name
string
code
string
asm
string
file
string
ncnn::float32_to_int8(float)
signed char float32_to_int8(float value) { float tmp; if (value >= 0.f) tmp = value + 0.5f; else tmp = value - 0.5f; if (tmp > 127) return 127; if (tmp < -128) return -128; return static_cast<signed char>(tmp); }
movss %xmm0, -0x8(%rsp) movss -0x8(%rsp), %xmm0 xorps %xmm1, %xmm1 ucomiss %xmm1, %xmm0 jb 0x16ee5ca movss 0x7a9a64(%rip), %xmm0 # 0x1e98020 addss -0x8(%rsp), %xmm0 movss %xmm0, -0xc(%rsp) jmp 0x16ee5e2 movss -0x8(%rsp), %xmm0 movss 0x7a9a48(%rip), %xmm1 # 0x1e98020 subss %xmm1, %xmm0 movss %xmm0, -0xc(%rsp) movss ...
/Tencent[P]ncnn/src/layer/cast.cpp
ncnn::Cast_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Cast_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (type_from == type_to) { top_blob = bottom_blob; return 0; } int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; int channels = bottom_blob.c; int dims = bottom_bl...
subq $0x378, %rsp # imm = 0x378 movq %rdi, 0x190(%rsp) movq %rsi, 0x188(%rsp) movq %rdx, 0x180(%rsp) movq %rcx, 0x178(%rsp) movq 0x190(%rsp), %rcx movq %rcx, 0x80(%rsp) movl 0xd0(%rcx), %eax cmpl 0xd4(%rcx), %eax jne 0x16f1d44 movq 0x188(%rsp), %rax movq 0x180(%rsp), %rcx movq %rcx, 0x1b0(%rsp) movq %rax, 0x...
/Tencent[P]ncnn/src/layer/x86/cast_x86.cpp
ncnn::cast_fp16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_F16C && __AVX__ && !__F16C__ if (ncnn::cpu_support_x86_f16c()) { cast_fp16_to_fp32_sse_f16c(bottom_blob, top_blob, opt); return; } #endif const int w = bottom_blob.w; const int h =...
subq $0x2d8, %rsp # imm = 0x2D8 movq %rdi, 0x140(%rsp) movq %rsi, 0x138(%rsp) movq %rdx, 0x130(%rsp) movq 0x140(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x12c(%rsp) movq 0x140(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x128(%rsp) movq 0x140(%rsp), %rax movl 0x34(%rax), %eax movl %eax, 0x124(%rsp) movq...
/Tencent[P]ncnn/src/layer/x86/cast_fp16.h
ncnn::cast_fp32_to_fp16_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp32_to_fp16_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_RUNTIME_CPU && NCNN_F16C && __AVX__ && !__F16C__ if (ncnn::cpu_support_x86_f16c()) { cast_fp32_to_fp16_sse_f16c(bottom_blob, top_blob, opt); return; } #endif const int w = bottom_blob....
pushq %rbp movq %rsp, %rbp andq $-0x40, %rsp subq $0x440, %rsp # imm = 0x440 movq %rdi, 0x210(%rsp) movq %rsi, 0x208(%rsp) movq %rdx, 0x200(%rsp) movq 0x210(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x1fc(%rsp) movq 0x210(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x1f8(%rsp) movq 0x210(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/x86/cast_fp16.h
ncnn::cast_fp32_to_bf16_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp32_to_bf16_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_RUNTIME_CPU && NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_fp32_to_bf16_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #...
pushq %rbp movq %rsp, %rbp andq $-0x40, %rsp subq $0xa40, %rsp # imm = 0xA40 movq %rdi, 0x220(%rsp) movq %rsi, 0x218(%rsp) movq %rdx, 0x210(%rsp) callq 0x7e2f0 cmpl $0x0, %eax je 0x16f7bd3 movq 0x220(%rsp), %rdi movq 0x218(%rsp), %rsi movq 0x210(%rsp), %rdx callq 0x1701b60 jmp 0x16f8ab2 movq 0x220(%rsp), %ra...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_bf16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_bf16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_bf16_to_fp32_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #if NCNN_RUNTIME_CPU ...
pushq %rbp movq %rsp, %rbp andq $-0x40, %rsp subq $0x780, %rsp # imm = 0x780 movq %rdi, 0x190(%rsp) movq %rsi, 0x188(%rsp) movq %rdx, 0x180(%rsp) callq 0x7e2f0 cmpl $0x0, %eax je 0x16f8b13 movq 0x190(%rsp), %rdi movq 0x188(%rsp), %rsi movq 0x180(%rsp), %rdx callq 0x17028a0 jmp 0x16f9995 movq 0x190(%rsp), %ra...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_fp16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_F16C && __AVX__ && !__F16C__ if (ncnn::cpu_support_x86_f16c()) { cast_fp16_to_fp32_sse_f16c(bottom_blob, top_blob, opt); return; } #endif const int w = bottom_blob.w; const int h =...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x3e0, %rsp # imm = 0x3E0 movq %rdi, 0x1a0(%rsp) movq %rsi, 0x198(%rsp) movq %rdx, 0x190(%rsp) movq 0x1a0(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x18c(%rsp) movq 0x1a0(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x188(%rsp) movq 0x1a0(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/x86/cast_fp16.h
ncnn::cast_fp32_to_bf16_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp32_to_bf16_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_RUNTIME_CPU && NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_fp32_to_bf16_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x660, %rsp # imm = 0x660 movq %rdi, 0x1b0(%rsp) movq %rsi, 0x1a8(%rsp) movq %rdx, 0x1a0(%rsp) callq 0x7e280 cmpl $0x0, %eax je 0x16fc093 movq 0x1b0(%rsp), %rdi movq 0x1a8(%rsp), %rsi movq 0x1a0(%rsp), %rdx callq 0x1703660 jmp 0x16fcf24 movq 0x1b0(%rsp), %ra...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_bf16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_bf16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_bf16_to_fp32_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #if NCNN_RUNTIME_CPU ...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x4c0, %rsp # imm = 0x4C0 movq %rdi, 0x150(%rsp) movq %rsi, 0x148(%rsp) movq %rdx, 0x140(%rsp) callq 0x7e280 cmpl $0x0, %eax je 0x16fcf83 movq 0x150(%rsp), %rdi movq 0x148(%rsp), %rsi movq 0x140(%rsp), %rdx callq 0x17043f0 jmp 0x16fdbeb movq 0x150(%rsp), %ra...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_bf16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_bf16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_bf16_to_fp32_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #if NCNN_RUNTIME_CPU ...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x4e0, %rsp # imm = 0x4E0 movq %rdi, 0x168(%rsp) movq %rsi, 0x160(%rsp) movq %rdx, 0x158(%rsp) callq 0x7e280 cmpl $0x0, %eax je 0x1700e83 movq 0x168(%rsp), %rdi movq 0x160(%rsp), %rsi movq 0x158(%rsp), %rdx callq 0x17043f0 jmp 0x1701b0f movq 0x168(%rsp), %ra...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_fp32_to_bf16_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_fp32_to_bf16_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_RUNTIME_CPU && NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_fp32_to_bf16_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x640, %rsp # imm = 0x640 movq %rdi, 0x1a8(%rsp) movq %rsi, 0x1a0(%rsp) movq %rdx, 0x198(%rsp) movq 0x1a8(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x194(%rsp) movq 0x1a8(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x190(%rsp) movq 0x1a8(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::cast_bf16_to_fp32_sse(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void cast_bf16_to_fp32_sse(const Mat& bottom_blob, Mat& top_blob, const Option& opt) { #if NCNN_AVX512BF16 && __AVX512F__ && !__AVX512BF16__ if (ncnn::cpu_support_x86_avx512_bf16()) { cast_bf16_to_fp32_sse_avx512bf16(bottom_blob, top_blob, opt); return; } #endif #if NCNN_RUNTIME_CPU ...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x500, %rsp # imm = 0x500 movq %rdi, 0x168(%rsp) movq %rsi, 0x160(%rsp) movq %rdx, 0x158(%rsp) movq 0x168(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x154(%rsp) movq 0x168(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x150(%rsp) movq 0x168(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/x86/cast_bf16.h
ncnn::HardSigmoid_x86_avx::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int HardSigmoid_x86_avx::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int channels = bottom_top_blob.c; int elempack = bottom_top_blob.elempack; int size = w * h * d * elempack; #pragma omp ...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x620, %rsp # imm = 0x620 movq %rdi, 0x1c0(%rsp) movq %rsi, 0x1b8(%rsp) movq %rdx, 0x1b0(%rsp) movq 0x1c0(%rsp), %rax movq %rax, 0x38(%rsp) movq 0x1b8(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x1ac(%rsp) movq 0x1b8(%rsp), %rax movl 0x30(%rax), %eax movl ...
