3 + 7]); compute[3] = compute[3] + (pad_temp_shared[index18] * placeholder_shared[index3 + 15]); int index19 = (int)threadIdx.x % 13 + 208; compute[4] = compute[4] + pad_temp_shared[index19] * placeholder_shared[index3]; compute[5] = compute[5] + (pad_temp_shared[index19] * placeholder_shared[index3 + 8]); int index20 = (int)threadIdx.x % 13 + 221; compute[4] = compute[4] + (pad_temp_shared[index20] * placeholder_shared[index3 + 1]); compute[5] = compute[5] + (pad_temp_shared[index20] * placeholder_shared[index3 + 9]); int index21 = (int)threadIdx.x % 13 + 234; compute[4] = compute[4] + (pad_temp_shared[index21] * placeholder_shared[index3 + 2]); compute[5] = compute[5] + (pad_temp_shared[index21] * placeholder_shared[index3 + 10]); int index22 = (int)threadIdx.x % 13 + 247; compute[4] = compute[4] + (pad_temp_shared[index22] * placeholder_shared[index3 + 3]); compute[5] = compute[5] + (pad_temp_shared[index22] * placeholder_shared[index3 + 11]); int index23 = (int)threadIdx.x % 13 + 260; compute[4] = compute[4] + (pad_temp_shared[index23] * placeholder_shared[index3 + 4]); compute[5] = compute[5] + (pad_temp_shared[index23] * placeholder_shared[index3 + 12]); int index24 = (int)threadIdx.x % 13 + 273; compute[4] = compute[4] + (pad_temp_shared[index24] * placeholder_shared[index3 + 5]); compute[5] = compute[5] + (pad_temp_shared[index24] * placeholder_shared[index3 + 13]); int index25 = (int)threadIdx.x % 13 + 286; compute[4] = compute[4] + (pad_temp_shared[index25] * placeholder_shared[index3 + 6]); compute[5] = compute[5] + (pad_temp_shared[index25] * placeholder_shared[index3 + 14]); int index26 = (int)threadIdx.x % 13 + 299; compute[4] = compute[4] + (pad_temp_shared[index26] * placeholder_shared[index3 + 7]); compute[5] = compute[5] + (pad_temp_shared[index26] * placeholder_shared[index3 + 15]); int index27 = (int)threadIdx.x % 13 + 312; compute[6] = compute[6] + pad_temp_shared[index27] * placeholder_shared[index3]; compute[7] = compute[7] + (pad_temp_shared[index27] * placeholder_shared[index3 + 8]); int index28 = (int)threadIdx.x % 13 + 325; compute[6] = compute[6] + (pad_temp_shared[index28] * placeholder_shared[index3 + 1]); compute[7] = compute[7] + (pad_temp_shared[index28] * placeholder_shared[index3 + 9]); int index29 = (int)threadIdx.x % 13 + 338; compute[6] = compute[6] + (pad_temp_shared[index29] * placeholder_shared[index3 + 2]); compute[7] = compute[7] + (pad_temp_shared[index29] * placeholder_shared[index3 + 10]); int index30 = (int)threadIdx.x % 13 + 351; compute[6] = compute[6] + (pad_temp_shared[index30] * placeholder_shared[index3 + 3]); compute[7] = compute[7] + (pad_temp_shared[index30] * placeholder_shared[index3 + 11]); int index31 = (int)threadIdx.x % 13 + 364; compute[6] = compute[6] + (pad_temp_shared[index31] * placeholder_shared[index3 + 4]); compute[7] = compute[7] + (pad_temp_shared[index31] * placeholder_shared[index3 + 12]); int index32 = (int)threadIdx.x % 13 + 377; compute[6] = compute[6] + (pad_temp_shared[index32] * placeholder_shared[index3 + 5]); compute[7] = compute[7] + (pad_temp_shared[index32] * placeholder_shared[index3 + 13]); int index33 = (int)threadIdx.x % 13 + 390; compute[6] = compute[6] + (pad_temp_shared[index33] * placeholder_shared[index3 + 6]); compute[7] = compute[7] + (pad_temp_shared[index33] * placeholder_shared[index3 + 14]); int index34 = (int)threadIdx.x % 13 + 403; compute[6] = compute[6] + (pad_temp_shared[index34] * placeholder_shared[index3 + 7]); compute[7] = compute[7] + (pad_temp_shared[index34] * placeholder_shared[index3 + 15]); } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index35 = (int)blockIdx.x / 39 * 32448 + ax0_inner * 8112 + (int)blockIdx.x % 39 / 13 * 2704 + (int)threadIdx.x / 13 * 338 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x % 13; int index36 = ax0_inner * 2 + ax1_inner; int index37 = (int)blockIdx.x % 39 / 13 * 16 + (int)threadIdx.x / 13 * 2 + ax1_inner; T_relu[index35] = max(compute[index36] + placeholder2[index37], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpbw3scr5g
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<11>; .reg .f32 %f<122>; .reg .b32 %r<92>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[346112]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd2, %rd6; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(31+i*4), %r1, (1626496491+i*-304968092); shr.u32 %r(32+i*4), %r(31+i*4), 31; shr.s32 %r(33+i*4), %r(31+i*4), (6+i*-4); add.s32 %r(34+i*-32), %r(33+i*4), %r(32+i*4); mov.u32 %r3, %ctaid.x; mul.lo.s32 %r38, %r34, 169; sub.s32 %r4, %r1, %r38; mad.lo.s32 %r5, %r3, 173056, %r4; shr.u32 %r39, %r1, 7; and.b32 %r40, %r1, 127; prmt.b32 %r6, %r39, %r40, 8452; shl.b32 %r41, %r1, 2; mov.u32 %r42, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r42, %r41; add.s32 %r43, %r1, 4056; shr.u32 %r44, %r43, 7; add.s32 %r45, %r1, 88; and.b32 %r46, %r45, 127; prmt.b32 %r8, %r44, %r46, 8452; mul.hi.s32 %r47, %r1, 542165497; shr.u32 %r48, %r47, 31; shr.s32 %r49, %r47, 8; add.s32 %r9, %r49, %r48; mad.lo.s32 %r10, %r9, 43264, %r4; mul.lo.s32 %r50, %r9, 2028; sub.s32 %r51, %r1, %r50; mul.hi.s32 %r52, %r51, 1626496491; shr.u32 %r53, %r52, 31; shr.s32 %r54, %r52, 6; add.s32 %r11, %r54, %r53; shl.b32 %r12, %r11, 9; add.s32 %r13, %r34, 24; mov.u32 %r55, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r14, %r55, %r41; mov.u32 %r30, 0; mov.f32 %f114, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(115+i*1), %f114; mov.u32 %r86, %r30; $L__BB0_1: bar.sync 0; mad.lo.s32 %r17, %r86, 21632, %r5; mov.u32 %r87, %r30; mov.u32 %r88, %r14; mov.u32 %r89, %r2; mov.u32 %r90, %r13; $L__BB0_2: mov.u32 %r18, %r87; add.s32 %r57, %r90, -24; setp.gt.s32 %p1, %r57, 511; @%p1 bra $L__BB0_6; setp.gt.s32 %p2, %r89, 6655; @%p2 bra $L__BB0_6; add.s32 %r58, %r1, %r18; setp.gt.s32 %p3, %r58, 86527; @%p3 bra $L__BB0_6; shr.s32 %r60, %r57, 7; and.b32 %r61, %r57, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(62+i*1), %r(60+i*1), (43264+i*-43095), %r(17+i*45); mul.wide.s32 %rd9, %r63, 4; add.s64 %rd10, %rd1, %rd9; ld.global.nc.f32 %f33, [%rd10]; st.shared.f32 [%r88], %f33; $L__BB0_6: setp.gt.s32 %p4, %r90, 511; @%p4 bra $L__BB0_10; add.s32 %r64, %r89, 312; setp.gt.s32 %p5, %r64, 6655; @%p5 bra $L__BB0_10; add.s32 %r65, %r1, %r18; add.s32 %r66, %r65, 4056; setp.gt.s32 %p6, %r66, 86527; @%p6 bra $L__BB0_10; shr.s32 %r67, %r90, 7; and.b32 %r68, %r90, 127; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(69+i*1), %r(67+i*1), (43264+i*-43095), %r(17+i*52); mul.wide.s32 %rd11, %r70, 4; add.s64 %rd12, %rd1, %rd11; ld.global.nc.f32 %f34, [%rd12]; st.shared.f32 [%r88+16224], %f34; $L__BB0_10: for.size.1 i in range(0, 2, 1): add.s32 %r(90+i*-1), %r(90+i*-1), (48+i*576); for.size.1 i in range(0, 2, 1): add.s32 %r(88+i*-1), %r(88+i*-70), (32448+i*-24336); setp.ne.s32 %p7, %r18, 81120; @%p7 bra $L__BB0_2; setp.gt.s32 %p8, %r1, 2087; shl.b32 %r26, %r86, 7; add.s32 %r71, %r6, %r26; mul.wide.s32 %rd13, %r71, 4; add.s64 %rd14, %rd2, %rd13; ld.global.nc.f32 %f35, [%rd14]; st.shared.f32 [%r7], %f35; @%p8 bra $L__BB0_13; add.s32 %r72, %r8, %r26; mul.wide.s32 %rd15, %r72, 4; add.s64 %rd16, %rd2, %rd15; ld.global.nc.f32 %f36, [%rd16]; st.shared.f32 [%r7+16224], %f36; $L__BB0_13: bar.sync 0; mov.u32 %r91, 0; $L__BB0_14: mad.lo.s32 %r74, %r91, 169, %r10; shl.b32 %r75, %r74, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(77+i*1), %r(55+i*36), %r(75+i*-63); shl.b32 %r79, %r78, 2; add.s32 %r81, %r42, %r79; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(37+i*1), [%r(81+i*-4)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(39+i*2), %f38, %f(37+i*3), %f(121+i*-1); ld.shared.f32 %f(40+i*2), [%r81+(512+i*512)]; for.size.14 i in range(0, 2, 1): fma.rn.f32 %f(43+i*14), %f(38+i*14), %f(42+i*14), %f(119+i*-76); ld.shared.f32 %f(44+i*14), [%r81+(1536+i*4)]; fma.rn.f32 %f(45+i*14), %f(38+i*14), %f(44+i*14), %f(118+i*-73); ld.shared.f32 %f(46+i*14), [%r77+(86528+i*676)]; fma.rn.f32 %f(47+i*14), %f(46+i*14), %f(37+i*14), %f(117+i*-70); fma.rn.f32 %f(48+i*14), %f(46+i*14), %f(40+i*14), %f(116+i*-68); fma.rn.f32 %f(49+i*14), %f(46+i*14), %f(42+i*14), %f(115+i*-66); fma.rn.f32 %f(50+i*14), %f(46+i*14), %f(44+i*14), %f(114+i*-64); ld.shared.f32 %f(51+i*14), [%r81+(4+i*4)]; ld.shared.f32 %f(52+i*14), [%r77+(676+i*676)]; fma.rn.f32 %f(53+i*14), %f(52+i*14), %f(51+i*14), %f(39+i*14); ld.shared.f32 %f(54+i*14), [%r81+(516+i*4)]; fma.rn.f32 %f(55+i*14), %f(52+i*14), %f(54+i*14), %f(41+i*14); ld.shared.f32 %f(56+i*14), [%r81+(1028+i*4)]; fma.rn.f32 %f71, %f66, %f70, %f57; ld.shared.f32 %f72, [%r81+1544]; fma.rn.f32 %f73, %f66, %f72, %f59; ld.shared.f32 %f74, [%r77+87880]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(75+i*1), %f74, %f(65+i*3), %f(61+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(77+i*1), %f74, %f(70+i*2), %f(63+i*1); ld.shared.f32 %f79, [%r81+12]; ld.shared.f32 %f80, [%r77+2028]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(121+i*-1), %f80, %f(79+i*2), %f(67+i*2); ld.shared.f32 %f(81+i*1), [%r81+(524+i*512)]; fma.rn.f32 %f119, %f80, %f82, %f71; ld.shared.f32 %f83, [%r81+1548]; fma.rn.f32 %f118, %f80, %f83, %f73; ld.shared.f32 %f84, [%r77+88556]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(117+i*-1), %f84, %f(79+i*2), %f(75+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(115+i*-1), %f84, %f(82+i*1), %f(77+i*1); add.s32 %r91, %r91, 4; setp.ne.s32 %p9, %r91, 128; @%p9 bra $L__BB0_14; add.s32 %r86, %r86, 1; setp.lt.u32 %p10, %r86, 2; @%p10 bra $L__BB0_1; shl.b32 %r82, %r11, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(83+i*1), %r(3+i*6), (32448+i*-16224), %r(4+i*79); mad.lo.s32 %r85, %r11, 676, %r84; mul.wide.s32 %rd17, %r82, 4; add.s64 %rd18, %rd4, %rd17; ld.global.nc.f32 %f85, [%rd18]; add.f32 %f86, %f121, %f85; mov.f32 %f87, 0f00000000; max.f32 %f88, %f86, %f87; mul.wide.s32 %rd19, %r85, 4; add.s64 %rd20, %rd3, %rd19; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd20+(0+i*676)], %f(88+i*3); ld.global.nc.f32 %f(89+i*3), [%rd18+(4+i*4)]; add.f32 %f(90+i*3), %f(120+i*-1), %f(89+i*3); max.f32 %f(91+i*3), %f(90+i*3), %f87; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(2028+i*30420)], %f(97+i*2); add.f32 %f(98+i*2), %f(117+i*-1), %f(85+i*4); max.f32 %f(99+i*2), %f(98+i*2), %f87; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(33124+i*676)], %f(101+i*2); add.f32 %f(102+i*2), %f(115+i*-1), %f(92+i*3); max.f32 %f(103+i*2), %f(102+i*2), %f87; st.global.f32 [%rd20+34476], %f105; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[86528]; __shared__ float placeholder_shared[6144]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 22; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 24 + (int)threadIdx.x / 169 < 512) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 312 + (int)threadIdx.x / 13 < 6656) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4056 + (int)threadIdx.x < 86528) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 4056 + (int)threadIdx.x; int index1 = (int)blockIdx.x * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 24 + (int)threadIdx.x / 169 >> 7) * 43264 + rc_outer_outer * 21632 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 24 + (int)threadIdx.x / 169 & 127) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } } } } int index2 = (int)threadIdx.x; int index3 = ((int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x & 127); placeholder_shared[index2] = placeholder1[index3]; if ((int)threadIdx.x < 2088) { int index4 = ((int)threadIdx.x + 4056 >> 7) * 256 + rc_outer_outer * 128 + ((int)threadIdx.x + 88 & 127); placeholder_shared[index2 + 4056] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index5 = (int)threadIdx.x / 2028 * 43264 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index6 = (int)threadIdx.x % 2028 / 169 * 512 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index5] * placeholder_shared[index6]; compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index6 + 128]); compute[2] = compute[2] + (pad_temp_shared[index5] * placeholder_shared[index6 + 256]); compute[3] = compute[3] + (pad_temp_shared[index5] * placeholder_shared[index6 + 384]); compute[4] = compute[4] + (pad_temp_shared[index5 + 21632] * placeholder_shared[index6]); compute[5] = compute[5] + (pad_temp_shared[index5 + 21632] * placeholder_shared[index6 + 128]); compute[6] = compute[6] + (pad_temp_shared[index5 + 21632] * placeholder_shared[index6 + 256]); compute[7] = compute[7] + (pad_temp_shared[index5 + 21632] * placeholder_shared[index6 + 384]); } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index7 = (int)blockIdx.x * 32448 + (int)threadIdx.x / 2028 * 16224 + ax0_inner * 8112 + (int)threadIdx.x % 2028 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index8 = ax0_inner * 4 + ax1_inner; int index9 = (int)threadIdx.x % 2028 / 169 * 4 + ax1_inner; T_relu[index7] = max(compute[index8] + placeholder2[index9], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpmnl0judd
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<453>; .reg .b32 %r<127>; .reg .b64 %rd<35>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[86528]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[3072]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd3, %rd5; mov.u32 %r1, %tid.x; shl.b32 %r28, %r1, 2; mov.u32 %r29, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r29, %r28; add.s32 %r30, %r1, 4056; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(31+i*4), %r(30+i*-29), 1626496491; shr.u32 %r(32+i*4), %r(31+i*4), 31; shr.s32 %r(33+i*4), %r(31+i*4), (10+i*-4); add.s32 %r(34+i*4), %r(33+i*4), %r(32+i*4); mul.lo.s32 %r39, %r38, 169; sub.s32 %r3, %r1, %r39; add.s32 %r40, %r39, 1352; for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(41+i*6), %r(34+i*12), 43264, %r3; add.s32 %r(4+i*1), %r(41+i*6), %r40; add.s32 %r(42+i*6), %r1, (6760+i*2704); mul.hi.s32 %r(43+i*6), %r(42+i*6), 1626496491; shr.u32 %r(44+i*6), %r(43+i*6), 31; shr.s32 %r(45+i*6), %r(43+i*6), 10; add.s32 %r(46+i*6), %r(45+i*6), %r(44+i*6); for.size.7 i in range(0, 4, 1): mad.lo.s32 %r(53+i*6), %r(52+i*6), 43264, %r3; add.s32 %r(6+i*1), %r(53+i*6), %r40; add.s32 %r(54+i*6), %r1, (12168+i*2704); mul.hi.s32 %r(55+i*6), %r(54+i*6), 1626496491; shr.u32 %r(56+i*6), %r(55+i*6), 31; shr.s32 %r(57+i*6), %r(55+i*6), 10; add.s32 %r(58+i*6), %r(57+i*6), %r(56+i*6); mad.lo.s32 %r77, %r76, 43264, %r3; add.s32 %r10, %r77, %r40; shr.u32 %r78, %r1, 4; and.b32 %r79, %r1, 15; prmt.b32 %r11, %r78, %r79, 8452; mov.u32 %r80, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r12, %r80, %r28; shr.s32 %r81, %r35, 7; add.s32 %r13, %r81, %r36; mul.lo.s32 %r82, %r13, 338; sub.s32 %r83, %r1, %r82; mul.hi.s32 %r84, %r83, 1626496491; shr.u32 %r85, %r84, 31; shr.s32 %r86, %r84, 6; add.s32 %r14, %r86, %r85; shl.b32 %r87, %r14, 9; add.s32 %r88, %r80, %r87; add.s32 %r15, %r88, 1280; mad.lo.s32 %r89, %r13, 2704, %r3; shl.b32 %r90, %r89, 2; add.s32 %r16, %r29, %r90; mov.u32 %r123, 0; mov.f32 %f405, 0f00000000; for.size.1 i in range(0, 47, 1): mov.f32 %f(406+i*1), %f405; $L__BB0_1: mov.u32 %r122, %tid.x; bar.sync 0; mul.lo.s32 %r91, %r123, 2704; add.s32 %r92, %r91, %r122; mul.wide.s32 %rd9, %r92, 4; add.s64 %rd10, %rd3, %rd9; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(193+i*1), [%rd10+(0+i*5408)]; st.shared.f32 [%r2+(0+i*5408)], %f(193+i*1); for.size.7 i in range(0, 7, 1): ld.global.nc.f32 %f(195+i*2), [%rd10+(173056+i*173056)]; st.shared.f32 [%r2+(10816+i*10816)], %f(195+i*2); add.s32 %r(93+i*1), %r(4+i*1), %r91; mul.wide.s32 %rd(11+i*2), %r(93+i*1), 4; add.s64 %rd(12+i*2), %rd3, %rd(11+i*2); ld.global.nc.f32 %f(196+i*2), [%rd(12+i*2)+0]; st.shared.f32 [%r2+(16224+i*10816)], %f(196+i*2); setp.gt.s32 %p1, %r122, 767; @%p1 bra $L__BB0_3; shl.b32 %r100, %r123, 4; add.s32 %r101, %r11, %r100; mul.wide.s32 %rd25, %r101, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f209, [%rd26]; st.shared.f32 [%r12], %f209; $L__BB0_3: bar.sync 0; mov.u32 %r124, 1280; mov.u32 %r125, %r16; mov.u32 %r126, %r15; $L__BB0_4: ld.shared.f32 %f210, [%r126+-1280]; ld.shared.f32 %f211, [%r125]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(452+i*-8), %f211, %f(210+i*2), %f(452+i*-8); ld.shared.f32 %f(212+i*1), [%r126+(-256+i*1024)]; fma.rn.f32 %f436, %f211, %f213, %f436; ld.shared.f32 %f214, [%r125+43264]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(428+i*-8), %f214, %f(210+i*2), %f(428+i*-8); for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(412+i*39), %f(214+i*-3), %f(213+i*2), %f(412+i*39); ld.shared.f32 %f(215+i*1), [%r126+(-1216+i*1024)]; for.size.9 i in range(0, 6, 1): fma.rn.f32 %f(443+i*-1), %f211, %f(216+i*3), %f(443+i*-1); ld.shared.f32 %f(217+i*3), [%r126+(832+i*64)]; fma.rn.f32 %f(435+i*-1), %f211, %f(217+i*3), %f(435+i*-1); fma.rn.f32 %f(427+i*-1), %f214, %f(215+i*3), %f(427+i*-1); fma.rn.f32 %f(419+i*-1), %f214, %f(216+i*3), %f(419+i*-1); fma.rn.f32 %f(411+i*-1), %f214, %f(217+i*3), %f(411+i*-1); ld.shared.f32 %f(218+i*3), [%r126+(-1152+i*64)]; fma.rn.f32 %f(450+i*-1), %f211, %f(218+i*3), %f(450+i*-1); ld.shared.f32 %f(219+i*3), [%r126+(-128+i*64)]; fma.rn.f32 %f437, %f211, %f234, %f437; ld.shared.f32 %f235, [%r126+1216]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(429+i*-8), %f(211+i*3), %f(235+i*-2), %f(429+i*-8); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(413+i*-8), %f214, %f(234+i*1), %f(413+i*-8); for.size.1 i in range(0, 2, 1): add.s32 %r(126+i*-1), %r(126+i*-1), (4+i*672); add.s32 %r124, %r124, 4; setp.ne.s32 %p2, %r124, 1344; @%p2 bra $L__BB0_4; add.s32 %r123, %r123, 1; setp.lt.u32 %p3, %r123, 16; @%p3 bra $L__BB0_1; mov.u32 %r121, %tid.x; mul.hi.s32 %r120, %r121, 1626496491; shr.u32 %r119, %r120, 31; ld.param.u64 %rd34, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd33, %rd34; ld.param.u64 %rd32, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd31, %rd32; shr.s32 %r118, %r120, 6; add.s32 %r117, %r118, %r119; mul.lo.s32 %r116, %r117, 169; sub.s32 %r115, %r121, %r116; add.s32 %r114, %r81, %r119; shr.s32 %r113, %r120, 7; add.s32 %r112, %r113, %r119; mul.lo.s32 %r111, %r112, 338; sub.s32 %r110, %r121, %r111; mul.hi.s32 %r109, %r110, 1626496491; shr.u32 %r108, %r109, 31; shr.s32 %r107, %r109, 6; add.s32 %r106, %r107, %r108; shl.b32 %r103, %r106, 3; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(104+i*1), %r(112+i*-6), (8112+i*-6760), %r(115+i*-11); mul.wide.s32 %rd27, %r103, 4; add.s64 %rd28, %rd31, %rd27; ld.global.nc.f32 %f236, [%rd28]; add.f32 %f237, %f452, %f236; mov.f32 %f238, 0f00000000; max.f32 %f239, %f237, %f238; mul.wide.s32 %rd29, %r105, 4; add.s64 %rd30, %rd33, %rd29; for.size.4 i in range(0, 2, 1): st.global.f32 [%rd30+(0+i*10816)], %f(239+i*3); ld.global.nc.f32 %f(240+i*3), [%rd28+(64+i*64)]; add.f32 %f(241+i*3), %f(444+i*-8), %f(240+i*3); max.f32 %f(242+i*3), %f(241+i*3), %f238; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd30+(21632+i*108160)], %f(245+i*2); add.f32 %f(246+i*2), %f(428+i*-8), %f(236+i*4); max.f32 %f(247+i*2), %f(246+i*2), %f238; for.size.21 i in range(0, 7, 1): st.global.f32 [%rd30+(140608+i*676)], %f(249+i*15); add.f32 %f(250+i*15), %f(412+i*-1), %f(243+i*15); max.f32 %f(251+i*15), %f(250+i*15), %f238; st.global.f32 [%rd30+(151424+i*676)], %f(251+i*15); ld.global.nc.f32 %f(252+i*15), [%rd28+(4+i*4)]; add.f32 %f(253+i*15), %f(451+i*-1), %f(252+i*15); max.f32 %f(254+i*15), %f(253+i*15), %f238; st.global.f32 [%rd30+(676+i*676)], %f(254+i*15); ld.global.nc.f32 %f(255+i*15), [%rd28+(68+i*4)]; add.f32 %f(256+i*15), %f(443+i*-1), %f(255+i*15); max.f32 %f(257+i*15), %f(256+i*15), %f238; st.global.f32 [%rd30+(11492+i*676)], %f(257+i*15); ld.global.nc.f32 %f(258+i*15), [%rd28+(132+i*4)]; add.f32 %f(259+i*15), %f(435+i*-1), %f(258+i*15); max.f32 %f(260+i*15), %f(259+i*15), %f238; st.global.f32 [%rd30+(22308+i*676)], %f(260+i*15); add.f32 %f(261+i*15), %f(427+i*-1), %f(252+i*15); max.f32 %f(262+i*15), %f(261+i*15), %f238; st.global.f32 [%rd30+(130468+i*676)], %f(262+i*15); add.f32 %f(263+i*15), %f(419+i*-1), %f(255+i*15); max.f32 %f(264+i*15), %f(263+i*15), %f238; st.global.f32 [%rd30+145340], %f354; add.f32 %f355, %f405, %f348; max.f32 %f356, %f355, %f238; st.global.f32 [%rd30+156156], %f356; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[48]; __shared__ float pad_temp_shared[21632]; __shared__ float placeholder_shared[768]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 8; ++ff_outer_inner_init) { int index0 = ff_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 16] = 0.000000e+00f; compute[index0 + 24] = 0.000000e+00f; compute[index0 + 32] = 0.000000e+00f; compute[index0 + 40] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 2704 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 1352] = placeholder[index2 + 1352]; pad_temp_shared[index1 + 2704] = placeholder[index2 + 43264]; int index3 = ((int)threadIdx.x + 4056) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 4056] = placeholder[index3]; pad_temp_shared[index1 + 5408] = placeholder[index2 + 86528]; int index4 = ((int)threadIdx.x + 6760) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 6760] = placeholder[index4]; pad_temp_shared[index1 + 8112] = placeholder[index2 + 129792]; int index5 = ((int)threadIdx.x + 9464) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 9464] = placeholder[index5]; pad_temp_shared[index1 + 10816] = placeholder[index2 + 173056]; int index6 = ((int)threadIdx.x + 12168) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 12168] = placeholder[index6]; pad_temp_shared[index1 + 13520] = placeholder[index2 + 216320]; int index7 = ((int)threadIdx.x + 14872) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 14872] = placeholder[index7]; pad_temp_shared[index1 + 16224] = placeholder[index2 + 259584]; int index8 = ((int)threadIdx.x + 17576) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 17576] = placeholder[index8]; pad_temp_shared[index1 + 18928] = placeholder[index2 + 302848]; int index9 = ((int)threadIdx.x + 20280) / 2704 * 43264 + rc_outer_outer * 2704 + ((int)threadIdx.x / 169 + 8) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 20280] = placeholder[index9]; if ((int)threadIdx.x < 768) { int index10 = ((int)threadIdx.x >> 4) * 256 + rc_outer_outer * 16 + ((int)threadIdx.x & 15); placeholder_shared[index1] = placeholder1[index10]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 8; ++ff_outer_inner) { int index11 = ff_outer_inner; int index12 = (int)threadIdx.x / 338 * 2704 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index13 = (int)threadIdx.x % 338 / 169 * 128 + ff_outer_inner * 16 + rc_outer_inner; compute[index11] = compute[index11] + pad_temp_shared[index12] * placeholder_shared[index13]; compute[index11 + 8] = compute[index11 + 8] + (pad_temp_shared[index12] * placeholder_shared[index13 + 256]); compute[index11 + 16] = compute[index11 + 16] + (pad_temp_shared[index12] * placeholder_shared[index13 + 512]); compute[index11 + 24] = compute[index11 + 24] + (pad_temp_shared[index12 + 10816] * placeholder_shared[index13]); compute[index11 + 32] = compute[index11 + 32] + (pad_temp_shared[index12 + 10816] * placeholder_shared[index13 + 256]); compute[index11 + 40] = compute[index11 + 40] + (pad_temp_shared[index12 + 10816] * placeholder_shared[index13 + 512]); } } } for (int ax1_inner = 0; ax1_inner < 8; ++ax1_inner) { int index14 = (int)threadIdx.x / 338 * 8112 + (int)threadIdx.x % 338 / 169 * 1352 + ax1_inner * 169 + (int)threadIdx.x % 169; int index15 = ax1_inner; int index16 = (int)threadIdx.x % 338 / 169 * 8 + ax1_inner; T_relu[index14] = max(compute[index15] + placeholder2[index16], 0.000000e+00f); T_relu[index14 + 2704] = max(compute[index15 + 8] + placeholder2[index16 + 16], 0.000000e+00f); T_relu[index14 + 5408] = max(compute[index15 + 16] + placeholder2[index16 + 32], 0.000000e+00f); T_relu[index14 + 32448] = max(compute[index15 + 24] + placeholder2[index16], 0.000000e+00f); T_relu[index14 + 35152] = max(compute[index15 + 32] + placeholder2[index16 + 16], 0.000000e+00f); T_relu[index14 + 37856] = max(compute[index15 + 40] + placeholder2[index16 + 32], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp0hmm0kbl
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<103>; .reg .b32 %r<59>; .reg .b64 %rd<24>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[832]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r14, %ctaid.x; mul.hi.s32 %r15, %r14, -770891565; add.s32 %r16, %r15, %r14; shr.u32 %r17, %r16, 31; shr.s32 %r18, %r16, 5; add.s32 %r1, %r18, %r17; mov.u32 %r2, %tid.x; for.size.6 i in range(0, 2, 1): mul.hi.s32 %r(19+i*6), %r(2+i*12), 1321528399; shr.u32 %r(20+i*6), %r(19+i*6), 31; shr.s32 %r(21+i*6), %r(19+i*6), (4+i*-2); add.s32 %r(22+i*6), %r(21+i*6), %r(20+i*6); mul.lo.s32 %r(23+i*6), %r(22+i*6), (52+i*-39); sub.s32 %r(24+i*-21), %r(2+i*12), %r(23+i*6); shl.b32 %r30, %r2, 2; mov.u32 %r31, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r31, %r30; mul.lo.s32 %r32, %r1, 39; sub.s32 %r33, %r14, %r32; mul.hi.s32 %r34, %r33, 1321528399; shr.u32 %r35, %r34, 31; shr.s32 %r36, %r34, 2; add.s32 %r5, %r36, %r35; shl.b32 %r37, %r5, 12; and.b32 %r38, %r2, 3; mov.u32 %r39, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r6, %r39, %r30; shr.s32 %r40, %r19, 2; add.s32 %r9, %r40, %r20; mul.lo.s32 %r41, %r9, 13; sub.s32 %r7, %r2, %r41; for.size.2 i in range(0, 2, 1): shl.b32 %r(42+i*1), %r(7+i*2), (2+i*3); add.s32 %r(8+i*2), %r(31+i*8), %r(42+i*1); shl.b32 %r44, %r2, 6; and.b32 %r45, %r44, -256; add.s32 %r46, %r37, %r45; or.b32 %r47, %r46, %r38; mul.wide.s32 %rd13, %r47, 4; add.s64 %rd23, %rd12, %rd13; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(48+i*1), %r(24+i*-23), (13+i*173043), %r(3+i*45); mad.lo.s32 %r50, %r22, 43264, %r49; cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r50, 4; add.s64 %rd16, %rd14, %rd15; add.s64 %rd22, %rd16, 346112; mov.u32 %r58, 0; mov.f32 %f95, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(96+i*1), %f95; $L__BB0_1: bar.sync 0; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(25+i*1), [%rd22+(-346112+i*346112)]; st.shared.f32 [%r4+(0+i*416)], %f(25+i*1); setp.gt.s32 %p1, %r2, 63; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f27, [%rd23]; st.shared.f32 [%r6], %f27; $L__BB0_3: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(28+i*1), [%r(10+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(30+i*2), %f29, %f(28+i*3), %f(102+i*-1); ld.shared.f32 %f(31+i*2), [%r10+(16+i*-12)]; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(34+i*5), [%r8+(52+i*52)]; fma.rn.f32 %f(35+i*5), %f(34+i*5), %f(33+i*5), %f(30+i*5); ld.shared.f32 %f(36+i*5), [%r10+(20+i*4)]; fma.rn.f32 %f(37+i*5), %f(34+i*5), %f(36+i*5), %f(32+i*5); ld.shared.f32 %f(38+i*5), [%r10+(8+i*4)]; ld.shared.f32 %f44, [%r8+156]; fma.rn.f32 %f102, %f44, %f43, %f40; ld.shared.f32 %f45, [%r10+28]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(101+i*-53), %f(44+i*2), %f(45+i*-14), %f(42+i*57); ld.shared.f32 %f(46+i*3), [%r8+(208+i*52)]; fma.rn.f32 %f(47+i*3), %f(46+i*3), %f(28+i*5), %f(100+i*-53); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(51+i*3), %f(49+i*3), %f(36+i*5), %f(48+i*3); ld.shared.f32 %f(52+i*3), [%r8+(312+i*52)]; fma.rn.f32 %f(53+i*47), %f(52+i*3), %f(38+i*5), %f(50+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(99+i*-41), %f(55+i*1), %f(45+i*-14), %f(54+i*43); ld.shared.f32 %f(56+i*3), [%r8+(416+i*52)]; fma.rn.f32 %f(57+i*3), %f(56+i*3), %f(28+i*5), %f(98+i*-41); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(61+i*3), %f(59+i*3), %f(36+i*5), %f(58+i*3); ld.shared.f32 %f(62+i*3), [%r8+(520+i*52)]; fma.rn.f32 %f(63+i*35), %f(62+i*3), %f(38+i*5), %f(60+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(97+i*-29), %f(65+i*1), %f(45+i*-14), %f(64+i*31); ld.shared.f32 %f(66+i*3), [%r8+(624+i*52)]; fma.rn.f32 %f(67+i*3), %f(66+i*3), %f(28+i*5), %f(96+i*-29); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(71+i*3), %f(69+i*3), %f(36+i*5), %f(68+i*3); ld.shared.f32 %f(72+i*3), [%r8+(728+i*52)]; fma.rn.f32 %f(73+i*23), %f(72+i*3), %f(38+i*5), %f(70+i*3); fma.rn.f32 %f95, %f75, %f45, %f74; for.size.1 i in range(0, 2, 1): add.s64 %rd(23+i*-1), %rd(23+i*-1), (16+i*2688); add.s32 %r58, %r58, 1; setp.ne.s32 %p2, %r58, 64; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd8; for.size.1 i in range(0, 2, 1): shl.b32 %r(51+i*1), %r(9+i*-4), (1+i*3); add.s32 %r53, %r52, %r51; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(54+i*1), %r(1+i*8), (32448+i*-32110), %r(3+i*51); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(56+i*1), %r(7+i*-2), (13+i*2691), %r(55+i*1); mul.wide.s32 %rd18, %r53, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f76, [%rd19]; add.f32 %f77, %f102, %f76; mov.f32 %f78, 0f00000000; max.f32 %f79, %f77, %f78; mul.wide.s32 %rd20, %r57, 4; add.s64 %rd21, %rd1, %rd20; st.global.f32 [%rd21], %f79; ld.global.nc.f32 %f80, [%rd19+4]; for.size.4 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): add.f32 %f(81+i*4+j*2), %f(101+i*-2+j*-1), %f(80+j*-4); max.f32 %f(82+i*4+j*2), %f(81+i*4+j*2), %f78; st.global.f32 [%rd21+(676+i*32448+j*31772)], %f(82+i*4+j*2); add.f32 %f93, %f95, %f80; max.f32 %f94, %f93, %f78; st.global.f32 [%rd21+98020], %f94; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[208]; __shared__ float placeholder_shared[64]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x / 39 * 173056 + (int)threadIdx.x / 52 * 43264 + rc_outer_outer * 676 + (int)threadIdx.x % 52 * 13 + (int)blockIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 104] = placeholder[index1 + 86528]; if ((int)threadIdx.x < 64) { int index2 = (int)blockIdx.x % 39 / 13 * 4096 + ((int)threadIdx.x >> 2) * 256 + rc_outer_outer * 4 + ((int)threadIdx.x & 3); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index3 = (int)threadIdx.x / 13 * 8; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index3]; compute[1] = compute[1] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 4]); int index4 = (int)threadIdx.x % 13 + 13; compute[0] = compute[0] + (pad_temp_shared[index4] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index3 + 5]); int index5 = (int)threadIdx.x % 13 + 26; compute[0] = compute[0] + (pad_temp_shared[index5] * placeholder_shared[index3 + 2]); compute[1] = compute[1] + (pad_temp_shared[index5] * placeholder_shared[index3 + 6]); int index6 = (int)threadIdx.x % 13 + 39; compute[0] = compute[0] + (pad_temp_shared[index6] * placeholder_shared[index3 + 3]); compute[1] = compute[1] + (pad_temp_shared[index6] * placeholder_shared[index3 + 7]); int index7 = (int)threadIdx.x % 13 + 52; compute[2] = compute[2] + pad_temp_shared[index7] * placeholder_shared[index3]; compute[3] = compute[3] + (pad_temp_shared[index7] * placeholder_shared[index3 + 4]); int index8 = (int)threadIdx.x % 13 + 65; compute[2] = compute[2] + (pad_temp_shared[index8] * placeholder_shared[index3 + 1]); compute[3] = compute[3] + (pad_temp_shared[index8] * placeholder_shared[index3 + 5]); int index9 = (int)threadIdx.x % 13 + 78; compute[2] = compute[2] + (pad_temp_shared[index9] * placeholder_shared[index3 + 2]); compute[3] = compute[3] + (pad_temp_shared[index9] * placeholder_shared[index3 + 6]); int index10 = (int)threadIdx.x % 13 + 91; compute[2] = compute[2] + (pad_temp_shared[index10] * placeholder_shared[index3 + 3]); compute[3] = compute[3] + (pad_temp_shared[index10] * placeholder_shared[index3 + 7]); int index11 = (int)threadIdx.x % 13 + 104; compute[4] = compute[4] + pad_temp_shared[index11] * placeholder_shared[index3]; compute[5] = compute[5] + (pad_temp_shared[index11] * placeholder_shared[index3 + 4]); int index12 = (int)threadIdx.x % 13 + 117; compute[4] = compute[4] + (pad_temp_shared[index12] * placeholder_shared[index3 + 1]); compute[5] = compute[5] + (pad_temp_shared[index12] * placeholder_shared[index3 + 5]); int index13 = (int)threadIdx.x % 13 + 130; compute[4] = compute[4] + (pad_temp_shared[index13] * placeholder_shared[index3 + 2]); compute[5] = compute[5] + (pad_temp_shared[index13] * placeholder_shared[index3 + 6]); int index14 = (int)threadIdx.x % 13 + 143; compute[4] = compute[4] + (pad_temp_shared[index14] * placeholder_shared[index3 + 3]); compute[5] = compute[5] + (pad_temp_shared[index14] * placeholder_shared[index3 + 7]); int index15 = (int)threadIdx.x % 13 + 156; compute[6] = compute[6] + pad_temp_shared[index15] * placeholder_shared[index3]; compute[7] = compute[7] + (pad_temp_shared[index15] * placeholder_shared[index3 + 4]); int index16 = (int)threadIdx.x % 13 + 169; compute[6] = compute[6] + (pad_temp_shared[index16] * placeholder_shared[index3 + 1]); compute[7] = compute[7] + (pad_temp_shared[index16] * placeholder_shared[index3 + 5]); int index17 = (int)threadIdx.x % 13 + 182; compute[6] = compute[6] + (pad_temp_shared[index17] * placeholder_shared[index3 + 2]); compute[7] = compute[7] + (pad_temp_shared[index17] * placeholder_shared[index3 + 6]); int index18 = (int)threadIdx.x % 13 + 195; compute[6] = compute[6] + (pad_temp_shared[index18] * placeholder_shared[index3 + 3]); compute[7] = compute[7] + (pad_temp_shared[index18] * placeholder_shared[index3 + 7]); } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index19 = (int)blockIdx.x / 39 * 32448 + ax0_inner * 8112 + (int)blockIdx.x % 39 / 13 * 2704 + (int)threadIdx.x / 13 * 338 + ax1_inner * 169 + (int)threadIdx.x % 13 * 13 + (int)blockIdx.x % 13; int index20 = ax0_inner * 2 + ax1_inner; int index21 = (int)blockIdx.x % 39 / 13 * 16 + (int)threadIdx.x / 13 * 2 + ax1_inner; T_relu[index19] = max(compute[index20] + placeholder2[index21], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpiokvumct
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<183>; .reg .b32 %r<200>; .reg .b64 %rd<47>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[43264]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[1536]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r37, %r1, 4; mul.hi.s32 %r38, %r37, 1626496491; shr.u32 %r39, %r38, 31; shr.s32 %r40, %r38, 9; add.s32 %r41, %r40, %r39; mul.lo.s32 %r42, %r41, 1352; sub.s32 %r43, %r37, %r42; mad.lo.s32 %r2, %r41, 43264, %r43; shl.b32 %r44, %r1, 6; mov.u32 %r45, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r45, %r44; for.size.8 i in range(0, 15, 1): or.b32 %r(46+i*7), %r37, (1+i*1); mul.hi.s32 %r(47+i*7), %r(46+i*7), 1626496491; shr.u32 %r(48+i*7), %r(47+i*7), 31; shr.s32 %r(49+i*7), %r(47+i*7), 9; add.s32 %r(50+i*7), %r(49+i*7), %r(48+i*7); mul.lo.s32 %r(51+i*7), %r(50+i*7), 1352; sub.s32 %r(52+i*7), %r(46+i*7), %r(51+i*7); mad.lo.s32 %r(4+i*1), %r(50+i*7), 43264, %r(52+i*7); shr.u32 %r151, %r1, 3; and.b32 %r152, %r1, 7; prmt.b32 %r19, %r151, %r152, 8452; shl.b32 %r153, %r1, 2; mov.u32 %r154, _ZZ24default_function_kernel0E18placeholder_shared; for.size.4 i in range(0, 2, 1): add.s32 %r(20+i*1), %r(154+i*3), %r(153+i*3); mul.hi.s32 %r(155+i*3), %r1, (542165497+i*1084330994); shr.u32 %r(156+i*3), %r(155+i*3), 31; shr.s32 %r(157+i*3), %r(155+i*3), (8+i*-2); add.s32 %r161, %r160, %r159; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(162+i*1), %r(161+i*-140), (169+i*1859); sub.s32 %r(22+i*142), %r1, %r(162+i*1); mul.hi.s32 %r165, %r164, 1626496491; shr.u32 %r166, %r165, 31; shr.s32 %r167, %r165, 6; add.s32 %r23, %r167, %r166; shl.b32 %r168, %r23, 7; add.s32 %r169, %r154, %r168; add.s32 %r24, %r169, 64; mad.lo.s32 %r170, %r21, 5408, %r22; shl.b32 %r171, %r170, 2; add.s32 %r172, %r45, %r171; add.s32 %r25, %r172, 10816; mov.u32 %r196, 0; mov.f32 %f167, 0f00000000; for.size.1 i in range(0, 15, 1): mov.f32 %f(168+i*1), %f167; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r1, 675; @%p1 bra $L__BB0_3; mul.lo.s32 %r173, %r196, 1352; for.size.5 i in range(0, 2, 1): add.s32 %r(174+i*1), %r(2+i*2), %r173; mul.wide.s32 %rd(9+i*2), %r(174+i*1), 4; add.s64 %rd(10+i*2), %rd2, %rd(9+i*2); ld.global.nc.f32 %f(65+i*1), [%rd(10+i*2)+0]; st.shared.f32 [%r3+(0+i*4)], %f(65+i*1); for.size.5 i in range(0, 14, 1): add.s32 %r(176+i*1), %r(5+i*1), %r173; mul.wide.s32 %rd(13+i*2), %r(176+i*1), 4; add.s64 %rd(14+i*2), %rd2, %rd(13+i*2); ld.global.nc.f32 %f(67+i*1), [%rd(14+i*2)+0]; st.shared.f32 [%r3+(8+i*4)], %f(67+i*1); $L__BB0_3: setp.gt.s32 %p2, %r1, 383; @%p2 bra $L__BB0_5; shl.b32 %r190, %r196, 3; add.s32 %r191, %r19, %r190; mul.wide.s32 %rd41, %r191, 4; add.s64 %rd42, %rd1, %rd41; ld.global.nc.f32 %f81, [%rd42]; st.shared.f32 [%r20], %f81; $L__BB0_5: bar.sync 0; mov.u32 %r199, 64; mov.u32 %r197, %r25; mov.u32 %r198, %r24; $L__BB0_6: for.size.24 i in range(0, 2, 1): ld.shared.f32 %f(82+i*24), [%r198+(-64+i*4)]; ld.shared.f32 %f(83+i*24), [%r197+(-10816+i*676)]; fma.rn.f32 %f(84+i*98), %f(83+i*24), %f(82+i*24), %f(182+i*-98); ld.shared.f32 %f(85+i*23), [%r198+(-32+i*4)]; fma.rn.f32 %f(86+i*95), %f(83+i*24), %f(85+i*23), %f(181+i*-95); ld.shared.f32 %f(87+i*22), [%r198+(0+i*4)]; fma.rn.f32 %f(88+i*92), %f(83+i*24), %f(87+i*22), %f(180+i*-92); ld.shared.f32 %f(89+i*21), [%r198+(32+i*4)]; fma.rn.f32 %f(90+i*89), %f(83+i*24), %f(89+i*21), %f(179+i*-89); ld.shared.f32 %f(91+i*20), [%r197+(-5408+i*676)]; fma.rn.f32 %f(92+i*86), %f(91+i*20), %f(82+i*24), %f(178+i*-86); fma.rn.f32 %f(93+i*84), %f(91+i*20), %f(85+i*23), %f(177+i*-84); fma.rn.f32 %f(94+i*82), %f(91+i*20), %f(87+i*22), %f(176+i*-82); fma.rn.f32 %f(95+i*80), %f(91+i*20), %f(89+i*21), %f(175+i*-80); ld.shared.f32 %f(96+i*16), [%r197+(0+i*676)]; fma.rn.f32 %f(97+i*77), %f(96+i*16), %f(82+i*24), %f(174+i*-77); fma.rn.f32 %f(98+i*75), %f(96+i*16), %f(85+i*23), %f(173+i*-75); fma.rn.f32 %f(99+i*73), %f(96+i*16), %f(87+i*22), %f(172+i*-73); fma.rn.f32 %f(100+i*71), %f(96+i*16), %f(89+i*21), %f(171+i*-71); ld.shared.f32 %f(101+i*12), [%r197+(5408+i*676)]; fma.rn.f32 %f(102+i*68), %f(101+i*12), %f(82+i*24), %f(170+i*-68); fma.rn.f32 %f(103+i*66), %f(101+i*12), %f(85+i*23), %f(169+i*-66); fma.rn.f32 %f(104+i*64), %f(101+i*12), %f(87+i*22), %f(168+i*-64); fma.rn.f32 %f(105+i*62), %f(101+i*12), %f(89+i*21), %f(167+i*-62); for.size.1 i in range(0, 2, 1): add.s32 %r(198+i*-1), %r(198+i*-1), (8+i*1344); add.s32 %r199, %r199, 8; setp.ne.s32 %p3, %r199, 96; @%p3 bra $L__BB0_6; add.s32 %r196, %r196, 1; setp.lt.u32 %p4, %r196, 32; @%p4 bra $L__BB0_1; shl.b32 %r193, %r23, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(194+i*1), %r(21+i*2), (32448+i*-31772), %r(22+i*172); mul.wide.s32 %rd43, %r193, 4; add.s64 %rd44, %rd4, %rd43; ld.global.nc.f32 %f114, [%rd44]; add.f32 %f115, %f182, %f114; mov.f32 %f116, 0f00000000; max.f32 %f117, %f115, %f116; mul.wide.s32 %rd45, %r195, 4; add.s64 %rd46, %rd3, %rd45; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd46+(0+i*676)], %f(117+i*3); ld.global.nc.f32 %f(118+i*3), [%rd44+(4+i*4)]; add.f32 %f(119+i*3), %f(181+i*-1), %f(118+i*3); max.f32 %f(120+i*3), %f(119+i*3), %f116; for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd46+(2028+i*32448+j*30420)], %f(126+i*8+j*2); add.f32 %f(127+i*8+j*2), %f(178+i*-4+j*-1), %f(114+j*4); max.f32 %f(128+i*8+j*2), %f(127+i*8+j*2), %f116; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd46+(33124+i*32448+j*676)], %f(130+i*8+j*2); add.f32 %f(131+i*8+j*2), %f(176+i*-4+j*-1), %f(121+j*3); max.f32 %f(132+i*8+j*2), %f(131+i*8+j*2), %f116; st.global.f32 [%rd46+99372], %f150; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16]; __shared__ float pad_temp_shared[10816]; __shared__ float placeholder_shared[384]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[15] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index1 = (int)threadIdx.x * 16 / 1352 * 43264 + rc_outer_outer * 1352 + (int)threadIdx.x * 16 % 1352; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index2 = ((int)threadIdx.x * 16 + 1) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 1) % 1352; pad_temp_shared[index0 + 1] = placeholder[index2]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index3 = ((int)threadIdx.x * 16 + 2) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 2) % 1352; pad_temp_shared[index0 + 2] = placeholder[index3]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index4 = ((int)threadIdx.x * 16 + 3) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 3) % 1352; pad_temp_shared[index0 + 3] = placeholder[index4]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index5 = ((int)threadIdx.x * 16 + 4) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 4) % 1352; pad_temp_shared[index0 + 4] = placeholder[index5]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index6 = ((int)threadIdx.x * 16 + 5) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 5) % 1352; pad_temp_shared[index0 + 5] = placeholder[index6]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index7 = ((int)threadIdx.x * 16 + 6) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 6) % 1352; pad_temp_shared[index0 + 6] = placeholder[index7]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index8 = ((int)threadIdx.x * 16 + 7) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 7) % 1352; pad_temp_shared[index0 + 7] = placeholder[index8]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index9 = ((int)threadIdx.x * 16 + 8) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 8) % 1352; pad_temp_shared[index0 + 8] = placeholder[index9]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index10 = ((int)threadIdx.x * 16 + 9) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 9) % 1352; pad_temp_shared[index0 + 9] = placeholder[index10]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index11 = ((int)threadIdx.x * 16 + 10) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 10) % 1352; pad_temp_shared[index0 + 10] = placeholder[index11]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index12 = ((int)threadIdx.x * 16 + 11) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 11) % 1352; pad_temp_shared[index0 + 11] = placeholder[index12]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index13 = ((int)threadIdx.x * 16 + 12) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 12) % 1352; pad_temp_shared[index0 + 12] = placeholder[index13]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index14 = ((int)threadIdx.x * 16 + 13) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 13) % 1352; pad_temp_shared[index0 + 13] = placeholder[index14]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index15 = ((int)threadIdx.x * 16 + 14) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 14) % 1352; pad_temp_shared[index0 + 14] = placeholder[index15]; } if ((int)threadIdx.x < 676) { int index0 = (int)threadIdx.x * 16; int index16 = ((int)threadIdx.x * 16 + 15) / 1352 * 43264 + rc_outer_outer * 1352 + ((int)threadIdx.x * 16 + 15) % 1352; pad_temp_shared[index0 + 15] = placeholder[index16]; } if ((int)threadIdx.x < 384) { int index17 = (int)threadIdx.x; int index18 = ((int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index17] = placeholder1[index18]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { int index19 = (int)threadIdx.x / 2028 * 5408 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index20 = (int)threadIdx.x % 2028 / 169 * 32 + rc_outer_inner; compute[0] = compute[0] + pad_temp_shared[index19] * placeholder_shared[index20]; compute[1] = compute[1] + (pad_temp_shared[index19] * placeholder_shared[index20 + 8]); compute[2] = compute[2] + (pad_temp_shared[index19] * placeholder_shared[index20 + 16]); compute[3] = compute[3] + (pad_temp_shared[index19] * placeholder_shared[index20 + 24]); compute[4] = compute[4] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20]); compute[5] = compute[5] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20 + 8]); compute[6] = compute[6] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20 + 16]); compute[7] = compute[7] + (pad_temp_shared[index19 + 1352] * placeholder_shared[index20 + 24]); compute[8] = compute[8] + (pad_temp_shared[index19 + 2704] * placeholder_shared[index20]); compute[9] = compute[9] + (pad_temp_shared[index19 + 2704] * placeholder_shared[index20 + 8]); compute[10] = compute[10] + (pad_temp_shared[index19 + 2704] * placeholder_shared[index20 + 16]); compute[11] = compute[11] + (pad_temp_shared[index19 + 2704] * placeholder_shared[index20 + 24]); compute[12] = compute[12] + (pad_temp_shared[index19 + 4056] * placeholder_shared[index20]); compute[13] = compute[13] + (pad_temp_shared[index19 + 4056] * placeholder_shared[index20 + 8]); compute[14] = compute[14] + (pad_temp_shared[index19 + 4056] * placeholder_shared[index20 + 16]); compute[15] = compute[15] + (pad_temp_shared[index19 + 4056] * placeholder_shared[index20 + 24]); } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index21 = (int)threadIdx.x / 2028 * 32448 + ax0_inner * 8112 + (int)threadIdx.x % 2028 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index22 = ax0_inner * 4 + ax1_inner; int index23 = (int)threadIdx.x % 2028 / 169 * 4 + ax1_inner; T_relu[index21] = max(compute[index22] + placeholder2[index23], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpf79svsfc
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[4056]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<9>; .reg .f32 %f<612>; .reg .b32 %r<99>; .reg .b64 %rd<52>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[21632]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[768]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd26, [default_function_kernel0_param_0]; ld.param.u64 %rd24, [default_function_kernel0_param_1]; ld.param.u64 %rd27, [default_function_kernel0_param_2]; ld.param.u64 %rd25, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd26; cvta.to.global.u64 %rd2, %rd27; add.u64 %rd50, %SPL, 0; mov.u32 %r34, 0; mov.u32 %r89, %r34; $L__BB0_1: mul.lo.s32 %r35, %r89, 13; mul.wide.s32 %rd29, %r35, 4; add.s64 %rd30, %rd50, %rd29; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 26, 1): st.local.u32 [%rd30+(0+i*4+j*156)], %r34; add.s32 %r89, %r89, 1; setp.ne.s32 %p1, %r89, 3; @%p1 bra $L__BB0_1; cvta.to.global.u64 %rd4, %rd24; cvta.to.global.u64 %rd5, %rd25; mov.u32 %r38, %ctaid.x; shr.s32 %r3, %r38, 1; mul.lo.s32 %r4, %r3, 173056; and.b32 %r5, %r38, 1; mov.u32 %r39, %tid.x; shr.u32 %r40, %r39, 3; and.b32 %r41, %r39, 7; prmt.b32 %r42, %r40, %r41, 8452; mad.lo.s32 %r6, %r5, 6144, %r42; shl.b32 %r43, %r39, 2; mov.u32 %r44, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r44, %r43; shr.s32 %r8, %r39, 2; mul.lo.s32 %r9, %r8, 1352; and.b32 %r10, %r39, 3; mul.lo.s32 %r11, %r10, 24; add.s32 %r12, %r39, 16; mov.u32 %r45, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r13, %r45, %r43; mov.u32 %r37, 0; mov.u32 %r90, %r37; $L__BB0_3: bar.sync 0; mad.lo.s32 %r15, %r90, 1352, %r4; mov.u32 %r93, 64; mov.u32 %r91, %r37; mov.u32 %r92, %r12; $L__BB0_4: add.s32 %r48, %r92, -16; mul.hi.s32 %r49, %r48, 1626496491; shr.u32 %r50, %r49, 31; shr.s32 %r51, %r49, 9; add.s32 %r52, %r51, %r50; mul.lo.s32 %r53, %r52, 1352; sub.s32 %r54, %r48, %r53; add.s32 %r55, %r15, %r54; mad.lo.s32 %r56, %r52, 43264, %r55; mul.wide.s32 %rd31, %r56, 4; add.s64 %rd32, %rd1, %rd31; ld.global.nc.f32 %f23, [%rd32]; add.s32 %r57, %r13, %r91; st.shared.f32 [%r57], %f23; mul.hi.s32 %r58, %r92, 1626496491; shr.u32 %r59, %r58, 31; shr.s32 %r60, %r58, 9; add.s32 %r61, %r60, %r59; mul.lo.s32 %r62, %r61, 1352; sub.s32 %r63, %r92, %r62; add.s32 %r64, %r15, %r63; mad.lo.s32 %r65, %r61, 43264, %r64; mul.wide.s32 %rd33, %r65, 4; add.s64 %rd34, %rd1, %rd33; ld.global.nc.f32 %f24, [%rd34]; st.shared.f32 [%r57+64], %f24; for.size.1 i in range(0, 2, 1): add.s32 %r(92+i*-1), %r(92+i*-1), (32+i*96); add.s32 %r93, %r93, 128; setp.ne.s32 %p2, %r93, 21696; @%p2 bra $L__BB0_4; shl.b32 %r67, %r90, 3; add.s32 %r68, %r6, %r67; mul.wide.s32 %rd35, %r68, 4; add.s64 %rd36, %rd4, %rd35; ld.global.nc.f32 %f25, [%rd36]; mov.u32 %r66, 0; for.size.2 i in range(0, 11, 1): st.shared.f32 [%r7+(0+i*64)], %f(25+i*1); ld.global.nc.f32 %f(26+i*1), [%rd36+(2048+i*2048)]; st.shared.f32 [%r7+704], %f36; bar.sync 0; mov.u32 %r94, %r66; $L__BB0_6: mul.lo.s32 %r23, %r94, 13; for.size.2 i in range(0, 2, 1): shl.b32 %r(70+i*2), %r(94+i*-23), (3+i*-1); add.s32 %r(71+i*3), %r(70+i*-26), %r(11+i*61); for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 2, 1): ld.shared.f32 %f(1+i*2+j*1), [%r74+(0+i*4+j*384)]; mov.u32 %r95, %r66; $L__BB0_7: add.s32 %r75, %r95, %r23; mul.wide.s32 %rd37, %r75, 4; add.s64 %rd38, %rd50, %rd37; add.s32 %r76, %r95, %r9; shl.b32 %r77, %r76, 2; add.s32 %r79, %r45, %r77; for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(37+i*3), [%r79+(0+i*52)]; ld.local.f32 %f(38+i*3), [%rd38+(0+i*156)]; fma.rn.f32 %f(39+i*3), %f(37+i*3), %f1, %f(38+i*3); for.size.2 i in range(0, 13, 1): ld.local.f32 %f(76+i*2), [%rd38+(2028+i*156)]; fma.rn.f32 %f(77+i*2), %f(37+i*3), %f2, %f(76+i*2); for.size.2 i in range(0, 13, 1): ld.shared.f32 %f(102+i*2), [%r79+(676+i*52)]; fma.rn.f32 %f(103+i*2), %f(102+i*2), %f3, %f(39+i*3); for.size.1 i in range(0, 13, 1): fma.rn.f32 %f(128+i*1), %f(102+i*2), %f4, %f(77+i*2); for.size.5 i in range(0, 5, 1): for.size.2 j in range(0, 13, 1): ld.shared.f32 %f(141+i*39+j*2), [%r79+(1352+i*676+j*52)]; fma.rn.f32 %f(142+i*39+j*2), %f(141+i*39+j*2), %f(5+j*2), %f(103+i*39+j*2); for.size.1 j in range(0, 13, 1): fma.rn.f32 %f(167+i*39+j*1), %f(141+i*39+j*2), %f(6+j*2), %f(128+i*39+j*1); for.size.3 i in range(0, 13, 1): ld.shared.f32 %f(336+i*2), [%r79+(4732+i*52)]; fma.rn.f32 %f(337+i*2), %f(336+i*2), %f15, %f(298+i*2); st.local.f32 [%rd38+(0+i*156)], %f(337+i*2); for.size.2 i in range(0, 13, 1): fma.rn.f32 %f(362+i*1), %f(336+i*2), %f16, %f(323+i*1); st.local.f32 [%rd38+(2028+i*156)], %f(362+i*1); add.s32 %r95, %r95, 1; setp.ne.s32 %p3, %r95, 13; @%p3 bra $L__BB0_7; add.s32 %r94, %r94, 1; setp.lt.u32 %p4, %r94, 3; @%p4 bra $L__BB0_6; add.s32 %r90, %r90, 1; setp.lt.u32 %p5, %r90, 32; @%p5 bra $L__BB0_3; mul.lo.s32 %r81, %r5, 24; mad.lo.s32 %r82, %r10, 3, %r81; mul.wide.u32 %rd39, %r82, 4; add.s64 %rd6, %rd5, %rd39; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(17+i*1), [%rd6+(0+i*48)]; mul.lo.s32 %r83, %r10, 507; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(84+i*1), %r(5+i*-2), (4056+i*28392), %r(83+i*1); mad.lo.s32 %r86, %r8, 8112, %r85; cvt.s64.s32 %rd7, %r86; mul.wide.s32 %rd40, %r86, 4; add.s64 %rd41, %rd2, %rd40; add.s64 %rd47, %rd41, 8112; mov.u32 %r96, 0; mov.u64 %rd46, %rd50; $L__BB0_11: ld.local.f32 %f375, [%rd46]; add.f32 %f376, %f375, %f17; mov.f32 %f377, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(378+i*3), %f(376+i*4), %f377; st.global.f32 [%rd47+(-8112+i*52)], %f(378+i*3); ld.local.f32 %f(379+i*3), [%rd46+(156+i*156)]; add.f32 %f(380+i*3), %f(379+i*3), %f17; for.size.4 i in range(0, 10, 1): max.f32 %f(384+i*3), %f(383+i*3), %f377; st.global.f32 [%rd47+(-8008+i*52)], %f(384+i*3); ld.local.f32 %f(385+i*3), [%rd46+(468+i*156)]; add.f32 %f(386+i*3), %f(385+i*3), %f17; for.size.4 i in range(0, 2, 1): max.f32 %f(414+i*3), %f(413+i*3), %f377; st.global.f32 [%rd47+(-7488+i*7488)], %f(414+i*3); ld.local.f32 %f(415+i*3), [%rd46+(2028+i*156)]; add.f32 %f(416+i*3), %f(415+i*3), %f18; for.size.4 i in range(0, 11, 1): max.f32 %f(420+i*3), %f(419+i*3), %f377; st.global.f32 [%rd47+(52+i*52)], %f(420+i*3); ld.local.f32 %f(421+i*3), [%rd46+(2340+i*156)]; add.f32 %f(422+i*3), %f(421+i*3), %f18; max.f32 %f453, %f452, %f377; st.global.f32 [%rd47+624], %f453; for.size.1 i in range(0, 2, 1): add.s64 %rd(47+i*-1), %rd(47+i*-1), 4; add.s32 %r96, %r96, 1; setp.ne.s32 %p6, %r96, 13; @%p6 bra $L__BB0_11; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(19+i*1), [%rd6+(4+i*48)]; shl.b64 %rd42, %rd7, 2; add.s64 %rd43, %rd2, %rd42; add.s64 %rd49, %rd43, 8788; mov.u32 %r97, 0; mov.u64 %rd48, %rd50; $L__BB0_13: ld.local.f32 %f454, [%rd48+52]; add.f32 %f455, %f454, %f19; mov.f32 %f456, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(457+i*3), %f(455+i*4), %f456; st.global.f32 [%rd49+(-8112+i*52)], %f(457+i*3); ld.local.f32 %f(458+i*3), [%rd48+(208+i*156)]; add.f32 %f(459+i*3), %f(458+i*3), %f19; for.size.4 i in range(0, 10, 1): max.f32 %f(463+i*3), %f(462+i*3), %f456; st.global.f32 [%rd49+(-8008+i*52)], %f(463+i*3); ld.local.f32 %f(464+i*3), [%rd48+(520+i*156)]; add.f32 %f(465+i*3), %f(464+i*3), %f19; for.size.4 i in range(0, 2, 1): max.f32 %f(493+i*3), %f(492+i*3), %f456; st.global.f32 [%rd49+(-7488+i*7488)], %f(493+i*3); ld.local.f32 %f(494+i*3), [%rd48+(2080+i*156)]; add.f32 %f(495+i*3), %f(494+i*3), %f20; for.size.4 i in range(0, 11, 1): max.f32 %f(499+i*3), %f(498+i*3), %f456; st.global.f32 [%rd49+(52+i*52)], %f(499+i*3); ld.local.f32 %f(500+i*3), [%rd48+(2392+i*156)]; add.f32 %f(501+i*3), %f(500+i*3), %f20; max.f32 %f532, %f531, %f456; st.global.f32 [%rd49+624], %f532; for.size.1 i in range(0, 2, 1): add.s64 %rd(49+i*-1), %rd(49+i*-1), 4; add.s32 %r97, %r97, 1; setp.ne.s32 %p7, %r97, 13; @%p7 bra $L__BB0_13; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(21+i*1), [%rd6+(8+i*48)]; add.s64 %rd51, %rd43, 9464; mov.u32 %r98, 0; $L__BB0_15: ld.local.f32 %f533, [%rd50+104]; add.f32 %f534, %f533, %f21; mov.f32 %f535, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(536+i*3), %f(534+i*4), %f535; st.global.f32 [%rd51+(-8112+i*52)], %f(536+i*3); ld.local.f32 %f(537+i*3), [%rd50+(260+i*156)]; add.f32 %f(538+i*3), %f(537+i*3), %f21; for.size.4 i in range(0, 10, 1): max.f32 %f(542+i*3), %f(541+i*3), %f535; st.global.f32 [%rd51+(-8008+i*52)], %f(542+i*3); ld.local.f32 %f(543+i*3), [%rd50+(572+i*156)]; add.f32 %f(544+i*3), %f(543+i*3), %f21; for.size.4 i in range(0, 2, 1): max.f32 %f(572+i*3), %f(571+i*3), %f535; st.global.f32 [%rd51+(-7488+i*7488)], %f(572+i*3); ld.local.f32 %f(573+i*3), [%rd50+(2132+i*156)]; add.f32 %f(574+i*3), %f(573+i*3), %f22; for.size.4 i in range(0, 11, 1): max.f32 %f(578+i*3), %f(577+i*3), %f535; st.global.f32 [%rd51+(52+i*52)], %f(578+i*3); ld.local.f32 %f(579+i*3), [%rd50+(2444+i*156)]; add.f32 %f(580+i*3), %f(579+i*3), %f22; max.f32 %f611, %f610, %f535; st.global.f32 [%rd51+624], %f611; for.size.1 i in range(0, 2, 1): add.s64 %rd(51+i*-1), %rd(51+i*-1), 4; add.s32 %r98, %r98, 1; setp.ne.s32 %p8, %r98, 13; @%p8 bra $L__BB0_15; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[1014]; __shared__ float pad_temp_shared[5408]; __shared__ float placeholder_shared[192]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 3; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 13; ++xx_outer_inner_init) { for (int vthread_s = 0; vthread_s < 26; ++vthread_s) { int index0 = vthread_s * 39 + ff_outer_inner_init * 13 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 338; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x; int index2 = ((int)blockIdx.x >> 1) * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) / 1352 * 43264 + rc_outer_outer * 1352 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x) % 1352; pad_temp_shared[index1] = placeholder[index2]; } int index3 = (int)threadIdx.x; int index4 = ((int)blockIdx.x & 1) * 6144 + ((int)threadIdx.x >> 3) * 256 + rc_outer_outer * 8 + ((int)threadIdx.x & 7); placeholder_shared[index3] = placeholder1[index4]; placeholder_shared[index3 + 16] = placeholder1[index4 + 512]; placeholder_shared[index3 + 32] = placeholder1[index4 + 1024]; placeholder_shared[index3 + 48] = placeholder1[index4 + 1536]; placeholder_shared[index3 + 64] = placeholder1[index4 + 2048]; placeholder_shared[index3 + 80] = placeholder1[index4 + 2560]; placeholder_shared[index3 + 96] = placeholder1[index4 + 3072]; placeholder_shared[index3 + 112] = placeholder1[index4 + 3584]; placeholder_shared[index3 + 128] = placeholder1[index4 + 4096]; placeholder_shared[index3 + 144] = placeholder1[index4 + 4608]; placeholder_shared[index3 + 160] = placeholder1[index4 + 5120]; placeholder_shared[index3 + 176] = placeholder1[index4 + 5632]; __syncthreads(); for (int ff_outer_inner = 0; ff_outer_inner < 3; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 13; ++xx_outer_inner) { for (int rc_inner = 0; rc_inner < 8; ++rc_inner) { for (int vthread_s1 = 0; vthread_s1 < 26; ++vthread_s1) { int index5 = vthread_s1 * 39 + ff_outer_inner * 13 + xx_outer_inner; int index6 = ((int)threadIdx.x >> 2) * 1352 + rc_inner * 169 + vthread_s1 % 13 * 13 + xx_outer_inner; int index7 = vthread_s1 / 13 * 96 + ((int)threadIdx.x & 3) * 24 + ff_outer_inner * 8 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; } } } } } for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 13; ++ax3_inner) { for (int vthread_s2 = 0; vthread_s2 < 26; ++vthread_s2) { int index8 = ((int)blockIdx.x >> 1) * 32448 + ((int)threadIdx.x >> 2) * 8112 + ((int)blockIdx.x & 1) * 4056 + vthread_s2 / 13 * 2028 + ((int)threadIdx.x & 3) * 507 + ax1_inner * 169 + vthread_s2 % 13 * 13 + ax3_inner; int index9 = vthread_s2 * 39 + ax1_inner * 13 + ax3_inner; int index10 = ((int)blockIdx.x & 1) * 24 + vthread_s2 / 13 * 12 + ((int)threadIdx.x & 3) * 3 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmp9b7oxejy