/Tencent[P]ncnn/build_O0/src/layer/x86/hardsigmoid_x86_avx.cpp
ncnn::SELU_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int SELU_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int elempack = bottom_top_blob.elempack; int channels = bottom_top_blob.c; int size = w * h * d * elempack; #pragma omp paralle...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x15c0, %rsp # imm = 0x15C0 movq %rdi, 0x280(%rsp) movq %rsi, 0x278(%rsp) movq %rdx, 0x270(%rsp) movq 0x280(%rsp), %rax movq %rax, 0x50(%rsp) movq 0x278(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x26c(%rsp) movq 0x278(%rsp), %rax movl 0x30(%rax), %eax movl...
/Tencent[P]ncnn/build_O0/src/layer/x86/selu_x86_fma.cpp
ncnn::HardSwish::load_param(ncnn::ParamDict const&)
int HardSwish::load_param(const ParamDict& pd) { // Note that tensorflow/pytorch use alpha,beta = 1/6, 0.5, not the default value here. // You can setup them manually in .param file. alpha = pd.get(0, 0.2f); beta = pd.get(1, 0.5f); lower = -beta / alpha; upper = (1.f / alpha) + lower; retur...
subq $0x18, %rsp movq %rdi, 0x10(%rsp) movq %rsi, 0x8(%rsp) movq 0x10(%rsp), %rax movq %rax, (%rsp) movq 0x8(%rsp), %rdi xorl %esi, %esi movss 0x79b6ba(%rip), %xmm0 # 0x1eada00 callq 0x78df0 movq (%rsp), %rax movss %xmm0, 0xd0(%rax) movq 0x8(%rsp), %rdi movl $0x1, %esi movss 0x785cb7(%rip), %xmm0 # 0x1e98020 callq ...
/Tencent[P]ncnn/src/layer/hardswish.cpp
ncnn::Mish_x86::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int Mish_x86::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int channels = bottom_top_blob.c; int elempack = bottom_top_blob.elempack; int size = w * h * d * elempack; #pragma omp parallel fo...
pushq %r14 pushq %rbx subq $0x15a8, %rsp # imm = 0x15A8 movq %rdi, 0xf8(%rsp) movq %rsi, 0xf0(%rsp) movq %rdx, 0xe8(%rsp) movq 0xf0(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0xe4(%rsp) movq 0xf0(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0xe0(%rsp) movq 0xf0(%rsp), %rax movl 0x34(%rax), %eax movl %eax, 0...
/Tencent[P]ncnn/src/layer/x86/mish_x86.cpp
ncnn::StatisticsPooling::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int StatisticsPooling::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; int size = w * h; size_t elemsize = bottom_blob.elemsize; int out_channels = channels; if (include_stddev) { ...
subq $0x318, %rsp # imm = 0x318 movq %rdi, 0x150(%rsp) movq %rsi, 0x148(%rsp) movq %rdx, 0x140(%rsp) movq %rcx, 0x138(%rsp) movq 0x150(%rsp), %rax movq 0x148(%rsp), %rcx movl 0x2c(%rcx), %ecx movl %ecx, 0x134(%rsp) movq 0x148(%rsp), %rcx movl 0x30(%rcx), %ecx movl %ecx, 0x130(%rsp) movq 0x148(%rsp), %rcx mov...
/Tencent[P]ncnn/src/layer/statisticspooling.cpp
ncnn::Swish_x86_avx::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int Swish_x86_avx::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int channels = bottom_top_blob.c; int elempack = bottom_top_blob.elempack; int size = w * h * d * elempack; #pragma omp parall...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x1500, %rsp # imm = 0x1500 movq %rdi, 0x178(%rsp) movq %rsi, 0x170(%rsp) movq %rdx, 0x168(%rsp) movq 0x170(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x164(%rsp) movq 0x170(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x160(%rsp) movq 0x170(%rsp), %rax mov...
/Tencent[P]ncnn/build_O0/src/layer/x86/swish_x86_avx.cpp
ncnn::Gemm::load_model(ncnn::ModelBin const&)
int Gemm::load_model(const ModelBin& mb) { if (constantA == 1) { if (transA == 0) A_data = mb.load(constantK, constantM, 0); else A_data = mb.load(constantM, constantK, 0); if (A_data.empty()) return -100; } if (constantB == 1) { i...
subq $0x9f8, %rsp # imm = 0x9F8 movq %rdi, 0x4b8(%rsp) movq %rsi, 0x4b0(%rsp) movq 0x4b8(%rsp), %rax movq %rax, 0x180(%rsp) cmpl $0x1, 0xe0(%rax) jne 0x17322c1 movq 0x180(%rsp), %rax cmpl $0x0, 0xd8(%rax) jne 0x1731d62 movq 0x180(%rsp), %rax movq 0x4b0(%rsp), %rsi movl 0xf4(%rax), %edx movl 0xec(%rax), %ecx ...
/Tencent[P]ncnn/src/layer/gemm.cpp
ncnn::Gemm::forward_int8(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int Gemm::forward_int8(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& A0 = constantA ? A_data : bottom_blobs[0]; const Mat& B0 = constantB ? B_data : constantA ? bottom_blobs[0] : bottom_blobs[1]; Mat A; if (transA == 0) { A = A0; ...
subq $0xa58, %rsp # imm = 0xA58 movq %rdi, 0x568(%rsp) movq %rsi, 0x560(%rsp) movq %rdx, 0x558(%rsp) movq %rcx, 0x550(%rsp) movq 0x568(%rsp), %rax movq %rax, 0x260(%rsp) cmpl $0x0, 0xe0(%rax) je 0x1737048 movq 0x260(%rsp), %rax addq $0x120, %rax # imm = 0x120 movq %rax, 0x258(%rsp) jmp 0x1737061 m...
/Tencent[P]ncnn/src/layer/gemm.cpp
ncnn::Gemm_x86::Gemm_x86()
Gemm_x86::Gemm_x86() { #if __SSE2__ support_packing = true; #endif // __SSE2__ nT = 0; }
subq $0x98, %rsp movq %rdi, 0x38(%rsp) movq 0x38(%rsp), %rdi movq %rdi, 0x18(%rsp) callq 0x1731200 movq 0x18(%rsp), %rax leaq 0x7e9739(%rip), %rcx # 0x1f24bb0 addq $0x10, %rcx movq %rcx, (%rax) addq $0x248, %rax # imm = 0x248 movq %rax, 0x40(%rsp) movq 0x40(%rsp), %rax movq %rax, 0x20(%rsp) movq $0x0, (%r...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::Gemm_x86::create_pipeline(ncnn::Option const&)
int Gemm_x86::create_pipeline(const Option& opt) { #if NCNN_INT8 if (int8_scale_term) { return create_pipeline_int8(opt); } #endif if (constantA) { const int M = constantM; const int K = constantK; int TILE_M, TILE_N, TILE_K; get_optimal_tile_mnk(M, 0, K, co...
pushq %rbx subq $0x730, %rsp # imm = 0x730 movq %rdi, 0x348(%rsp) movq %rsi, 0x340(%rsp) movq 0x348(%rsp), %rax movq %rax, 0x138(%rsp) cmpl $0x0, 0x10c(%rax) je 0x173b632 movq 0x138(%rsp), %rdi movq 0x340(%rsp), %rsi callq 0x173d910 movl %eax, 0x354(%rsp) jmp 0x173d8ec movq 0x138(%rsp), %rax cmpl $0x0, 0xe0(...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::pack_A_tile(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { const int elempack = A.elempack; const int A_hstep = A.dims == 3 ? (int)A.cstep : A.w; float* pp = AT; int ii = 0; #if __SSE2__ #if __AVX__ #if __AVX512F__ for (; ii + 15 < max_ii; ii += 16) { if (el...
subq $0x398, %rsp # imm = 0x398 movq %rdi, 0xd8(%rsp) movq %rsi, 0xd0(%rsp) movl %edx, 0xcc(%rsp) movl %ecx, 0xc8(%rsp) movl %r8d, 0xc4(%rsp) movl %r9d, 0xc0(%rsp) movq 0xd8(%rsp), %rax movl 0x18(%rax), %eax movl %eax, 0xbc(%rsp) movq 0xd8(%rsp), %rax cmpl $0x3, 0x28(%rax) jne 0x17405f7 movq 0xd8(%rsp), %rax...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::Gemm_x86::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int Gemm_x86::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { #if NCNN_INT8 if (int8_scale_term) { // return Gemm::forward_int8(bottom_blobs, top_blobs, opt); return forward_int8(bottom_blobs, top_blobs, opt); } #endif int M; int...
pushq %rbp pushq %r15 pushq %r14 pushq %r12 pushq %rbx subq $0x890, %rsp # imm = 0x890 movq %rdi, 0x4a0(%rsp) movq %rsi, 0x498(%rsp) movq %rdx, 0x490(%rsp) movq %rcx, 0x488(%rsp) movq 0x4a0(%rsp), %rax movq %rax, 0x238(%rsp) cmpl $0x0, 0x10c(%rax) je 0x1746039 movq 0x238(%rsp), %rdi movq 0x498(%rsp), %rsi mo...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::Gemm_x86::forward_int8(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int Gemm_x86::forward_int8(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { int M; int N; if (constantA && constantB) { M = constantM; N = constantN; } else if (constantA) { const Mat& B = bottom_blobs[0]; M = const...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x718, %rsp # imm = 0x718 movq %rdi, 0x3e8(%rsp) movq %rsi, 0x3e0(%rsp) movq %rdx, 0x3d8(%rsp) movq %rcx, 0x3d0(%rsp) movq 0x3e8(%rsp), %rax movq %rax, 0x1d8(%rsp) cmpl $0x0, 0xe0(%rax) je 0x1749372 movq 0x1d8(%rsp), %rax cmpl $0x0, 0xe4...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::gemm_AT_BT_x86(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int, int, int, int, ncnn::Option const&)
static int gemm_AT_BT_x86(const Mat& AT, const Mat& BT, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int N, int K, int output_transpose, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { // NCNN_LOGE("M/N/K = %d %d %d", M, N, K); int TILE_M, TILE_N, TILE_K...
pushq %rbp pushq %rbx subq $0x7f8, %rsp # imm = 0x7F8 movq 0x848(%rsp), %rax movl 0x840(%rsp), %eax movl 0x838(%rsp), %eax movl 0x830(%rsp), %eax movl 0x828(%rsp), %eax movl 0x820(%rsp), %eax movl 0x818(%rsp), %eax movl 0x810(%rsp), %eax movq %rdi, 0x3c0(%rsp) movq %rsi, 0x3b8(%rsp) movq %rdx, 0x3b0(%rsp) mo...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::get_optimal_tile_mnk_int8(int, int, int, int, int, int, int&, int&, int&, int)
static void get_optimal_tile_mnk_int8(int M, int N, int K, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int& TILE_M, int& TILE_N, int& TILE_K, int nT) { // resolve optimal tile size from cache size const size_t l2_cache_size = get_cpu_level2_cache_size(); if (nT == 0) nT = get_phy...
subq $0x98, %rsp movl 0xb8(%rsp), %eax movq 0xb0(%rsp), %rax movq 0xa8(%rsp), %rax movq 0xa0(%rsp), %rax movl %edi, 0x94(%rsp) movl %esi, 0x90(%rsp) movl %edx, 0x8c(%rsp) movl %ecx, 0x88(%rsp) movl %r8d, 0x84(%rsp) movl %r9d, 0x80(%rsp) callq 0x7e540 cltq movq %rax, 0x78(%rsp) cmpl $0x0, 0xb8(%rsp) jne 0x1756a44 callq ...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::transpose_pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { transpose_pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; ...
subq $0x20, %rsp movq %rdi, -0x20(%rsp) movq %rsi, -0x28(%rsp) movl %edx, -0x2c(%rsp) movl %ecx, -0x30(%rsp) movl %r8d, -0x34(%rsp) movl %r9d, -0x38(%rsp) movq -0x20(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, -0x3c(%rsp) movq -0x28(%rsp), %rax movq %rax, -0x18(%rsp) movq -0x18(%rsp), %rax movq (%rax), %rax movq %rax,...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; } #endif #if NCN...
subq $0x78, %rsp movq %rdi, -0x8(%rsp) movq %rsi, -0x10(%rsp) movl %edx, -0x14(%rsp) movl %ecx, -0x18(%rsp) movl %r8d, -0x1c(%rsp) movl %r9d, -0x20(%rsp) movq -0x10(%rsp), %rax movq %rax, (%rsp) movq (%rsp), %rax movq (%rax), %rax movq %rax, -0x28(%rsp) movl $0x0, -0x2c(%rsp) movl -0x2c(%rsp), %eax addl $0x3, %eax cmpl...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; } #endif #if NCN...
subq $0x140, %rsp # imm = 0x140 movq %rdi, 0x40(%rsp) movq %rsi, 0x38(%rsp) movl %edx, 0x34(%rsp) movl %ecx, 0x30(%rsp) movl %r8d, 0x2c(%rsp) movl %r9d, 0x28(%rsp) movq 0x38(%rsp), %rax movq %rax, 0x48(%rsp) movq 0x48(%rsp), %rax movq (%rax), %rax movq %rax, 0x20(%rsp) movl $0x0, 0x1c(%rsp) movl 0x1c(%rsp), ...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::transpose_unpack_output_tile(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_unpack_output_tile(const Mat& topT, Mat& top_blob, int i, int max_ii, int j, int max_jj) { const int out_elempack = top_blob.elempack; const int out_hstep = top_blob.dims == 3 ? (int)top_blob.cstep : top_blob.w; const float* pp = topT; int ii = 0; #if __SSE2__ #if __AVX__ #if __A...
subq $0x2d8, %rsp # imm = 0x2D8 movq %rdi, 0xb8(%rsp) movq %rsi, 0xb0(%rsp) movl %edx, 0xac(%rsp) movl %ecx, 0xa8(%rsp) movl %r8d, 0xa4(%rsp) movl %r9d, 0xa0(%rsp) movq 0xb0(%rsp), %rax movl 0x18(%rax), %eax movl %eax, 0x9c(%rsp) movq 0xb0(%rsp), %rax cmpl $0x3, 0x28(%rax) jne 0x1773967 movq 0xb0(%rsp), %rax...