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<2>; .reg .f32 %f<176>; .reg .b32 %r<41>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[128]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[512]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd11, %rd10; mov.u32 %r10, %ctaid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(11+i*3), %r10, (542165497+i*1084330994); shr.u32 %r(12+i*3), %r(11+i*3), 31; shr.s32 %r(13+i*3), %r(11+i*3), 6; add.s32 %r(1+i*16), %r(13+i*3), %r(12+i*3); mul.lo.s32 %r18, %r17, 169; sub.s32 %r2, %r10, %r18; mov.u32 %r19, %tid.x; shl.b32 %r20, %r19, 2; mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r21, %r20; mul.lo.s32 %r22, %r1, 507; sub.s32 %r23, %r10, %r22; mul.hi.s32 %r24, %r23, 1626496491; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 6; add.s32 %r4, %r26, %r25; shl.b32 %r27, %r4, 12; mov.u32 %r28, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r28, %r20; shl.b32 %r29, %r19, 6; add.s32 %r6, %r28, %r29; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(30+i*1), %r(1+i*18), (173056+i*-172887), %r(2+i*28); cvta.to.global.u64 %rd12, %rd9; mul.wide.s32 %rd13, %r31, 4; add.s64 %rd14, %rd12, %rd13; add.s64 %rd24, %rd14, 346112; add.s32 %r32, %r19, %r27; mul.wide.s32 %rd15, %r32, 4; add.s64 %rd16, %rd11, %rd15; add.s64 %rd23, %rd16, 8192; mov.u32 %r40, 0; mov.f32 %f168, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(169+i*1), %f168; $L__BB0_1: bar.sync 0; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(25+i*1), [%rd24+(-346112+i*173056)]; st.shared.f32 [%r3+(0+i*32)], %f(25+i*1); for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(29+i*1), [%rd23+(-8192+i*1024)]; st.shared.f32 [%r5+(0+i*32)], %f(29+i*1); bar.sync 0; for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(45+i*5), [%r6+(0+i*4)]; ld.shared.f32 %f(46+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(0+i*4)]; fma.rn.f32 %f(47+i*5), %f(46+i*5), %f(45+i*5), %f(175+i*-128); ld.shared.f32 %f(48+i*5), [%r6+(32+i*4)]; fma.rn.f32 %f(49+i*5), %f(46+i*5), %f(48+i*5), %f(174+i*-125); for.size.5 i in range(0, 5, 1): ld.shared.f32 %f(55+i*5), [%r6+(8+i*4)]; ld.shared.f32 %f(56+i*5), [_ZZ24default_function_kernel0E15pad_temp_shared+(8+i*4)]; fma.rn.f32 %f(57+i*5), %f(56+i*5), %f(55+i*5), %f(52+i*5); ld.shared.f32 %f(58+i*5), [%r6+(40+i*4)]; fma.rn.f32 %f(59+i*5), %f(56+i*5), %f(58+i*5), %f(54+i*5); ld.shared.f32 %f80, [%r6+28]; ld.shared.f32 %f81, [_ZZ24default_function_kernel0E15pad_temp_shared+28]; fma.rn.f32 %f175, %f81, %f80, %f77; ld.shared.f32 %f82, [%r6+60]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(174+i*-89), %f(81+i*2), %f(82+i*-34), %f(79+i*93); ld.shared.f32 %f(83+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(32+i*4)]; fma.rn.f32 %f(84+i*3), %f(83+i*3), %f(45+i*5), %f(173+i*-89); for.size.3 i in range(0, 5, 1): fma.rn.f32 %f(88+i*3), %f(86+i*3), %f(53+i*5), %f(85+i*3); ld.shared.f32 %f(89+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(40+i*4)]; fma.rn.f32 %f(90+i*3), %f(89+i*3), %f(55+i*5), %f(87+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(103+i*69), %f(101+i*3), %f(78+i*4), %f(100+i*3); ld.shared.f32 %f(104+i*1), [_ZZ24default_function_kernel0E15pad_temp_shared+(60+i*4)]; fma.rn.f32 %f(173+i*-67), %f(104+i*1), %f(80+i*-35), %f(102+i*69); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(107+i*3), %f(105+i*3), %f(48+i*5), %f(170+i*-63); ld.shared.f32 %f(108+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(68+i*4)]; fma.rn.f32 %f(109+i*3), %f(108+i*3), %f(50+i*5), %f(106+i*3); for.size.3 i in range(0, 4, 1): fma.rn.f32 %f(113+i*3), %f(111+i*3), %f(58+i*5), %f(110+i*3); ld.shared.f32 %f(114+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(76+i*4)]; fma.rn.f32 %f(115+i*3), %f(114+i*3), %f(60+i*5), %f(112+i*3); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(125+i*45), %f(123+i*3), %f(78+i*4), %f(122+i*3); ld.shared.f32 %f(126+i*1), [_ZZ24default_function_kernel0E15pad_temp_shared+(92+i*4)]; fma.rn.f32 %f(171+i*-43), %f(126+i*1), %f(80+i*-35), %f(124+i*45); for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(129+i*3), %f(127+i*3), %f(48+i*5), %f(168+i*-39); ld.shared.f32 %f(130+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(100+i*4)]; fma.rn.f32 %f(131+i*3), %f(130+i*3), %f(50+i*5), %f(128+i*3); for.size.3 i in range(0, 4, 1): fma.rn.f32 %f(135+i*3), %f(133+i*3), %f(58+i*5), %f(132+i*3); ld.shared.f32 %f(136+i*3), [_ZZ24default_function_kernel0E15pad_temp_shared+(108+i*4)]; fma.rn.f32 %f(137+i*3), %f(136+i*3), %f(60+i*5), %f(134+i*3); fma.rn.f32 %f147, %f145, %f78, %f144; ld.shared.f32 %f148, [_ZZ24default_function_kernel0E15pad_temp_shared+124]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(169+i*-1), %f148, %f(80+i*2), %f(146+i*1); for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (5408+i*-5376); add.s32 %r40, %r40, 1; setp.ne.s32 %p1, %r40, 32; @%p1 bra $L__BB0_1; cvta.to.global.u64 %rd17, %rd8; for.size.1 i in range(0, 2, 1): shl.b32 %r(34+i*1), %r(19+i*-15), (1+i*3); add.s32 %r36, %r35, %r34; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(37+i*1), %r(19+i*-18), (338+i*32110), %r(2+i*35); mad.lo.s32 %r39, %r4, 2704, %r38; mul.wide.s32 %rd18, %r36, 4; add.s64 %rd19, %rd17, %rd18; ld.global.nc.f32 %f149, [%rd19]; add.f32 %f150, %f175, %f149; mov.f32 %f151, 0f00000000; max.f32 %f152, %f150, %f151; cvta.to.global.u64 %rd20, %rd7; mul.wide.s32 %rd21, %r39, 4; add.s64 %rd22, %rd20, %rd21; st.global.f32 [%rd22], %f152; ld.global.nc.f32 %f153, [%rd19+4]; for.size.4 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): add.f32 %f(154+i*4+j*2), %f(174+i*-2+j*-1), %f(153+j*-4); max.f32 %f(155+i*4+j*2), %f(154+i*4+j*2), %f151; st.global.f32 [%rd22+(676+i*32448+j*31772)], %f(155+i*4+j*2); add.f32 %f166, %f168, %f153; max.f32 %f167, %f166, %f151; st.global.f32 [%rd22+98020], %f167; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[32]; __shared__ float placeholder_shared[128]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 32; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x / 507 * 173056 + rc_outer_outer * 1352 + (int)threadIdx.x * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 8] = placeholder[index1 + 43264]; pad_temp_shared[index0 + 16] = placeholder[index1 + 86528]; pad_temp_shared[index0 + 24] = placeholder[index1 + 129792]; int index2 = (int)blockIdx.x % 507 / 169 * 4096 + rc_outer_outer * 8 + (int)threadIdx.x; placeholder_shared[index0] = placeholder1[index2]; placeholder_shared[index0 + 8] = placeholder1[index2 + 256]; placeholder_shared[index0 + 16] = placeholder1[index2 + 512]; placeholder_shared[index0 + 24] = placeholder1[index2 + 768]; placeholder_shared[index0 + 32] = placeholder1[index2 + 1024]; placeholder_shared[index0 + 40] = placeholder1[index2 + 1280]; placeholder_shared[index0 + 48] = placeholder1[index2 + 1536]; placeholder_shared[index0 + 56] = placeholder1[index2 + 1792]; placeholder_shared[index0 + 64] = placeholder1[index2 + 2048]; placeholder_shared[index0 + 72] = placeholder1[index2 + 2304]; placeholder_shared[index0 + 80] = placeholder1[index2 + 2560]; placeholder_shared[index0 + 88] = placeholder1[index2 + 2816]; placeholder_shared[index0 + 96] = placeholder1[index2 + 3072]; placeholder_shared[index0 + 104] = placeholder1[index2 + 3328]; placeholder_shared[index0 + 112] = placeholder1[index2 + 3584]; placeholder_shared[index0 + 120] = placeholder1[index2 + 3840]; __syncthreads(); compute[0] = compute[0] + pad_temp_shared[0] * placeholder_shared[index0 * 16]; int index3 = (int)threadIdx.x * 16 + 8; compute[1] = compute[1] + pad_temp_shared[0] * placeholder_shared[index3]; int index4 = (int)threadIdx.x * 16 + 1; compute[0] = compute[0] + pad_temp_shared[1] * placeholder_shared[index4]; int index5 = (int)threadIdx.x * 16 + 9; compute[1] = compute[1] + pad_temp_shared[1] * placeholder_shared[index5]; int index6 = (int)threadIdx.x * 16 + 2; compute[0] = compute[0] + pad_temp_shared[2] * placeholder_shared[index6]; int index7 = (int)threadIdx.x * 16 + 10; compute[1] = compute[1] + pad_temp_shared[2] * placeholder_shared[index7]; int index8 = (int)threadIdx.x * 16 + 3; compute[0] = compute[0] + pad_temp_shared[3] * placeholder_shared[index8]; int index9 = (int)threadIdx.x * 16 + 11; compute[1] = compute[1] + pad_temp_shared[3] * placeholder_shared[index9]; int index10 = (int)threadIdx.x * 16 + 4; compute[0] = compute[0] + pad_temp_shared[4] * placeholder_shared[index10]; int index11 = (int)threadIdx.x * 16 + 12; compute[1] = compute[1] + pad_temp_shared[4] * placeholder_shared[index11]; int index12 = (int)threadIdx.x * 16 + 5; compute[0] = compute[0] + pad_temp_shared[5] * placeholder_shared[index12]; int index13 = (int)threadIdx.x * 16 + 13; compute[1] = compute[1] + pad_temp_shared[5] * placeholder_shared[index13]; int index14 = (int)threadIdx.x * 16 + 6; compute[0] = compute[0] + pad_temp_shared[6] * placeholder_shared[index14]; int index15 = (int)threadIdx.x * 16 + 14; compute[1] = compute[1] + pad_temp_shared[6] * placeholder_shared[index15]; int index16 = (int)threadIdx.x * 16 + 7; compute[0] = compute[0] + pad_temp_shared[7] * placeholder_shared[index16]; int index17 = (int)threadIdx.x * 16 + 15; compute[1] = compute[1] + pad_temp_shared[7] * placeholder_shared[index17]; compute[2] = compute[2] + pad_temp_shared[8] * placeholder_shared[index0 * 16]; compute[3] = compute[3] + pad_temp_shared[8] * placeholder_shared[index3]; compute[2] = compute[2] + pad_temp_shared[9] * placeholder_shared[index4]; compute[3] = compute[3] + pad_temp_shared[9] * placeholder_shared[index5]; compute[2] = compute[2] + pad_temp_shared[10] * placeholder_shared[index6]; compute[3] = compute[3] + pad_temp_shared[10] * placeholder_shared[index7]; compute[2] = compute[2] + pad_temp_shared[11] * placeholder_shared[index8]; compute[3] = compute[3] + pad_temp_shared[11] * placeholder_shared[index9]; compute[2] = compute[2] + pad_temp_shared[12] * placeholder_shared[index10]; compute[3] = compute[3] + pad_temp_shared[12] * placeholder_shared[index11]; compute[2] = compute[2] + pad_temp_shared[13] * placeholder_shared[index12]; compute[3] = compute[3] + pad_temp_shared[13] * placeholder_shared[index13]; compute[2] = compute[2] + pad_temp_shared[14] * placeholder_shared[index14]; compute[3] = compute[3] + pad_temp_shared[14] * placeholder_shared[index15]; compute[2] = compute[2] + pad_temp_shared[15] * placeholder_shared[index16]; compute[3] = compute[3] + pad_temp_shared[15] * placeholder_shared[index17]; compute[4] = compute[4] + pad_temp_shared[16] * placeholder_shared[index0 * 16]; compute[5] = compute[5] + pad_temp_shared[16] * placeholder_shared[index3]; compute[4] = compute[4] + pad_temp_shared[17] * placeholder_shared[index4]; compute[5] = compute[5] + pad_temp_shared[17] * placeholder_shared[index5]; compute[4] = compute[4] + pad_temp_shared[18] * placeholder_shared[index6]; compute[5] = compute[5] + pad_temp_shared[18] * placeholder_shared[index7]; compute[4] = compute[4] + pad_temp_shared[19] * placeholder_shared[index8]; compute[5] = compute[5] + pad_temp_shared[19] * placeholder_shared[index9]; compute[4] = compute[4] + pad_temp_shared[20] * placeholder_shared[index10]; compute[5] = compute[5] + pad_temp_shared[20] * placeholder_shared[index11]; compute[4] = compute[4] + pad_temp_shared[21] * placeholder_shared[index12]; compute[5] = compute[5] + pad_temp_shared[21] * placeholder_shared[index13]; compute[4] = compute[4] + pad_temp_shared[22] * placeholder_shared[index14]; compute[5] = compute[5] + pad_temp_shared[22] * placeholder_shared[index15]; compute[4] = compute[4] + pad_temp_shared[23] * placeholder_shared[index16]; compute[5] = compute[5] + pad_temp_shared[23] * placeholder_shared[index17]; compute[6] = compute[6] + pad_temp_shared[24] * placeholder_shared[index0 * 16]; compute[7] = compute[7] + pad_temp_shared[24] * placeholder_shared[index3]; compute[6] = compute[6] + pad_temp_shared[25] * placeholder_shared[index4]; compute[7] = compute[7] + pad_temp_shared[25] * placeholder_shared[index5]; compute[6] = compute[6] + pad_temp_shared[26] * placeholder_shared[index6]; compute[7] = compute[7] + pad_temp_shared[26] * placeholder_shared[index7]; compute[6] = compute[6] + pad_temp_shared[27] * placeholder_shared[index8]; compute[7] = compute[7] + pad_temp_shared[27] * placeholder_shared[index9]; compute[6] = compute[6] + pad_temp_shared[28] * placeholder_shared[index10]; compute[7] = compute[7] + pad_temp_shared[28] * placeholder_shared[index11]; compute[6] = compute[6] + pad_temp_shared[29] * placeholder_shared[index12]; compute[7] = compute[7] + pad_temp_shared[29] * placeholder_shared[index13]; compute[6] = compute[6] + pad_temp_shared[30] * placeholder_shared[index14]; compute[7] = compute[7] + pad_temp_shared[30] * placeholder_shared[index15]; compute[6] = compute[6] + pad_temp_shared[31] * placeholder_shared[index16]; compute[7] = compute[7] + pad_temp_shared[31] * placeholder_shared[index17]; } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 2; ++ax1_inner) { int index18 = (int)blockIdx.x / 507 * 32448 + ax0_inner * 8112 + (int)blockIdx.x % 507 / 169 * 2704 + (int)threadIdx.x * 338 + ax1_inner * 169 + (int)blockIdx.x % 169; int index19 = ax0_inner * 2 + ax1_inner; int index20 = (int)blockIdx.x % 507 / 169 * 16 + (int)threadIdx.x * 2 + ax1_inner; T_relu[index18] = max(compute[index19] + placeholder2[index20], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpl1l456ae
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<1111>; .reg .b32 %r<174>; .reg .b64 %rd<39>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[86528]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048]; ld.param.u64 %rd2, [default_function_kernel0_param_0]; ld.param.u64 %rd3, [default_function_kernel0_param_1]; mov.u32 %r17, %ctaid.x; mul.hi.s32 %r18, %r17, 1431655766; shr.u32 %r19, %r18, 31; add.s32 %r1, %r18, %r19; mov.u32 %r20, %tid.x; mul.hi.s32 %r21, %r20, 1321528399; shr.u32 %r22, %r21, 31; shr.s32 %r23, %r21, 2; add.s32 %r2, %r23, %r22; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(24+i*1), %r(2+i*-1), (13+i*-10); sub.s32 %r(3+i*1), %r(20+i*-3), %r(24+i*1); mov.u32 %r170, 0; mov.f32 %f1007, 0f00000000; cvta.to.global.u64 %rd6, %rd2; cvta.to.global.u64 %rd23, %rd3; for.size.1 i in range(0, 103, 1): mov.f32 %f(1008+i*1), %f1007; $L__BB0_1: mov.u32 %r171, 0; bar.sync 0; $L__BB0_2: for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(161+i*-1), %r(1+i*169), (173056+i*-167648), %r(3+i*158); shl.b32 %r28, %r171, 3; add.s32 %r29, %r28, %r2; mul.hi.s32 %r30, %r29, 1321528399; shr.u32 %r31, %r30, 31; shr.s32 %r32, %r30, 7; add.s32 %r33, %r32, %r31; mul.lo.s32 %r34, %r33, 416; sub.s32 %r35, %r29, %r34; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(36+i*1), %r(33+i*2), (43264+i*-43251), %r(160+i*-124); mul.wide.s32 %rd7, %r37, 4; add.s64 %rd8, %rd6, %rd7; ld.global.nc.f32 %f521, [%rd8]; mad.lo.s32 %r39, %r171, 104, %r20; shl.b32 %r40, %r39, 2; mov.u32 %r41, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r42, %r41, %r40; for.size.13 i in range(0, 7, 1): st.shared.f32 [%r42+(0+i*416)], %f(521+i*1); add.s32 %r(43+i*9), %r29, (8+i*8); mul.hi.s32 %r(44+i*9), %r(43+i*9), 1321528399; shr.u32 %r(45+i*9), %r(44+i*9), 31; shr.s32 %r(46+i*9), %r(44+i*9), 7; add.s32 %r(47+i*9), %r(46+i*9), %r(45+i*9); mul.lo.s32 %r(48+i*9), %r(47+i*9), 416; sub.s32 %r(49+i*9), %r(43+i*9), %r(48+i*9); mad.lo.s32 %r(50+i*9), %r(47+i*9), 43264, %r160; mad.lo.s32 %r(51+i*9), %r(49+i*9), 13, %r(50+i*9); mul.wide.s32 %rd(9+i*2), %r(51+i*9), 4; add.s64 %rd(10+i*2), %rd6, %rd(9+i*2); ld.global.nc.f32 %f(522+i*1), [%rd(10+i*2)+0]; st.shared.f32 [%r42+2912], %f528; add.s32 %r171, %r171, 8; setp.ne.s32 %p1, %r171, 208; @%p1 bra $L__BB0_2; setp.gt.s32 %p2, %r20, 95; shr.u32 %r107, %r20, 5; and.b32 %r108, %r20, 31; prmt.b32 %r109, %r107, %r108, 8452; for.size.2 i in range(0, 2, 1): shl.b32 %r(110+i*2), %r(4+i*166), (12+i*-7); add.s32 %r(111+i*2), %r(109+i*2), %r(110+i*2); mul.wide.s32 %rd24, %r113, 4; add.s64 %rd1, %rd23, %rd24; ld.global.nc.f32 %f529, [%rd1]; shl.b32 %r114, %r20, 2; mov.u32 %r115, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r116, %r115, %r114; st.shared.f32 [%r116], %f529; add.s32 %r117, %r20, 8; and.b32 %r118, %r117, 31; bfi.b32 %r119, %r4, %r118, 12, 20; shl.b32 %r120, %r20, 3; for.size.11 i in range(0, 2, 1): add.s32 %r(121+i*7), %r120, (832+i*832); and.b32 %r(122+i*7), %r(121+i*7), -256; add.s32 %r(123+i*7), %r(119+i*8), %r(122+i*7); add.s32 %r(124+i*7), %r(123+i*7), %r112; mul.wide.s32 %rd(25+i*2), %r(124+i*7), 4; add.s64 %rd(26+i*2), %rd23, %rd(25+i*2); ld.global.nc.f32 %f(530+i*1), [%rd(26+i*2)+0]; st.shared.f32 [%r116+(416+i*416)], %f(530+i*1); add.s32 %r(125+i*7), %r20, (16+i*8); and.b32 %r(126+i*7), %r(125+i*7), 31; bfi.b32 %r(127+i*7), %r4, %r(126+i*7), 12, 20; add.s32 %r135, %r120, 2496; and.b32 %r136, %r135, -256; for.size.1 i in range(0, 2, 1): add.s32 %r(137+i*1), %r(134+i*3), %r(136+i*-24); mul.wide.s32 %rd29, %r138, 4; add.s64 %rd30, %rd23, %rd29; ld.global.nc.f32 %f532, [%rd30]; st.shared.f32 [%r116+1248], %f532; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f533, [%rd1+13312]; st.shared.f32 [%r116+1664], %f533; $L__BB0_5: bar.sync 0; mov.u32 %r172, 0; $L__BB0_6: mov.u32 %r173, 0; $L__BB0_7: for.size.1 i in range(0, 2, 1): shl.b32 %r(168+i*-1), %r(2+i*170), (5+i*-3); add.s32 %r166, %r167, %r168; mad.lo.s32 %r165, %r172, 676, %r3; mov.u32 %r164, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r162, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r147, %r173, 169, %r165; for.size.2 i in range(0, 2, 1): add.s32 %r(148+i*3), %r(166+i*-2), %r(173+i*-24); shl.b32 %r(149+i*3), %r(148+i*-1), 2; add.s32 %r154, %r162, %r152; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(534+i*1), [%r(154+i*-3)+0]; fma.rn.f32 %f1110, %f534, %f535, %f1110; ld.shared.f32 %f536, [%r151+1024]; for.size.3 i in range(0, 2, 1): fma.rn.f32 %f(1058+i*-1), %f(534+i*3), %f536, %f(1058+i*-1); ld.shared.f32 %f(537+i*1), [%r154+(52+i*52)]; fma.rn.f32 %f(1109+i*-1), %f(537+i*1), %f535, %f(1109+i*-1); for.size.3 i in range(0, 10, 1): fma.rn.f32 %f(1056+i*-1), %f(538+i*1), %f536, %f(1056+i*-1); ld.shared.f32 %f(539+i*1), [%r154+(156+i*52)]; fma.rn.f32 %f(1107+i*-1), %f(539+i*1), %f535, %f(1107+i*-1); for.size.4 i in range(0, 3, 1): for.size.3 j in range(0, 13, 1): fma.rn.f32 %f(1046+i*-13+j*-1), %f(548+i*13+j*1), %f536, %f(1046+i*-13+j*-1); ld.shared.f32 %f(549+i*13+j*1), [%r154+(21632+i*21632+j*52)]; fma.rn.f32 %f(1097+i*-13+j*-1), %f(549+i*13+j*1), %f535, %f(1097+i*-13+j*-1); fma.rn.f32 %f1007, %f587, %f536, %f1007; add.s32 %r173, %r173, 1; setp.ne.s32 %p3, %r173, 4; @%p3 bra $L__BB0_7; add.s32 %r172, %r172, 1; setp.lt.u32 %p4, %r172, 8; @%p4 bra $L__BB0_6; add.s32 %r170, %r170, 1; setp.lt.u32 %p5, %r170, 8; @%p5 bra $L__BB0_1; ld.param.u64 %rd38, [default_function_kernel0_param_2]; ld.param.u64 %rd37, [default_function_kernel0_param_3]; shl.b32 %r155, %r4, 4; add.s32 %r156, %r155, %r2; cvta.to.global.u64 %rd31, %rd37; mul.wide.s32 %rd32, %r156, 4; add.s64 %rd33, %rd31, %rd32; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(157+i*1), %r(1+i*3), (32448+i*-29744), %r(3+i*154); mad.lo.s32 %r159, %r2, 169, %r158; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(588+i*1), [%rd33+(32+i*-32)]; add.f32 %f590, %f1110, %f589; mov.f32 %f591, 0f00000000; max.f32 %f592, %f590, %f591; cvta.to.global.u64 %rd34, %rd38; mul.wide.s32 %rd35, %r159, 4; add.s64 %rd36, %rd34, %rd35; for.size.4 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd36+(0+i*52+j*5408)], %f(592+i*4+j*2); add.f32 %f(593+i*4+j*2), %f(1058+i*-1+j*51), %f(588+j*1); max.f32 %f(594+i*4+j*2), %f(593+i*4+j*2), %f591; for.size.4 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd36+(32448+i*52+j*5408)], %f(644+i*4+j*2); add.f32 %f(645+i*4+j*2), %f(1045+i*-1+j*51), %f(588+j*1); max.f32 %f(646+i*4+j*2), %f(645+i*4+j*2), %f591; for.size.4 i in range(0, 13, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd36+(64896+i*52+j*5408)], %f(696+i*4+j*2); add.f32 %f(697+i*4+j*2), %f(1032+i*-1+j*51), %f(588+j*1); max.f32 %f(698+i*4+j*2), %f(697+i*4+j*2), %f591; for.size.4 i in range(0, 12, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd36+(97344+i*52+j*5408)], %f(748+i*4+j*2); add.f32 %f(749+i*4+j*2), %f(1019+i*-1+j*51), %f(588+j*1); max.f32 %f(750+i*4+j*2), %f(749+i*4+j*2), %f591; st.global.f32 [%rd36+97968], %f796; add.f32 %f797, %f1007, %f588; max.f32 %f798, %f797, %f591; st.global.f32 [%rd36+103376], %f798; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[104]; __shared__ float pad_temp_shared[21632]; __shared__ float placeholder_shared[512]; for (int nn_inner_init = 0; nn_inner_init < 4; ++nn_inner_init) { int index0 = nn_inner_init * 13; compute[index0] = 0.000000e+00f; compute[index0 + 52] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 53] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 54] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 55] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 56] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 57] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; compute[index0 + 58] = 0.000000e+00f; compute[index0 + 7] = 0.000000e+00f; compute[index0 + 59] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; compute[index0 + 60] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 61] = 0.000000e+00f; compute[index0 + 10] = 0.000000e+00f; compute[index0 + 62] = 0.000000e+00f; compute[index0 + 11] = 0.000000e+00f; compute[index0 + 63] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; compute[index0 + 64] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 208; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 104 + (int)threadIdx.x; int index2 = (int)blockIdx.x / 3 * 173056 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 8 + (int)threadIdx.x / 13) / 416 * 43264 + rc_outer_outer * 5408 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 8 + (int)threadIdx.x / 13) % 416 * 13 + (int)threadIdx.x % 13; pad_temp_shared[index1] = placeholder[index2]; } int index3 = (int)threadIdx.x; int index4 = (int)blockIdx.x % 3 * 4096 + ((int)threadIdx.x >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index3] = placeholder1[index4]; int index5 = (int)blockIdx.x % 3 * 4096 + ((int)threadIdx.x + 104 >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x + 8 & 31); placeholder_shared[index3 + 104] = placeholder1[index5]; int index6 = (int)blockIdx.x % 3 * 4096 + ((int)threadIdx.x + 208 >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x + 16 & 31); placeholder_shared[index3 + 208] = placeholder1[index6]; int index7 = (int)blockIdx.x % 3 * 4096 + ((int)threadIdx.x + 312 >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x + 24 & 31); placeholder_shared[index3 + 312] = placeholder1[index7]; if ((int)threadIdx.x < 96) { placeholder_shared[index3 + 416] = placeholder1[index4 + 3328]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 4; ++rc_inner) { for (int nn_inner = 0; nn_inner < 4; ++nn_inner) { int index8 = nn_inner * 13; int index9 = nn_inner * 5408 + rc_outer_inner * 676 + rc_inner * 169 + (int)threadIdx.x % 13; int index10 = (int)threadIdx.x / 13 * 32 + rc_outer_inner * 4 + rc_inner; compute[index8] = compute[index8] + pad_temp_shared[index9] * placeholder_shared[index10]; compute[index8 + 52] = compute[index8 + 52] + (pad_temp_shared[index9] * placeholder_shared[index10 + 256]); compute[index8 + 1] = compute[index8 + 1] + (pad_temp_shared[index9 + 13] * placeholder_shared[index10]); compute[index8 + 53] = compute[index8 + 53] + (pad_temp_shared[index9 + 13] * placeholder_shared[index10 + 256]); compute[index8 + 2] = compute[index8 + 2] + (pad_temp_shared[index9 + 26] * placeholder_shared[index10]); compute[index8 + 54] = compute[index8 + 54] + (pad_temp_shared[index9 + 26] * placeholder_shared[index10 + 256]); compute[index8 + 3] = compute[index8 + 3] + (pad_temp_shared[index9 + 39] * placeholder_shared[index10]); compute[index8 + 55] = compute[index8 + 55] + (pad_temp_shared[index9 + 39] * placeholder_shared[index10 + 256]); compute[index8 + 4] = compute[index8 + 4] + (pad_temp_shared[index9 + 52] * placeholder_shared[index10]); compute[index8 + 56] = compute[index8 + 56] + (pad_temp_shared[index9 + 52] * placeholder_shared[index10 + 256]); compute[index8 + 5] = compute[index8 + 5] + (pad_temp_shared[index9 + 65] * placeholder_shared[index10]); compute[index8 + 57] = compute[index8 + 57] + (pad_temp_shared[index9 + 65] * placeholder_shared[index10 + 256]); compute[index8 + 6] = compute[index8 + 6] + (pad_temp_shared[index9 + 78] * placeholder_shared[index10]); compute[index8 + 58] = compute[index8 + 58] + (pad_temp_shared[index9 + 78] * placeholder_shared[index10 + 256]); compute[index8 + 7] = compute[index8 + 7] + (pad_temp_shared[index9 + 91] * placeholder_shared[index10]); compute[index8 + 59] = compute[index8 + 59] + (pad_temp_shared[index9 + 91] * placeholder_shared[index10 + 256]); compute[index8 + 8] = compute[index8 + 8] + (pad_temp_shared[index9 + 104] * placeholder_shared[index10]); compute[index8 + 60] = compute[index8 + 60] + (pad_temp_shared[index9 + 104] * placeholder_shared[index10 + 256]); compute[index8 + 9] = compute[index8 + 9] + (pad_temp_shared[index9 + 117] * placeholder_shared[index10]); compute[index8 + 61] = compute[index8 + 61] + (pad_temp_shared[index9 + 117] * placeholder_shared[index10 + 256]); compute[index8 + 10] = compute[index8 + 10] + (pad_temp_shared[index9 + 130] * placeholder_shared[index10]); compute[index8 + 62] = compute[index8 + 62] + (pad_temp_shared[index9 + 130] * placeholder_shared[index10 + 256]); compute[index8 + 11] = compute[index8 + 11] + (pad_temp_shared[index9 + 143] * placeholder_shared[index10]); compute[index8 + 63] = compute[index8 + 63] + (pad_temp_shared[index9 + 143] * placeholder_shared[index10 + 256]); compute[index8 + 12] = compute[index8 + 12] + (pad_temp_shared[index9 + 156] * placeholder_shared[index10]); compute[index8 + 64] = compute[index8 + 64] + (pad_temp_shared[index9 + 156] * placeholder_shared[index10 + 256]); } } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index11 = (int)blockIdx.x / 3 * 32448 + ax0_inner * 8112 + (int)blockIdx.x % 3 * 2704 + (int)threadIdx.x / 13 * 169 + ax2_inner * 13 + (int)threadIdx.x % 13; int index12 = ax0_inner * 13 + ax2_inner; int index13 = (int)blockIdx.x % 3 * 16 + (int)threadIdx.x / 13; T_relu[index11] = max(compute[index12] + placeholder2[index13], 0.000000e+00f); T_relu[index11 + 1352] = max(compute[index12 + 52] + placeholder2[index13 + 8], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpydsd_9ch
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 8 .b8 __local_depot0[1352]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<10>; .reg .f32 %f<1367>; .reg .b32 %r<149>; .reg .b64 %rd<45>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[24576]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd4, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; add.u64 %rd1, %SPL, 0; mov.u64 %rd9, 0; mov.f32 %f509, 0f00000000; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.v2.f32 [%rd1+(0+i*8+j*104)], {%f509, %f509}; mov.u32 %r141, 0; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 12, 1): st.local.u32 [%rd1+(16+i*4+j*104)], %r141; for.size.1 i in range(0, 2, 1): st.local.v2.f32 [%rd1+(1264+i*-1240)], {%f509, %f509}; for.size.1 i in range(0, 12, 1): st.local.v2.f32 [%rd1+(128+i*104)], {%f509, %f509}; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd1+(1176+i*104)], %r141; for.size.1 i in range(0, 11, 1): st.local.v2.f32 [%rd1+(32+i*104)], {%f509, %f509}; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd1+(1180+i*104)], %r141; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.v2.f32 [%rd1+(40+i*8+j*104)], {%f509, %f509}; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd1+(56+i*4+j*104)], %r141; st.local.u64 [%rd1+992], %rd9; for.size.1 i in range(0, 3, 1): st.local.v2.f32 [%rd1+(1096+i*104)], {%f509, %f509}; for.size.1 i in range(0, 13, 1): st.local.v2.f32 [%rd1+(64+i*104)], {%f509, %f509}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd1+(1008+i*104)], %r141; for.size.1 i in range(0, 9, 1): st.local.v2.f32 [%rd1+(72+i*104)], {%f509, %f509}; for.size.1 i in range(0, 4, 1): st.local.u32 [%rd1+(1012+i*104)], %r141; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 13, 1): st.local.v2.f32 [%rd1+(80+i*8+j*104)], {%f509, %f509}; for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 6, 1): st.local.u32 [%rd1+(96+i*4+j*104)], %r141; for.size.1 i in range(0, 2, 1): st.local.u64 [%rd1+(720+i*104)], %rd9; for.size.1 i in range(0, 5, 1): st.local.v2.f32 [%rd1+(928+i*104)], {%f509, %f509}; mov.u32 %r33, %ctaid.x; mul.lo.s32 %r34, %r33, 86528; mov.u32 %r35, %tid.x; cvta.to.global.u64 %rd16, %rd5; $L__BB0_1: mov.u32 %r138, %tid.x; add.s32 %r143, %r138, 48; mov.u32 %r142, 0; bar.sync 0; mad.lo.s32 %r2, %r141, 21632, %r34; $L__BB0_2: add.s32 %r36, %r143, -48; setp.gt.s32 %p1, %r36, 43263; shl.b32 %r38, %r35, 2; mov.u32 %r39, _ZZ24default_function_kernel0E15pad_temp_shared; for.size.1 i in range(0, 2, 1): add.s32 %r(40+i*-34), %r(39+i*1), %r(38+i*104); @%p1 bra $L__BB0_4; mul.hi.s32 %r42, %r36, 1626496491; shr.u32 %r43, %r42, 31; shr.s32 %r44, %r42, 13; add.s32 %r45, %r44, %r43; mul.lo.s32 %r46, %r45, 21632; sub.s32 %r47, %r36, %r46; add.s32 %r48, %r2, %r47; mad.lo.s32 %r49, %r45, 43264, %r48; cvta.to.global.u64 %rd10, %rd4; mul.wide.s32 %rd11, %r49, 4; add.s64 %rd12, %rd10, %rd11; ld.global.nc.f32 %f510, [%rd12]; st.shared.f32 [%r6], %f510; $L__BB0_4: setp.gt.s32 %p2, %r143, 43263; @%p2 bra $L__BB0_6; mul.hi.s32 %r50, %r143, 1626496491; shr.u32 %r51, %r50, 31; shr.s32 %r52, %r50, 13; add.s32 %r53, %r52, %r51; mul.lo.s32 %r54, %r53, 21632; sub.s32 %r55, %r143, %r54; add.s32 %r56, %r2, %r55; mad.lo.s32 %r57, %r53, 43264, %r56; cvta.to.global.u64 %rd13, %rd4; mul.wide.s32 %rd14, %r57, 4; add.s64 %rd15, %rd13, %rd14; ld.global.nc.f32 %f511, [%rd15]; st.shared.f32 [%r6+192], %f511; $L__BB0_6: for.size.1 i in range(0, 2, 1): add.s32 %r(143+i*-135), %r(143+i*-1), (96+i*288); setp.ne.s32 %p3, %r142, 172800; mov.u32 %r142, %r8; @%p3 bra $L__BB0_2; shl.b32 %r59, %r141, 7; for.size.3 i in range(0, 2, 1): and.b32 %r(61+i*2), %r(35+i*27), 127; bfi.b32 %r(9+i*1), %r141, %r(61+i*2), 7, 25; add.s32 %r(62+i*2), %r35, (48+i*48); and.b32 %r65, %r64, 127; bfi.b32 %r11, %r141, %r65, 7, 25; add.s32 %r66, %r35, 16; and.b32 %r67, %r66, 127; bfi.b32 %r12, %r141, %r67, 7, 25; xor.b32 %r68, %r61, 64; or.b32 %r13, %r68, %r59; for.size.3 i in range(0, 2, 1): add.s32 %r(69+i*2), %r35, (112+i*-80); and.b32 %r(70+i*2), %r(69+i*2), 127; bfi.b32 %r(14+i*1), %r141, %r(70+i*2), 7, 25; add.s32 %r73, %r35, 80; and.b32 %r74, %r73, 127; bfi.b32 %r16, %r141, %r74, 7, 25; mov.u32 %r144, 0; $L__BB0_8: mad.lo.s32 %r76, %r144, 48, %r35; shl.b32 %r77, %r76, 1; and.b32 %r78, %r77, -256; add.s32 %r79, %r9, %r78; mul.wide.s32 %rd17, %r79, 4; add.s64 %rd18, %rd16, %rd17; ld.global.nc.f32 %f512, [%rd18]; shl.b32 %r80, %r76, 2; mov.u32 %r81, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r82, %r81, %r80; for.size.7 i in range(0, 7, 1): st.shared.f32 [%r82+(0+i*192)], %f(512+i*1); add.s32 %r(83+i*3), %r77, (96+i*96); and.b32 %r(84+i*3), %r(83+i*3), -256; add.s32 %r(85+i*3), %r(10+i*1), %r(84+i*3); mul.wide.s32 %rd(19+i*2), %r(85+i*3), 4; add.s64 %rd(20+i*2), %rd16, %rd(19+i*2); ld.global.nc.f32 %f(513+i*1), [%rd(20+i*2)+0]; st.shared.f32 [%r82+1344], %f519; add.s32 %r144, %r144, 8; setp.ne.s32 %p4, %r144, 128; @%p4 bra $L__BB0_8; bar.sync 0; mov.u32 %r145, 0; $L__BB0_10: mov.u32 %r146, 0; $L__BB0_11: mul.lo.s32 %r110, %r146, 13; mul.wide.s32 %rd33, %r110, 4; add.s64 %rd2, %rd1, %rd33; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 13, 1): ld.local.f32 %f(1366+i*-13+j*-1), [%rd2+(0+i*4+j*104)]; mov.u32 %r147, 0; $L__BB0_12: mov.u32 %r133, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r132, _ZZ24default_function_kernel0E18placeholder_shared; for.size.1 i in range(0, 2, 1): shl.b32 %r(131+i*-1), %r(35+i*110), (7+i*-1); add.s32 %r129, %r130, %r131; mul.lo.s32 %r128, %r145, 10816; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(127+i*-16), %r(146+i*1), (21632+i*-21463), %r(128+i*-1); for.size.2 i in range(0, 2, 1): add.s32 %r(112+i*3), %r(129+i*3), %r(147+i*-34); shl.b32 %r(113+i*3), %r(112+i*-1), 2; add.s32 %r118, %r133, %r116; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(520+i*1), [%r(118+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1366+i*-1), %f(520+i*2), %f521, %f(1366+i*-1); ld.shared.f32 %f(522+i*1), [%r118+(4+i*4)]; for.size.2 i in range(0, 166, 1): fma.rn.f32 %f(1364+i*-1), %f(523+i*1), %f521, %f(1364+i*-1); ld.shared.f32 %f(524+i*1), [%r118+(12+i*4)]; fma.rn.f32 %f1198, %f689, %f521, %f1198; add.s32 %r147, %r147, 1; setp.ne.s32 %p5, %r147, 64; @%p5 bra $L__BB0_12; mul.lo.s32 %r134, %r146, 13; mul.wide.s32 %rd42, %r134, 4; add.s64 %rd41, %rd1, %rd42; for.size.2 i in range(0, 13, 1): for.size.1 j in range(0, 13, 1): st.local.f32 [%rd41+(0+i*4+j*104)], %f(1366+i*-13+j*-1); add.s32 %r146, %r146, 1; setp.lt.u32 %p6, %r146, 2; @%p6 bra $L__BB0_11; add.s32 %r145, %r145, 1; setp.lt.u32 %p7, %r145, 2; @%p7 bra $L__BB0_10; add.s32 %r141, %r141, 1; setp.lt.u32 %p8, %r141, 2; @%p8 bra $L__BB0_1; ld.param.u64 %rd44, [default_function_kernel0_param_2]; mov.u32 %r135, %ctaid.x; ld.param.u64 %rd43, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd34, %rd43; mul.wide.s32 %rd35, %r35, 4; add.s64 %rd36, %rd34, %rd35; ld.global.nc.f32 %f508, [%rd36]; mul.lo.s32 %r122, %r135, 16224; mad.lo.s32 %r29, %r35, 169, %r122; cvta.to.global.u64 %rd3, %rd44; mov.u32 %r148, 0; $L__BB0_17: mad.lo.s32 %r124, %r148, 8112, %r29; mul.lo.s32 %r125, %r148, 13; mul.wide.s32 %rd37, %r125, 4; add.s64 %rd38, %rd1, %rd37; ld.local.f32 %f690, [%rd38]; add.f32 %f691, %f690, %f508; mov.f32 %f692, 0f00000000; max.f32 %f693, %f691, %f692; mul.wide.s32 %rd39, %r124, 4; add.s64 %rd40, %rd3, %rd39; for.size.4 i in range(0, 12, 1): st.global.f32 [%rd40+(0+i*4)], %f(693+i*3); ld.local.f32 %f(694+i*3), [%rd38+(104+i*104)]; add.f32 %f(695+i*3), %f(694+i*3), %f508; max.f32 %f(696+i*3), %f(695+i*3), %f692; for.size.5 i in range(0, 12, 1): for.size.4 j in range(0, 13, 1): st.global.f32 [%rd40+(48+i*52+j*4)], %f(729+i*39+j*3); ld.local.f32 %f(730+i*39+j*3), [%rd38+(4+i*4+j*104)]; add.f32 %f(731+i*39+j*3), %f(730+i*39+j*3), %f508; max.f32 %f(732+i*39+j*3), %f(731+i*39+j*3), %f692; st.global.f32 [%rd40+672], %f1197; add.s32 %r126, %r148, 1; mov.u32 %r148, 1; setp.lt.u32 %p9, %r126, 2; @%p9 bra $L__BB0_17; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[338]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[6144]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 2; ++nn_outer_inner_init) { for (int yy_inner_init = 0; yy_inner_init < 13; ++yy_inner_init) { int index0 = nn_outer_inner_init * 13 + yy_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 26] = 0.000000e+00f; compute[index0 + 52] = 0.000000e+00f; compute[index0 + 78] = 0.000000e+00f; compute[index0 + 104] = 0.000000e+00f; compute[index0 + 130] = 0.000000e+00f; compute[index0 + 156] = 0.000000e+00f; compute[index0 + 182] = 0.000000e+00f; compute[index0 + 208] = 0.000000e+00f; compute[index0 + 234] = 0.000000e+00f; compute[index0 + 260] = 0.000000e+00f; compute[index0 + 286] = 0.000000e+00f; compute[index0 + 312] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 902; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 48 + (int)threadIdx.x < 43264) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 48 + (int)threadIdx.x; int index2 = (int)blockIdx.x * 86528 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 48 + (int)threadIdx.x) / 21632 * 43264 + rc_outer_outer * 21632 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 48 + (int)threadIdx.x) % 21632; pad_temp_shared[index1] = placeholder[index2]; } } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 128; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 48 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 48 + (int)threadIdx.x >> 7) * 256 + rc_outer_outer * 128 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 48 + (int)threadIdx.x & 127); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 2; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 2; ++nn_outer_inner) { for (int rc_inner = 0; rc_inner < 64; ++rc_inner) { for (int yy_inner = 0; yy_inner < 13; ++yy_inner) { int index5 = nn_outer_inner * 13 + yy_inner; int index6 = nn_outer_inner * 21632 + rc_outer_inner * 10816 + rc_inner * 169 + yy_inner * 13; int index7 = (int)threadIdx.x * 128 + rc_outer_inner * 64 + rc_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 26] = compute[index5 + 26] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 52] = compute[index5 + 52] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 78] = compute[index5 + 78] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 104] = compute[index5 + 104] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 130] = compute[index5 + 130] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 156] = compute[index5 + 156] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 182] = compute[index5 + 182] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 208] = compute[index5 + 208] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); compute[index5 + 234] = compute[index5 + 234] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 260] = compute[index5 + 260] + (pad_temp_shared[index6 + 10] * placeholder_shared[index7]); compute[index5 + 286] = compute[index5 + 286] + (pad_temp_shared[index6 + 11] * placeholder_shared[index7]); compute[index5 + 312] = compute[index5 + 312] + (pad_temp_shared[index6 + 12] * placeholder_shared[index7]); } } } } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax2_inner = 0; ax2_inner < 13; ++ax2_inner) { int index8 = (int)blockIdx.x * 16224 + ax0_inner * 8112 + (int)threadIdx.x * 169 + ax2_inner * 13; int index9 = ax0_inner * 13 + ax2_inner; int index10 = (int)threadIdx.x; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 1] = max(compute[index9 + 26] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 2] = max(compute[index9 + 52] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 3] = max(compute[index9 + 78] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 4] = max(compute[index9 + 104] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 5] = max(compute[index9 + 130] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 6] = max(compute[index9 + 156] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 7] = max(compute[index9 + 182] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 8] = max(compute[index9 + 208] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 9] = max(compute[index9 + 234] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 10] = max(compute[index9 + 260] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 11] = max(compute[index9 + 286] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 12] = max(compute[index9 + 312] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpih6o2kbb
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<260>; .reg .b32 %r<100>; .reg .b64 %rd<31>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[173056]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[6144]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd5; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %tid.x; shl.b32 %r27, %r1, 2; mov.u32 %r28, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r2, %r28, %r27; add.s32 %r29, %r1, 8112; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(30+i*4), %r(29+i*-28), 1626496491; shr.u32 %r(31+i*4), %r(30+i*4), 31; shr.s32 %r(32+i*4), %r(30+i*4), (11+i*-5); add.s32 %r(33+i*-20), %r(32+i*4), %r(31+i*4); mul.lo.s32 %r37, %r13, 169; sub.s32 %r3, %r1, %r37; add.s32 %r38, %r37, 2704; for.size.7 i in range(0, 2, 1): mad.lo.s32 %r(39+i*6), %r(33+i*11), 43264, %r3; add.s32 %r(4+i*1), %r(39+i*6), %r38; add.s32 %r(40+i*6), %r1, (13520+i*5408); mul.hi.s32 %r(41+i*6), %r(40+i*6), 1626496491; shr.u32 %r(42+i*6), %r(41+i*6), 31; shr.s32 %r(43+i*6), %r(41+i*6), 11; add.s32 %r(44+i*6), %r(43+i*6), %r(42+i*6); for.size.7 i in range(0, 4, 1): mad.lo.s32 %r(51+i*6), %r(50+i*6), 43264, %r3; add.s32 %r(6+i*1), %r(51+i*6), %r38; add.s32 %r(52+i*6), %r1, (24336+i*5408); mul.hi.s32 %r(53+i*6), %r(52+i*6), 1626496491; shr.u32 %r(54+i*6), %r(53+i*6), 31; shr.s32 %r(55+i*6), %r(53+i*6), 11; add.s32 %r(56+i*6), %r(55+i*6), %r(54+i*6); mad.lo.s32 %r75, %r74, 43264, %r3; add.s32 %r10, %r75, %r38; shr.u32 %r76, %r1, 5; and.b32 %r77, %r1, 31; prmt.b32 %r11, %r76, %r77, 8452; mov.u32 %r78, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r12, %r78, %r27; mad.lo.s32 %r79, %r13, 384, %r78; add.s32 %r14, %r79, 132; shl.b32 %r80, %r3, 2; add.s32 %r81, %r28, %r80; add.s32 %r15, %r81, 86528; mov.u32 %r96, 0; mov.f32 %f236, 0f00000000; for.size.1 i in range(0, 23, 1): mov.f32 %f(237+i*1), %f236; $L__BB0_1: bar.sync 0; mul.lo.s32 %r82, %r96, 5408; add.s32 %r83, %r82, %r1; mul.wide.s32 %rd9, %r83, 4; add.s64 %rd10, %rd3, %rd9; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(97+i*1), [%rd10+(0+i*10816)]; st.shared.f32 [%r2+(0+i*10816)], %f(97+i*1); for.size.7 i in range(0, 7, 1): ld.global.nc.f32 %f(99+i*2), [%rd10+(173056+i*173056)]; st.shared.f32 [%r2+(21632+i*21632)], %f(99+i*2); add.s32 %r(84+i*1), %r(4+i*1), %r82; mul.wide.s32 %rd(11+i*2), %r(84+i*1), 4; add.s64 %rd(12+i*2), %rd3, %rd(11+i*2); ld.global.nc.f32 %f(100+i*2), [%rd(12+i*2)+0]; st.shared.f32 [%r2+(32448+i*21632)], %f(100+i*2); setp.gt.s32 %p1, %r1, 1535; @%p1 bra $L__BB0_3; shl.b32 %r91, %r96, 5; add.s32 %r92, %r11, %r91; mul.wide.s32 %rd25, %r92, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f113, [%rd26]; st.shared.f32 [%r12], %f113; $L__BB0_3: bar.sync 0; mov.u32 %r99, 132; mov.u32 %r97, %r15; mov.u32 %r98, %r14; $L__BB0_4: for.size.35 i in range(0, 2, 1): ld.shared.f32 %f(114+i*35), [%r97+(-86528+i*676)]; ld.shared.f32 %f(115+i*35), [%r98+(-132+i*4)]; fma.rn.f32 %f(116+i*143), %f(114+i*35), %f(115+i*35), %f(259+i*-143); ld.shared.f32 %f(117+i*34), [%r97+(0+i*676)]; fma.rn.f32 %f(118+i*129), %f(117+i*34), %f(115+i*35), %f(247+i*-129); ld.shared.f32 %f(119+i*33), [%r98+(-4+i*4)]; fma.rn.f32 %f(120+i*138), %f(114+i*35), %f(119+i*33), %f(258+i*-138); fma.rn.f32 %f(121+i*125), %f(117+i*34), %f(119+i*33), %f(246+i*-125); ld.shared.f32 %f(122+i*31), [%r98+(124+i*4)]; fma.rn.f32 %f(123+i*134), %f(114+i*35), %f(122+i*31), %f(257+i*-134); fma.rn.f32 %f(124+i*121), %f(117+i*34), %f(122+i*31), %f(245+i*-121); ld.shared.f32 %f(125+i*29), [%r97+(-64896+i*676)]; fma.rn.f32 %f(126+i*130), %f(125+i*29), %f(115+i*35), %f(256+i*-130); ld.shared.f32 %f(127+i*28), [%r97+(21632+i*676)]; fma.rn.f32 %f(128+i*116), %f(127+i*28), %f(115+i*35), %f(244+i*-116); fma.rn.f32 %f(129+i*126), %f(125+i*29), %f(119+i*33), %f(255+i*-126); fma.rn.f32 %f(130+i*113), %f(127+i*28), %f(119+i*33), %f(243+i*-113); fma.rn.f32 %f(131+i*123), %f(125+i*29), %f(122+i*31), %f(254+i*-123); fma.rn.f32 %f(132+i*110), %f(127+i*28), %f(122+i*31), %f(242+i*-110); ld.shared.f32 %f(133+i*23), [%r97+(-43264+i*676)]; fma.rn.f32 %f(134+i*119), %f(133+i*23), %f(115+i*35), %f(253+i*-119); ld.shared.f32 %f(135+i*22), [%r97+(43264+i*676)]; fma.rn.f32 %f(136+i*105), %f(135+i*22), %f(115+i*35), %f(241+i*-105); fma.rn.f32 %f(137+i*115), %f(133+i*23), %f(119+i*33), %f(252+i*-115); fma.rn.f32 %f(138+i*102), %f(135+i*22), %f(119+i*33), %f(240+i*-102); fma.rn.f32 %f(139+i*112), %f(133+i*23), %f(122+i*31), %f(251+i*-112); fma.rn.f32 %f(140+i*99), %f(135+i*22), %f(122+i*31), %f(239+i*-99); ld.shared.f32 %f(141+i*17), [%r97+(-21632+i*676)]; fma.rn.f32 %f(142+i*108), %f(141+i*17), %f(115+i*35), %f(250+i*-108); ld.shared.f32 %f(143+i*16), [%r97+(64896+i*676)]; fma.rn.f32 %f(144+i*94), %f(143+i*16), %f(115+i*35), %f(238+i*-94); fma.rn.f32 %f(145+i*104), %f(141+i*17), %f(119+i*33), %f(249+i*-104); fma.rn.f32 %f(146+i*91), %f(143+i*16), %f(119+i*33), %f(237+i*-91); fma.rn.f32 %f(147+i*101), %f(141+i*17), %f(122+i*31), %f(248+i*-101); fma.rn.f32 %f(148+i*88), %f(143+i*16), %f(122+i*31), %f(236+i*-88); for.size.1 i in range(0, 2, 1): add.s32 %r(98+i*-1), %r(98+i*-1), (8+i*1344); add.s32 %r99, %r99, 8; setp.ne.s32 %p2, %r99, 260; @%p2 bra $L__BB0_4; add.s32 %r96, %r96, 1; setp.lt.u32 %p3, %r96, 8; @%p3 bra $L__BB0_1; mul.lo.s32 %r94, %r13, 3; mad.lo.s32 %r95, %r13, 507, %r3; mul.wide.s32 %rd27, %r94, 4; add.s64 %rd28, %rd4, %rd27; ld.global.nc.f32 %f160, [%rd28]; add.f32 %f161, %f259, %f160; mov.f32 %f162, 0f00000000; max.f32 %f163, %f161, %f162; mul.wide.s32 %rd29, %r95, 4; add.s64 %rd30, %rd2, %rd29; for.size.7 i in range(0, 2, 1): st.global.f32 [%rd30+(0+i*676)], %f(163+i*5); add.f32 %f(164+i*5), %f(247+i*-1), %f(160+i*6); max.f32 %f(165+i*5), %f(164+i*5), %f162; st.global.f32 [%rd30+(129792+i*676)], %f(165+i*5); ld.global.nc.f32 %f(166+i*5), [%rd28+(4+i*4)]; add.f32 %f(167+i*5), %f(258+i*-1), %f(166+i*5); max.f32 %f(168+i*5), %f(167+i*5), %f162; for.size.12 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(1352+i*32448+j*129792)], %f(173+i*12+j*2); add.f32 %f(174+i*12+j*2), %f(245+i*-3+j*11), %f(171+j*-11); max.f32 %f(175+i*12+j*2), %f(174+i*12+j*2), %f162; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(32448+i*32448+j*129792)], %f(177+i*12+j*2); add.f32 %f(178+i*12+j*2), %f(244+i*-3+j*11), %f(160+j*6); max.f32 %f(179+i*12+j*2), %f(178+i*12+j*2), %f162; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(33124+i*32448+j*129792)], %f(181+i*12+j*2); add.f32 %f(182+i*12+j*2), %f(243+i*-3+j*11), %f(166+j*5); max.f32 %f(183+i*12+j*2), %f(182+i*12+j*2), %f162; st.global.f32 [%rd30+98696], %f209; add.f32 %f210, %f236, %f171; max.f32 %f211, %f210, %f162; st.global.f32 [%rd30+228488], %f211; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[24]; __shared__ float pad_temp_shared[43264]; __shared__ float placeholder_shared[1536]; for (int nn_inner_init = 0; nn_inner_init < 4; ++nn_inner_init) { int index0 = nn_inner_init * 3; compute[index0] = 0.000000e+00f; compute[index0 + 12] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 13] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 14] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); int index1 = (int)threadIdx.x; int index2 = rc_outer_outer * 5408 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; pad_temp_shared[index1 + 2704] = placeholder[index2 + 2704]; pad_temp_shared[index1 + 5408] = placeholder[index2 + 43264]; int index3 = ((int)threadIdx.x + 8112) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 8112] = placeholder[index3]; pad_temp_shared[index1 + 10816] = placeholder[index2 + 86528]; int index4 = ((int)threadIdx.x + 13520) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 13520] = placeholder[index4]; pad_temp_shared[index1 + 16224] = placeholder[index2 + 129792]; int index5 = ((int)threadIdx.x + 18928) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 18928] = placeholder[index5]; pad_temp_shared[index1 + 21632] = placeholder[index2 + 173056]; int index6 = ((int)threadIdx.x + 24336) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 24336] = placeholder[index6]; pad_temp_shared[index1 + 27040] = placeholder[index2 + 216320]; int index7 = ((int)threadIdx.x + 29744) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 29744] = placeholder[index7]; pad_temp_shared[index1 + 32448] = placeholder[index2 + 259584]; int index8 = ((int)threadIdx.x + 35152) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 35152] = placeholder[index8]; pad_temp_shared[index1 + 37856] = placeholder[index2 + 302848]; int index9 = ((int)threadIdx.x + 40560) / 5408 * 43264 + rc_outer_outer * 5408 + ((int)threadIdx.x / 169 + 16) * 169 + (int)threadIdx.x % 169; pad_temp_shared[index1 + 40560] = placeholder[index9]; if ((int)threadIdx.x < 1536) { int index10 = ((int)threadIdx.x >> 5) * 256 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index1] = placeholder1[index10]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 32; ++rc_outer_inner) { for (int nn_inner = 0; nn_inner < 4; ++nn_inner) { int index11 = nn_inner * 3; int index12 = nn_inner * 5408 + rc_outer_inner * 169 + (int)threadIdx.x % 169; int index13 = (int)threadIdx.x / 169 * 96 + rc_outer_inner; compute[index11] = compute[index11] + pad_temp_shared[index12] * placeholder_shared[index13]; compute[index11 + 12] = compute[index11 + 12] + (pad_temp_shared[index12 + 21632] * placeholder_shared[index13]); compute[index11 + 1] = compute[index11 + 1] + (pad_temp_shared[index12] * placeholder_shared[index13 + 32]); compute[index11 + 13] = compute[index11 + 13] + (pad_temp_shared[index12 + 21632] * placeholder_shared[index13 + 32]); compute[index11 + 2] = compute[index11 + 2] + (pad_temp_shared[index12] * placeholder_shared[index13 + 64]); compute[index11 + 14] = compute[index11 + 14] + (pad_temp_shared[index12 + 21632] * placeholder_shared[index13 + 64]); } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { int index14 = ax0_inner * 8112 + (int)threadIdx.x / 169 * 507 + ax1_inner * 169 + (int)threadIdx.x % 169; int index15 = ax0_inner * 3 + ax1_inner; int index16 = (int)threadIdx.x / 169 * 3 + ax1_inner; T_relu[index14] = max(compute[index15] + placeholder2[index16], 0.000000e+00f); T_relu[index14 + 32448] = max(compute[index15 + 12] + placeholder2[index16], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpttqfzb59
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<10>; .reg .f32 %f<142>; .reg .b32 %r<70>; .reg .b64 %rd<22>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[1384448]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[12288]; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd8, [default_function_kernel0_param_1]; ld.param.u64 %rd9, [default_function_kernel0_param_2]; ld.param.u64 %rd11, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(29+i*3), %r1, (1626496491+i*-304968092); shr.u32 %r(30+i*3), %r(29+i*3), 31; shr.s32 %r(31+i*3), %r(29+i*3), (6+i*-4); add.s32 %r(2+i*62), %r(31+i*3), %r(30+i*3); add.s32 %r65, %r2, 6; mov.u32 %r66, 1014; shl.b32 %r35, %r1, 2; mov.u32 %r36, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r37, %r36, %r35; add.s32 %r62, %r37, 4056; mul.wide.s32 %rd13, %r1, 4; add.s64 %rd14, %rd12, %rd13; add.s64 %rd21, %rd14, 4056; mov.u32 %r63, 0; $L__BB0_1: add.s32 %r38, %r65, -6; setp.gt.s32 %p1, %r38, 2047; add.s64 %rd4, %rd21, -4056; @%p1 bra $L__BB0_5; setp.gt.s32 %p2, %r64, 26623; @%p2 bra $L__BB0_5; add.s32 %r39, %r1, %r63; setp.gt.s32 %p3, %r39, 346111; @%p3 bra $L__BB0_5; ld.global.nc.f32 %f33, [%rd4]; add.s32 %r60, %r62, -4056; st.shared.f32 [%r60], %f33; $L__BB0_5: setp.gt.s32 %p4, %r65, 2047; @%p4 bra $L__BB0_9; add.s32 %r40, %r64, 78; setp.gt.s32 %p5, %r40, 26623; @%p5 bra $L__BB0_9; add.s32 %r41, %r1, %r66; setp.gt.s32 %p6, %r41, 346111; @%p6 bra $L__BB0_9; ld.global.nc.f32 %f34, [%rd4+4056]; add.s32 %r61, %r62, -4056; st.shared.f32 [%r61+4056], %f34; $L__BB0_9: for.size.1 i in range(0, 2, 1): add.s32 %r(66+i*-1), %r(66+i*-1), (2028+i*-2016); for.size.1 i in range(0, 2, 1): add.s32 %r(64+i*-1), %r(64+i*-1), (156+i*1872); add.s32 %r62, %r62, 8112; add.s64 %rd21, %rd21, 8112; setp.ne.s32 %p7, %r66, 347802; @%p7 bra $L__BB0_1; mov.u32 %r17, %ctaid.x; mad.lo.s32 %r42, %r17, 3072, %r1; cvta.to.global.u64 %rd15, %rd8; mul.wide.s32 %rd16, %r42, 4; add.s64 %rd6, %rd15, %rd16; ld.global.nc.f32 %f35, [%rd6]; mov.u32 %r44, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r18, %r44, %r35; for.size.2 i in range(0, 2, 1): st.shared.f32 [%r18+(0+i*4056)], %f(35+i*1); ld.global.nc.f32 %f(36+i*1), [%rd6+(4056+i*4056)]; st.shared.f32 [%r18+8112], %f37; setp.gt.s32 %p8, %r1, 29; @%p8 bra $L__BB0_12; ld.global.nc.f32 %f38, [%rd6+12168]; st.shared.f32 [%r18+12168], %f38; $L__BB0_12: bar.sync 0; shl.b32 %r46, %r2, 10; add.s32 %r48, %r44, %r46; add.s32 %r69, %r48, 6148; mul.lo.s32 %r53, %r2, 169; sub.s32 %r54, %r1, %r53; shl.b32 %r55, %r54, 2; add.s32 %r57, %r36, %r55; add.s32 %r67, %r57, 692224; cvta.to.global.u64 %rd7, %rd9; mov.f32 %f126, 0f00000000; mov.u32 %r68, 6144; for.size.1 i in range(0, 15, 1): mov.f32 %f(127+i*1), %f126; $L__BB0_13: for.size.8 i in range(0, 2, 1): ld.shared.f32 %f(55+i*8), [%r67+(-692224+i*676)]; ld.shared.f32 %f(56+i*8), [%r69+(-6148+i*4)]; fma.rn.f32 %f(57+i*84), %f(55+i*8), %f(56+i*8), %f(141+i*-84); ld.shared.f32 %f(58+i*7), [%r69+(-4+i*4)]; fma.rn.f32 %f(59+i*78), %f(55+i*8), %f(58+i*7), %f(137+i*-78); ld.shared.f32 %f(60+i*6), [%r67+(0+i*676)]; fma.rn.f32 %f(61+i*72), %f(60+i*6), %f(56+i*8), %f(133+i*-72); fma.rn.f32 %f(62+i*67), %f(60+i*6), %f(58+i*7), %f(129+i*-67); for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): ld.shared.f32 %f(67+i*8+j*3), [%r67+(-519168+i*173056+j*692224)]; fma.rn.f32 %f(68+i*8+j*3), %f(67+i*8+j*3), %f56, %f(140+i*-1+j*-8); fma.rn.f32 %f(69+i*8+j*3), %f(67+i*8+j*3), %f58, %f(136+i*-1+j*-8); for.size.3 j in range(0, 2, 1): ld.shared.f32 %f(73+i*8+j*1), [%r67+(-518492+i*173056+j*692224)]; fma.rn.f32 %f(140+i*-1+j*-8), %f(73+i*8+j*1), %f64, %f(68+i*8+j*3); fma.rn.f32 %f(136+i*-1+j*-8), %f(73+i*8+j*1), %f65, %f(69+i*8+j*3); for.size.1 i in range(0, 2, 1): add.s32 %r(69+i*-2), %r(69+i*-2), (8+i*1344); add.s32 %r68, %r68, 8; setp.ne.s32 %p9, %r68, 7168; @%p9 bra $L__BB0_13; mad.lo.s32 %r58, %r17, 12, %r2; mul.wide.s32 %rd17, %r58, 4; add.s64 %rd18, %rd1, %rd17; mad.lo.s32 %r59, %r17, 2028, %r1; for.size.1 i in range(0, 2, 1): ld.global.nc.f32 %f(91+i*1), [%rd18+(24+i*-24)]; add.f32 %f93, %f141, %f92; mov.f32 %f94, 0f00000000; max.f32 %f95, %f93, %f94; mul.wide.s32 %rd19, %r59, 4; add.s64 %rd20, %rd7, %rd19; for.size.8 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd20+(0+i*32448+j*4056)], %f(95+i*8+j*2); add.f32 %f(96+i*8+j*2), %f(137+i*-1+j*-4), %f(91+j*1); max.f32 %f(97+i*8+j*2), %f(96+i*8+j*2), %f94; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd20+(129792+i*32448+j*4056)], %f(99+i*8+j*2); add.f32 %f(100+i*8+j*2), %f(129+i*-1+j*11), %f(91+j*1); max.f32 %f(101+i*8+j*2), %f(100+i*8+j*2), %f94; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(97344+i*4056)], %f(119+i*2); add.f32 %f(120+i*2), %f(134+i*-4), %f(91+i*1); max.f32 %f(121+i*2), %f(120+i*2), %f94; st.global.f32 [%rd20+227136], %f123; add.f32 %f124, %f126, %f91; max.f32 %f125, %f124, %f94; st.global.f32 [%rd20+231192], %f125; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[16]; __shared__ float pad_temp_shared[346112]; __shared__ float placeholder_shared[3072]; compute[0] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[15] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 342; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 169 < 2048) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 78 + (int)threadIdx.x / 13 < 26624) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x < 346112) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index0]; } } } } int index1 = (int)threadIdx.x; int index2 = (int)blockIdx.x * 3072 + (int)threadIdx.x; placeholder_shared[index1] = placeholder1[index2]; placeholder_shared[index1 + 1014] = placeholder1[index2 + 1014]; placeholder_shared[index1 + 2028] = placeholder1[index2 + 2028]; if ((int)threadIdx.x < 30) { placeholder_shared[index1 + 3042] = placeholder1[index2 + 3042]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 4; ++nn_outer_inner) { int index3 = nn_outer_inner; int index4 = nn_outer_inner * 43264 + rc_outer_inner * 338 + (int)threadIdx.x % 169; int index5 = (int)threadIdx.x / 169 * 256 + rc_outer_inner * 2; compute[index3] = compute[index3] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[index3 + 4] = compute[index3 + 4] + (pad_temp_shared[index4] * placeholder_shared[index5 + 1536]); compute[index3 + 8] = compute[index3 + 8] + (pad_temp_shared[index4 + 173056] * placeholder_shared[index5]); compute[index3 + 12] = compute[index3 + 12] + (pad_temp_shared[index4 + 173056] * placeholder_shared[index5 + 1536]); compute[index3] = compute[index3] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 1]); compute[index3 + 4] = compute[index3 + 4] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 1537]); compute[index3 + 8] = compute[index3 + 8] + (pad_temp_shared[index4 + 173225] * placeholder_shared[index5 + 1]); compute[index3 + 12] = compute[index3 + 12] + (pad_temp_shared[index4 + 173225] * placeholder_shared[index5 + 1537]); } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { int index6 = ax0_inner * 8112 + (int)blockIdx.x * 2028 + (int)threadIdx.x; int index7 = ax0_inner; int index8 = (int)blockIdx.x * 12 + (int)threadIdx.x / 169; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 1014] = max(compute[index7 + 4] + placeholder2[index8 + 6], 0.000000e+00f); T_relu[index6 + 32448] = max(compute[index7 + 8] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 33462] = max(compute[index7 + 12] + placeholder2[index8 + 6], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpc_jz6er4
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<197>; .reg .b32 %r<39>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[832]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[384]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r1, %tid.x; mul.hi.s32 %r11, %r1, 1321528399; shr.u32 %r12, %r11, 31; shr.s32 %r13, %r11, 3; add.s32 %r14, %r13, %r12; mul.lo.s32 %r15, %r14, 26; sub.s32 %r16, %r1, %r15; mul.hi.s32 %r17, %r16, 1321528399; shr.u32 %r18, %r17, 31; shr.s32 %r19, %r17, 2; add.s32 %r20, %r19, %r18; mov.u32 %r21, %ctaid.x; mul.lo.s32 %r22, %r21, 13; shr.s32 %r23, %r11, 2; add.s32 %r6, %r23, %r12; mul.lo.s32 %r24, %r6, 13; sub.s32 %r25, %r1, %r24; add.s32 %r2, %r25, %r22; shl.b32 %r26, %r1, 2; mov.u32 %r27, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r27, %r26; shr.u32 %r28, %r1, 1; and.b32 %r29, %r1, 1; mov.u32 %r30, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r4, %r30, %r26; shl.b32 %r31, %r25, 2; add.s32 %r5, %r27, %r31; mad.lo.s32 %r7, %r6, 24, %r30; prmt.b32 %r32, %r28, %r29, 8452; mul.wide.s32 %rd13, %r32, 4; add.s64 %rd22, %rd12, %rd13; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(33+i*1), %r(20+i*-6), (169+i*43095), %r(25+i*8); add.s32 %r35, %r34, %r22; cvta.to.global.u64 %rd14, %rd9; mul.wide.s32 %rd15, %r35, 4; add.s64 %rd21, %rd14, %rd15; mov.u32 %r38, 0; mov.f32 %f173, 0f00000000; for.size.1 i in range(0, 23, 1): mov.f32 %f(174+i*1), %f173; $L__BB0_1: bar.sync 0; ld.global.nc.f32 %f73, [%rd21]; st.shared.f32 [%r3], %f73; setp.gt.s32 %p1, %r1, 95; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f74, [%rd22]; st.shared.f32 [%r4], %f74; $L__BB0_3: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(75+i*1), [%r(7+i*-2)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(77+i*2), %f(76+i*2), %f75, %f(196+i*-1); ld.shared.f32 %f(78+i*2), [%r5+(416+i*-312)]; fma.rn.f32 %f81, %f80, %f75, %f194; ld.shared.f32 %f82, [%r5+520]; fma.rn.f32 %f83, %f82, %f75, %f193; ld.shared.f32 %f84, [%r7+4]; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(85+i*2+j*1), [%r5+(52+i*104+j*416)]; fma.rn.f32 %f(196+i*-2+j*-1), %f(85+i*2+j*1), %f84, %f(77+i*4+j*2); for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(89+i*5), [%r7+(8+i*4)]; fma.rn.f32 %f(90+i*102), %f(76+i*9), %f(89+i*5), %f(192+i*-102); fma.rn.f32 %f(91+i*100), %f(89+i*5), %f(78+i*8), %f(191+i*-100); fma.rn.f32 %f(92+i*98), %f(80+i*7), %f(89+i*5), %f(190+i*-98); fma.rn.f32 %f(93+i*96), %f(89+i*5), %f(82+i*6), %f(189+i*-96); for.size.5 i in range(0, 2, 1): ld.shared.f32 %f(95+i*5), [%r7+(16+i*4)]; fma.rn.f32 %f(96+i*92), %f(76+i*9), %f(95+i*5), %f(188+i*-92); fma.rn.f32 %f(97+i*90), %f(95+i*5), %f(78+i*8), %f(187+i*-90); fma.rn.f32 %f(98+i*88), %f(80+i*7), %f(95+i*5), %f(186+i*-88); fma.rn.f32 %f(99+i*86), %f(95+i*5), %f(82+i*6), %f(185+i*-86); for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(101+i*4+j*2), [%r5+(208+i*104+j*416)]; fma.rn.f32 %f(102+i*4+j*2), %f(101+i*4+j*2), %f75, %f(184+i*-2+j*-1); for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 2, 1): ld.shared.f32 %f(109+i*2+j*1), [%r5+(260+i*104+j*416)]; fma.rn.f32 %f(184+i*-2+j*-1), %f(109+i*2+j*1), %f84, %f(102+i*4+j*2); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(113+i*1), %f(101+i*-12), %f(89+i*14), %f(180+i*-1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(115+i*1), %f(105+i*-16), %f(89+i*18), %f(178+i*-1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(180+i*-1), %f(109+i*-15), %f(94+i*16), %f(113+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(178+i*-1), %f(111+i*-17), %f(94+i*18), %f(115+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(117+i*1), %f(101+i*-6), %f(95+i*8), %f(176+i*-1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(119+i*1), %f(105+i*-10), %f(95+i*12), %f(174+i*-1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(176+i*-1), %f(109+i*-9), %f(100+i*10), %f(117+i*1); for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(174+i*-1), %f(111+i*-11), %f(100+i*12), %f(119+i*1); for.size.1 i in range(0, 2, 1): add.s64 %rd(22+i*-1), %rd(22+i*-1), (8+i*1344); add.s32 %r38, %r38, 1; setp.ne.s32 %p2, %r38, 128; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd16, %rd8; mad.lo.s32 %r36, %r6, 507, %r2; mul.lo.s32 %r37, %r6, 3; mul.wide.s32 %rd17, %r37, 4; add.s64 %rd18, %rd16, %rd17; ld.global.nc.f32 %f121, [%rd18]; add.f32 %f122, %f196, %f121; mov.f32 %f123, 0f00000000; max.f32 %f124, %f122, %f123; mul.wide.s32 %rd19, %r36, 4; add.s64 %rd20, %rd1, %rd19; for.size.7 i in range(0, 2, 1): st.global.f32 [%rd20+(0+i*676)], %f(124+i*5); add.f32 %f(125+i*5), %f(195+i*-4), %f(121+i*6); max.f32 %f(126+i*5), %f(125+i*5), %f123; st.global.f32 [%rd20+(129792+i*676)], %f(126+i*5); ld.global.nc.f32 %f(127+i*5), [%rd18+(4+i*4)]; add.f32 %f(128+i*5), %f(192+i*-4), %f(127+i*5); max.f32 %f(129+i*5), %f(128+i*5), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(1352+i*129792)], %f(134+i*2); add.f32 %f(135+i*2), %f(187+i*7), %f(132+i*-11); max.f32 %f(136+i*2), %f(135+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(32448+i*129792)], %f(138+i*2); add.f32 %f(139+i*2), %f(193+i*-3), %f(121+i*6); max.f32 %f(140+i*2), %f(139+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(33124+i*129792)], %f(142+i*2); add.f32 %f(143+i*2), %f(189+i*-3), %f(127+i*5); max.f32 %f(144+i*2), %f(143+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(33800+i*129792)], %f(146+i*2); add.f32 %f(147+i*2), %f(185+i*-1), %f(132+i*-11); max.f32 %f(148+i*2), %f(147+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(64896+i*129792)], %f(150+i*2); add.f32 %f(151+i*2), %f(183+i*-3), %f(121+i*6); max.f32 %f(152+i*2), %f(151+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(65572+i*129792)], %f(154+i*2); add.f32 %f(155+i*2), %f(179+i*-3), %f(127+i*5); max.f32 %f(156+i*2), %f(155+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(66248+i*129792)], %f(158+i*2); add.f32 %f(159+i*2), %f(175+i*7), %f(132+i*-11); max.f32 %f(160+i*2), %f(159+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(97344+i*129792)], %f(162+i*2); add.f32 %f(163+i*2), %f(181+i*-3), %f(121+i*6); max.f32 %f(164+i*2), %f(163+i*2), %f123; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd20+(98020+i*129792)], %f(166+i*2); add.f32 %f(167+i*2), %f(177+i*-3), %f(127+i*5); max.f32 %f(168+i*2), %f(167+i*2), %f123; st.global.f32 [%rd20+98696], %f170; add.f32 %f171, %f173, %f132; max.f32 %f172, %f171, %f123; st.global.f32 [%rd20+228488], %f172; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[24]; __shared__ float pad_temp_shared[208]; __shared__ float placeholder_shared[96]; compute[0] = 0.000000e+00f; compute[12] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[15] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[13] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[16] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[14] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[17] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[18] = 0.000000e+00f; compute[9] = 0.000000e+00f; compute[21] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[19] = 0.000000e+00f; compute[10] = 0.000000e+00f; compute[22] = 0.000000e+00f; compute[8] = 0.000000e+00f; compute[20] = 0.000000e+00f; compute[11] = 0.000000e+00f; compute[23] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)threadIdx.x / 26 * 43264 + rc_outer_outer * 338 + (int)threadIdx.x % 26 / 13 * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; pad_temp_shared[index0] = placeholder[index1]; if ((int)threadIdx.x < 96) { int index2 = ((int)threadIdx.x >> 1) * 256 + rc_outer_outer * 2 + ((int)threadIdx.x & 1); placeholder_shared[index0] = placeholder1[index2]; } __syncthreads(); int index3 = (int)threadIdx.x / 13 * 6; compute[0] = compute[0] + pad_temp_shared[index0 % 13] * placeholder_shared[index3]; int index4 = (int)threadIdx.x % 13 + 104; compute[12] = compute[12] + pad_temp_shared[index4] * placeholder_shared[index3]; int index5 = (int)threadIdx.x % 13 + 26; compute[3] = compute[3] + pad_temp_shared[index5] * placeholder_shared[index3]; int index6 = (int)threadIdx.x % 13 + 130; compute[15] = compute[15] + pad_temp_shared[index6] * placeholder_shared[index3]; int index7 = (int)threadIdx.x % 13 + 13; compute[0] = compute[0] + (pad_temp_shared[index7] * placeholder_shared[index3 + 1]); int index8 = (int)threadIdx.x % 13 + 117; compute[12] = compute[12] + (pad_temp_shared[index8] * placeholder_shared[index3 + 1]); int index9 = (int)threadIdx.x % 13 + 39; compute[3] = compute[3] + (pad_temp_shared[index9] * placeholder_shared[index3 + 1]); int index10 = (int)threadIdx.x % 13 + 143; compute[15] = compute[15] + (pad_temp_shared[index10] * placeholder_shared[index3 + 1]); compute[1] = compute[1] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 2]); compute[13] = compute[13] + (pad_temp_shared[index4] * placeholder_shared[index3 + 2]); compute[4] = compute[4] + (pad_temp_shared[index5] * placeholder_shared[index3 + 2]); compute[16] = compute[16] + (pad_temp_shared[index6] * placeholder_shared[index3 + 2]); compute[1] = compute[1] + (pad_temp_shared[index7] * placeholder_shared[index3 + 3]); compute[13] = compute[13] + (pad_temp_shared[index8] * placeholder_shared[index3 + 3]); compute[4] = compute[4] + (pad_temp_shared[index9] * placeholder_shared[index3 + 3]); compute[16] = compute[16] + (pad_temp_shared[index10] * placeholder_shared[index3 + 3]); compute[2] = compute[2] + (pad_temp_shared[index0 % 13] * placeholder_shared[index3 + 4]); compute[14] = compute[14] + (pad_temp_shared[index4] * placeholder_shared[index3 + 4]); compute[5] = compute[5] + (pad_temp_shared[index5] * placeholder_shared[index3 + 4]); compute[17] = compute[17] + (pad_temp_shared[index6] * placeholder_shared[index3 + 4]); compute[2] = compute[2] + (pad_temp_shared[index7] * placeholder_shared[index3 + 5]); compute[14] = compute[14] + (pad_temp_shared[index8] * placeholder_shared[index3 + 5]); compute[5] = compute[5] + (pad_temp_shared[index9] * placeholder_shared[index3 + 5]); compute[17] = compute[17] + (pad_temp_shared[index10] * placeholder_shared[index3 + 5]); int index11 = (int)threadIdx.x % 13 + 52; compute[6] = compute[6] + pad_temp_shared[index11] * placeholder_shared[index3]; int index12 = (int)threadIdx.x % 13 + 156; compute[18] = compute[18] + pad_temp_shared[index12] * placeholder_shared[index3]; int index13 = (int)threadIdx.x % 13 + 78; compute[9] = compute[9] + pad_temp_shared[index13] * placeholder_shared[index3]; int index14 = (int)threadIdx.x % 13 + 182; compute[21] = compute[21] + pad_temp_shared[index14] * placeholder_shared[index3]; int index15 = (int)threadIdx.x % 13 + 65; compute[6] = compute[6] + (pad_temp_shared[index15] * placeholder_shared[index3 + 1]); int index16 = (int)threadIdx.x % 13 + 169; compute[18] = compute[18] + (pad_temp_shared[index16] * placeholder_shared[index3 + 1]); int index17 = (int)threadIdx.x % 13 + 91; compute[9] = compute[9] + (pad_temp_shared[index17] * placeholder_shared[index3 + 1]); int index18 = (int)threadIdx.x % 13 + 195; compute[21] = compute[21] + (pad_temp_shared[index18] * placeholder_shared[index3 + 1]); compute[7] = compute[7] + (pad_temp_shared[index11] * placeholder_shared[index3 + 2]); compute[19] = compute[19] + (pad_temp_shared[index12] * placeholder_shared[index3 + 2]); compute[10] = compute[10] + (pad_temp_shared[index13] * placeholder_shared[index3 + 2]); compute[22] = compute[22] + (pad_temp_shared[index14] * placeholder_shared[index3 + 2]); compute[7] = compute[7] + (pad_temp_shared[index15] * placeholder_shared[index3 + 3]); compute[19] = compute[19] + (pad_temp_shared[index16] * placeholder_shared[index3 + 3]); compute[10] = compute[10] + (pad_temp_shared[index17] * placeholder_shared[index3 + 3]); compute[22] = compute[22] + (pad_temp_shared[index18] * placeholder_shared[index3 + 3]); compute[8] = compute[8] + (pad_temp_shared[index11] * placeholder_shared[index3 + 4]); compute[20] = compute[20] + (pad_temp_shared[index12] * placeholder_shared[index3 + 4]); compute[11] = compute[11] + (pad_temp_shared[index13] * placeholder_shared[index3 + 4]); compute[23] = compute[23] + (pad_temp_shared[index14] * placeholder_shared[index3 + 4]); compute[8] = compute[8] + (pad_temp_shared[index15] * placeholder_shared[index3 + 5]); compute[20] = compute[20] + (pad_temp_shared[index16] * placeholder_shared[index3 + 5]); compute[11] = compute[11] + (pad_temp_shared[index17] * placeholder_shared[index3 + 5]); compute[23] = compute[23] + (pad_temp_shared[index18] * placeholder_shared[index3 + 5]); } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 3; ++ax1_inner) { int index19 = ax0_inner * 8112 + (int)threadIdx.x / 13 * 507 + ax1_inner * 169 + (int)blockIdx.x * 13 + (int)threadIdx.x % 13; int index20 = ax0_inner * 3 + ax1_inner; int index21 = (int)threadIdx.x / 13 * 3 + ax1_inner; T_relu[index19] = max(compute[index20] + placeholder2[index21], 0.000000e+00f); T_relu[index19 + 32448] = max(compute[index20 + 12] + placeholder2[index21], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpgf1y_f2t
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<16>; .reg .f32 %f<110>; .reg .b32 %r<109>; .reg .b64 %rd<21>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[692224]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[12288]; ld.param.u64 %rd8, [default_function_kernel0_param_0]; ld.param.u64 %rd5, [default_function_kernel0_param_1]; ld.param.u64 %rd6, [default_function_kernel0_param_2]; ld.param.u64 %rd7, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd9, %rd8; mov.u32 %r38, %tid.x; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(39+i*3), %r38, (1626496491+i*-304968092); shr.u32 %r(40+i*3), %r(39+i*3), 31; shr.s32 %r(41+i*3), %r(39+i*3), (6+i*-4); add.s32 %r(1+i*99), %r(41+i*3), %r(40+i*3); mov.u32 %r45, %ctaid.x; shr.s32 %r46, %r45, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(105+i*-1), %r(100+i*-62), (78+i*2964); add.s32 %r103, %r1, 18; mad.lo.s32 %r47, %r46, 173056, %r38; mul.wide.s32 %rd10, %r47, 4; add.s64 %rd11, %rd9, %rd10; add.s64 %rd20, %rd11, 8112; mov.u32 %r101, 2028; shl.b32 %r48, %r38, 2; mov.u32 %r49, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r50, %r49, %r48; add.s32 %r98, %r50, 8112; mov.u32 %r102, 12; mov.u32 %r99, 0; bra.uni $L__BB0_1; $L__BB0_18: for.size.1 i in range(0, 2, 1): add.s32 %r(105+i*-1), %r(105+i*-1), (312+i*3744); for.size.1 i in range(0, 2, 1): add.s32 %r(103+i*-1), %r(103+i*-1), 24; for.size.1 i in range(0, 2, 1): add.s32 %r(101+i*-1), %r(101+i*-1), (4056+i*-3744); add.s32 %r99, %r99, 24; add.s64 %rd20, %rd20, 16224; add.s32 %r98, %r98, 16224; $L__BB0_1: add.s32 %r51, %r1, %r99; setp.gt.s32 %p1, %r51, 1023; @%p1 bra $L__BB0_5; setp.gt.s32 %p2, %r100, 13311; @%p2 bra $L__BB0_5; add.s32 %r52, %r104, -3042; setp.gt.s32 %p3, %r52, 173055; @%p3 bra $L__BB0_5; ld.global.nc.f32 %f17, [%rd20+-8112]; st.shared.f32 [%r98+-8112], %f17; $L__BB0_5: add.s32 %r54, %r51, 6; setp.gt.s32 %p4, %r54, 1023; @%p4 bra $L__BB0_9; setp.gt.s32 %p5, %r105, 13311; @%p5 bra $L__BB0_9; add.s32 %r55, %r104, -2028; setp.gt.s32 %p6, %r55, 173055; @%p6 bra $L__BB0_9; ld.global.nc.f32 %f18, [%rd20+-4056]; st.shared.f32 [%r98+-4056], %f18; $L__BB0_9: add.s32 %r56, %r1, %r102; setp.gt.s32 %p7, %r56, 1023; @%p7 bra $L__BB0_13; add.s32 %r57, %r100, 156; setp.gt.s32 %p8, %r57, 13311; @%p8 bra $L__BB0_13; add.s32 %r59, %r38, %r101; setp.gt.s32 %p9, %r59, 173055; @%p9 bra $L__BB0_13; ld.global.nc.f32 %f19, [%rd20]; st.shared.f32 [%r98], %f19; $L__BB0_13: setp.eq.s32 %p10, %r102, 1020; @%p10 bra $L__BB0_19; setp.gt.s32 %p11, %r103, 1023; @%p11 bra $L__BB0_18; add.s32 %r60, %r105, 156; setp.gt.s32 %p12, %r60, 13311; @%p12 bra $L__BB0_18; setp.gt.s32 %p13, %r104, 173055; @%p13 bra $L__BB0_18; ld.global.nc.f32 %f20, [%rd20+4056]; st.shared.f32 [%r98+4056], %f20; bra.uni $L__BB0_18; $L__BB0_19: cvta.to.global.u64 %rd12, %rd5; and.b32 %r62, %r45, 3; mad.lo.s32 %r64, %r62, 3072, %r38; mul.wide.s32 %rd13, %r64, 4; add.s64 %rd4, %rd12, %rd13; ld.global.nc.f32 %f21, [%rd4]; mov.u32 %r66, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r23, %r66, %r48; for.size.2 i in range(0, 2, 1): st.shared.f32 [%r23+(0+i*4056)], %f(21+i*1); ld.global.nc.f32 %f(22+i*1), [%rd4+(4056+i*4056)]; st.shared.f32 [%r23+8112], %f23; setp.gt.s32 %p14, %r38, 29; @%p14 bra $L__BB0_21; ld.global.nc.f32 %f24, [%rd4+12168]; st.shared.f32 [%r23+12168], %f24; $L__BB0_21: bar.sync 0; mul.hi.s32 %r69, %r38, 542165497; shr.u32 %r70, %r69, 31; shr.s32 %r71, %r69, 6; add.s32 %r24, %r71, %r70; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(76+i*1), %r(1+i*23), (169+i*338); sub.s32 %r(25+i*53), %r38, %r(76+i*1); mul.hi.s32 %r79, %r78, 1626496491; shr.u32 %r80, %r79, 31; shr.s32 %r81, %r79, 6; add.s32 %r26, %r81, %r80; shl.b32 %r82, %r26, 12; add.s32 %r84, %r66, %r82; add.s32 %r107, %r84, 2048; mov.u32 %r108, 2048; mad.lo.s32 %r85, %r24, 86528, %r25; shl.b32 %r86, %r85, 2; add.s32 %r88, %r49, %r86; add.s32 %r106, %r88, 173056; mov.f32 %f102, 0f00000000; for.size.1 i in range(0, 7, 1): mov.f32 %f(103+i*1), %f102; $L__BB0_22: for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(33+i*14), [%r107+(-2048+i*4)]; ld.shared.f32 %f(34+i*14), [%r106+(-173056+i*676)]; fma.rn.f32 %f(35+i*14), %f(34+i*14), %f(33+i*14), %f(109+i*-74); ld.shared.f32 %f(36+i*14), [%r107+(-1024+i*4)]; fma.rn.f32 %f(37+i*14), %f(34+i*14), %f(36+i*14), %f(108+i*-71); ld.shared.f32 %f(38+i*14), [%r107+(0+i*4)]; fma.rn.f32 %f(39+i*14), %f(34+i*14), %f(38+i*14), %f(107+i*-68); ld.shared.f32 %f(40+i*14), [%r107+(1024+i*4)]; fma.rn.f32 %f(41+i*14), %f(34+i*14), %f(40+i*14), %f(106+i*-65); ld.shared.f32 %f(42+i*14), [%r106+(0+i*676)]; fma.rn.f32 %f(43+i*14), %f(42+i*14), %f(33+i*14), %f(105+i*-62); fma.rn.f32 %f(44+i*14), %f(42+i*14), %f(36+i*14), %f(104+i*-60); fma.rn.f32 %f(45+i*14), %f(42+i*14), %f(38+i*14), %f(103+i*-58); fma.rn.f32 %f(46+i*14), %f(42+i*14), %f(40+i*14), %f(102+i*-56); for.size.14 i in range(0, 2, 1): ld.shared.f32 %f(61+i*14), [%r107+(-2040+i*4)]; ld.shared.f32 %f(62+i*14), [%r106+(-171704+i*676)]; fma.rn.f32 %f(63+i*46), %f(62+i*14), %f(61+i*14), %f(49+i*14); ld.shared.f32 %f(64+i*13), [%r107+(-1016+i*4)]; fma.rn.f32 %f(65+i*43), %f(62+i*14), %f(64+i*13), %f(51+i*14); ld.shared.f32 %f(66+i*12), [%r107+(8+i*4)]; fma.rn.f32 %f(67+i*40), %f(62+i*14), %f(66+i*12), %f(53+i*14); ld.shared.f32 %f(68+i*11), [%r107+(1032+i*4)]; fma.rn.f32 %f(69+i*37), %f(62+i*14), %f(68+i*11), %f(55+i*14); ld.shared.f32 %f(70+i*10), [%r106+(1352+i*676)]; fma.rn.f32 %f(71+i*34), %f(70+i*10), %f(61+i*14), %f(57+i*14); fma.rn.f32 %f(72+i*32), %f(70+i*10), %f(64+i*13), %f(58+i*14); fma.rn.f32 %f(73+i*30), %f(70+i*10), %f(66+i*12), %f(59+i*14); fma.rn.f32 %f(74+i*28), %f(70+i*10), %f(68+i*11), %f(60+i*14); for.size.1 i in range(0, 2, 1): add.s32 %r(107+i*-1), %r(107+i*-1), (16+i*2688); add.s32 %r108, %r108, 16; setp.ne.s32 %p15, %r108, 3072; @%p15 bra $L__BB0_22; shl.b32 %r91, %r26, 2; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(92+i*2), %r(62+i*-38), (12+i*16212), %r(91+i*-66); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(95+i*1), %r(46+i*16), (32448+i*-30420), %r(94+i*1); mad.lo.s32 %r97, %r26, 676, %r96; cvta.to.global.u64 %rd14, %rd7; mul.wide.s32 %rd15, %r92, 4; add.s64 %rd16, %rd14, %rd15; ld.global.nc.f32 %f81, [%rd16]; add.f32 %f82, %f109, %f81; mov.f32 %f83, 0f00000000; max.f32 %f84, %f82, %f83; cvta.to.global.u64 %rd17, %rd6; mul.wide.s32 %rd18, %r97, 4; add.s64 %rd19, %rd17, %rd18; for.size.4 i in range(0, 3, 1): st.global.f32 [%rd19+(0+i*676)], %f(84+i*3); ld.global.nc.f32 %f(85+i*3), [%rd16+(4+i*4)]; add.f32 %f(86+i*3), %f(108+i*-1), %f(85+i*3); max.f32 %f(87+i*3), %f(86+i*3), %f83; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd19+(2028+i*30420)], %f(93+i*2); add.f32 %f(94+i*2), %f(105+i*-1), %f(81+i*4); max.f32 %f(95+i*2), %f(94+i*2), %f83; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd19+(33124+i*676)], %f(97+i*2); add.f32 %f(98+i*2), %f(103+i*-1), %f(88+i*3); max.f32 %f(99+i*2), %f(98+i*2), %f83; st.global.f32 [%rd19+34476], %f101; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[8]; __shared__ float pad_temp_shared[173056]; __shared__ float placeholder_shared[3072]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 171; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 169 < 1024) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 78 + (int)threadIdx.x / 13 < 13312) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x < 173056) { int index0 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x; int index1 = ((int)blockIdx.x >> 2) * 173056 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 1014 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; } } } } int index2 = (int)threadIdx.x; int index3 = ((int)blockIdx.x & 3) * 3072 + (int)threadIdx.x; placeholder_shared[index2] = placeholder1[index3]; placeholder_shared[index2 + 1014] = placeholder1[index3 + 1014]; placeholder_shared[index2 + 2028] = placeholder1[index3 + 2028]; if ((int)threadIdx.x < 30) { placeholder_shared[index2 + 3042] = placeholder1[index3 + 3042]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 128; ++rc_outer_inner) { int index4 = (int)threadIdx.x / 507 * 86528 + rc_outer_inner * 338 + (int)threadIdx.x % 169; int index5 = (int)threadIdx.x % 507 / 169 * 1024 + rc_outer_inner * 2; compute[0] = compute[0] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[1] = compute[1] + (pad_temp_shared[index4] * placeholder_shared[index5 + 256]); compute[2] = compute[2] + (pad_temp_shared[index4] * placeholder_shared[index5 + 512]); compute[3] = compute[3] + (pad_temp_shared[index4] * placeholder_shared[index5 + 768]); compute[4] = compute[4] + (pad_temp_shared[index4 + 43264] * placeholder_shared[index5]); compute[5] = compute[5] + (pad_temp_shared[index4 + 43264] * placeholder_shared[index5 + 256]); compute[6] = compute[6] + (pad_temp_shared[index4 + 43264] * placeholder_shared[index5 + 512]); compute[7] = compute[7] + (pad_temp_shared[index4 + 43264] * placeholder_shared[index5 + 768]); compute[0] = compute[0] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 1]); compute[1] = compute[1] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 257]); compute[2] = compute[2] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 513]); compute[3] = compute[3] + (pad_temp_shared[index4 + 169] * placeholder_shared[index5 + 769]); compute[4] = compute[4] + (pad_temp_shared[index4 + 43433] * placeholder_shared[index5 + 1]); compute[5] = compute[5] + (pad_temp_shared[index4 + 43433] * placeholder_shared[index5 + 257]); compute[6] = compute[6] + (pad_temp_shared[index4 + 43433] * placeholder_shared[index5 + 513]); compute[7] = compute[7] + (pad_temp_shared[index4 + 43433] * placeholder_shared[index5 + 769]); } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { int index6 = ((int)blockIdx.x >> 2) * 32448 + (int)threadIdx.x / 507 * 16224 + ax0_inner * 8112 + ((int)blockIdx.x & 3) * 2028 + (int)threadIdx.x % 507 / 169 * 676 + ax1_inner * 169 + (int)threadIdx.x % 169; int index7 = ax0_inner * 4 + ax1_inner; int index8 = ((int)blockIdx.x & 3) * 12 + (int)threadIdx.x % 507 / 169 * 4 + ax1_inner; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpxvhzy8h3
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<74>; .reg .b32 %r<33>; .reg .b64 %rd<26>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[32]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[96]; ld.param.u64 %rd10, [default_function_kernel0_param_0]; ld.param.u64 %rd11, [default_function_kernel0_param_1]; ld.param.u64 %rd12, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd13, %rd11; cvta.to.global.u64 %rd1, %rd12; mov.u32 %r14, %ctaid.x; mul.hi.s32 %r15, %r14, 1626496491; shr.u32 %r16, %r15, 31; shr.s32 %r17, %r15, 6; add.s32 %r4, %r17, %r16; mul.lo.s32 %r18, %r4, 169; sub.s32 %r1, %r14, %r18; mov.u32 %r2, %tid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(19+i*1), %r2, (2+i*2); mov.u32 %r21, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r21, %r20; mov.u32 %r22, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r22, %r19; shr.s32 %r6, %r2, 2; and.b32 %r23, %r2, -4; add.s32 %r7, %r21, %r23; and.b32 %r8, %r2, 3; mul.lo.s32 %r9, %r8, 6; shl.b32 %r24, %r9, 2; add.s32 %r10, %r22, %r24; shl.b32 %r25, %r2, 8; mad.lo.s32 %r26, %r4, 6144, %r25; mul.wide.s32 %rd14, %r26, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd25, %rd15, 4; mad.lo.s32 %r27, %r2, 173056, %r1; cvta.to.global.u64 %rd16, %rd10; mul.wide.s32 %rd17, %r27, 4; add.s64 %rd18, %rd16, %rd17; add.s64 %rd24, %rd18, 346112; mov.u32 %r32, 0; mov.f32 %f68, 0f00000000; for.size.1 i in range(0, 5, 1): mov.f32 %f(69+i*1), %f68; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r2, 1; @%p1 bra $L__BB0_3; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(25+i*1), [%rd24+(-346112+i*173056)]; st.shared.f32 [%r3+(0+i*4)], %f(25+i*1); $L__BB0_3: setp.gt.s32 %p2, %r2, 23; add.s64 %rd6, %rd25, -4; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f29, [%rd6]; st.shared.f32 [%r5], %f29; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(30+i*1), [%r(10+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(7+i*1), %f31, %f(30+i*2), %f(73+i*-1); ld.shared.f32 %f(32+i*1), [%r10+(4+i*4)]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(9+i*1), %f31, %f(33+i*1), %f(71+i*-1); ld.shared.f32 %f(34+i*1), [%r10+(12+i*4)]; fma.rn.f32 %f12, %f31, %f36, %f68; bar.sync 0; @%p1 bra $L__BB0_7; for.size.2 i in range(0, 4, 1): ld.global.nc.f32 %f(37+i*1), [%rd24+(-345436+i*173056)]; st.shared.f32 [%r3+(0+i*4)], %f(37+i*1); $L__BB0_7: @%p2 bra $L__BB0_9; ld.global.nc.f32 %f41, [%rd6+4]; st.shared.f32 [%r5], %f41; $L__BB0_9: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(42+i*1), [%r(10+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(73+i*-1), %f43, %f(42+i*2), %f(7+i*1); ld.shared.f32 %f(44+i*1), [%r10+(4+i*4)]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(71+i*-1), %f43, %f(45+i*1), %f(9+i*1); ld.shared.f32 %f(46+i*1), [%r10+(12+i*4)]; fma.rn.f32 %f68, %f43, %f48, %f12; for.size.1 i in range(0, 2, 1): add.s64 %rd(25+i*-1), %rd(25+i*-1), (8+i*1344); add.s32 %r32, %r32, 2; setp.ne.s32 %p5, %r32, 256; @%p5 bra $L__BB0_1; cvta.to.global.u64 %rd19, %rd9; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(28+i*1), %r4, (24+i*4032), %r(9+i*-8); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(30+i*1), %r(6+i*2), (8112+i*-7098), %r(29+i*1); mul.wide.s32 %rd20, %r28, 4; add.s64 %rd21, %rd19, %rd20; ld.global.nc.f32 %f49, [%rd21]; add.f32 %f50, %f73, %f49; mov.f32 %f51, 0f00000000; max.f32 %f52, %f50, %f51; mul.wide.s32 %rd22, %r31, 4; add.s64 %rd23, %rd1, %rd22; for.size.4 i in range(0, 5, 1): st.global.f32 [%rd23+(0+i*676)], %f(52+i*3); ld.global.nc.f32 %f(53+i*3), [%rd21+(4+i*4)]; add.f32 %f(54+i*3), %f(72+i*-1), %f(53+i*3); max.f32 %f(55+i*3), %f(54+i*3), %f51; st.global.f32 [%rd23+3380], %f67; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[6]; __shared__ float pad_temp_shared[8]; __shared__ float placeholder_shared[24]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 256; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 4; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0] = placeholder[index1]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 4; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 1] = placeholder[index1 + 43264]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 4; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 2] = placeholder[index1 + 86528]; } if ((int)threadIdx.x < 2) { int index0 = (int)threadIdx.x * 4; int index1 = (int)threadIdx.x * 173056 + rc_outer_outer * 169 + (int)blockIdx.x % 169; pad_temp_shared[index0 + 3] = placeholder[index1 + 129792]; } if ((int)threadIdx.x < 24) { int index2 = (int)threadIdx.x; int index3 = (int)blockIdx.x / 169 * 6144 + (int)threadIdx.x * 256 + rc_outer_outer; placeholder_shared[index2] = placeholder1[index3]; } __syncthreads(); int index2 = (int)threadIdx.x; int index4 = ((int)threadIdx.x & 3) * 6; compute[0] = compute[0] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4]); compute[1] = compute[1] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4 + 1]); compute[2] = compute[2] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4 + 2]); compute[3] = compute[3] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4 + 3]); compute[4] = compute[4] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4 + 4]); compute[5] = compute[5] + (pad_temp_shared[index2 >> 2] * placeholder_shared[index4 + 5]); } for (int ax1_inner = 0; ax1_inner < 6; ++ax1_inner) { int index5 = ((int)threadIdx.x >> 2) * 8112 + (int)blockIdx.x / 169 * 4056 + ((int)threadIdx.x & 3) * 1014 + ax1_inner * 169 + (int)blockIdx.x % 169; int index6 = ax1_inner; int index7 = (int)blockIdx.x / 169 * 24 + ((int)threadIdx.x & 3) * 6 + ax1_inner; T_relu[index5] = max(compute[index6] + placeholder2[index7], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,8,256,13,13,48,256,1,1,1,48,1,1,8,48,13,13]
tmpp_4jikil
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<3>; .reg .f32 %f<97>; .reg .b32 %r<33>; .reg .b64 %rd<25>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[128]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[384]; ld.param.u64 %rd9, [default_function_kernel0_param_0]; ld.param.u64 %rd10, [default_function_kernel0_param_1]; ld.param.u64 %rd11, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd12, %rd10; cvta.to.global.u64 %rd1, %rd11; mov.u32 %r13, %ctaid.x; mul.hi.s32 %r14, %r13, 1626496491; shr.u32 %r15, %r14, 31; shr.s32 %r16, %r14, 6; add.s32 %r4, %r16, %r15; mul.lo.s32 %r17, %r4, 169; sub.s32 %r1, %r13, %r17; mov.u32 %r2, %tid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(18+i*1), %r2, (2+i*2); mov.u32 %r20, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r20, %r19; shr.s32 %r5, %r2, 2; and.b32 %r6, %r2, 3; mov.u32 %r21, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r7, %r21, %r18; and.b32 %r22, %r18, -16; add.s32 %r8, %r20, %r22; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(9+i*14), %r(6+i*-4), (96+i*43168), %r(21+i*-20); cvta.to.global.u64 %rd13, %rd9; mul.wide.s32 %rd14, %r23, 4; add.s64 %rd15, %rd13, %rd14; add.s64 %rd24, %rd15, 1352; shl.b32 %r24, %r5, 8; mad.lo.s32 %r25, %r4, 6144, %r24; or.b32 %r26, %r25, %r6; mul.wide.s32 %rd16, %r26, 4; add.s64 %rd17, %rd12, %rd16; add.s64 %rd23, %rd17, 16384; mov.u32 %r32, 0; mov.f32 %f91, 0f00000000; for.size.1 i in range(0, 5, 1): mov.f32 %f(92+i*1), %f91; $L__BB0_1: bar.sync 0; setp.gt.s32 %p1, %r2, 7; @%p1 bra $L__BB0_3; for.size.1 i in range(0, 4, 1): ld.global.nc.f32 %f(19+i*1), [%rd24+(676+i*-676)]; st.shared.v4.f32 [%r3], {%f22, %f21, %f20, %f19}; $L__BB0_3: for.size.2 i in range(0, 3, 1): ld.global.nc.f32 %f(23+i*1), [%rd23+(-16384+i*8192)]; st.shared.f32 [%r7+(0+i*128)], %f(23+i*1); bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(26+i*1), [%r(9+i*-1)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(28+i*2), %f27, %f(26+i*3), %f(96+i*-1); ld.shared.f32 %f(29+i*2), [%r9+(16+i*16)]; for.size.7 i in range(0, 2, 1): fma.rn.f32 %f(32+i*7), %f(27+i*7), %f(31+i*7), %f(94+i*-62); ld.shared.f32 %f(33+i*7), [%r9+(4+i*4)]; ld.shared.f32 %f(34+i*7), [%r8+(4+i*4)]; fma.rn.f32 %f(35+i*7), %f(34+i*7), %f(33+i*7), %f(28+i*7); ld.shared.f32 %f(36+i*7), [%r9+(20+i*4)]; fma.rn.f32 %f(37+i*7), %f(34+i*7), %f(36+i*7), %f(30+i*7); ld.shared.f32 %f(38+i*7), [%r9+(36+i*4)]; fma.rn.f32 %f46, %f41, %f45, %f39; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(47+i*1), [%r(9+i*-1)+12]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(96+i*-1), %f48, %f(47+i*2), %f(42+i*2); ld.shared.f32 %f(49+i*1), [%r9+(28+i*16)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(94+i*-42), %f(48+i*-21), %f(50+i*1), %f(46+i*47); ld.shared.f32 %f(51+i*2), [%r9+(48+i*16)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(54+i*2), %f27, %f(53+i*2), %f(92+i*-1); ld.shared.f32 %f(55+i*2), [%r9+(80+i*-28)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(58+i*2), %f34, %f(57+i*2), %f(52+i*2); ld.shared.f32 %f(59+i*2), [%r9+(68+i*16)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(62+i*2), %f(34+i*7), %f(61+i*2), %f(56+i*2); ld.shared.f32 %f(63+i*2), [%r9+(56+i*16)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(66+i*2), %f41, %f(65+i*2), %f(60+i*2); ld.shared.f32 %f(67+i*2), [%r9+(88+i*-28)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(93+i*-1), %f48, %f(69+i*1), %f(64+i*2); ld.shared.f32 %f(70+i*1), [%r9+(76+i*16)]; fma.rn.f32 %f91, %f48, %f71, %f68; for.size.1 i in range(0, 2, 1): add.s64 %rd(24+i*-1), %rd(24+i*-1), (2704+i*-2688); add.s32 %r32, %r32, 1; setp.ne.s32 %p2, %r32, 64; @%p2 bra $L__BB0_1; cvta.to.global.u64 %rd18, %rd8; mul.lo.s32 %r27, %r6, 6; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(28+i*1), %r4, (24+i*4032), %r(27+i*-26); for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(30+i*1), %r(5+i*1), (8112+i*-7098), %r(29+i*1); mul.wide.s32 %rd19, %r28, 4; add.s64 %rd20, %rd18, %rd19; ld.global.nc.f32 %f72, [%rd20]; add.f32 %f73, %f96, %f72; mov.f32 %f74, 0f00000000; max.f32 %f75, %f73, %f74; mul.wide.s32 %rd21, %r31, 4; add.s64 %rd22, %rd1, %rd21; for.size.4 i in range(0, 5, 1): st.global.f32 [%rd22+(0+i*676)], %f(75+i*3); ld.global.nc.f32 %f(76+i*3), [%rd20+(4+i*4)]; add.f32 %f(77+i*3), %f(95+i*-1), %f(76+i*3); max.f32 %f(78+i*3), %f(77+i*3), %f74; st.global.f32 [%rd22+3380], %f90; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[6]; __shared__ float pad_temp_shared[32]; __shared__ float placeholder_shared[96]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) { __syncthreads(); if ((int)threadIdx.x < 8) { int4 _1 = make_int4((int)threadIdx.x * 43264 + rc_outer_outer * 676 + (int)blockIdx.x % 169+169*0, (int)threadIdx.x * 43264 + rc_outer_outer * 676 + (int)blockIdx.x % 169+169*1, (int)threadIdx.x * 43264 + rc_outer_outer * 676 + (int)blockIdx.x % 169+169*2, (int)threadIdx.x * 43264 + rc_outer_outer * 676 + (int)blockIdx.x % 169+169*3); int index0 = _1.x; int index1 = _1.y; int index2 = _1.z; int index3 = _1.w; ((float4*)(pad_temp_shared + (int)threadIdx.x * 4))[0] = make_float4(placeholder[index0],placeholder[index1],placeholder[index2],placeholder[index3]); } int index4 = (int)threadIdx.x; int index5 = (int)blockIdx.x / 169 * 6144 + ((int)threadIdx.x >> 2) * 256 + rc_outer_outer * 4 + ((int)threadIdx.x & 3); placeholder_shared[index4] = placeholder1[index5]; placeholder_shared[index4 + 32] = placeholder1[index5 + 2048]; placeholder_shared[index4 + 64] = placeholder1[index5 + 4096]; __syncthreads(); int index6 = ((int)threadIdx.x >> 2) * 4; int index7 = ((int)threadIdx.x & 3) * 24; compute[0] = compute[0] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[1] = compute[1] + (pad_temp_shared[index6] * placeholder_shared[index7 + 4]); compute[2] = compute[2] + (pad_temp_shared[index6] * placeholder_shared[index7 + 8]); compute[0] = compute[0] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 1]); compute[1] = compute[1] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 5]); compute[2] = compute[2] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 9]); compute[0] = compute[0] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 2]); compute[1] = compute[1] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 6]); compute[2] = compute[2] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 10]); compute[0] = compute[0] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 3]); compute[1] = compute[1] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 7]); compute[2] = compute[2] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 11]); compute[3] = compute[3] + (pad_temp_shared[index6] * placeholder_shared[index7 + 12]); compute[4] = compute[4] + (pad_temp_shared[index6] * placeholder_shared[index7 + 16]); compute[5] = compute[5] + (pad_temp_shared[index6] * placeholder_shared[index7 + 20]); compute[3] = compute[3] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 13]); compute[4] = compute[4] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 17]); compute[5] = compute[5] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7 + 21]); compute[3] = compute[3] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 14]); compute[4] = compute[4] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 18]); compute[5] = compute[5] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7 + 22]); compute[3] = compute[3] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 15]); compute[4] = compute[4] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 19]); compute[5] = compute[5] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7 + 23]); } for (int ax1_inner = 0; ax1_inner < 6; ++ax1_inner) { int index8 = ((int)threadIdx.x >> 2) * 8112 + (int)blockIdx.x / 169 * 4056 + ((int)threadIdx.x & 3) * 1014 + ax1_inner * 169 + (int)blockIdx.x % 169; int index9 = ax1_inner; int index10 = (int)blockIdx.x / 169 * 24 + ((int)threadIdx.x & 3) * 6 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmppa_v8egc