/Tencent[P]ncnn/src/layer/x86/gemm_x86.cpp
ncnn::Gemm_x86_avx512::create_pipeline_int8(ncnn::Option const&)
int Gemm_x86_avx512::create_pipeline_int8(const Option& opt) { if (constantA) { const int M = constantM; const int K = constantK; int TILE_M, TILE_N, TILE_K; get_optimal_tile_mnk_int8(M, 0, K, constant_TILE_M, constant_TILE_N, constant_TILE_K, TILE_M, TILE_N, TILE_K, opt.num_thr...
pushq %rbx subq $0x5f0, %rsp # imm = 0x5F0 movq %rdi, 0x2b0(%rsp) movq %rsi, 0x2a8(%rsp) movq 0x2b0(%rsp), %rax movq %rax, 0x100(%rsp) cmpl $0x0, 0xe0(%rax) je 0x1798bbf movq 0x100(%rsp), %rax movl 0xec(%rax), %ecx movl %ecx, 0x2a4(%rsp) movl 0xf4(%rax), %ecx movl %ecx, 0x2a0(%rsp) movl 0x2a4(%rsp), %edi mov...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx512.cpp
ncnn::gemm_AT_BT_x86(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int, int, int, int, ncnn::Option const&)
static int gemm_AT_BT_x86(const Mat& AT, const Mat& BT, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int N, int K, int output_transpose, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { // NCNN_LOGE("M/N/K = %d %d %d", M, N, K); int TILE_M, TILE_N, TILE_K...
pushq %rbp pushq %rbx subq $0x7f8, %rsp # imm = 0x7F8 movq 0x848(%rsp), %rax movl 0x840(%rsp), %eax movl 0x838(%rsp), %eax movl 0x830(%rsp), %eax movl 0x828(%rsp), %eax movl 0x820(%rsp), %eax movl 0x818(%rsp), %eax movl 0x810(%rsp), %eax movq %rdi, 0x3c0(%rsp) movq %rsi, 0x3b8(%rsp) movq %rdx, 0x3b0(%rsp) mo...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx512.cpp
ncnn::gemm_AT_x86(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int, int, int, int, ncnn::Option const&)
static int gemm_AT_x86(const Mat& AT, const Mat& B, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int K, int transB, int output_transpose, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { const int N = transB ? (B.dims == 3 ? B.c : B.h) * B.elempack : B.w; ...
pushq %rbp pushq %rbx subq $0xb28, %rsp # imm = 0xB28 movq 0xb78(%rsp), %rax movl 0xb70(%rsp), %eax movl 0xb68(%rsp), %eax movl 0xb60(%rsp), %eax movl 0xb58(%rsp), %eax movl 0xb50(%rsp), %eax movl 0xb48(%rsp), %eax movl 0xb40(%rsp), %eax movq %rdi, 0x530(%rsp) movq %rsi, 0x528(%rsp) movq %rdx, 0x520(%rsp) mo...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx512.cpp
ncnn::gemm_x86(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int, int, int, ncnn::Option const&)
static int gemm_x86(const Mat& A, const Mat& B, const Mat& C, Mat& top_blob, int broadcast_type_C, int transA, int transB, int output_transpose, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { const int M = transA ? A.w : (A.dims == 3 ? A.c : A.h) * A.elempack; const ...
pushq %rbp pushq %rbx subq $0xc58, %rsp # imm = 0xC58 movq 0xca0(%rsp), %rax movl 0xc98(%rsp), %eax movl 0xc90(%rsp), %eax movl 0xc88(%rsp), %eax movl 0xc80(%rsp), %eax movl 0xc78(%rsp), %eax movl 0xc70(%rsp), %eax movq %rdi, 0x5e8(%rsp) movq %rsi, 0x5e0(%rsp) movq %rdx, 0x5d8(%rsp) movq %rcx, 0x5d0(%rsp) mo...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx512.cpp
ncnn::pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; } #endif #if NCN...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %rbx andq $-0x40, %rsp subq $0x8c0, %rsp # imm = 0x8C0 movq %rdi, 0x1d0(%rsp) movq %rsi, 0x1c8(%rsp) movl %edx, 0x1c4(%rsp) movl %ecx, 0x1c0(%rsp) movl %r8d, 0x1bc(%rsp) movl %r9d, 0x1b8(%rsp) callq 0x7e2e0 cmpl $0x0, %eax je 0x17c4094 movq 0x1d0(%rsp), ...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::gemm_BT_x86(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int, int, int, int, ncnn::Option const&)
static int gemm_BT_x86(const Mat& A, const Mat& BT, const Mat& C, Mat& top_blob, int broadcast_type_C, int N, int K, int transA, int output_transpose, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { const int M = transA ? A.w : (A.dims == 3 ? A.c : A.h) * A.elempack; ...
pushq %rbp pushq %rbx subq $0x908, %rsp # imm = 0x908 movq 0x958(%rsp), %rax movl 0x950(%rsp), %eax movl 0x948(%rsp), %eax movl 0x940(%rsp), %eax movl 0x938(%rsp), %eax movl 0x930(%rsp), %eax movl 0x928(%rsp), %eax movl 0x920(%rsp), %eax movq %rdi, 0x458(%rsp) movq %rsi, 0x450(%rsp) movq %rdx, 0x448(%rsp) mo...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_fma.cpp
ncnn::transpose_pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { transpose_pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; ...
subq $0x1b8, %rsp # imm = 0x1B8 movq %rdi, 0xb0(%rsp) movq %rsi, 0xa8(%rsp) movl %edx, 0xa4(%rsp) movl %ecx, 0xa0(%rsp) movl %r8d, 0x9c(%rsp) movl %r9d, 0x98(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x18d0977 movq 0xb0(%rsp), %rdi movq 0xa8(%rsp), %rsi movl 0xa4(%rsp), %edx movl 0xa0(%rsp), %ecx movl 0x9c(%rsp...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::gemm_AT_BT_x86_int8(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, float, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, float, float, int, int, int, int, ncnn::Option const&)
static int gemm_AT_BT_x86_int8(const Mat& AT, const Mat& A_int8_scales, const Mat& BT, float B_int8_scale, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int N, int K, int output_transpose, float alpha, float beta, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { ...
pushq %rbx subq $0x850, %rsp # imm = 0x850 movq 0x8a0(%rsp), %rax movl 0x898(%rsp), %eax movl 0x890(%rsp), %eax movl 0x888(%rsp), %eax movl 0x880(%rsp), %eax movl 0x878(%rsp), %eax movl 0x870(%rsp), %eax movl 0x868(%rsp), %eax movl 0x860(%rsp), %eax movq %rdi, 0x3e8(%rsp) movq %rsi, 0x3e0(%rsp) movq %rdx, 0x...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_fma.cpp
ncnn::gemm_AT_x86_int8(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, float, float, int, int, int, int, ncnn::Option const&)
static int gemm_AT_x86_int8(const Mat& AT, const Mat& A_int8_scales, const Mat& B, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int K, int transB, int output_transpose, float alpha, float beta, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { // NCNN_LOGE("gem...
pushq %rbx subq $0xb70, %rsp # imm = 0xB70 movq 0xbc0(%rsp), %rax movl 0xbb8(%rsp), %eax movl 0xbb0(%rsp), %eax movl 0xba8(%rsp), %eax movl 0xba0(%rsp), %eax movl 0xb98(%rsp), %eax movl 0xb90(%rsp), %eax movl 0xb88(%rsp), %eax movl 0xb80(%rsp), %eax movq %rdi, 0x548(%rsp) movq %rsi, 0x540(%rsp) movq %rdx, 0x...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_fma.cpp
ncnn::Gemm_x86_avx::create_pipeline(ncnn::Option const&)
int Gemm_x86_avx::create_pipeline(const Option& opt) { #if NCNN_INT8 if (int8_scale_term) { return create_pipeline_int8(opt); } #endif if (constantA) { const int M = constantM; const int K = constantK; int TILE_M, TILE_N, TILE_K; get_optimal_tile_mnk(M, 0, K...
pushq %rbx subq $0x730, %rsp # imm = 0x730 movq %rdi, 0x348(%rsp) movq %rsi, 0x340(%rsp) movq 0x348(%rsp), %rax movq %rax, 0x138(%rsp) cmpl $0x0, 0x10c(%rax) je 0x193fe92 movq 0x138(%rsp), %rdi movq 0x340(%rsp), %rsi callq 0x19421b0 movl %eax, 0x354(%rsp) jmp 0x194218b movq 0x138(%rsp), %rax cmpl $0x0, 0xe0(...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx.cpp
ncnn::Gemm_x86_avx::create_pipeline_int8(ncnn::Option const&)
int Gemm_x86_avx::create_pipeline_int8(const Option& opt) { if (constantA) { const int M = constantM; const int K = constantK; int TILE_M, TILE_N, TILE_K; get_optimal_tile_mnk_int8(M, 0, K, constant_TILE_M, constant_TILE_N, constant_TILE_K, TILE_M, TILE_N, TILE_K, opt.num_thread...
pushq %rbx subq $0x5f0, %rsp # imm = 0x5F0 movq %rdi, 0x2b0(%rsp) movq %rsi, 0x2a8(%rsp) movq 0x2b0(%rsp), %rax movq %rax, 0x100(%rsp) cmpl $0x0, 0xe0(%rax) je 0x1942f4f movq 0x100(%rsp), %rax movl 0xec(%rax), %ecx movl %ecx, 0x2a4(%rsp) movl 0xf4(%rax), %ecx movl %ecx, 0x2a0(%rsp) movl 0x2a4(%rsp), %edi mov...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx.cpp
ncnn::Gemm_x86_avx::forward_int8(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int Gemm_x86_avx::forward_int8(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { int M; int N; if (constantA && constantB) { M = constantM; N = constantN; } else if (constantA) { const Mat& B = bottom_blobs[0]; M = c...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x718, %rsp # imm = 0x718 movq %rdi, 0x3e8(%rsp) movq %rsi, 0x3e0(%rsp) movq %rdx, 0x3d8(%rsp) movq %rcx, 0x3d0(%rsp) movq 0x3e8(%rsp), %rax movq %rax, 0x1d8(%rsp) cmpl $0x0, 0xe0(%rax) je 0x1955952 movq 0x1d8(%rsp), %rax cmpl $0x0, 0xe4...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx.cpp
ncnn::get_optimal_tile_mnk_int8(int, int, int, int, int, int, int&, int&, int&, int)
static void get_optimal_tile_mnk_int8(int M, int N, int K, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int& TILE_M, int& TILE_N, int& TILE_K, int nT) { // resolve optimal tile size from cache size const size_t l2_cache_size = get_cpu_level2_cache_size(); if (nT == 0) nT = get_phy...
subq $0x98, %rsp movl 0xb8(%rsp), %eax movq 0xb0(%rsp), %rax movq 0xa8(%rsp), %rax movq 0xa0(%rsp), %rax movl %edi, 0x94(%rsp) movl %esi, 0x90(%rsp) movl %edx, 0x8c(%rsp) movl %ecx, 0x88(%rsp) movl %r8d, 0x84(%rsp) movl %r9d, 0x80(%rsp) callq 0x7e540 cltq movq %rax, 0x78(%rsp) cmpl $0x0, 0xb8(%rsp) jne 0x1963304 callq ...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::transpose_pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { transpose_pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; ...
subq $0xa8, %rsp movq %rdi, 0x68(%rsp) movq %rsi, 0x60(%rsp) movl %edx, 0x5c(%rsp) movl %ecx, 0x58(%rsp) movl %r8d, 0x54(%rsp) movl %r9d, 0x50(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x19637c3 movq 0x68(%rsp), %rdi movq 0x60(%rsp), %rsi movl 0x5c(%rsp), %edx movl 0x58(%rsp), %ecx movl 0x54(%rsp), %r8d movl 0x50(%rsp), %...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; } #endif #if NCN...
subq $0xf8, %rsp movq %rdi, 0x78(%rsp) movq %rsi, 0x70(%rsp) movl %edx, 0x6c(%rsp) movl %ecx, 0x68(%rsp) movl %r8d, 0x64(%rsp) movl %r9d, 0x60(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1963ca3 movq 0x78(%rsp), %rdi movq 0x70(%rsp), %rsi movl 0x6c(%rsp), %edx movl 0x68(%rsp), %ecx movl 0x64(%rsp), %r8d movl 0x60(%rsp), %...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; } #endif #if NCN...
subq $0x1c8, %rsp # imm = 0x1C8 movq %rdi, 0xc8(%rsp) movq %rsi, 0xc0(%rsp) movl %edx, 0xbc(%rsp) movl %ecx, 0xb8(%rsp) movl %r8d, 0xb4(%rsp) movl %r9d, 0xb0(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x19642f7 movq 0xc8(%rsp), %rdi movq 0xc0(%rsp), %rsi movl 0xbc(%rsp), %edx movl 0xb8(%rsp), %ecx movl 0xb4(%rsp...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::gemm_AT_x86_int8(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, float, float, int, int, int, int, ncnn::Option const&)
static int gemm_AT_x86_int8(const Mat& AT, const Mat& A_int8_scales, const Mat& B, const Mat& C, Mat& top_blob, int broadcast_type_C, int M, int K, int transB, int output_transpose, float alpha, float beta, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { // NCNN_LOGE("gem...
pushq %rbx subq $0xb70, %rsp # imm = 0xB70 movq 0xbc0(%rsp), %rax movl 0xbb8(%rsp), %eax movl 0xbb0(%rsp), %eax movl 0xba8(%rsp), %eax movl 0xba0(%rsp), %eax movl 0xb98(%rsp), %eax movl 0xb90(%rsp), %eax movl 0xb88(%rsp), %eax movl 0xb80(%rsp), %eax movq %rdi, 0x548(%rsp) movq %rsi, 0x540(%rsp) movq %rdx, 0x...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx.cpp
ncnn::gemm_x86_int8(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, int, int, int, float, float, int, int, int, int, ncnn::Option const&)
static int gemm_x86_int8(const Mat& A, const Mat& B, const Mat& C, Mat& top_blob, int broadcast_type_C, int transA, int transB, int output_transpose, float alpha, float beta, int constant_TILE_M, int constant_TILE_N, int constant_TILE_K, int nT, const Option& opt) { // NCNN_LOGE("gemm_x86_int8"); const int M =...
pushq %rbx subq $0xd70, %rsp # imm = 0xD70 movq 0xdb0(%rsp), %rax movl 0xda8(%rsp), %eax movl 0xda0(%rsp), %eax movl 0xd98(%rsp), %eax movl 0xd90(%rsp), %eax movl 0xd88(%rsp), %eax movl 0xd80(%rsp), %eax movq %rdi, 0x6a0(%rsp) movq %rsi, 0x698(%rsp) movq %rdx, 0x690(%rsp) movq %rcx, 0x688(%rsp) movl %r8d, 0x...