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[1728]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<9>; .reg .f32 %f<569>; .reg .b32 %r<75>; .reg .b64 %rd<35>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[20736]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[4096]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd16, [default_function_kernel0_param_0]; ld.param.u64 %rd17, [default_function_kernel0_param_1]; ld.param.u64 %rd18, [default_function_kernel0_param_2]; ld.param.u64 %rd19, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd20, %rd16; cvta.to.global.u64 %rd1, %rd17; cvta.to.global.u64 %rd2, %rd18; cvta.to.global.u64 %rd34, %rd19; add.u64 %rd32, %SPL, 0; mov.u32 %r68, 0; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(0+i*12+j*192)], %r68; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(4+i*12+j*192)], %r68; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(8+i*12+j*192)], %r68; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(96+i*12+j*192)], %r68; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(100+i*12+j*192)], %r68; for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 9, 1): st.local.u32 [%rd32+(104+i*12+j*192)], %r68; mov.u32 %r26, %ctaid.x; mul.hi.s32 %r27, %r26, 954437177; shr.u32 %r28, %r27, 31; shr.s32 %r29, %r27, 1; add.s32 %r1, %r29, %r28; mul.lo.s32 %r30, %r1, 9; sub.s32 %r31, %r26, %r30; mul.lo.s32 %r2, %r31, 3; mov.u32 %r3, %tid.x; mul.lo.s32 %r4, %r3, 3; shl.b32 %r32, %r3, 2; mov.u32 %r33, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r34, %r33, %r32; add.s32 %r5, %r34, 96; add.s64 %rd5, %rd20, 864; add.s32 %r35, %r3, %r2; mad.lo.s32 %r6, %r1, 93312, %r35; $L__BB0_1: bar.sync 0; mad.lo.s32 %r37, %r68, 46656, %r6; mul.wide.s32 %rd22, %r37, 4; add.s64 %rd31, %rd5, %rd22; mov.u32 %r70, 96; mov.u32 %r69, %r5; $L__BB0_2: for.size.2 i in range(0, 16, 1): ld.global.nc.f32 %f(28+i*1), [%rd31+(-864+i*108)]; st.shared.f32 [%r69+(-96+i*12)], %f(28+i*1); add.s32 %r69, %r69, 192; add.s64 %rd31, %rd31, 1728; add.s32 %r70, %r70, 192; setp.ne.s32 %p1, %r70, 20832; @%p1 bra $L__BB0_2; shl.b32 %r13, %r68, 6; mov.u32 %r71, 0; $L__BB0_4: mad.lo.s32 %r15, %r71, 3, %r3; setp.gt.s32 %p2, %r15, 1023; shl.b32 %r39, %r15, 2; mov.u32 %r40, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r16, %r40, %r39; @%p2 bra $L__BB0_6; shl.b32 %r41, %r15, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(42+i*1), %r(41+i*-26), (-128+i*191); for.size.1 i in range(0, 2, 1): add.s32 %r(44+i*1), %r(43+i*1), %r(13+i*29); mul.wide.s32 %rd23, %r45, 4; add.s64 %rd24, %rd1, %rd23; ld.global.nc.f32 %f44, [%rd24]; st.shared.f32 [%r16], %f44; $L__BB0_6: add.s32 %r17, %r15, 3; setp.gt.s32 %p3, %r17, 1023; @%p3 bra $L__BB0_8; shl.b32 %r46, %r17, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(47+i*1), %r(46+i*-29), (-128+i*191); for.size.1 i in range(0, 2, 1): add.s32 %r(49+i*1), %r(48+i*1), %r(13+i*34); mul.wide.s32 %rd25, %r50, 4; add.s64 %rd26, %rd1, %rd25; ld.global.nc.f32 %f45, [%rd26]; st.shared.f32 [%r16+12], %f45; $L__BB0_8: add.s32 %r71, %r71, 2; setp.ne.s32 %p4, %r71, 342; @%p4 bra $L__BB0_4; bar.sync 0; mov.u32 %r51, 0; mov.u32 %r72, %r51; $L__BB0_10: mad.lo.s32 %r53, %r72, 81, %r4; shl.b32 %r54, %r53, 2; add.s32 %r56, %r33, %r54; for.size.2 i in range(0, 3, 1): for.size.1 j in range(0, 9, 1): ld.shared.f32 %f(1+i*9+j*1), [%r56+(0+i*4+j*36)]; mov.u32 %r73, %r51; $L__BB0_11: shl.b32 %r58, %r73, 9; add.s32 %r59, %r58, %r72; mul.lo.s32 %r60, %r73, 24; mul.wide.s32 %rd27, %r60, 4; add.s64 %rd28, %rd32, %rd27; shl.b32 %r61, %r59, 2; add.s32 %r63, %r40, %r61; for.size.28 i in range(0, 8, 1): ld.shared.f32 %f(46+i*19), [%r63+(0+i*256)]; ld.local.f32 %f(47+i*19), [%rd28+(0+i*12)]; fma.rn.f32 %f(48+i*19), %f1, %f(46+i*19), %f(47+i*19); st.local.f32 [%rd28+(0+i*12)], %f(48+i*19); ld.local.f32 %f(49+i*19), [%rd28+(192+i*12)]; fma.rn.f32 %f(50+i*19), %f2, %f(46+i*19), %f(49+i*19); st.local.f32 [%rd28+(192+i*12)], %f(50+i*19); ld.local.f32 %f(51+i*19), [%rd28+(384+i*12)]; fma.rn.f32 %f(52+i*19), %f3, %f(46+i*19), %f(51+i*19); st.local.f32 [%rd28+(384+i*12)], %f(52+i*19); ld.local.f32 %f(53+i*19), [%rd28+(576+i*12)]; fma.rn.f32 %f(54+i*19), %f4, %f(46+i*19), %f(53+i*19); st.local.f32 [%rd28+(576+i*12)], %f(54+i*19); ld.local.f32 %f(55+i*19), [%rd28+(768+i*12)]; fma.rn.f32 %f(56+i*19), %f5, %f(46+i*19), %f(55+i*19); st.local.f32 [%rd28+(768+i*12)], %f(56+i*19); ld.local.f32 %f(57+i*19), [%rd28+(960+i*12)]; fma.rn.f32 %f(58+i*19), %f6, %f(46+i*19), %f(57+i*19); st.local.f32 [%rd28+(960+i*12)], %f(58+i*19); ld.local.f32 %f(59+i*19), [%rd28+(1152+i*12)]; fma.rn.f32 %f(60+i*19), %f7, %f(46+i*19), %f(59+i*19); st.local.f32 [%rd28+(1152+i*12)], %f(60+i*19); ld.local.f32 %f(61+i*19), [%rd28+(1344+i*12)]; fma.rn.f32 %f(62+i*19), %f8, %f(46+i*19), %f(61+i*19); st.local.f32 [%rd28+(1344+i*12)], %f(62+i*19); ld.local.f32 %f(63+i*19), [%rd28+(1536+i*12)]; fma.rn.f32 %f(64+i*19), %f9, %f(46+i*19), %f(63+i*19); st.local.f32 [%rd28+(1536+i*12)], %f(64+i*19); for.size.4 i in range(0, 8, 1): for.size.3 j in range(0, 9, 1): ld.local.f32 %f(198+i*18+j*2), [%rd28+(4+i*12+j*192)]; fma.rn.f32 %f(199+i*18+j*2), %f(10+j*1), %f(46+j*19), %f(198+i*18+j*2); st.local.f32 [%rd28+(4+i*12+j*192)], %f(199+i*18+j*2); for.size.4 i in range(0, 8, 1): for.size.3 j in range(0, 9, 1): ld.local.f32 %f(342+i*18+j*2), [%rd28+(8+i*12+j*192)]; fma.rn.f32 %f(343+i*18+j*2), %f(19+j*1), %f(46+j*19), %f(342+i*18+j*2); st.local.f32 [%rd28+(8+i*12+j*192)], %f(343+i*18+j*2); add.s32 %r64, %r73, 1; mov.u32 %r73, 1; setp.lt.u32 %p5, %r64, 2; @%p5 bra $L__BB0_11; add.s32 %r72, %r72, 1; setp.lt.u32 %p6, %r72, 64; @%p6 bra $L__BB0_10; add.s32 %r68, %r68, 1; setp.lt.u32 %p7, %r68, 2; @%p7 bra $L__BB0_1; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(66+i*1), %r(1+i*2), (11664+i*-11637), %r(2+i*64); mul.wide.s32 %rd29, %r67, 4; add.s64 %rd30, %rd2, %rd29; add.s64 %rd33, %rd30, 1300; mov.u32 %r74, 0; $L__BB0_15: ld.local.f32 %f486, [%rd32]; ld.global.nc.f32 %f487, [%rd34]; add.f32 %f488, %f486, %f487; mov.f32 %f489, 0f00000000; for.size.4 i in range(0, 2, 1): max.f32 %f(490+i*3), %f(488+i*4), %f489; st.global.f32 [%rd33+(-1300+i*324)], %f(490+i*3); ld.local.f32 %f(491+i*3), [%rd32+(192+i*192)]; add.f32 %f(492+i*3), %f(491+i*3), %f487; for.size.4 i in range(0, 6, 1): max.f32 %f(496+i*3), %f(495+i*3), %f489; st.global.f32 [%rd33+(-652+i*324)], %f(496+i*3); ld.local.f32 %f(497+i*3), [%rd32+(576+i*192)]; add.f32 %f(498+i*3), %f(497+i*3), %f487; for.size.10 i in range(0, 2, 1): for.size.4 j in range(0, 2, 1): max.f32 %f(514+i*27+j*3), %f(513+i*27+j*3), %f489; st.global.f32 [%rd33+(1292+i*4+j*-2588)], %f(514+i*27+j*3); ld.local.f32 %f(515+i*27+j*3), [%rd32+(4+i*4+j*192)]; add.f32 %f(516+i*27+j*3), %f(515+i*27+j*3), %f487; for.size.4 j in range(0, 7, 1): max.f32 %f(520+i*27+j*3), %f(519+i*27+j*3), %f489; st.global.f32 [%rd33+(-972+i*4+j*324)], %f(520+i*27+j*3); ld.local.f32 %f(521+i*27+j*3), [%rd32+(388+i*4+j*192)]; add.f32 %f(522+i*27+j*3), %f(521+i*27+j*3), %f487; max.f32 %f568, %f567, %f489; st.global.f32 [%rd33+1300], %f568; for.size.1 i in range(0, 2, 1): add.s64 %rd(34+i*-1), %rd(34+i*-1), (4+i*2912); add.s64 %rd32, %rd32, 12; add.s32 %r74, %r74, 1; setp.ne.s32 %p8, %r74, 16; @%p8 bra $L__BB0_15; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[432]; __shared__ float pad_temp_shared[5184]; __shared__ float placeholder_shared[1024]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 3; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 8; ++ff_inner_init) { int index0 = ff_outer_inner_init * 24 + ff_inner_init * 3 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 48] = 0.000000e+00f; compute[index0 + 96] = 0.000000e+00f; compute[index0 + 144] = 0.000000e+00f; compute[index0 + 192] = 0.000000e+00f; compute[index0 + 240] = 0.000000e+00f; compute[index0 + 288] = 0.000000e+00f; compute[index0 + 336] = 0.000000e+00f; compute[index0 + 384] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 1728; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3 + (int)threadIdx.x; int index2 = (int)blockIdx.x / 9 * 93312 + rc_outer_outer * 46656 + ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 27 + (int)blockIdx.x % 9 * 3 + (int)threadIdx.x; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 342; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x < 1024) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x >> 6) * 128 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 3; ++xx_outer_inner) { for (int ff_inner = 0; ff_inner < 8; ++ff_inner) { int index5 = ff_outer_inner * 24 + ff_inner * 3 + xx_outer_inner; int index6 = rc_outer_inner * 81 + (int)threadIdx.x * 3 + xx_outer_inner; int index7 = ff_outer_inner * 512 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 48] = compute[index5 + 48] + (pad_temp_shared[index6 + 9] * placeholder_shared[index7]); compute[index5 + 96] = compute[index5 + 96] + (pad_temp_shared[index6 + 18] * placeholder_shared[index7]); compute[index5 + 144] = compute[index5 + 144] + (pad_temp_shared[index6 + 27] * placeholder_shared[index7]); compute[index5 + 192] = compute[index5 + 192] + (pad_temp_shared[index6 + 36] * placeholder_shared[index7]); compute[index5 + 240] = compute[index5 + 240] + (pad_temp_shared[index6 + 45] * placeholder_shared[index7]); compute[index5 + 288] = compute[index5 + 288] + (pad_temp_shared[index6 + 54] * placeholder_shared[index7]); compute[index5 + 336] = compute[index5 + 336] + (pad_temp_shared[index6 + 63] * placeholder_shared[index7]); compute[index5 + 384] = compute[index5 + 384] + (pad_temp_shared[index6 + 72] * placeholder_shared[index7]); } } } } } for (int ax1_inner = 0; ax1_inner < 16; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 3; ++ax3_inner) { int index8 = (int)blockIdx.x / 9 * 11664 + ax1_inner * 729 + (int)threadIdx.x * 27 + (int)blockIdx.x % 9 * 3 + ax3_inner; int index9 = ax1_inner * 3 + ax3_inner; int index10 = ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 81] = max(compute[index9 + 48] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 162] = max(compute[index9 + 96] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 243] = max(compute[index9 + 144] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 324] = max(compute[index9 + 192] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 405] = max(compute[index9 + 240] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 486] = max(compute[index9 + 288] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 567] = max(compute[index9 + 336] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 648] = max(compute[index9 + 384] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmp802yp5ur
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<1801>; .reg .b32 %r<188>; .reg .b64 %rd<39>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[248832]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[4096]; ld.param.u64 %rd1, [default_function_kernel0_param_0]; mov.u32 %r16, %tid.x; mul.hi.s32 %r17, %r16, 715827883; shr.u32 %r18, %r17, 31; shr.s32 %r19, %r17, 3; add.s32 %r20, %r19, %r18; mul.lo.s32 %r21, %r20, 15552; mul.hi.s32 %r22, %r16, 1431655766; shr.u32 %r23, %r22, 31; add.s32 %r24, %r22, %r23; mul.lo.s32 %r25, %r24, 3; sub.s32 %r26, %r16, %r25; mad.lo.s32 %r1, %r26, 81, %r21; mul.lo.s32 %r27, %r20, 48; sub.s32 %r28, %r16, %r27; mul.hi.s32 %r29, %r28, 1431655766; shr.u32 %r30, %r29, 31; add.s32 %r2, %r29, %r30; mov.u32 %r184, 0; mov.f32 %f1639, 0f00000000; cvta.to.global.u64 %rd5, %rd1; for.size.1 i in range(0, 161, 1): mov.f32 %f(1640+i*1), %f1639; $L__BB0_1: mov.u32 %r185, 0; bar.sync 0; $L__BB0_2: mov.u32 %r178, %ctaid.x; mul.lo.s32 %r177, %r178, 9; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(176+i*-141), %r(184+i*1), (46656+i*-46560), %r(177+i*-161); mul.hi.s32 %r36, %r35, -2032597691; add.s32 %r37, %r36, %r35; shr.u32 %r38, %r37, 31; shr.s32 %r39, %r37, 13; add.s32 %r40, %r39, %r38; mul.lo.s32 %r41, %r40, 15552; sub.s32 %r42, %r35, %r41; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(43+i*4), %r(42+i*-7), 954437177; shr.u32 %r(44+i*4), %r(43+i*4), 31; shr.s32 %r(45+i*4), %r(43+i*4), 1; add.s32 %r(46+i*4), %r(45+i*4), %r(44+i*4); mul.lo.s32 %r51, %r50, 9; sub.s32 %r52, %r35, %r51; add.s32 %r53, %r176, %r52; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(54+i*1), %r(40+i*6), (93312+i*-93285), %r(53+i*1); mul.wide.s32 %rd6, %r55, 4; add.s64 %rd7, %rd5, %rd6; ld.global.nc.f32 %f811, [%rd7]; shl.b32 %r56, %r35, 2; mov.u32 %r57, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r58, %r57, %r56; for.size.25 i in range(0, 3, 1): st.shared.f32 [%r58+(0+i*384)], %f(811+i*1); add.s32 %r(59+i*21), %r35, (96+i*96); mul.hi.s32 %r(60+i*21), %r(59+i*21), -2032597691; add.s32 %r(61+i*21), %r(60+i*21), %r(59+i*21); shr.u32 %r(62+i*21), %r(61+i*21), 31; shr.s32 %r(63+i*21), %r(61+i*21), 13; add.s32 %r(64+i*21), %r(63+i*21), %r(62+i*21); mul.lo.s32 %r(65+i*21), %r(64+i*21), 15552; sub.s32 %r(66+i*21), %r(59+i*21), %r(65+i*21); mul.hi.s32 %r(67+i*21), %r(66+i*21), 954437177; shr.u32 %r(68+i*21), %r(67+i*21), 31; shr.s32 %r(69+i*21), %r(67+i*21), 1; add.s32 %r(70+i*21), %r(69+i*21), %r(68+i*21); mul.hi.s32 %r(71+i*21), %r(59+i*21), 954437177; shr.u32 %r(72+i*21), %r(71+i*21), 31; shr.s32 %r(73+i*21), %r(71+i*21), 1; add.s32 %r(74+i*21), %r(73+i*21), %r(72+i*21); mul.lo.s32 %r(75+i*21), %r(74+i*21), 9; sub.s32 %r(76+i*21), %r(59+i*21), %r(75+i*21); add.s32 %r(77+i*21), %r176, %r(76+i*21); mad.lo.s32 %r(78+i*21), %r(64+i*21), 93312, %r(77+i*21); mad.lo.s32 %r(79+i*21), %r(70+i*21), 27, %r(78+i*21); mul.wide.s32 %rd(8+i*2), %r(79+i*21), 4; add.s64 %rd(9+i*2), %rd5, %rd(8+i*2); ld.global.nc.f32 %f(812+i*1), [%rd(9+i*2)+0]; st.shared.f32 [%r58+1152], %f814; add.s32 %r185, %r185, 4; setp.ne.s32 %p1, %r185, 648; @%p1 bra $L__BB0_2; ld.param.u64 %rd36, [default_function_kernel0_param_1]; setp.gt.s32 %p2, %r16, 63; shr.u32 %r123, %r16, 6; and.b32 %r124, %r16, 63; bfi.b32 %r125, %r123, %r124, 7, 25; shl.b32 %r7, %r184, 6; add.s32 %r126, %r125, %r7; cvta.to.global.u64 %rd14, %rd36; mul.wide.s32 %rd15, %r126, 4; add.s64 %rd16, %rd14, %rd15; ld.global.nc.f32 %f815, [%rd16]; shl.b32 %r127, %r16, 2; mov.u32 %r128, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r129, %r128, %r127; st.shared.f32 [%r129], %f815; add.s32 %r130, %r16, 96; shr.u32 %r131, %r130, 6; add.s32 %r132, %r16, 32; and.b32 %r133, %r132, 63; for.size.10 i in range(0, 2, 1): bfi.b32 %r(134+i*4), %r(131+i*6), %r133, 7, 25; add.s32 %r(135+i*4), %r(134+i*4), %r7; mul.wide.s32 %rd(17+i*2), %r(135+i*4), 4; add.s64 %rd(18+i*2), %rd14, %rd(17+i*2); ld.global.nc.f32 %f(816+i*2), [%rd(18+i*2)+0]; st.shared.f32 [%r129+(384+i*768)], %f(816+i*2); ld.global.nc.f32 %f(817+i*2), [%rd16+(1536+i*1536)]; st.shared.f32 [%r129+(768+i*768)], %f(817+i*2); add.s32 %r(136+i*4), %r16, (288+i*192); shr.u32 %r(137+i*4), %r(136+i*4), 6; for.size.10 i in range(0, 2, 1): bfi.b32 %r(142+i*4), %r(141+i*4), %r133, 7, 25; add.s32 %r(143+i*4), %r(142+i*4), %r7; mul.wide.s32 %rd(21+i*2), %r(143+i*4), 4; add.s64 %rd(22+i*2), %rd14, %rd(21+i*2); ld.global.nc.f32 %f(820+i*2), [%rd(22+i*2)+0]; st.shared.f32 [%r129+(1920+i*768)], %f(820+i*2); ld.global.nc.f32 %f(821+i*2), [%rd16+(4608+i*1536)]; st.shared.f32 [%r129+(2304+i*768)], %f(821+i*2); add.s32 %r(144+i*4), %r16, (672+i*192); shr.u32 %r(145+i*4), %r(144+i*4), 6; bfi.b32 %r150, %r149, %r133, 7, 25; add.s32 %r151, %r150, %r7; mul.wide.s32 %rd25, %r151, 4; add.s64 %rd26, %rd14, %rd25; ld.global.nc.f32 %f824, [%rd26]; st.shared.f32 [%r129+3456], %f824; @%p2 bra $L__BB0_5; add.s32 %r153, %r16, %r7; mul.wide.s32 %rd28, %r153, 4; add.s64 %rd29, %rd14, %rd28; ld.global.nc.f32 %f825, [%rd29+7680]; st.shared.f32 [%r129+3840], %f825; $L__BB0_5: bar.sync 0; mov.u32 %r186, 0; $L__BB0_6: mov.u32 %r187, 0; mad.lo.s32 %r9, %r186, 3888, %r1; $L__BB0_7: for.size.1 i in range(0, 2, 1): shl.b32 %r(182+i*-1), %r(2+i*184), (6+i*-2); add.s32 %r180, %r181, %r182; mov.u32 %r179, _ZZ24default_function_kernel0E18placeholder_shared; mov.u32 %r173, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r161, %r187, 243, %r9; for.size.2 i in range(0, 2, 1): add.s32 %r(162+i*3), %r(180+i*-1), %r(187+i*-24); shl.b32 %r(163+i*3), %r(162+i*-1), 2; add.s32 %r168, %r173, %r166; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(826+i*1), [%r(168+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1800+i*-81), %f(826+i*2), %f827, %f(1800+i*-81); ld.shared.f32 %f(828+i*1), [%r168+(124416+i*-124412)]; for.size.3 i in range(0, 79, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(1799+i*-1+j*-81), %f(829+i*2+j*1), %f827, %f(1799+i*-1+j*-81); ld.shared.f32 %f(830+i*2+j*1), [%r168+(124420+i*4+j*-124412)]; fma.rn.f32 %f1720, %f987, %f827, %f1720; ld.shared.f32 %f988, [%r168+124736]; fma.rn.f32 %f1639, %f988, %f827, %f1639; add.s32 %r187, %r187, 1; setp.ne.s32 %p3, %r187, 16; @%p3 bra $L__BB0_7; add.s32 %r186, %r186, 1; setp.lt.u32 %p4, %r186, 4; @%p4 bra $L__BB0_6; add.s32 %r184, %r184, 1; setp.lt.u32 %p5, %r184, 2; @%p5 bra $L__BB0_1; ld.param.u64 %rd38, [default_function_kernel0_param_2]; mov.u32 %r174, %ctaid.x; ld.param.u64 %rd37, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd30, %rd37; mul.wide.s32 %rd31, %r2, 4; add.s64 %rd32, %rd30, %rd31; mul.lo.s32 %r170, %r16, 243; mad.lo.s32 %r172, %r174, 9, %r170; ld.global.nc.f32 %f989, [%rd32]; add.f32 %f990, %f1800, %f989; mov.f32 %f991, 0f00000000; max.f32 %f992, %f990, %f991; cvta.to.global.u64 %rd33, %rd38; mul.wide.s32 %rd34, %r172, 4; add.s64 %rd35, %rd33, %rd34; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(0+i*4+j*93312)], %f(992+i*4+j*2); add.f32 %f(993+i*4+j*2), %f(1719+i*-1+j*80), %f989; max.f32 %f(994+i*4+j*2), %f(993+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(108+i*4+j*93312)], %f(1028+i*4+j*2); add.f32 %f(1029+i*4+j*2), %f(1710+i*-1+j*80), %f989; max.f32 %f(1030+i*4+j*2), %f(1029+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(216+i*4+j*93312)], %f(1064+i*4+j*2); add.f32 %f(1065+i*4+j*2), %f(1701+i*-1+j*80), %f989; max.f32 %f(1066+i*4+j*2), %f(1065+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(324+i*4+j*93312)], %f(1100+i*4+j*2); add.f32 %f(1101+i*4+j*2), %f(1692+i*-1+j*80), %f989; max.f32 %f(1102+i*4+j*2), %f(1101+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(432+i*4+j*93312)], %f(1136+i*4+j*2); add.f32 %f(1137+i*4+j*2), %f(1683+i*-1+j*80), %f989; max.f32 %f(1138+i*4+j*2), %f(1137+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(540+i*4+j*93312)], %f(1172+i*4+j*2); add.f32 %f(1173+i*4+j*2), %f(1674+i*-1+j*80), %f989; max.f32 %f(1174+i*4+j*2), %f(1173+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(648+i*4+j*93312)], %f(1208+i*4+j*2); add.f32 %f(1209+i*4+j*2), %f(1665+i*-1+j*80), %f989; max.f32 %f(1210+i*4+j*2), %f(1209+i*4+j*2), %f991; for.size.4 i in range(0, 9, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(756+i*4+j*93312)], %f(1244+i*4+j*2); add.f32 %f(1245+i*4+j*2), %f(1656+i*-1+j*80), %f989; max.f32 %f(1246+i*4+j*2), %f(1245+i*4+j*2), %f991; for.size.4 i in range(0, 8, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd35+(864+i*4+j*93312)], %f(1280+i*4+j*2); add.f32 %f(1281+i*4+j*2), %f(1647+i*-1+j*80), %f989; max.f32 %f(1282+i*4+j*2), %f(1281+i*4+j*2), %f991; st.global.f32 [%rd35+896], %f1312; add.f32 %f1313, %f1639, %f989; max.f32 %f1314, %f1313, %f991; st.global.f32 [%rd35+94208], %f1314; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[162]; __shared__ float pad_temp_shared[62208]; __shared__ float placeholder_shared[1024]; for (int yy_inner_init = 0; yy_inner_init < 9; ++yy_inner_init) { for (int xx_inner_init = 0; xx_inner_init < 9; ++xx_inner_init) { int index0 = yy_inner_init * 9 + xx_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 81] = 0.000000e+00f; } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 648; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 96 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 96 + (int)threadIdx.x) / 15552 * 93312 + rc_outer_outer * 46656 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 96 + (int)threadIdx.x) % 15552 / 9 * 27 + (int)blockIdx.x * 9 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 96 + (int)threadIdx.x) % 9; pad_temp_shared[index1] = placeholder[index2]; } int index3 = (int)threadIdx.x; int index4 = ((int)threadIdx.x >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; int index5 = ((int)threadIdx.x + 96 >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x + 32 & 63); placeholder_shared[index3 + 96] = placeholder1[index5]; placeholder_shared[index3 + 192] = placeholder1[index4 + 384]; int index6 = ((int)threadIdx.x + 288 >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x + 32 & 63); placeholder_shared[index3 + 288] = placeholder1[index6]; placeholder_shared[index3 + 384] = placeholder1[index4 + 768]; int index7 = ((int)threadIdx.x + 480 >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x + 32 & 63); placeholder_shared[index3 + 480] = placeholder1[index7]; placeholder_shared[index3 + 576] = placeholder1[index4 + 1152]; int index8 = ((int)threadIdx.x + 672 >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x + 32 & 63); placeholder_shared[index3 + 672] = placeholder1[index8]; placeholder_shared[index3 + 768] = placeholder1[index4 + 1536]; int index9 = ((int)threadIdx.x + 864 >> 6) * 128 + rc_outer_outer * 64 + ((int)threadIdx.x + 32 & 63); placeholder_shared[index3 + 864] = placeholder1[index9]; if ((int)threadIdx.x < 64) { int index10 = rc_outer_outer * 64 + (int)threadIdx.x + 1920; placeholder_shared[index3 + 960] = placeholder1[index10]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 4; ++rc_outer_inner) { for (int rc_inner = 0; rc_inner < 16; ++rc_inner) { for (int yy_inner = 0; yy_inner < 9; ++yy_inner) { for (int xx_inner = 0; xx_inner < 9; ++xx_inner) { int index11 = yy_inner * 9 + xx_inner; int index12 = (int)threadIdx.x / 48 * 15552 + rc_outer_inner * 3888 + rc_inner * 243 + (int)threadIdx.x % 3 * 81 + yy_inner * 9 + xx_inner; int index13 = (int)threadIdx.x % 48 / 3 * 64 + rc_outer_inner * 16 + rc_inner; compute[index11] = compute[index11] + pad_temp_shared[index12] * placeholder_shared[index13]; compute[index11 + 81] = compute[index11 + 81] + (pad_temp_shared[index12 + 31104] * placeholder_shared[index13]); } } } } } for (int ax2_inner = 0; ax2_inner < 9; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 9; ++ax3_inner) { int index14 = (int)threadIdx.x * 243 + ax2_inner * 27 + (int)blockIdx.x * 9 + ax3_inner; int index15 = ax2_inner * 9 + ax3_inner; int index16 = (int)threadIdx.x % 48 / 3; T_relu[index14] = max(compute[index15] + placeholder2[index16], 0.000000e+00f); T_relu[index14 + 23328] = max(compute[index15 + 81] + placeholder2[index16], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmp4zc3bzb0
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<7>; .reg .f32 %f<1592>; .reg .b32 %r<143>; .reg .b64 %rd<34>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[82944]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[4096]; ld.param.u64 %rd1, [default_function_kernel0_param_0]; ld.param.u64 %rd2, [default_function_kernel0_param_1]; mov.u32 %r16, %tid.x; mul.hi.s32 %r17, %r16, 1431655766; shr.u32 %r18, %r17, 31; add.s32 %r1, %r17, %r18; mov.u32 %r19, %ctaid.x; mul.lo.s32 %r20, %r1, 3; sub.s32 %r21, %r16, %r20; mad.lo.s32 %r2, %r19, 3, %r21; mov.u32 %r139, 0; mov.f32 %f1208, 0f00000000; cvta.to.global.u64 %rd5, %rd1; for.size.1 i in range(0, 191, 1): mov.f32 %f(1209+i*1), %f1208; $L__BB0_1: mov.u32 %r140, 0; bar.sync 0; mad.lo.s32 %r4, %r139, 46656, %r2; $L__BB0_2: mad.lo.s32 %r23, %r140, 9, %r1; mul.hi.s32 %r24, %r23, 1272582903; shr.u32 %r25, %r24, 31; shr.s32 %r26, %r24, 9; add.s32 %r27, %r26, %r25; mul.lo.s32 %r28, %r27, 1728; sub.s32 %r29, %r23, %r28; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(30+i*1), %r(27+i*2), (93312+i*-93285), %r(4+i*26); mul.wide.s32 %rd6, %r31, 4; add.s64 %rd7, %rd5, %rd6; ld.global.nc.f32 %f769, [%rd7]; mad.lo.s32 %r33, %r140, 27, %r16; shl.b32 %r34, %r33, 2; mov.u32 %r35, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r36, %r35, %r34; for.size.13 i in range(0, 7, 1): st.shared.f32 [%r36+(0+i*108)], %f(769+i*1); add.s32 %r(37+i*9), %r23, (9+i*9); mul.hi.s32 %r(38+i*9), %r(37+i*9), 1272582903; shr.u32 %r(39+i*9), %r(38+i*9), 31; shr.s32 %r(40+i*9), %r(38+i*9), 9; add.s32 %r(41+i*9), %r(40+i*9), %r(39+i*9); mul.lo.s32 %r(42+i*9), %r(41+i*9), 1728; sub.s32 %r(43+i*9), %r(37+i*9), %r(42+i*9); mad.lo.s32 %r(44+i*9), %r(41+i*9), 93312, %r4; mad.lo.s32 %r(45+i*9), %r(43+i*9), 27, %r(44+i*9); mul.wide.s32 %rd(8+i*2), %r(45+i*9), 4; add.s64 %rd(9+i*2), %rd5, %rd(8+i*2); ld.global.nc.f32 %f(770+i*1), [%rd(9+i*2)+0]; st.shared.f32 [%r36+756], %f776; add.s32 %r140, %r140, 8; setp.ne.s32 %p1, %r140, 768; @%p1 bra $L__BB0_2; mov.u32 %r141, 0; $L__BB0_4: mad.lo.s32 %r8, %r141, 27, %r16; setp.gt.s32 %p2, %r8, 1023; @%p2 bra $L__BB0_6; shl.b32 %r104, %r8, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(105+i*1), %r(104+i*-96), (-128+i*191); bfi.b32 %r107, %r139, %r106, 6, 26; add.s32 %r108, %r107, %r105; cvta.to.global.u64 %rd22, %rd2; mul.wide.s32 %rd23, %r108, 4; add.s64 %rd24, %rd22, %rd23; ld.global.nc.f32 %f777, [%rd24]; shl.b32 %r109, %r8, 2; mov.u32 %r110, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r111, %r110, %r109; st.shared.f32 [%r111], %f777; $L__BB0_6: add.s32 %r9, %r8, 27; setp.gt.s32 %p3, %r9, 1023; @%p3 bra $L__BB0_8; shl.b32 %r112, %r9, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(113+i*1), %r(112+i*-103), (-128+i*191); bfi.b32 %r115, %r139, %r114, 6, 26; add.s32 %r116, %r115, %r113; cvta.to.global.u64 %rd25, %rd2; mul.wide.s32 %rd26, %r116, 4; add.s64 %rd27, %rd25, %rd26; ld.global.nc.f32 %f778, [%rd27]; shl.b32 %r119, %r8, 2; mov.u32 %r120, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r121, %r120, %r119; st.shared.f32 [%r121+108], %f778; $L__BB0_8: add.s32 %r141, %r141, 2; setp.ne.s32 %p4, %r141, 38; @%p4 bra $L__BB0_4; bar.sync 0; mov.u32 %r142, 0; $L__BB0_10: mul.lo.s32 %r138, %r16, 3; mov.u32 %r135, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r124, %r142, 81, %r138; shl.b32 %r125, %r124, 2; add.s32 %r127, %r135, %r125; shl.b32 %r128, %r142, 2; mov.u32 %r129, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r130, %r129, %r128; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(779+i*1), [%r(130+i*-3)+0]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(1399+i*-48), %f(780+i*1), %f779, %f(1399+i*-48); ld.shared.f32 %f(781+i*1), [%r127+(20736+i*20736)]; for.size.5 i in range(0, 2, 1): fma.rn.f32 %f(1255+i*-3), %f783, %f(779+i*5), %f(1255+i*-3); ld.shared.f32 %f(784+i*1), [%r130+(256+i*256)]; fma.rn.f32 %f(1396+i*-3), %f780, %f(784+i*1), %f(1396+i*-3); fma.rn.f32 %f(1348+i*-3), %f781, %f(784+i*1), %f(1348+i*-3); fma.rn.f32 %f(1300+i*-3), %f782, %f(784+i*1), %f(1300+i*-3); for.size.5 i in range(0, 5, 1): fma.rn.f32 %f(1249+i*-3), %f783, %f(785+i*1), %f(1249+i*-3); ld.shared.f32 %f(786+i*1), [%r130+(768+i*256)]; fma.rn.f32 %f(1390+i*-3), %f780, %f(786+i*1), %f(1390+i*-3); fma.rn.f32 %f(1342+i*-3), %f781, %f(786+i*1), %f(1342+i*-3); fma.rn.f32 %f(1294+i*-3), %f782, %f(786+i*1), %f(1294+i*-3); for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1234+i*164), %f(783+i*8), %f(790+i*-11), %f(1234+i*164); ld.shared.f32 %f(791+i*1), [%r127+(4+i*20736)]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(1350+i*-48), %f(792+i*1), %f779, %f(1350+i*-48); ld.shared.f32 %f(793+i*1), [%r127+(41476+i*20736)]; for.size.1 i in range(0, 2, 1): fma.rn.f32 %f(1254+i*141), %f(794+i*-3), %f(779+i*5), %f(1254+i*141); for.size.1 i in range(0, 3, 1): fma.rn.f32 %f(1347+i*-48), %f(792+i*1), %f784, %f(1347+i*-48); for.size.2 i in range(0, 6, 1): for.size.1 j in range(0, 4, 1): fma.rn.f32 %f(1392+i*-3+j*-48), %f(791+j*1), %f(785+j*1), %f(1392+i*-3+j*-48); for.size.2 i in range(0, 4, 1): ld.shared.f32 %f(795+i*1), [%r127+(8+i*20736)]; fma.rn.f32 %f(1397+i*-48), %f(795+i*1), %f779, %f(1397+i*-48); for.size.2 i in range(0, 7, 1): for.size.1 j in range(0, 4, 1): fma.rn.f32 %f(1394+i*-3+j*-48), %f(795+j*1), %f(784+j*1), %f(1394+i*-3+j*-48); for.size.5 i in range(0, 8, 1): ld.shared.f32 %f(799+i*1), [%r130+(2048+i*256)]; fma.rn.f32 %f(1375+i*-3), %f780, %f(799+i*1), %f(1375+i*-3); fma.rn.f32 %f(1327+i*-3), %f781, %f(799+i*1), %f(1327+i*-3); fma.rn.f32 %f(1279+i*-3), %f782, %f(799+i*1), %f(1279+i*-3); fma.rn.f32 %f(1231+i*-3), %f783, %f(799+i*1), %f(1231+i*-3); for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 4, 1): fma.rn.f32 %f(1374+i*-3+j*-48), %f(791+j*1), %f(799+j*1), %f(1374+i*-3+j*-48); for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 4, 1): fma.rn.f32 %f(1373+i*-3+j*-48), %f(795+j*1), %f(799+j*1), %f(1373+i*-3+j*-48); add.s32 %r142, %r142, 1; setp.ne.s32 %p5, %r142, 64; @%p5 bra $L__BB0_10; add.s32 %r139, %r139, 1; setp.lt.u32 %p6, %r139, 2; @%p6 bra $L__BB0_1; ld.param.u64 %rd33, [default_function_kernel0_param_2]; ld.param.u64 %rd32, [default_function_kernel0_param_3]; mov.u32 %r136, %ctaid.x; mul.lo.s32 %r132, %r16, 27; mad.lo.s32 %r134, %r136, 3, %r132; cvta.to.global.u64 %rd28, %rd32; ld.global.nc.f32 %f807, [%rd28]; add.f32 %f808, %f1399, %f807; mov.f32 %f809, 0f00000000; max.f32 %f810, %f808, %f809; cvta.to.global.u64 %rd29, %rd33; mul.wide.s32 %rd30, %r134, 4; add.s64 %rd31, %rd29, %rd30; for.size.3 i in range(0, 3, 1): st.global.f32 [%rd31+(0+i*46656)], %f(810+i*2); add.f32 %f(811+i*2), %f(1351+i*-48), %f807; max.f32 %f(812+i*2), %f(811+i*2), %f809; for.size.8 i in range(0, 2, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd31+(139968+i*4+j*-139964)], %f(816+i*8+j*2); add.f32 %f(817+i*8+j*2), %f(1398+i*-1+j*-48), %f807; max.f32 %f(818+i*8+j*2), %f(817+i*8+j*2), %f809; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd31+(46660+i*4+j*46656)], %f(820+i*8+j*2); add.f32 %f(821+i*8+j*2), %f(1302+i*-1+j*-48), %f807; max.f32 %f(822+i*8+j*2), %f(821+i*8+j*2), %f809; for.size.37 i in range(0, 15, 1): st.global.f32 [%rd31+(139976+i*2916)], %f(832+i*25); ld.global.nc.f32 %f(833+i*25), [%rd28+(4+i*4)]; add.f32 %f(834+i*25), %f(1396+i*-3), %f(833+i*25); max.f32 %f(835+i*25), %f(834+i*25), %f809; st.global.f32 [%rd31+(2916+i*2916)], %f(835+i*25); add.f32 %f(836+i*25), %f(1348+i*-3), %f(833+i*25); max.f32 %f(837+i*25), %f(836+i*25), %f809; st.global.f32 [%rd31+(49572+i*2916)], %f(837+i*25); add.f32 %f(838+i*25), %f(1300+i*-3), %f(833+i*25); max.f32 %f(839+i*25), %f(838+i*25), %f809; st.global.f32 [%rd31+(96228+i*2916)], %f(839+i*25); add.f32 %f(840+i*25), %f(1252+i*-3), %f(833+i*25); max.f32 %f(841+i*25), %f(840+i*25), %f809; st.global.f32 [%rd31+(142884+i*2916)], %f(841+i*25); add.f32 %f(842+i*25), %f(1395+i*-3), %f(833+i*25); max.f32 %f(843+i*25), %f(842+i*25), %f809; st.global.f32 [%rd31+(2920+i*2916)], %f(843+i*25); add.f32 %f(844+i*25), %f(1347+i*-3), %f(833+i*25); max.f32 %f(845+i*25), %f(844+i*25), %f809; st.global.f32 [%rd31+(49576+i*2916)], %f(845+i*25); add.f32 %f(846+i*25), %f(1299+i*-3), %f(833+i*25); max.f32 %f(847+i*25), %f(846+i*25), %f809; st.global.f32 [%rd31+(96232+i*2916)], %f(847+i*25); add.f32 %f(848+i*25), %f(1251+i*-3), %f(833+i*25); max.f32 %f(849+i*25), %f(848+i*25), %f809; st.global.f32 [%rd31+(142888+i*2916)], %f(849+i*25); add.f32 %f(850+i*25), %f(1394+i*-3), %f(833+i*25); max.f32 %f(851+i*25), %f(850+i*25), %f809; st.global.f32 [%rd31+(2924+i*2916)], %f(851+i*25); add.f32 %f(852+i*25), %f(1346+i*-3), %f(833+i*25); max.f32 %f(853+i*25), %f(852+i*25), %f809; st.global.f32 [%rd31+(49580+i*2916)], %f(853+i*25); add.f32 %f(854+i*25), %f(1298+i*-3), %f(833+i*25); max.f32 %f(855+i*25), %f(854+i*25), %f809; st.global.f32 [%rd31+(96236+i*2916)], %f(855+i*25); add.f32 %f(856+i*25), %f(1250+i*-3), %f(833+i*25); max.f32 %f(857+i*25), %f(856+i*25), %f809; st.global.f32 [%rd31+183716], %f1207; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[192]; __shared__ float pad_temp_shared[20736]; __shared__ float placeholder_shared[1024]; for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 3; ++xx_outer_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 8; ++ff_inner_init) { int index0 = ff_outer_inner_init * 24 + ff_inner_init * 3 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 48] = 0.000000e+00f; compute[index0 + 96] = 0.000000e+00f; compute[index0 + 144] = 0.000000e+00f; } } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 768; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 27 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 9 + (int)threadIdx.x / 3) / 1728 * 93312 + rc_outer_outer * 46656 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 9 + (int)threadIdx.x / 3) % 1728 * 27 + (int)blockIdx.x * 3 + (int)threadIdx.x % 3; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 38; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 27 + (int)threadIdx.x < 1024) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 27 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 27 + (int)threadIdx.x >> 6) * 128 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 27 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 64; ++rc_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 3; ++xx_outer_inner) { for (int ff_inner = 0; ff_inner < 8; ++ff_inner) { int index5 = ff_outer_inner * 24 + ff_inner * 3 + xx_outer_inner; int index6 = rc_outer_inner * 81 + (int)threadIdx.x * 3 + xx_outer_inner; int index7 = ff_outer_inner * 512 + ff_inner * 64 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 48] = compute[index5 + 48] + (pad_temp_shared[index6 + 5184] * placeholder_shared[index7]); compute[index5 + 96] = compute[index5 + 96] + (pad_temp_shared[index6 + 10368] * placeholder_shared[index7]); compute[index5 + 144] = compute[index5 + 144] + (pad_temp_shared[index6 + 15552] * placeholder_shared[index7]); } } } } } for (int ax1_inner = 0; ax1_inner < 16; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 3; ++ax3_inner) { int index8 = ax1_inner * 729 + (int)threadIdx.x * 27 + (int)blockIdx.x * 3 + ax3_inner; int index9 = ax1_inner * 3 + ax3_inner; int index10 = ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 11664] = max(compute[index9 + 48] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 23328] = max(compute[index9 + 96] + placeholder2[index10], 0.000000e+00f); T_relu[index8 + 34992] = max(compute[index9 + 144] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmpugqhoyzx
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<492>; .reg .b32 %r<121>; .reg .b64 %rd<23>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[93312]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[256]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; cvta.to.global.u64 %rd1, %rd5; cvta.to.global.u64 %rd3, %rd6; mov.u32 %r1, %tid.x; mul.hi.s32 %r26, %r1, 1272582903; shr.u32 %r27, %r26, 31; shr.s32 %r28, %r26, 3; add.s32 %r29, %r28, %r27; mul.lo.s32 %r30, %r29, 27; sub.s32 %r2, %r1, %r30; shl.b32 %r31, %r1, 5; mov.u32 %r32, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r3, %r32, %r31; mov.u32 %r4, %ctaid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(33+i*1), %r(4+i*-3), (10+i*-3); add.s32 %r5, %r33, %r34; shr.s32 %r35, %r26, 6; add.s32 %r6, %r35, %r27; mul.lo.s32 %r7, %r2, 9; mad.lo.s32 %r8, %r6, 5832, %r7; mul.lo.s32 %r36, %r6, 216; sub.s32 %r37, %r1, %r36; mul.hi.s32 %r38, %r37, 1272582903; shr.u32 %r39, %r38, 31; shr.s32 %r40, %r38, 3; add.s32 %r9, %r40, %r39; for.size.1 i in range(0, 2, 1): shl.b32 %r(10+i*31), %r(9+i*-8), (3+i*-1); mov.u32 %r42, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r43, %r42, %r41; for.size.1 i in range(0, 2, 1): add.s32 %r(11+i*1), %r(43+i*-14), (1728+i*-1712); mov.u32 %r116, 0; mov.f32 %f438, 0f00000000; for.size.1 i in range(0, 53, 1): mov.f32 %f(439+i*1), %f438; $L__BB0_1: bar.sync 0; mad.lo.s32 %r15, %r116, 5832, %r2; mov.u32 %r119, 16; mov.u32 %r117, %r12; mov.u32 %r118, %r11; $L__BB0_2: add.s32 %r45, %r117, -16; for.size.12 i in range(0, 2, 1): mul.hi.s32 %r(46+i*8), %r(45+i*72), 1272582903; shr.u32 %r(47+i*8), %r(46+i*8), 31; shr.s32 %r(48+i*8), %r(46+i*8), 6; add.s32 %r(49+i*8), %r(48+i*8), %r(47+i*8); mul.lo.s32 %r(50+i*8), %r(49+i*8), 216; sub.s32 %r(51+i*8), %r(45+i*72), %r(50+i*8); mad.lo.s32 %r(52+i*8), %r(49+i*8), 93312, %r15; mad.lo.s32 %r(53+i*8), %r(51+i*8), 27, %r(52+i*8); mul.wide.s32 %rd(9+i*2), %r(53+i*8), 4; add.s64 %rd(10+i*2), %rd1, %rd(9+i*2); ld.global.nc.f32 %f(217+i*1), [%rd(10+i*2)+0]; st.shared.f32 [%r118+(-1728+i*1728)], %f(217+i*1); for.size.1 i in range(0, 2, 1): add.s32 %r(118+i*-1), %r(118+i*-1), (3456+i*-3424); add.s32 %r119, %r119, 32; setp.ne.s32 %p1, %r119, 880; @%p1 bra $L__BB0_2; mov.u32 %r95, %tid.x; setp.gt.s32 %p2, %r95, 7; @%p2 bra $L__BB0_5; shl.b32 %r62, %r116, 3; add.s32 %r63, %r5, %r62; mul.wide.s32 %rd13, %r63, 4; add.s64 %rd14, %rd3, %rd13; for.size.1 i in range(0, 2, 1): ld.global.nc.v4.u32 {%r(64+i*8), %r(65+i*8), %r(66+i*8), %r(67+i*8)}, [%rd14+(0+i*16)]; for.size.1 i in range(0, 2, 1): st.shared.v4.u32 [%r3+(0+i*16)], {%r(64+i*8), %r(65+i*8), %r(66+i*8), %r(67+i*8)}; $L__BB0_5: bar.sync 0; mov.u32 %r120, 0; $L__BB0_6: mov.u32 %r97, _ZZ24default_function_kernel0E15pad_temp_shared; mov.u32 %r96, _ZZ24default_function_kernel0E18placeholder_shared; mad.lo.s32 %r81, %r120, 729, %r8; for.size.2 i in range(0, 2, 1): add.s32 %r(82+i*3), %r(120+i*-24), %r(10+i*73); shl.b32 %r(83+i*3), %r(82+i*-1), 2; add.s32 %r88, %r97, %r86; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(219+i*1), [%r(88+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(491+i*-9), %f(219+i*2), %f220, %f(491+i*-9); ld.shared.f32 %f(221+i*1), [%r88+(972+i*972)]; for.size.2 i in range(0, 3, 1): fma.rn.f32 %f(473+i*-9), %f(222+i*1), %f220, %f(473+i*-9); ld.shared.f32 %f(223+i*1), [%r88+(46656+i*972)]; for.size.9 i in range(0, 8, 1): for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(446+i*-1+j*44), %f(225+i*6+j*1), %f220, %f(446+i*-1+j*44); ld.shared.f32 %f(226+i*6+j*1), [%r88+(4+i*4+j*972)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(481+i*-1+j*-9), %f(227+i*6+j*1), %f220, %f(481+i*-1+j*-9); ld.shared.f32 %f(228+i*6+j*1), [%r88+(1948+i*4+j*44712)]; for.size.2 j in range(0, 2, 1): fma.rn.f32 %f(463+i*-1+j*-9), %f(229+i*6+j*1), %f220, %f(463+i*-1+j*-9); ld.shared.f32 %f(230+i*6+j*1), [%r88+(47632+i*4+j*972)]; fma.rn.f32 %f438, %f273, %f220, %f438; add.s32 %r120, %r120, 1; setp.ne.s32 %p3, %r120, 8; @%p3 bra $L__BB0_6; add.s32 %r116, %r116, 1; setp.lt.u32 %p4, %r116, 16; @%p4 bra $L__BB0_1; mov.u32 %r115, %tid.x; mul.hi.s32 %r114, %r115, 1272582903; shr.u32 %r113, %r114, 31; ld.param.u64 %rd22, [default_function_kernel0_param_2]; cvta.to.global.u64 %rd21, %rd22; shr.s32 %r112, %r114, 3; add.s32 %r111, %r112, %r113; mul.lo.s32 %r110, %r111, 27; sub.s32 %r109, %r115, %r110; mul.lo.s32 %r108, %r109, 9; add.s32 %r107, %r35, %r113; ld.param.u64 %rd20, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd19, %rd20; shr.s32 %r106, %r114, 6; add.s32 %r105, %r106, %r113; mul.lo.s32 %r104, %r105, 216; sub.s32 %r103, %r115, %r104; mul.hi.s32 %r102, %r103, 1272582903; shr.u32 %r101, %r102, 31; shr.s32 %r100, %r102, 3; add.s32 %r99, %r100, %r101; mov.u32 %r98, %ctaid.x; shl.b32 %r89, %r98, 3; add.s32 %r90, %r99, %r89; mul.wide.s32 %rd15, %r90, 4; add.s64 %rd16, %rd19, %rd15; mul.lo.s32 %r91, %r98, 5832; mad.lo.s32 %r92, %r105, 11664, %r91; add.s32 %r93, %r92, %r108; mad.lo.s32 %r94, %r99, 729, %r93; ld.global.nc.f32 %f274, [%rd16]; add.f32 %f275, %f491, %f274; mov.f32 %f276, 0f00000000; max.f32 %f277, %f275, %f276; mul.wide.s32 %rd17, %r94, 4; add.s64 %rd18, %rd21, %rd17; for.size.3 i in range(0, 3, 1): st.global.f32 [%rd18+(0+i*972)], %f(277+i*2); add.f32 %f(278+i*2), %f(482+i*-9), %f274; max.f32 %f(279+i*2), %f(278+i*2), %f276; for.size.12 i in range(0, 8, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd18+(93312+i*4+j*972)], %f(283+i*12+j*2); add.f32 %f(284+i*12+j*2), %f(455+i*-1+j*-9), %f274; max.f32 %f(285+i*12+j*2), %f(284+i*12+j*2), %f276; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd18+(95256+i*4+j*-95252)], %f(287+i*12+j*2); add.f32 %f(288+i*12+j*2), %f(490+i*-1+j*-9), %f274; max.f32 %f(289+i*12+j*2), %f(288+i*12+j*2), %f276; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd18+(976+i*4+j*972)], %f(291+i*12+j*2); add.f32 %f(292+i*12+j*2), %f(472+i*-1+j*-9), %f274; max.f32 %f(293+i*12+j*2), %f(292+i*12+j*2), %f276; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd18+(93344+i*972)], %f(379+i*2); add.f32 %f(380+i*2), %f(447+i*-9), %f274; max.f32 %f(381+i*2), %f(380+i*2), %f276; st.global.f32 [%rd18+95288], %f383; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[54]; __shared__ float pad_temp_shared[23328]; __shared__ float placeholder_shared[64]; for (int xx_inner_init = 0; xx_inner_init < 9; ++xx_inner_init) { int index0 = xx_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 9] = 0.000000e+00f; compute[index0 + 18] = 0.000000e+00f; compute[index0 + 27] = 0.000000e+00f; compute[index0 + 36] = 0.000000e+00f; compute[index0 + 45] = 0.000000e+00f; } for (int rc_outer_outer = 0; rc_outer_outer < 16; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 54; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 432 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x / 27) / 216 * 93312 + rc_outer_outer * 5832 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 16 + (int)threadIdx.x / 27) % 216 * 27 + (int)threadIdx.x % 27; pad_temp_shared[index1] = placeholder[index2]; } if ((int)threadIdx.x < 8) { ulonglong4*(placeholder_shared + (int)threadIdx.x * 8)[0] = ulonglong4*(placeholder1 + (int)blockIdx.x * 1024 + (int)threadIdx.x * 128 + rc_outer_outer * 8)[0]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int xx_inner = 0; xx_inner < 9; ++xx_inner) { int index3 = xx_inner; int index4 = (int)threadIdx.x / 216 * 5832 + rc_outer_inner * 729 + (int)threadIdx.x % 27 * 9 + xx_inner; int index5 = (int)threadIdx.x % 216 / 27 * 8 + rc_outer_inner; compute[index3] = compute[index3] + pad_temp_shared[index4] * placeholder_shared[index5]; compute[index3 + 9] = compute[index3 + 9] + (pad_temp_shared[index4 + 243] * placeholder_shared[index5]); compute[index3 + 18] = compute[index3 + 18] + (pad_temp_shared[index4 + 486] * placeholder_shared[index5]); compute[index3 + 27] = compute[index3 + 27] + (pad_temp_shared[index4 + 11664] * placeholder_shared[index5]); compute[index3 + 36] = compute[index3 + 36] + (pad_temp_shared[index4 + 11907] * placeholder_shared[index5]); compute[index3 + 45] = compute[index3 + 45] + (pad_temp_shared[index4 + 12150] * placeholder_shared[index5]); } } } for (int ax3_inner = 0; ax3_inner < 9; ++ax3_inner) { int index6 = (int)threadIdx.x / 216 * 11664 + (int)blockIdx.x * 5832 + (int)threadIdx.x % 216 / 27 * 729 + (int)threadIdx.x % 27 * 9 + ax3_inner; int index7 = ax3_inner; int index8 = (int)blockIdx.x * 8 + (int)threadIdx.x % 216 / 27; T_relu[index6] = max(compute[index7] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 243] = max(compute[index7 + 9] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 486] = max(compute[index7 + 18] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 23328] = max(compute[index7 + 27] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 23571] = max(compute[index7 + 36] + placeholder2[index8], 0.000000e+00f); T_relu[index6 + 23814] = max(compute[index7 + 45] + placeholder2[index8], 0.000000e+00f); } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmpw_8ivh2v
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<6>; .reg .f32 %f<98>; .reg .b32 %r<95>; .reg .b64 %rd<43>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[5832]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[8]; ld.param.u64 %rd19, [default_function_kernel0_param_0]; ld.param.u64 %rd20, [default_function_kernel0_param_1]; ld.param.u64 %rd21, [default_function_kernel0_param_2]; ld.param.u64 %rd18, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd22, %rd20; cvta.to.global.u64 %rd1, %rd21; mov.u32 %r15, %ctaid.x; shr.s32 %r1, %r15, 3; mov.u32 %r2, %tid.x; shl.b32 %r16, %r2, 2; mov.u32 %r17, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r3, %r17, %r16; add.s32 %r18, %r2, 648; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(19+i*4), %r(18+i*-16), (1508246403+i*-235663500); shr.u32 %r(20+i*4), %r(19+i*4), 31; shr.s32 %r(21+i*4), %r(19+i*4), (8+i*-5); add.s32 %r(22+i*4), %r(21+i*4), %r(20+i*4); add.s32 %r27, %r26, 24; mul.hi.s32 %r28, %r27, 1272582903; shr.u32 %r29, %r28, 31; shr.s32 %r30, %r28, 3; add.s32 %r31, %r30, %r29; for.size.2 i in range(0, 2, 1): mul.lo.s32 %r(32+i*2), %r(31+i*-5), 27; sub.s32 %r(33+i*2), %r(27+i*-25), %r(32+i*2); for.size.5 i in range(0, 2, 1): add.s32 %r(36+i*5), %r2, (972+i*324); mul.hi.s32 %r(37+i*5), %r(36+i*5), 1508246403; shr.u32 %r(38+i*5), %r(37+i*5), 31; shr.s32 %r(39+i*5), %r(37+i*5), 8; add.s32 %r(40+i*5), %r(39+i*5), %r(38+i*5); and.b32 %r4, %r15, 7; shl.b32 %r46, %r15, 8; and.b32 %r47, %r46, 1792; mov.u32 %r48, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r5, %r48, %r16; mul.hi.s32 %r49, %r2, 424194301; shr.u32 %r50, %r49, 31; for.size.2 i in range(0, 2, 1): shr.s32 %r(51+i*1), %r49, (4+i*-1); add.s32 %r(6+i*47), %r(51+i*1), %r50; mul.lo.s32 %r54, %r53, 81; sub.s32 %r55, %r2, %r54; for.size.4 i in range(0, 2, 1): mul.hi.s32 %r(56+i*4), %r(55+i*-53), 954437177; shr.u32 %r(57+i*4), %r(56+i*4), 31; shr.s32 %r(58+i*4), %r(56+i*4), 1; add.s32 %r(59+i*4), %r(58+i*4), %r(57+i*4); mul.lo.s32 %r64, %r63, 9; sub.s32 %r65, %r2, %r64; mul.lo.s32 %r7, %r65, 3; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(66+i*1), %r(6+i*53), (729+i*-648), %r(7+i*59); shl.b32 %r68, %r67, 2; add.s32 %r8, %r17, %r68; mul.lo.s32 %r69, %r6, 162; sub.s32 %r9, %r2, %r69; mul.hi.s32 %r70, %r9, 424194301; shr.u32 %r71, %r70, 31; shr.s32 %r72, %r70, 3; for.size.2 i in range(0, 2, 1): add.s32 %r(10+i*1), %r(72+i*-24), %r(71+i*2); shl.b32 %r(73+i*1), %r(10+i*-8), (2+i*5); add.s32 %r75, %r47, %r74; mul.wide.s32 %rd23, %r75, 4; add.s64 %rd24, %rd22, %rd23; add.s64 %rd42, %rd24, 4; mul.lo.s32 %r76, %r1, 186624; add.s32 %r77, %r2, %r76; cvta.to.global.u64 %rd25, %rd19; mul.wide.s32 %rd26, %r77, 4; add.s64 %rd27, %rd25, %rd26; add.s64 %rd41, %rd27, 2916; mad.lo.s32 %r78, %r33, 27, %r35; add.s32 %r79, %r78, %r76; for.size.3 i in range(0, 2, 1): mad.lo.s32 %r(80+i*1), %r(22+i*18), 93312, %r(79+i*-2); mul.wide.s32 %rd(28+i*1), %r(80+i*1), 4; add.s64 %rd(40+i*-10), %rd25, %rd(28+i*1); add.s64 %rd39, %rd30, 3888; mad.lo.s32 %r82, %r45, 93312, %r77; mul.wide.s32 %rd31, %r82, 4; add.s64 %rd32, %rd25, %rd31; add.s64 %rd38, %rd32, 5184; mov.u32 %r94, 0; mov.f32 %f89, 0f00000000; for.size.1 i in range(0, 8, 1): mov.f32 %f(90+i*1), %f89; $L__BB0_1: bar.sync 0; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(37+i*1), [%rd41+(-2916+i*1296)]; st.shared.f32 [%r3+(0+i*1296)], %f(37+i*1); ld.global.nc.f32 %f39, [%rd40]; st.shared.f32 [%r3+2592], %f39; ld.global.nc.f32 %f40, [%rd39+-2916]; st.shared.f32 [%r3+3888], %f40; setp.gt.s32 %p1, %r2, 161; @%p1 bra $L__BB0_3; ld.global.nc.f32 %f41, [%rd38+-2916]; st.shared.f32 [%r3+5184], %f41; $L__BB0_3: setp.gt.s32 %p2, %r2, 1; add.s64 %rd12, %rd42, -4; @%p2 bra $L__BB0_5; ld.global.nc.f32 %f42, [%rd12]; st.shared.f32 [%r5], %f42; $L__BB0_5: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(43+i*1), [%r(11+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(10+i*1), %f(44+i*1), %f43, %f(97+i*-1); ld.shared.f32 %f(45+i*1), [%r8+(4+i*4)]; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 3, 1): fma.rn.f32 %f(12+i*3+j*1), %f(46+i*3+j*1), %f43, %f(95+i*-3+j*-1); ld.shared.f32 %f(47+i*3+j*1), [%r8+(108+i*108+j*4)]; fma.rn.f32 %f18, %f52, %f43, %f89; bar.sync 0; for.size.2 i in range(0, 2, 1): ld.global.nc.f32 %f(53+i*1), [%rd41+(0+i*1296)]; st.shared.f32 [%r3+(0+i*1296)], %f(53+i*1); ld.global.nc.f32 %f55, [%rd40+2916]; st.shared.f32 [%r3+2592], %f55; ld.global.nc.f32 %f56, [%rd39]; st.shared.f32 [%r3+3888], %f56; @%p1 bra $L__BB0_7; ld.global.nc.f32 %f57, [%rd38]; st.shared.f32 [%r3+5184], %f57; $L__BB0_7: @%p2 bra $L__BB0_9; ld.global.nc.f32 %f58, [%rd12+4]; st.shared.f32 [%r5], %f58; $L__BB0_9: bar.sync 0; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(59+i*1), [%r(11+i*-3)+0]; for.size.2 i in range(0, 2, 1): fma.rn.f32 %f(97+i*-1), %f(60+i*1), %f59, %f(10+i*1); ld.shared.f32 %f(61+i*1), [%r8+(4+i*4)]; for.size.3 i in range(0, 2, 1): for.size.2 j in range(0, 3, 1): fma.rn.f32 %f(95+i*-3+j*-1), %f(62+i*3+j*1), %f59, %f(12+i*3+j*1); ld.shared.f32 %f(63+i*3+j*1), [%r8+(108+i*108+j*4)]; fma.rn.f32 %f89, %f68, %f59, %f18; for.size.1 i in range(0, 2, 1): add.s64 %rd(42+i*-1), %rd(42+i*-1), (8+i*5824); for.size.1 i in range(0, 3, 1): add.s64 %rd(40+i*-1), %rd(40+i*-1), 5832; add.s32 %r94, %r94, 2; setp.ne.s32 %p5, %r94, 128; @%p5 bra $L__BB0_1; cvta.to.global.u64 %rd33, %rd18; shl.b32 %r83, %r4, 1; add.s32 %r84, %r83, %r10; mul.wide.s32 %rd34, %r84, 4; add.s64 %rd35, %rd33, %rd34; mul.lo.s32 %r85, %r6, 11664; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(86+i*1), %r(1+i*3), (23328+i*-21870), %r(85+i*1); mul.hi.s32 %r88, %r9, 954437177; shr.u32 %r89, %r88, 31; shr.s32 %r90, %r88, 1; for.size.1 i in range(0, 2, 1): add.s32 %r(91+i*1), %r(90+i*-3), %r(89+i*-82); mad.lo.s32 %r93, %r91, 81, %r92; ld.global.nc.f32 %f69, [%rd35]; add.f32 %f70, %f97, %f69; mov.f32 %f71, 0f00000000; max.f32 %f72, %f70, %f71; mul.wide.s32 %rd36, %r93, 4; add.s64 %rd37, %rd1, %rd36; for.size.4 i in range(0, 2, 1): for.size.3 j in range(0, 3, 1): st.global.f32 [%rd37+(0+i*108+j*4)], %f(72+i*6+j*2); add.f32 %f(73+i*6+j*2), %f(96+i*-3+j*-1), %f69; max.f32 %f(74+i*6+j*2), %f(73+i*6+j*2), %f71; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd37+(216+i*4)], %f(84+i*2); add.f32 %f(85+i*2), %f(90+i*-1), %f69; max.f32 %f(86+i*2), %f(85+i*2), %f71; st.global.f32 [%rd37+224], %f88; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[9]; __shared__ float pad_temp_shared[1458]; __shared__ float placeholder_shared[2]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; compute[6] = 0.000000e+00f; compute[7] = 0.000000e+00f; compute[8] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 128; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = ((int)blockIdx.x >> 3) * 186624 + rc_outer_outer * 729 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 324] = placeholder[index1 + 324]; int index2 = ((int)blockIdx.x >> 3) * 186624 + ((int)threadIdx.x + 648) / 729 * 93312 + rc_outer_outer * 729 + ((int)threadIdx.x / 27 + 24) % 27 * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 648] = placeholder[index2]; int index3 = ((int)blockIdx.x >> 3) * 186624 + ((int)threadIdx.x + 972) / 729 * 93312 + rc_outer_outer * 729 + ((int)threadIdx.x / 27 + 9) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 972] = placeholder[index3]; if ((int)threadIdx.x < 162) { int index4 = ((int)blockIdx.x >> 3) * 186624 + ((int)threadIdx.x + 1296) / 729 * 93312 + rc_outer_outer * 729 + ((int)threadIdx.x / 27 + 21) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 1296] = placeholder[index4]; } if ((int)threadIdx.x < 2) { int index5 = ((int)blockIdx.x & 7) * 256 + (int)threadIdx.x * 128 + rc_outer_outer; placeholder_shared[index0] = placeholder1[index5]; } __syncthreads(); int index6 = (int)threadIdx.x / 162 * 729 + (int)threadIdx.x % 81 / 9 * 81 + (int)threadIdx.x % 9 * 3; int index7 = (int)threadIdx.x % 162 / 81; compute[0] = compute[0] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[1] = compute[1] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[2] = compute[2] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[3] = compute[3] + (pad_temp_shared[index6 + 27] * placeholder_shared[index7]); compute[4] = compute[4] + (pad_temp_shared[index6 + 28] * placeholder_shared[index7]); compute[5] = compute[5] + (pad_temp_shared[index6 + 29] * placeholder_shared[index7]); compute[6] = compute[6] + (pad_temp_shared[index6 + 54] * placeholder_shared[index7]); compute[7] = compute[7] + (pad_temp_shared[index6 + 55] * placeholder_shared[index7]); compute[8] = compute[8] + (pad_temp_shared[index6 + 56] * placeholder_shared[index7]); } for (int ax2_inner = 0; ax2_inner < 3; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 3; ++ax3_inner) { int index8 = ((int)blockIdx.x >> 3) * 23328 + (int)threadIdx.x / 162 * 11664 + ((int)blockIdx.x & 7) * 1458 + (int)threadIdx.x % 162 / 9 * 81 + ax2_inner * 27 + (int)threadIdx.x % 9 * 3 + ax3_inner; int index9 = ax2_inner * 3 + ax3_inner; int index10 = ((int)blockIdx.x & 7) * 2 + (int)threadIdx.x % 162 / 81; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmpx0bz3usz
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<5>; .reg .f32 %f<1207>; .reg .b32 %r<137>; .reg .b64 %rd<33>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[186624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[512]; ld.param.u64 %rd1, [default_function_kernel0_param_0]; ld.param.u64 %rd2, [default_function_kernel0_param_1]; mov.u32 %r13, %tid.x; for.size.6 i in range(0, 2, 1): mul.hi.s32 %r(14+i*4), %r13, (1272582903+i*-848388602); shr.u32 %r(15+i*4), %r(14+i*4), 31; shr.s32 %r(16+i*4), %r(14+i*4), 3; add.s32 %r(1+i*3), %r(16+i*4), %r(15+i*4); mul.lo.s32 %r(17+i*4), %r(1+i*3), (27+i*54); sub.s32 %r(2+i*20), %r13, %r(17+i*4); mul.lo.s32 %r3, %r22, 9; mov.u32 %r134, 0; mov.f32 %f1063, 0f00000000; cvta.to.global.u64 %rd5, %rd1; cvta.to.global.u64 %rd22, %rd2; for.size.1 i in range(0, 143, 1): mov.f32 %f(1064+i*1), %f1063; $L__BB0_1: mov.u32 %r135, 0; bar.sync 0; $L__BB0_2: mad.lo.s32 %r24, %r135, 6, %r1; mul.hi.s32 %r25, %r24, 1272582903; shr.u32 %r26, %r25, 31; shr.s32 %r27, %r25, 7; add.s32 %r28, %r27, %r26; mul.lo.s32 %r29, %r28, 432; sub.s32 %r30, %r24, %r29; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(31+i*1), %r(134+i*-106), (11664+i*81648), %r(2+i*29); mad.lo.s32 %r33, %r30, 27, %r32; mul.wide.s32 %rd6, %r33, 4; add.s64 %rd7, %rd5, %rd6; ld.global.nc.f32 %f577, [%rd7]; mad.lo.s32 %r35, %r135, 162, %r13; shl.b32 %r36, %r35, 2; mov.u32 %r37, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r38, %r37, %r36; for.size.13 i in range(0, 7, 1): st.shared.f32 [%r38+(0+i*648)], %f(577+i*1); add.s32 %r(39+i*9), %r24, (6+i*6); mul.hi.s32 %r(40+i*9), %r(39+i*9), 1272582903; shr.u32 %r(41+i*9), %r(40+i*9), 31; shr.s32 %r(42+i*9), %r(40+i*9), 7; add.s32 %r(43+i*9), %r(42+i*9), %r(41+i*9); mul.lo.s32 %r(44+i*9), %r(43+i*9), 432; sub.s32 %r(45+i*9), %r(39+i*9), %r(44+i*9); mad.lo.s32 %r(46+i*9), %r(43+i*9), 93312, %r31; mad.lo.s32 %r(47+i*9), %r(45+i*9), 27, %r(46+i*9); mul.wide.s32 %rd(8+i*2), %r(47+i*9), 4; add.s64 %rd(9+i*2), %rd5, %rd(8+i*2); ld.global.nc.f32 %f(578+i*1), [%rd(9+i*2)+0]; st.shared.f32 [%r38+4536], %f584; add.s32 %r135, %r135, 8; setp.ne.s32 %p1, %r135, 288; @%p1 bra $L__BB0_2; setp.gt.s32 %p2, %r13, 127; @%p2 bra $L__BB0_5; shl.b32 %r103, %r134, 4; and.b32 %r105, %r13, 15; mov.u32 %r106, %ctaid.x; bfi.b32 %r107, %r106, %r105, 10, 22; shl.b32 %r108, %r13, 3; and.b32 %r109, %r108, -128; for.size.1 i in range(0, 2, 1): add.s32 %r(110+i*1), %r(107+i*3), %r(109+i*-6); mul.wide.s32 %rd23, %r111, 4; add.s64 %rd24, %rd22, %rd23; ld.global.nc.f32 %f585, [%rd24]; shl.b32 %r112, %r13, 2; mov.u32 %r113, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r114, %r113, %r112; st.shared.f32 [%r114], %f585; $L__BB0_5: bar.sync 0; mov.u32 %r136, 0; $L__BB0_6: shl.b32 %r133, %r4, 6; mov.u32 %r131, _ZZ24default_function_kernel0E15pad_temp_shared; mad.lo.s32 %r116, %r136, 729, %r3; shl.b32 %r117, %r116, 2; for.size.1 i in range(0, 2, 1): add.s32 %r(119+i*1), %r(131+i*5), %r(117+i*16); shl.b32 %r121, %r120, 2; mov.u32 %r122, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r123, %r122, %r121; for.size.1 i in range(0, 2, 1): ld.shared.f32 %f(586+i*1), [%r(123+i*-4)+0]; for.size.2 i in range(0, 8, 1): fma.rn.f32 %f(1206+i*-1), %f(587+i*1), %f586, %f(1206+i*-1); ld.shared.f32 %f(588+i*1), [%r119+(4+i*4)]; fma.rn.f32 %f1198, %f595, %f586, %f1198; ld.shared.f32 %f596, [%r123+64]; for.size.1 i in range(0, 9, 1): fma.rn.f32 %f(1197+i*-1), %f(587+i*1), %f596, %f(1197+i*-1); for.size.2 i in range(0, 9, 1): ld.shared.f32 %f(597+i*1), [%r119+(46656+i*4)]; fma.rn.f32 %f(1170+i*-1), %f(597+i*1), %f586, %f(1170+i*-1); for.size.1 i in range(0, 9, 1): fma.rn.f32 %f(1161+i*-1), %f(597+i*1), %f596, %f(1161+i*-1); for.size.10 i in range(0, 2, 1): ld.shared.f32 %f(606+i*1), [%r123+(128+i*64)]; fma.rn.f32 %f(1188+i*-9), %f587, %f(606+i*1), %f(1188+i*-9); fma.rn.f32 %f(1187+i*-9), %f588, %f(606+i*1), %f(1187+i*-9); fma.rn.f32 %f(1186+i*-9), %f589, %f(606+i*1), %f(1186+i*-9); fma.rn.f32 %f(1185+i*-9), %f590, %f(606+i*1), %f(1185+i*-9); fma.rn.f32 %f(1184+i*-9), %f591, %f(606+i*1), %f(1184+i*-9); fma.rn.f32 %f(1183+i*-9), %f592, %f(606+i*1), %f(1183+i*-9); fma.rn.f32 %f(1182+i*-9), %f593, %f(606+i*1), %f(1182+i*-9); fma.rn.f32 %f(1181+i*-9), %f594, %f(606+i*1), %f(1181+i*-9); fma.rn.f32 %f(1180+i*-9), %f595, %f(606+i*1), %f(1180+i*-9); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 9, 1): fma.rn.f32 %f(1152+i*-9+j*-1), %f(597+j*1), %f(606+j*1), %f(1152+i*-9+j*-1); for.size.5 i in range(0, 2, 1): for.size.2 j in range(0, 9, 1): ld.shared.f32 %f(608+i*9+j*1), [%r119+(93312+i*46656+j*4)]; fma.rn.f32 %f(1134+i*-36+j*-1), %f(608+i*9+j*1), %f586, %f(1134+i*-36+j*-1); for.size.1 j in range(0, 9, 1): fma.rn.f32 %f(1125+i*-36+j*-1), %f(608+i*9+j*1), %f596, %f(1125+i*-36+j*-1); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 9, 1): fma.rn.f32 %f(1116+i*-9+j*-1), %f(608+j*1), %f(606+j*1), %f(1116+i*-9+j*-1); for.size.2 i in range(0, 2, 1): for.size.1 j in range(0, 9, 1): fma.rn.f32 %f(1080+i*-9+j*-1), %f(617+j*1), %f(606+j*1), %f(1080+i*-9+j*-1); add.s32 %r136, %r136, 1; setp.ne.s32 %p3, %r136, 16; @%p3 bra $L__BB0_6; add.s32 %r134, %r134, 1; setp.lt.u32 %p4, %r134, 8; @%p4 bra $L__BB0_1; ld.param.u64 %rd32, [default_function_kernel0_param_2]; ld.param.u64 %rd31, [default_function_kernel0_param_3]; mov.u32 %r124, %ctaid.x; for.size.1 i in range(0, 2, 1): shl.b32 %r(125+i*1), %r(124+i*-120), (3+i*-1); add.s32 %r127, %r126, %r125; mul.lo.s32 %r128, %r124, 5832; mad.lo.s32 %r129, %r4, 2916, %r128; add.s32 %r130, %r129, %r3; cvta.to.global.u64 %rd25, %rd31; mul.wide.s32 %rd26, %r127, 4; add.s64 %rd27, %rd25, %rd26; ld.global.nc.f32 %f626, [%rd27]; add.f32 %f627, %f1206, %f626; mov.f32 %f628, 0f00000000; max.f32 %f629, %f627, %f628; cvta.to.global.u64 %rd28, %rd32; mul.wide.s32 %rd29, %r130, 4; add.s64 %rd30, %rd28, %rd29; for.size.3 i in range(0, 8, 1): st.global.f32 [%rd30+(0+i*4)], %f(629+i*2); add.f32 %f(630+i*2), %f(1205+i*-1), %f626; max.f32 %f(631+i*2), %f(630+i*2), %f628; for.size.28 i in range(0, 3, 1): st.global.f32 [%rd30+(32+i*2916)], %f(645+i*19); ld.global.nc.f32 %f(646+i*19), [%rd27+(4+i*4)]; add.f32 %f(647+i*19), %f(1197+i*-9), %f(646+i*19); max.f32 %f(648+i*19), %f(647+i*19), %f628; st.global.f32 [%rd30+(2916+i*2916)], %f(648+i*19); add.f32 %f(649+i*19), %f(1196+i*-9), %f(646+i*19); max.f32 %f(650+i*19), %f(649+i*19), %f628; st.global.f32 [%rd30+(2920+i*2916)], %f(650+i*19); add.f32 %f(651+i*19), %f(1195+i*-9), %f(646+i*19); max.f32 %f(652+i*19), %f(651+i*19), %f628; st.global.f32 [%rd30+(2924+i*2916)], %f(652+i*19); add.f32 %f(653+i*19), %f(1194+i*-9), %f(646+i*19); max.f32 %f(654+i*19), %f(653+i*19), %f628; st.global.f32 [%rd30+(2928+i*2916)], %f(654+i*19); add.f32 %f(655+i*19), %f(1193+i*-9), %f(646+i*19); max.f32 %f(656+i*19), %f(655+i*19), %f628; st.global.f32 [%rd30+(2932+i*2916)], %f(656+i*19); add.f32 %f(657+i*19), %f(1192+i*-9), %f(646+i*19); max.f32 %f(658+i*19), %f(657+i*19), %f628; st.global.f32 [%rd30+(2936+i*2916)], %f(658+i*19); add.f32 %f(659+i*19), %f(1191+i*-9), %f(646+i*19); max.f32 %f(660+i*19), %f(659+i*19), %f628; st.global.f32 [%rd30+(2940+i*2916)], %f(660+i*19); add.f32 %f(661+i*19), %f(1190+i*-9), %f(646+i*19); max.f32 %f(662+i*19), %f(661+i*19), %f628; st.global.f32 [%rd30+(2944+i*2916)], %f(662+i*19); add.f32 %f(663+i*19), %f(1189+i*-9), %f(646+i*19); max.f32 %f(664+i*19), %f(663+i*19), %f628; for.size.32 i in range(0, 3, 1): for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(8780+i*46656+j*37876)], %f(702+i*72+j*2); add.f32 %f(703+i*72+j*2), %f(1170+i*-36+j*-1), %f626; max.f32 %f(704+i*72+j*2), %f(703+i*72+j*2), %f628; for.size.3 j in range(0, 7, 1): st.global.f32 [%rd30+(46660+i*46656+j*4)], %f(706+i*72+j*2); add.f32 %f(707+i*72+j*2), %f(1168+i*-36+j*-1), %f626; max.f32 %f(708+i*72+j*2), %f(707+i*72+j*2), %f628; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(46688+i*46656+j*2884)], %f(720+i*72+j*2); add.f32 %f(721+i*72+j*2), %f(1161+i*-36+j*-1), %f646; max.f32 %f(722+i*72+j*2), %f(721+i*72+j*2), %f628; for.size.3 j in range(0, 7, 1): st.global.f32 [%rd30+(49576+i*46656+j*4)], %f(724+i*72+j*2); add.f32 %f(725+i*72+j*2), %f(1159+i*-36+j*-1), %f646; max.f32 %f(726+i*72+j*2), %f(725+i*72+j*2), %f628; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(49604+i*46656+j*2884)], %f(738+i*72+j*2); add.f32 %f(739+i*72+j*2), %f(1152+i*-36+j*-1), %f665; max.f32 %f(740+i*72+j*2), %f(739+i*72+j*2), %f628; for.size.3 j in range(0, 7, 1): st.global.f32 [%rd30+(52492+i*46656+j*4)], %f(742+i*72+j*2); add.f32 %f(743+i*72+j*2), %f(1150+i*-36+j*-1), %f665; max.f32 %f(744+i*72+j*2), %f(743+i*72+j*2), %f628; for.size.3 j in range(0, 2, 1): st.global.f32 [%rd30+(52520+i*46656+j*2884)], %f(756+i*72+j*2); add.f32 %f(757+i*72+j*2), %f(1143+i*-36+j*-1), %f684; max.f32 %f(758+i*72+j*2), %f(757+i*72+j*2), %f628; for.size.3 j in range(0, 7, 1): st.global.f32 [%rd30+(55408+i*46656+j*4)], %f(760+i*72+j*2); add.f32 %f(761+i*72+j*2), %f(1141+i*-36+j*-1), %f684; max.f32 %f(762+i*72+j*2), %f(761+i*72+j*2), %f628; st.global.f32 [%rd30+148748], %f918; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[144]; __shared__ float pad_temp_shared[46656]; __shared__ float placeholder_shared[128]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 2; ++nn_outer_inner_init) { for (int ff_outer_inner_init = 0; ff_outer_inner_init < 2; ++ff_outer_inner_init) { for (int nn_inner_init = 0; nn_inner_init < 2; ++nn_inner_init) { for (int ff_inner_init = 0; ff_inner_init < 2; ++ff_inner_init) { int index0 = nn_outer_inner_init * 72 + nn_inner_init * 36 + ff_outer_inner_init * 18 + ff_inner_init * 9; compute[index0] = 0.000000e+00f; compute[index0 + 1] = 0.000000e+00f; compute[index0 + 2] = 0.000000e+00f; compute[index0 + 3] = 0.000000e+00f; compute[index0 + 4] = 0.000000e+00f; compute[index0 + 5] = 0.000000e+00f; compute[index0 + 6] = 0.000000e+00f; compute[index0 + 7] = 0.000000e+00f; compute[index0 + 8] = 0.000000e+00f; } } } } for (int rc_outer_outer = 0; rc_outer_outer < 8; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 288; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 162 + (int)threadIdx.x; int index2 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 27) / 432 * 93312 + rc_outer_outer * 11664 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 6 + (int)threadIdx.x / 27) % 432 * 27 + (int)threadIdx.x % 27; pad_temp_shared[index1] = placeholder[index2]; } if ((int)threadIdx.x < 128) { int index3 = (int)threadIdx.x; int index4 = (int)blockIdx.x * 1024 + ((int)threadIdx.x >> 4) * 128 + rc_outer_outer * 16 + ((int)threadIdx.x & 15); placeholder_shared[index3] = placeholder1[index4]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 16; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 2; ++nn_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 2; ++ff_outer_inner) { for (int nn_inner = 0; nn_inner < 2; ++nn_inner) { for (int ff_inner = 0; ff_inner < 2; ++ff_inner) { int index5 = nn_outer_inner * 72 + nn_inner * 36 + ff_outer_inner * 18 + ff_inner * 9; int index6 = nn_outer_inner * 23328 + nn_inner * 11664 + rc_outer_inner * 729 + (int)threadIdx.x % 81 * 9; int index7 = (int)threadIdx.x / 81 * 64 + ff_outer_inner * 32 + ff_inner * 16 + rc_outer_inner; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 1] = compute[index5 + 1] + (pad_temp_shared[index6 + 1] * placeholder_shared[index7]); compute[index5 + 2] = compute[index5 + 2] + (pad_temp_shared[index6 + 2] * placeholder_shared[index7]); compute[index5 + 3] = compute[index5 + 3] + (pad_temp_shared[index6 + 3] * placeholder_shared[index7]); compute[index5 + 4] = compute[index5 + 4] + (pad_temp_shared[index6 + 4] * placeholder_shared[index7]); compute[index5 + 5] = compute[index5 + 5] + (pad_temp_shared[index6 + 5] * placeholder_shared[index7]); compute[index5 + 6] = compute[index5 + 6] + (pad_temp_shared[index6 + 6] * placeholder_shared[index7]); compute[index5 + 7] = compute[index5 + 7] + (pad_temp_shared[index6 + 7] * placeholder_shared[index7]); compute[index5 + 8] = compute[index5 + 8] + (pad_temp_shared[index6 + 8] * placeholder_shared[index7]); } } } } } } for (int ax0_inner = 0; ax0_inner < 4; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 4; ++ax1_inner) { for (int ax3_inner = 0; ax3_inner < 9; ++ax3_inner) { int index8 = ax0_inner * 11664 + (int)blockIdx.x * 5832 + (int)threadIdx.x / 81 * 2916 + ax1_inner * 729 + (int)threadIdx.x % 81 * 9 + ax3_inner; int index9 = ax0_inner * 36 + ax1_inner * 9 + ax3_inner; int index10 = (int)blockIdx.x * 8 + (int)threadIdx.x / 81 * 4 + ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmp7kcynwfq
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .local .align 16 .b8 __local_depot0[31104]; .reg .b64 %SP; .reg .b64 %SPL; .reg .pred %p<14>; .reg .f32 %f<1461>; .reg .b32 %r<165>; .reg .b64 %rd<41>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[373248]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[4096]; mov.u64 %SPL, __local_depot0; ld.param.u64 %rd6, [default_function_kernel0_param_0]; ld.param.u64 %rd7, [default_function_kernel0_param_1]; ld.param.u64 %rd8, [default_function_kernel0_param_2]; ld.param.u64 %rd9, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd7; cvta.to.global.u64 %rd3, %rd8; cvta.to.global.u64 %rd4, %rd9; add.u64 %rd5, %SPL, 0; mov.u32 %r41, 0; mov.u32 %r154, %r41; $L__BB0_1: mul.lo.s32 %r2, %r154, 3888; mov.u32 %r155, %r41; $L__BB0_2: mad.lo.s32 %r43, %r155, 486, %r2; mul.wide.s32 %rd11, %r43, 4; add.s64 %rd12, %rd5, %rd11; mov.u32 %r44, 0; st.local.u32 [%rd12+972], %r44; mov.f32 %f17, 0f00000000; for.size.3 i in range(0, 120, 1): st.local.v2.f32 [%rd12+(0+i*8)], {%f17, %f17}; st.local.u32 [%rd12+(976+i*8)], %r44; st.local.u32 [%rd12+(980+i*8)], %r44; st.local.v2.f32 [%rd12+960], {%f17, %f17}; for.size.1 i in range(0, 2, 1): st.local.u32 [%rd12+(1936+i*-968)], %r44; st.local.u32 [%rd12+1940], %r44; add.s32 %r155, %r155, 1; setp.ne.s32 %p1, %r155, 8; @%p1 bra $L__BB0_2; add.s32 %r154, %r154, 1; setp.lt.u32 %p2, %r154, 2; @%p2 bra $L__BB0_1; mov.u32 %r6, %ctaid.x; mul.lo.s32 %r7, %r6, 186624; mov.u32 %r8, %tid.x; mul.lo.s32 %r9, %r8, 243; mov.u32 %r45, 0; mov.u32 %r156, %r45; $L__BB0_5: bar.sync 0; mad.lo.s32 %r11, %r156, 46656, %r7; mov.u32 %r157, %r45; $L__BB0_6: mad.lo.s32 %r47, %r157, 3, %r8; mul.hi.s32 %r48, %r47, 1508246403; shr.u32 %r49, %r48, 31; shr.s32 %r50, %r48, 14; add.s32 %r51, %r50, %r49; mul.lo.s32 %r52, %r51, 46656; sub.s32 %r53, %r47, %r52; add.s32 %r54, %r11, %r53; mad.lo.s32 %r55, %r51, 93312, %r54; mul.wide.s32 %rd13, %r55, 4; add.s64 %rd14, %rd1, %rd13; ld.global.nc.f32 %f18, [%rd14]; shl.b32 %r56, %r47, 2; mov.u32 %r57, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r58, %r57, %r56; for.size.13 i in range(0, 7, 1): st.shared.f32 [%r58+(0+i*12)], %f(18+i*1); add.s32 %r(59+i*9), %r47, (3+i*3); mul.hi.s32 %r(60+i*9), %r(59+i*9), 1508246403; shr.u32 %r(61+i*9), %r(60+i*9), 31; shr.s32 %r(62+i*9), %r(60+i*9), 14; add.s32 %r(63+i*9), %r(62+i*9), %r(61+i*9); mul.lo.s32 %r(64+i*9), %r(63+i*9), 46656; sub.s32 %r(65+i*9), %r(59+i*9), %r(64+i*9); add.s32 %r(66+i*9), %r11, %r(65+i*9); mad.lo.s32 %r(67+i*9), %r(63+i*9), 93312, %r(66+i*9); mul.wide.s32 %rd(15+i*2), %r(67+i*9), 4; add.s64 %rd(16+i*2), %rd1, %rd(15+i*2); ld.global.nc.f32 %f(19+i*1), [%rd(16+i*2)+0]; st.shared.f32 [%r58+84], %f25; add.s32 %r157, %r157, 8; setp.ne.s32 %p3, %r157, 31104; @%p3 bra $L__BB0_6; shl.b32 %r14, %r156, 6; mov.u32 %r158, 0; $L__BB0_8: mad.lo.s32 %r16, %r158, 3, %r8; setp.gt.s32 %p4, %r16, 1023; shl.b32 %r123, %r16, 2; mov.u32 %r124, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r17, %r124, %r123; @%p4 bra $L__BB0_10; shl.b32 %r125, %r16, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(126+i*1), %r(125+i*-109), (-128+i*191); for.size.1 i in range(0, 2, 1): add.s32 %r(128+i*1), %r(127+i*1), %r(14+i*112); mul.wide.s32 %rd29, %r129, 4; add.s64 %rd30, %rd2, %rd29; ld.global.nc.f32 %f26, [%rd30]; st.shared.f32 [%r17], %f26; $L__BB0_10: add.s32 %r18, %r16, 3; setp.gt.s32 %p5, %r18, 1023; @%p5 bra $L__BB0_12; shl.b32 %r130, %r18, 1; for.size.1 i in range(0, 2, 1): and.b32 %r(131+i*1), %r(130+i*-112), (-128+i*191); for.size.1 i in range(0, 2, 1): add.s32 %r(133+i*1), %r(132+i*1), %r(14+i*117); mul.wide.s32 %rd31, %r134, 4; add.s64 %rd32, %rd2, %rd31; ld.global.nc.f32 %f27, [%rd32]; st.shared.f32 [%r17+12], %f27; $L__BB0_12: add.s32 %r158, %r158, 2; setp.ne.s32 %p6, %r158, 342; @%p6 bra $L__BB0_8; bar.sync 0; mov.u32 %r135, 0; mov.u32 %r159, %r135; $L__BB0_14: mad.lo.s32 %r21, %r159, 5832, %r9; shl.b32 %r22, %r159, 3; mov.u32 %r160, %r135; $L__BB0_15: mul.lo.s32 %r24, %r160, 3888; mad.lo.s32 %r25, %r160, 46656, %r21; mov.u32 %r137, 0; mov.u32 %r161, %r137; $L__BB0_16: mad.lo.s32 %r27, %r161, 486, %r24; for.size.2 i in range(0, 2, 1): shl.b32 %r(139+i*2), %r(161+i*-21), (7+i*-5); add.s32 %r(140+i*3), %r(139+i*-15), %r(22+i*119); for.size.2 i in range(0, 8, 1): for.size.1 j in range(0, 2, 1): ld.shared.f32 %f(1+i*2+j*1), [%r143+(0+i*4+j*256)]; mov.u32 %r162, %r137; $L__BB0_17: mul.lo.s32 %r144, %r162, 27; for.size.1 i in range(0, 2, 1): add.s32 %r(145+i*1), %r(27+i*-2), %r144; mul.wide.s32 %rd33, %r145, 4; add.s64 %rd34, %rd5, %rd33; shl.b32 %r147, %r146, 2; add.s32 %r149, %r57, %r147; for.size.28 i in range(0, 27, 1): ld.shared.f32 %f(28+i*26), [%r149+(0+i*4)]; ld.local.f32 %f(29+i*26), [%rd34+(0+i*4)]; fma.rn.f32 %f(30+i*26), %f(28+i*26), %f1, %f(29+i*26); ld.local.f32 %f(31+i*26), [%rd34+(972+i*4)]; fma.rn.f32 %f(32+i*26), %f(28+i*26), %f2, %f(31+i*26); ld.shared.f32 %f(33+i*26), [%r149+(2916+i*4)]; fma.rn.f32 %f(34+i*26), %f(33+i*26), %f3, %f(30+i*26); fma.rn.f32 %f(35+i*26), %f(33+i*26), %f4, %f(32+i*26); ld.shared.f32 %f(36+i*26), [%r149+(5832+i*4)]; fma.rn.f32 %f(37+i*26), %f(36+i*26), %f5, %f(34+i*26); fma.rn.f32 %f(38+i*26), %f(36+i*26), %f6, %f(35+i*26); ld.shared.f32 %f(39+i*26), [%r149+(8748+i*4)]; fma.rn.f32 %f(40+i*26), %f(39+i*26), %f7, %f(37+i*26); fma.rn.f32 %f(41+i*26), %f(39+i*26), %f8, %f(38+i*26); ld.shared.f32 %f(42+i*26), [%r149+(11664+i*4)]; fma.rn.f32 %f(43+i*26), %f(42+i*26), %f9, %f(40+i*26); fma.rn.f32 %f(44+i*26), %f(42+i*26), %f10, %f(41+i*26); ld.shared.f32 %f(45+i*26), [%r149+(14580+i*4)]; fma.rn.f32 %f(46+i*26), %f(45+i*26), %f11, %f(43+i*26); fma.rn.f32 %f(47+i*26), %f(45+i*26), %f12, %f(44+i*26); ld.shared.f32 %f(48+i*26), [%r149+(17496+i*4)]; fma.rn.f32 %f(49+i*26), %f(48+i*26), %f13, %f(46+i*26); fma.rn.f32 %f(50+i*26), %f(48+i*26), %f14, %f(47+i*26); ld.shared.f32 %f(51+i*26), [%r149+(20412+i*4)]; fma.rn.f32 %f(52+i*26), %f(51+i*26), %f15, %f(49+i*26); st.local.f32 [%rd34+(0+i*4)], %f(52+i*26); fma.rn.f32 %f(53+i*26), %f(51+i*26), %f16, %f(50+i*26); st.local.f32 [%rd34+(972+i*4)], %f(53+i*26); add.s32 %r162, %r162, 1; setp.ne.s32 %p7, %r162, 9; @%p7 bra $L__BB0_17; add.s32 %r161, %r161, 1; setp.lt.u32 %p8, %r161, 8; @%p8 bra $L__BB0_16; add.s32 %r160, %r160, 1; setp.lt.u32 %p9, %r160, 2; @%p9 bra $L__BB0_15; add.s32 %r159, %r159, 1; setp.lt.u32 %p10, %r159, 8; @%p10 bra $L__BB0_14; add.s32 %r156, %r156, 1; setp.lt.u32 %p11, %r156, 2; @%p11 bra $L__BB0_5; mad.lo.s32 %r34, %r6, 23328, %r9; mov.u32 %r150, 0; mov.u32 %r163, %r150; $L__BB0_23: mul.lo.s32 %r36, %r163, 3888; mad.lo.s32 %r37, %r163, 11664, %r34; mov.u32 %r164, %r150; $L__BB0_24: for.size.3 i in range(0, 2, 1): mad.lo.s32 %r(152+i*1), %r164, (243+i*486), %r(36+i*1); mul.wide.s32 %rd(35+i*2), %r(164+i*-12), 4; add.s64 %rd(36+i*2), %rd(4+i*1), %rd(35+i*2); ld.local.f32 %f730, [%rd38]; ld.global.nc.f32 %f731, [%rd36]; add.f32 %f732, %f730, %f731; mov.f32 %f733, 0f00000000; max.f32 %f734, %f732, %f733; mul.wide.s32 %rd39, %r153, 4; add.s64 %rd40, %rd3, %rd39; for.size.4 i in range(0, 242, 1): st.global.f32 [%rd40+(0+i*4)], %f(734+i*3); ld.local.f32 %f(735+i*3), [%rd38+(4+i*4)]; add.f32 %f(736+i*3), %f(735+i*3), %f731; max.f32 %f(737+i*3), %f(736+i*3), %f733; st.global.f32 [%rd40+968], %f1460; add.s32 %r164, %r164, 1; setp.ne.s32 %p12, %r164, 16; @%p12 bra $L__BB0_24; add.s32 %r163, %r163, 1; setp.lt.u32 %p13, %r163, 2; @%p13 bra $L__BB0_23; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[7776]; __shared__ float pad_temp_shared[93312]; __shared__ float placeholder_shared[1024]; for (int nn_outer_inner_init = 0; nn_outer_inner_init < 2; ++nn_outer_inner_init) { for (int ff_outer_inner_init = 0; ff_outer_inner_init < 8; ++ff_outer_inner_init) { for (int yy_outer_inner_init = 0; yy_outer_inner_init < 9; ++yy_outer_inner_init) { for (int xx_outer_inner_init = 0; xx_outer_inner_init < 27; ++xx_outer_inner_init) { int index0 = nn_outer_inner_init * 3888 + ff_outer_inner_init * 486 + yy_outer_inner_init * 27 + xx_outer_inner_init; compute[index0] = 0.000000e+00f; compute[index0 + 243] = 0.000000e+00f; } } } } for (int rc_outer_outer = 0; rc_outer_outer < 2; ++rc_outer_outer) { __syncthreads(); for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer < 31104; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer) { int index1 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3 + (int)threadIdx.x; int index2 = (int)blockIdx.x * 186624 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3 + (int)threadIdx.x) / 46656 * 93312 + rc_outer_outer * 46656 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer * 3 + (int)threadIdx.x) % 46656; pad_temp_shared[index1] = placeholder[index2]; } for (int ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 = 0; ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 < 342; ++ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1) { if (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x < 1024) { int index3 = ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x; int index4 = (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x >> 6) * 128 + rc_outer_outer * 64 + (ax0_ax1_fused_ax2_fused_ax3_fused_outer_outer1 * 3 + (int)threadIdx.x & 63); placeholder_shared[index3] = placeholder1[index4]; } } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { for (int nn_outer_inner = 0; nn_outer_inner < 2; ++nn_outer_inner) { for (int ff_outer_inner = 0; ff_outer_inner < 8; ++ff_outer_inner) { for (int yy_outer_inner = 0; yy_outer_inner < 9; ++yy_outer_inner) { for (int xx_outer_inner = 0; xx_outer_inner < 27; ++xx_outer_inner) { int index5 = nn_outer_inner * 3888 + ff_outer_inner * 486 + yy_outer_inner * 27 + xx_outer_inner; int index6 = nn_outer_inner * 46656 + rc_outer_inner * 5832 + (int)threadIdx.x * 243 + yy_outer_inner * 27 + xx_outer_inner; int index7 = ff_outer_inner * 128 + rc_outer_inner * 8; compute[index5] = compute[index5] + pad_temp_shared[index6] * placeholder_shared[index7]; compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6] * placeholder_shared[index7 + 64]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 729] * placeholder_shared[index7 + 1]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 729] * placeholder_shared[index7 + 65]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 1458] * placeholder_shared[index7 + 2]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 1458] * placeholder_shared[index7 + 66]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 2187] * placeholder_shared[index7 + 3]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 2187] * placeholder_shared[index7 + 67]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 2916] * placeholder_shared[index7 + 4]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 2916] * placeholder_shared[index7 + 68]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 3645] * placeholder_shared[index7 + 5]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 3645] * placeholder_shared[index7 + 69]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 4374] * placeholder_shared[index7 + 6]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 4374] * placeholder_shared[index7 + 70]); compute[index5] = compute[index5] + (pad_temp_shared[index6 + 5103] * placeholder_shared[index7 + 7]); compute[index5 + 243] = compute[index5 + 243] + (pad_temp_shared[index6 + 5103] * placeholder_shared[index7 + 71]); } } } } } } for (int ax0_inner = 0; ax0_inner < 2; ++ax0_inner) { for (int ax1_inner = 0; ax1_inner < 16; ++ax1_inner) { for (int ax2_inner = 0; ax2_inner < 9; ++ax2_inner) { for (int ax3_inner = 0; ax3_inner < 27; ++ax3_inner) { int index8 = (int)blockIdx.x * 23328 + ax0_inner * 11664 + ax1_inner * 729 + (int)threadIdx.x * 243 + ax2_inner * 27 + ax3_inner; int index9 = ax0_inner * 3888 + ax1_inner * 243 + ax2_inner * 27 + ax3_inner; int index10 = ax1_inner; T_relu[index8] = max(compute[index9] + placeholder2[index10], 0.000000e+00f); } } } } }
[c1a8e09f49679b4bb850cfe0944e7e1e,4,128,27,27,16,128,1,1,1,16,1,1,4,16,27,27]
tmpfk8xzt5e
default_function_kernel0
.visible .entry default_function_kernel0( .param .u64 default_function_kernel0_param_0, .param .u64 default_function_kernel0_param_1, .param .u64 default_function_kernel0_param_2, .param .u64 default_function_kernel0_param_3 ) { .reg .pred %p<4>; .reg .f32 %f<110>; .reg .b32 %r<87>; .reg .b64 %rd<27>; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E15pad_temp_shared[186624]; // demoted variable .shared .align 4 .b8 _ZZ24default_function_kernel0E18placeholder_shared[2048]; ld.param.u64 %rd5, [default_function_kernel0_param_0]; ld.param.u64 %rd6, [default_function_kernel0_param_1]; ld.param.u64 %rd7, [default_function_kernel0_param_2]; ld.param.u64 %rd8, [default_function_kernel0_param_3]; cvta.to.global.u64 %rd1, %rd6; cvta.to.global.u64 %rd2, %rd5; cvta.to.global.u64 %rd3, %rd7; cvta.to.global.u64 %rd4, %rd8; mov.u32 %r1, %ctaid.x; mov.u32 %r2, %tid.x; mad.lo.s32 %r3, %r1, 186624, %r2; shl.b32 %r27, %r2, 2; mov.u32 %r28, _ZZ24default_function_kernel0E15pad_temp_shared; add.s32 %r4, %r28, %r27; for.size.7 i in range(0, 5, 1): add.s32 %r(29+i*6), %r2, (27216+i*3888); mul.hi.s32 %r(30+i*6), %r(29+i*6), 1508246403; shr.u32 %r(31+i*6), %r(30+i*6), 31; shr.s32 %r(32+i*6), %r(30+i*6), 13; add.s32 %r(33+i*6), %r(32+i*6), %r(31+i*6); mad.lo.s32 %r(34+i*6), %r(33+i*6), 93312, %r3; add.s32 %r(5+i*1), %r(34+i*6), (3888+i*3888); shr.u32 %r59, %r2, 5; and.b32 %r60, %r2, 31; bfi.b32 %r10, %r59, %r60, 7, 25; mov.u32 %r61, _ZZ24default_function_kernel0E18placeholder_shared; add.s32 %r11, %r61, %r27; mul.hi.s32 %r62, %r2, -2032597691; add.s32 %r63, %r62, %r2; shr.u32 %r64, %r63, 31; shr.s32 %r65, %r63, 7; add.s32 %r13, %r65, %r64; mul.lo.s32 %r66, %r13, 243; sub.s32 %r12, %r2, %r66; for.size.3 i in range(0, 2, 1): shl.b32 %r(67+i*2), %r(13+i*-1), (7+i*-5); add.s32 %r(68+i*2), %r(61+i*-33), %r(67+i*2); add.s32 %r(14+i*1), %r(68+i*2), (8+i*93304); mov.f32 %f104, 0f00000000; mov.u32 %r83, 0; for.size.1 i in range(0, 5, 1): mov.f32 %f(105+i*1), %f104; $L__BB0_1: bar.sync 0; mul.lo.s32 %r71, %r83, 23328; add.s32 %r72, %r3, %r71; mul.wide.s32 %rd9, %r72, 4; add.s64 %rd10, %rd2, %rd9; for.size.2 i in range(0, 6, 1): ld.global.nc.f32 %f(25+i*1), [%rd10+(0+i*15552)]; st.shared.f32 [%r4+(0+i*15552)], %f(25+i*1); ld.global.nc.f32 %f31, [%rd10+373248]; for.size.5 i in range(0, 5, 1): st.shared.f32 [%r4+(93312+i*15552)], %f(31+i*1); add.s32 %r(73+i*1), %r(5+i*1), %r71; mul.wide.s32 %rd(11+i*2), %r(73+i*1), 4; add.s64 %rd(12+i*2), %rd2, %rd(11+i*2); ld.global.nc.f32 %f(32+i*1), [%rd(12+i*2)+0]; st.shared.f32 [%r4+171072], %f36; setp.gt.s32 %p1, %r2, 511; @%p1 bra $L__BB0_3; shl.b32 %r78, %r83, 5; add.s32 %r79, %r10, %r78; mul.wide.s32 %rd21, %r79, 4; add.s64 %rd22, %rd1, %rd21; ld.global.nc.f32 %f37, [%rd22]; st.shared.f32 [%r11], %f37; $L__BB0_3: bar.sync 0; mov.u32 %r85, 8; mov.u32 %r84, %r15; mov.u32 %r86, %r14; $L__BB0_4: for.size.13 i in range(0, 2, 1): ld.shared.f32 %f(38+i*13), [%r86+(-8+i*4)]; ld.shared.f32 %f(39+i*13), [%r84+(-93312+i*2916)]; fma.rn.f32 %f(40+i*13), %f(39+i*13), %f(38+i*13), %f(104+i*-64); ld.shared.f32 %f(41+i*13), [%r84+(-92340+i*2916)]; fma.rn.f32 %f(42+i*13), %f(38+i*13), %f(41+i*13), %f(105+i*-63); ld.shared.f32 %f(43+i*13), [%r84+(-91368+i*2916)]; fma.rn.f32 %f(44+i*13), %f(43+i*13), %f(38+i*13), %f(106+i*-62); ld.shared.f32 %f(45+i*13), [%r84+(0+i*2916)]; fma.rn.f32 %f(46+i*13), %f(38+i*13), %f(45+i*13), %f(107+i*-61); ld.shared.f32 %f(47+i*13), [%r84+(972+i*2916)]; fma.rn.f32 %f(48+i*13), %f(47+i*13), %f(38+i*13), %f(108+i*-60); ld.shared.f32 %f(49+i*13), [%r84+(1944+i*2916)]; fma.rn.f32 %f(50+i*13), %f(38+i*13), %f(49+i*13), %f(109+i*-59); for.size.13 i in range(0, 2, 1): ld.shared.f32 %f(64+i*13), [%r86+(0+i*4)]; ld.shared.f32 %f(65+i*13), [%r84+(-87480+i*2916)]; fma.rn.f32 %f(66+i*38), %f(65+i*13), %f(64+i*13), %f(53+i*13); ld.shared.f32 %f(67+i*12), [%r84+(-86508+i*2916)]; fma.rn.f32 %f(68+i*37), %f(64+i*13), %f(67+i*12), %f(55+i*13); ld.shared.f32 %f(69+i*11), [%r84+(-85536+i*2916)]; fma.rn.f32 %f(70+i*36), %f(69+i*11), %f(64+i*13), %f(57+i*13); ld.shared.f32 %f(71+i*10), [%r84+(5832+i*2916)]; fma.rn.f32 %f(72+i*35), %f(64+i*13), %f(71+i*10), %f(59+i*13); ld.shared.f32 %f(73+i*9), [%r84+(6804+i*2916)]; fma.rn.f32 %f(74+i*34), %f(73+i*9), %f(64+i*13), %f(61+i*13); ld.shared.f32 %f(75+i*8), [%r84+(7776+i*2916)]; fma.rn.f32 %f(76+i*33), %f(64+i*13), %f(75+i*8), %f(63+i*13); for.size.1 i in range(0, 2, 1): add.s32 %r(86+i*-2), %r(86+i*-2), (16+i*11648); add.s32 %r85, %r85, 16; setp.ne.s32 %p2, %r85, 136; @%p2 bra $L__BB0_4; add.s32 %r83, %r83, 1; setp.lt.u32 %p3, %r83, 4; @%p3 bra $L__BB0_1; mul.wide.s32 %rd23, %r13, 4; add.s64 %rd24, %rd4, %rd23; ld.global.nc.f32 %f84, [%rd24]; add.f32 %f85, %f104, %f84; mov.f32 %f86, 0f00000000; max.f32 %f87, %f85, %f86; for.size.1 i in range(0, 2, 1): mad.lo.s32 %r(81+i*1), %r(1+i*12), (23328+i*-22599), %r(12+i*69); mul.wide.s32 %rd25, %r82, 4; add.s64 %rd26, %rd3, %rd25; for.size.3 i in range(0, 3, 1): st.global.f32 [%rd26+(0+i*972)], %f(87+i*2); add.f32 %f(88+i*2), %f(105+i*1), %f84; max.f32 %f(89+i*2), %f(88+i*2), %f86; for.size.3 i in range(0, 2, 1): st.global.f32 [%rd26+(46656+i*972)], %f(93+i*2); add.f32 %f(94+i*2), %f(108+i*1), %f84; max.f32 %f(95+i*2), %f(94+i*2), %f86; st.global.f32 [%rd26+48600], %f97; ret; }
extern "C" __global__ void default_function_kernel0(float* __restrict__ placeholder, float* __restrict__ placeholder1, float* __restrict__ T_relu, float* __restrict__ placeholder2) { float compute[6]; __shared__ float pad_temp_shared[46656]; __shared__ float placeholder_shared[512]; compute[0] = 0.000000e+00f; compute[1] = 0.000000e+00f; compute[2] = 0.000000e+00f; compute[3] = 0.000000e+00f; compute[4] = 0.000000e+00f; compute[5] = 0.000000e+00f; for (int rc_outer_outer = 0; rc_outer_outer < 4; ++rc_outer_outer) { __syncthreads(); int index0 = (int)threadIdx.x; int index1 = (int)blockIdx.x * 186624 + rc_outer_outer * 23328 + (int)threadIdx.x; pad_temp_shared[index0] = placeholder[index1]; pad_temp_shared[index0 + 3888] = placeholder[index1 + 3888]; pad_temp_shared[index0 + 7776] = placeholder[index1 + 7776]; pad_temp_shared[index0 + 11664] = placeholder[index1 + 11664]; pad_temp_shared[index0 + 15552] = placeholder[index1 + 15552]; pad_temp_shared[index0 + 19440] = placeholder[index1 + 19440]; pad_temp_shared[index0 + 23328] = placeholder[index1 + 93312]; int index2 = (int)blockIdx.x * 186624 + ((int)threadIdx.x + 27216) / 23328 * 93312 + rc_outer_outer * 23328 + ((int)threadIdx.x / 27 + 144) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 27216] = placeholder[index2]; int index3 = (int)blockIdx.x * 186624 + ((int)threadIdx.x + 31104) / 23328 * 93312 + rc_outer_outer * 23328 + ((int)threadIdx.x / 27 + 288) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 31104] = placeholder[index3]; int index4 = (int)blockIdx.x * 186624 + ((int)threadIdx.x + 34992) / 23328 * 93312 + rc_outer_outer * 23328 + ((int)threadIdx.x / 27 + 432) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 34992] = placeholder[index4]; int index5 = (int)blockIdx.x * 186624 + ((int)threadIdx.x + 38880) / 23328 * 93312 + rc_outer_outer * 23328 + ((int)threadIdx.x / 27 + 576) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 38880] = placeholder[index5]; int index6 = (int)blockIdx.x * 186624 + ((int)threadIdx.x + 42768) / 23328 * 93312 + rc_outer_outer * 23328 + ((int)threadIdx.x / 27 + 720) * 27 + (int)threadIdx.x % 27; pad_temp_shared[index0 + 42768] = placeholder[index6]; if ((int)threadIdx.x < 512) { int index7 = ((int)threadIdx.x >> 5) * 128 + rc_outer_outer * 32 + ((int)threadIdx.x & 31); placeholder_shared[index0] = placeholder1[index7]; } __syncthreads(); for (int rc_outer_inner = 0; rc_outer_inner < 8; ++rc_outer_inner) { int index8 = rc_outer_inner * 2916 + (int)threadIdx.x % 243; int index9 = (int)threadIdx.x / 243 * 32 + rc_outer_inner * 4; compute[0] = compute[0] + pad_temp_shared[index8] * placeholder_shared[index9]; compute[1] = compute[1] + (pad_temp_shared[index8 + 243] * placeholder_shared[index9]); compute[2] = compute[2] + (pad_temp_shared[index8 + 486] * placeholder_shared[index9]); compute[3] = compute[3] + (pad_temp_shared[index8 + 23328] * placeholder_shared[index9]); compute[4] = compute[4] + (pad_temp_shared[index8 + 23571] * placeholder_shared[index9]); compute[5] = compute[5] + (pad_temp_shared[index8 + 23814] * placeholder_shared[index9]); compute[0] = compute[0] + (pad_temp_shared[index8 + 729] * placeholder_shared[index9 + 1]); compute[1] = compute[1] + (pad_temp_shared[index8 + 972] * placeholder_shared[index9 + 1]); compute[2] = compute[2] + (pad_temp_shared[index8 + 1215] * placeholder_shared[index9 + 1]); compute[3] = compute[3] + (pad_temp_shared[index8 + 24057] * placeholder_shared[index9 + 1]); compute[4] = compute[4] + (pad_temp_shared[index8 + 24300] * placeholder_shared[index9 + 1]); compute[5] = compute[5] + (pad_temp_shared[index8 + 24543] * placeholder_shared[index9 + 1]); compute[0] = compute[0] + (pad_temp_shared[index8 + 1458] * placeholder_shared[index9 + 2]); compute[1] = compute[1] + (pad_temp_shared[index8 + 1701] * placeholder_shared[index9 + 2]); compute[2] = compute[2] + (pad_temp_shared[index8 + 1944] * placeholder_shared[index9 + 2]); compute[3] = compute[3] + (pad_temp_shared[index8 + 24786] * placeholder_shared[index9 + 2]); compute[4] = compute[4] + (pad_temp_shared[index8 + 25029] * placeholder_shared[index9 + 2]); compute[5] = compute[5] + (pad_temp_shared[index8 + 25272] * placeholder_shared[index9 + 2]); compute[0] = compute[0] + (pad_temp_shared[index8 + 2187] * placeholder_shared[index9 + 3]); compute[1] = compute[1] + (pad_temp_shared[index8 + 2430] * placeholder_shared[index9 + 3]); compute[2] = compute[2] + (pad_temp_shared[index8 + 2673] * placeholder_shared[index9 + 3]); compute[3] = compute[3] + (pad_temp_shared[index8 + 25515] * placeholder_shared[index9 + 3]); compute[4] = compute[4] + (pad_temp_shared[index8 + 25758] * placeholder_shared[index9 + 3]); compute[5] = compute[5] + (pad_temp_shared[index8 + 26001] * placeholder_shared[index9 + 3]); } } int index0 = (int)threadIdx.x; int index10 = (int)blockIdx.x * 23328 + (int)threadIdx.x / 243 * 729 + (int)threadIdx.x % 243; T_relu[index10] = max(compute[0] + placeholder2[index0 / 243], 0.000000e+00f); T_relu[index10 + 243] = max(compute[1] + placeholder2[index0 / 243], 0.000000e+00f); T_relu[index10 + 486] = max(compute[2] + placeholder2[index0 / 243], 0.000000e+00f); T_relu[index10 + 11664] = max(compute[3] + placeholder2[index0 / 243], 0.000000e+00f); T_relu[index10 + 11907] = max(compute[4] + placeholder2[index0 / 243], 0.000000e+00f); T_relu[index10 + 12150] = max(compute[5] + placeholder2[index0 / 243], 0.000000e+00f); }