/Tencent[P]ncnn/build_O0/src/layer/x86/gemm_x86_avx.cpp
ncnn::compute_B_fp32_int8_scale(ncnn::Mat const&, float&)
static void compute_B_fp32_int8_scale(const Mat& B, float& scale) { // NCNN_LOGE("compute_B_fp32_int8_scale"); float absmax = 0.f; #if __SSE2__ #if __AVX__ #if __AVX512F__ __m512 _absmax_avx512 = _mm512_setzero_ps(); #endif // __AVX512F__ __m256 _absmax_avx = _mm256_setzero_ps(); #endif // __AVX__ ...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x500, %rsp # imm = 0x500 movq %rdi, 0xe0(%rsp) movq %rsi, 0xd8(%rsp) movl $0x0, 0xd4(%rsp) vxorps %xmm0, %xmm0, %xmm0 vmovaps %ymm0, 0x100(%rsp) vmovaps 0x100(%rsp), %ymm0 vmovaps %ymm0, 0xa0(%rsp) vxorps %xmm0, %xmm0, %xmm0 vmovaps %xmm0, 0x130(%rsp) vmova...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; } #endif #if NCN...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0xb60, %rsp # imm = 0xB60 movq %rdi, 0x1e0(%rsp) movq %rsi, 0x1d8(%rsp) movl %edx, 0x1d4(%rsp) movl %ecx, 0x1d0(%rsp) movl %r8d, 0x1cc(%rsp) movl %r9d, 0x1c8(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1a60...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; } #endif #if NCN...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x9a0, %rsp # imm = 0x9A0 movq %rdi, 0x1a0(%rsp) movq %rsi, 0x198(%rsp) movl %edx, 0x194(%rsp) movl %ecx, 0x190(%rsp) movl %r8d, 0x18c(%rsp) movl %r9d, 0x188(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1a64...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::transpose_pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { transpose_pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; ...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x660, %rsp # imm = 0x660 movq %rdi, 0x1a8(%rsp) movq %rsi, 0x1a0(%rsp) movl %edx, 0x19c(%rsp) movl %ecx, 0x198(%rsp) movl %r8d, 0x194(%rsp) movl %r9d, 0x190(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1a66...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::pack_A_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void pack_A_tile_int8(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { pack_A_tile_int8_avx512vnni(A, AT, i, max_ii, k, max_kk); return; } #endif #if NCN...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x7a0, %rsp # imm = 0x7A0 movq %rdi, 0x120(%rsp) movq %rsi, 0x118(%rsp) movl %edx, 0x114(%rsp) movl %ecx, 0x110(%rsp) movl %r8d, 0x10c(%rsp) movl %r9d, 0x108(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1af2...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::transpose_pack_B_tile_int8(ncnn::Mat const&, ncnn::Mat&, int, int, int, int)
static void transpose_pack_B_tile_int8(const Mat& B, Mat& BT, int j, int max_jj, int k, int max_kk) { #if NCNN_RUNTIME_CPU && NCNN_AVX512VNNI && __AVX512F__ && !__AVX512VNNI__ if (ncnn::cpu_support_x86_avx512_vnni()) { transpose_pack_B_tile_int8_avx512vnni(B, BT, j, max_jj, k, max_kk); return; ...
subq $0x1c8, %rsp # imm = 0x1C8 movq %rdi, 0xb0(%rsp) movq %rsi, 0xa8(%rsp) movl %edx, 0xa4(%rsp) movl %ecx, 0xa0(%rsp) movl %r8d, 0x9c(%rsp) movl %r9d, 0x98(%rsp) callq 0x7e2a0 cmpl $0x0, %eax je 0x1af5ec7 movq 0xb0(%rsp), %rdi movq 0xa8(%rsp), %rsi movl 0xa4(%rsp), %edx movl 0xa0(%rsp), %ecx movl 0x9c(%rsp...
/Tencent[P]ncnn/src/layer/x86/gemm_int8.h
ncnn::GroupNorm::GroupNorm()
GroupNorm::GroupNorm() { one_blob_only = true; support_inplace = true; }
subq $0x68, %rsp movq %rdi, 0x30(%rsp) movq 0x30(%rsp), %rdi movq %rdi, 0x10(%rsp) callq 0x88680 movq 0x10(%rsp), %rax leaq 0x3e5ebc(%rip), %rcx # 0x1f24d90 addq $0x10, %rcx movq %rcx, (%rax) addq $0xe0, %rax movq %rax, 0x38(%rsp) movq 0x38(%rsp), %rax movq %rax, 0x18(%rsp) movq $0x0, (%rax) movq $0x0, 0x8(%rax) mov...
/Tencent[P]ncnn/src/layer/groupnorm.cpp
ncnn::GroupNorm::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int GroupNorm::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { const int dims = bottom_top_blob.dims; const int channels_per_group = channels / group; if (dims == 1) { #pragma omp parallel for num_threads(opt.num_threads) for (int g = 0; g < group; g++) { ...
subq $0xe58, %rsp # imm = 0xE58 movq %rdi, 0x620(%rsp) movq %rsi, 0x618(%rsp) movq %rdx, 0x610(%rsp) movq 0x620(%rsp), %rcx movq %rcx, 0x190(%rsp) movq 0x618(%rsp), %rax movl 0x28(%rax), %eax movl %eax, 0x60c(%rsp) movl 0xd4(%rcx), %eax cltd idivl 0xd0(%rcx) movl %eax, 0x608(%rsp) cmpl $0x1, 0x60c(%rsp) jne ...
/Tencent[P]ncnn/src/layer/groupnorm.cpp
ncnn::LayerNorm_x86::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LayerNorm_x86::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { const int dims = bottom_top_blob.dims; const int elempack = bottom_top_blob.elempack; const int w = bottom_top_blob.w; const int h = bottom_top_blob.h; const int channels = bottom_top_blob.c; if (dims == 1) ...
subq $0x348, %rsp # imm = 0x348 movq %rdi, 0x150(%rsp) movq %rsi, 0x148(%rsp) movq %rdx, 0x140(%rsp) movq 0x150(%rsp), %rax movq %rax, 0x50(%rsp) movq 0x148(%rsp), %rax movl 0x28(%rax), %eax movl %eax, 0x13c(%rsp) movq 0x148(%rsp), %rax movl 0x18(%rax), %eax movl %eax, 0x138(%rsp) movq 0x148(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/x86/layernorm_x86.cpp
ncnn::layernorm(float*, float const*, float const*, float, int, int)
static void layernorm(float* ptr, const float* gamma_ptr, const float* beta_ptr, float eps, int elemcount, int elempack) { const int size = elemcount * elempack; #if __SSE2__ #if __AVX__ #if __AVX512F__ __m512 _mean_avx512 = _mm512_set1_ps(0.f); #endif // __AVX512F__ __m256 _mean_avx = _mm256_set1_ps(0.f);...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0x18a0, %rsp # imm = 0x18A0 movq %rdi, 0x4b8(%rsp) movq %rsi, 0x4b0(%rsp) movq %rdx, 0x4a8(%rsp) vmovss %xmm0, 0x4a4(%rsp) movl %ecx, 0x4a0(%rsp) movl %r8d, 0x49c(%rsp) movl 0x4a0(%rsp), %eax movl 0x49c(%rsp), %ecx imull %ecx, %eax movl %eax, 0x498(%rsp) movl...
/Tencent[P]ncnn/build_O0/src/layer/x86/layernorm_x86_avx.cpp
ncnn::GRU::load_model(ncnn::ModelBin const&)
int GRU::load_model(const ModelBin& mb) { int num_directions = direction == 2 ? 2 : 1; int size = weight_data_size / num_directions / num_output / 3; // raw weight data weight_xc_data = mb.load(size, num_output * 3, num_directions, 0); if (weight_xc_data.empty()) return -100; bias_c_d...
subq $0x4f8, %rsp # imm = 0x4F8 movq %rdi, 0x260(%rsp) movq %rsi, 0x258(%rsp) movq 0x260(%rsp), %rcx movq %rcx, 0xc8(%rsp) movl 0xd8(%rcx), %esi movl $0x1, %eax movl $0x2, %edx cmpl $0x2, %esi cmovel %edx, %eax movl %eax, 0x254(%rsp) movl 0xd4(%rcx), %eax cltd idivl 0x254(%rsp) cltd idivl 0xd0(%rcx) movl $0x...
/Tencent[P]ncnn/src/layer/gru.cpp
ncnn::gru(ncnn::Mat const&, ncnn::Mat&, int, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static int gru(const Mat& bottom_blob, Mat& top_blob, int reverse, const Mat& weight_xc, const Mat& bias_c, const Mat& weight_hc, Mat& hidden_state, const Option& opt) { int size = bottom_blob.w; int T = bottom_blob.h; int num_output = top_blob.w; // 2 x num_output Mat gates(2, num_output, 4u, opt...
subq $0x338, %rsp # imm = 0x338 movq 0x348(%rsp), %rax movq 0x340(%rsp), %rax movq %rdi, 0x198(%rsp) movq %rsi, 0x190(%rsp) movl %edx, 0x18c(%rsp) movq %rcx, 0x180(%rsp) movq %r8, 0x178(%rsp) movq %r9, 0x170(%rsp) movq 0x198(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x16c(%rsp) movq 0x198(%rsp), %rax movl...
/Tencent[P]ncnn/src/layer/gru.cpp
ncnn::MultiHeadAttention::forward_int8(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int MultiHeadAttention::forward_int8(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& q_blob = bottom_blobs[0]; const Mat& k_blob = (bottom_blobs.size() == 1 || (bottom_blobs.size() == 2 && attn_mask)) ? q_blob : bottom_blobs[1]; const Mat& v_blob = (b...
subq $0x2698, %rsp # imm = 0x2698 movq %rdi, 0x11e0(%rsp) movq %rsi, 0x11d8(%rsp) movq %rdx, 0x11d0(%rsp) movq %rcx, 0x11c8(%rsp) movq 0x11e0(%rsp), %rax movq %rax, 0x650(%rsp) movq 0x11d8(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq %rax, 0x11c0(%rsp) movq 0x11d8(%rsp), %rdi callq 0x704d0 c...
/Tencent[P]ncnn/src/layer/multiheadattention.cpp
ncnn::MultiHeadAttention_x86::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int MultiHeadAttention_x86::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& _opt) const { const Mat& q_blob = bottom_blobs[0]; const Mat& k_blob = (bottom_blobs.size() == 1 || (bottom_blobs.size() == 2 && attn_mask)) ? q_blob : bottom_blobs[1]; const Mat& v_blob = (b...
subq $0x1158, %rsp # imm = 0x1158 movq %rdi, 0x8f0(%rsp) movq %rsi, 0x8e8(%rsp) movq %rdx, 0x8e0(%rsp) movq %rcx, 0x8d8(%rsp) movq 0x8f0(%rsp), %rax movq %rax, 0x2e0(%rsp) movq 0x8e8(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq %rax, 0x8d0(%rsp) movq 0x8e8(%rsp), %rdi callq 0x704d0 cmpq $0x1...
/Tencent[P]ncnn/src/layer/x86/multiheadattention_x86.cpp
ncnn::MultiHeadAttention_x86_avx512::create_pipeline(ncnn::Option const&)
int MultiHeadAttention_x86_avx512::create_pipeline(const Option& _opt) { Option opt = _opt; if (int8_scale_term) { support_packing = false; opt.use_packing_layout = false; // TODO enable packing } { qk_softmax = ncnn::create_layer_cpu(ncnn::LayerType::Softmax); ncnn...
subq $0xe28, %rsp # imm = 0xE28 movq %rdi, 0x8b0(%rsp) movq %rsi, 0x8a8(%rsp) movq 0x8b0(%rsp), %rax movq %rax, 0x318(%rsp) movq 0x8a8(%rsp), %rsi leaq 0x868(%rsp), %rdi movl $0x40, %edx callq 0x24580 movq 0x318(%rsp), %rax cmpl $0x0, 0xec(%rax) je 0x1ba3816 movq 0x318(%rsp), %rax movb $0x0, 0xb(%rax) movb $...
/Tencent[P]ncnn/build_O0/src/layer/x86/multiheadattention_x86_avx512.cpp
ncnn::MultiHeadAttention_x86_fma::create_pipeline(ncnn::Option const&)
int MultiHeadAttention_x86_fma::create_pipeline(const Option& _opt) { Option opt = _opt; if (int8_scale_term) { support_packing = false; opt.use_packing_layout = false; // TODO enable packing } { qk_softmax = ncnn::create_layer_cpu(ncnn::LayerType::Softmax); ncnn::P...
subq $0xe28, %rsp # imm = 0xE28 movq %rdi, 0x8b0(%rsp) movq %rsi, 0x8a8(%rsp) movq 0x8b0(%rsp), %rax movq %rax, 0x318(%rsp) movq 0x8a8(%rsp), %rsi leaq 0x868(%rsp), %rdi movl $0x40, %edx callq 0x24580 movq 0x318(%rsp), %rax cmpl $0x0, 0xec(%rax) je 0x1bad2e6 movq 0x318(%rsp), %rax movb $0x0, 0xb(%rax) movb $...
/Tencent[P]ncnn/build_O0/src/layer/x86/multiheadattention_x86_fma.cpp
ncnn::MultiHeadAttention_x86_fma::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int MultiHeadAttention_x86_fma::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& _opt) const { const Mat& q_blob = bottom_blobs[0]; const Mat& k_blob = (bottom_blobs.size() == 1 || (bottom_blobs.size() == 2 && attn_mask)) ? q_blob : bottom_blobs[1]; const Mat& v_blob ...
subq $0x1158, %rsp # imm = 0x1158 movq %rdi, 0x8f0(%rsp) movq %rsi, 0x8e8(%rsp) movq %rdx, 0x8e0(%rsp) movq %rcx, 0x8d8(%rsp) movq 0x8f0(%rsp), %rax movq %rax, 0x2e0(%rsp) movq 0x8e8(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq %rax, 0x8d0(%rsp) movq 0x8e8(%rsp), %rdi callq 0x704d0 cmpq $0x1...
/Tencent[P]ncnn/build_O0/src/layer/x86/multiheadattention_x86_fma.cpp
ncnn::MultiHeadAttention_x86_avx::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int MultiHeadAttention_x86_avx::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& _opt) const { const Mat& q_blob = bottom_blobs[0]; const Mat& k_blob = (bottom_blobs.size() == 1 || (bottom_blobs.size() == 2 && attn_mask)) ? q_blob : bottom_blobs[1]; const Mat& v_blob ...
subq $0x1158, %rsp # imm = 0x1158 movq %rdi, 0x8f0(%rsp) movq %rsi, 0x8e8(%rsp) movq %rdx, 0x8e0(%rsp) movq %rcx, 0x8d8(%rsp) movq 0x8f0(%rsp), %rax movq %rax, 0x2e0(%rsp) movq 0x8e8(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq %rax, 0x8d0(%rsp) movq 0x8e8(%rsp), %rdi callq 0x704d0 cmpq $0x1...
/Tencent[P]ncnn/build_O0/src/layer/x86/multiheadattention_x86_avx.cpp
ncnn::GELU_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int GELU_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { if (!fast_gelu) { return GELU::forward_inplace(bottom_top_blob, opt); } int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int elempack = bottom_top_blob.elempack; in...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0xe60, %rsp # imm = 0xE60 movq %rdi, 0x240(%rsp) movq %rsi, 0x238(%rsp) movq %rdx, 0x230(%rsp) movq 0x240(%rsp), %rax movq %rax, 0x48(%rsp) cmpl $0x0, 0xd0(%rax) jne 0x1bc4893 movq 0x48(%rsp), %rdi movq 0x238(%rsp), %rsi movq 0x230(%rsp), %rdx callq 0x1bc087...
/Tencent[P]ncnn/build_O0/src/layer/x86/gelu_x86_fma.cpp
ncnn::GELU_x86_avx::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int GELU_x86_avx::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { if (!fast_gelu) { return GELU::forward_inplace(bottom_top_blob, opt); } int w = bottom_top_blob.w; int h = bottom_top_blob.h; int d = bottom_top_blob.d; int elempack = bottom_top_blob.elempack; in...
pushq %rbp movq %rsp, %rbp andq $-0x20, %rsp subq $0xfc0, %rsp # imm = 0xFC0 movq %rdi, 0x240(%rsp) movq %rsi, 0x238(%rsp) movq %rdx, 0x230(%rsp) movq 0x240(%rsp), %rax movq %rax, 0x48(%rsp) cmpl $0x0, 0xd0(%rax) jne 0x1bc65b3 movq 0x48(%rsp), %rdi movq 0x238(%rsp), %rsi movq 0x230(%rsp), %rdx callq 0x1bc087...
/Tencent[P]ncnn/build_O0/src/layer/x86/gelu_x86_avx.cpp
ncnn::Convolution1D::load_model(ncnn::ModelBin const&)
int Convolution1D::load_model(const ModelBin& mb) { if (dynamic_weight) return 0; weight_data = mb.load(weight_data_size, 0); if (weight_data.empty()) return -100; if (bias_term) { bias_data = mb.load(num_output, 1); if (bias_data.empty()) return -100; ...
subq $0x238, %rsp # imm = 0x238 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq 0x118(%rsp), %rax movq %rax, 0x68(%rsp) cmpl $0x0, 0x140(%rax) je 0x1bc8d8d movl $0x0, 0x124(%rsp) jmp 0x1bc9794 movq 0x68(%rsp), %rax movq 0x110(%rsp), %rsi movl 0xf0(%rax), %edx movq (%rsi), %rax leaq 0xc8(%rsp), %rdi xorl %...
/Tencent[P]ncnn/src/layer/convolution1d.cpp
ncnn::Convolution1D_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Convolution1D_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; size_t elemsize = bottom_blob.elemsize; int elempack = bottom_blob.elempack; const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1; Mat bottom_blob_bordered; make_paddi...
pushq %rbx subq $0x190, %rsp # imm = 0x190 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq %rdx, 0x108(%rsp) movq %rcx, 0x100(%rsp) movq 0x118(%rsp), %rdi movq %rdi, 0x68(%rsp) movq 0x110(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0xfc(%rsp) movq 0x110(%rsp), %rax movq 0x10(%rax), %rax movq %rax, 0xf0(%...
/Tencent[P]ncnn/src/layer/x86/convolution1d_x86.cpp
ncnn::Convolution1D_x86_fma::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Convolution1D_x86_fma::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; size_t elemsize = bottom_blob.elemsize; int elempack = bottom_blob.elempack; const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1; Mat bottom_blob_bordered; make_p...
pushq %rbx subq $0x190, %rsp # imm = 0x190 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq %rdx, 0x108(%rsp) movq %rcx, 0x100(%rsp) movq 0x118(%rsp), %rdi movq %rdi, 0x68(%rsp) movq 0x110(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0xfc(%rsp) movq 0x110(%rsp), %rax movq 0x10(%rax), %rax movq %rax, 0xf0(%...
/Tencent[P]ncnn/build_O0/src/layer/x86/convolution1d_x86_fma.cpp
ncnn::ConvolutionDepthWise1D::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise1D::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int _num_o...
subq $0x358, %rsp # imm = 0x358 movq %rdi, 0x230(%rsp) movq %rsi, 0x228(%rsp) movq %rdx, 0x220(%rsp) movq %rcx, 0x218(%rsp) movq 0x230(%rsp), %rax movq %rax, 0xd0(%rsp) movq 0x228(%rsp), %rdi xorl %eax, %eax movl %eax, %esi movq %rsi, 0xd8(%rsp) callq 0x89920 movq %rax, 0x210(%rsp) movq 0x228(%rsp), %rdi mov...
/Tencent[P]ncnn/src/layer/convolutiondepthwise1d.cpp
ncnn::Pooling3D::load_param(ncnn::ParamDict const&)
int Pooling3D::load_param(const ParamDict& pd) { pooling_type = pd.get(0, 0); kernel_w = pd.get(1, 0); kernel_h = pd.get(11, kernel_w); kernel_d = pd.get(21, kernel_w); stride_w = pd.get(2, 1); stride_h = pd.get(12, stride_w); stride_d = pd.get(22, stride_w); pad_left = pd.get(3, 0); ...
subq $0x18, %rsp movq %rdi, 0x10(%rsp) movq %rsi, 0x8(%rsp) movq 0x10(%rsp), %rax movq %rax, (%rsp) movq 0x8(%rsp), %rdi xorl %edx, %edx movl %edx, %esi callq 0x78d90 movl %eax, %ecx movq (%rsp), %rax movl %ecx, 0xd0(%rax) movq 0x8(%rsp), %rdi movl $0x1, %esi xorl %edx, %edx callq 0x78d90 movl %eax, %ecx movq (%rsp), %...
/Tencent[P]ncnn/src/layer/pooling3d.cpp
ncnn::Pooling3D::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Pooling3D::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { // max value in NxN window // avg value in NxN window int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; int channels = bottom_blob.c; size_t elemsize = bottom_blob.elemsize; // ...
subq $0x15e8, %rsp # imm = 0x15E8 movq %rdi, 0x9e8(%rsp) movq %rsi, 0x9e0(%rsp) movq %rdx, 0x9d8(%rsp) movq %rcx, 0x9d0(%rsp) movq 0x9e8(%rsp), %rax movq %rax, 0x2f0(%rsp) movq 0x9e0(%rsp), %rcx movl 0x2c(%rcx), %ecx movl %ecx, 0x9cc(%rsp) movq 0x9e0(%rsp), %rcx movl 0x30(%rcx), %ecx movl %ecx, 0x9c8(%rsp) mo...
/Tencent[P]ncnn/src/layer/pooling3d.cpp
ncnn::Pooling3D::make_padding(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
void Pooling3D::make_padding(const Mat& bottom_blob, Mat& bottom_blob_bordered, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; bottom_blob_bordered = bottom_blob; float pad_value = 0.f; if (pooling_type == PoolMethod_MAX) { pad_value...
pushq %rbx subq $0x210, %rsp # imm = 0x210 movq %rdi, 0x1d0(%rsp) movq %rsi, 0x1c8(%rsp) movq %rdx, 0x1c0(%rsp) movq %rcx, 0x1b8(%rsp) movq 0x1d0(%rsp), %rax movq %rax, 0x60(%rsp) movq 0x1c8(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, 0x1b4(%rsp) movq 0x1c8(%rsp), %rax movl 0x30(%rax), %eax movl %eax, 0x1b0...
/Tencent[P]ncnn/src/layer/pooling3d.cpp
ncnn::matmul_transb(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static void matmul_transb(const Mat& A, const Mat& B, Mat& top_blob, const Option& opt) { const int M = A.h; const int K = A.w; // assert A.w == B.w const int N = B.h; const float* pA = A; const float* pB = B; float* pOut = top_blob; #pragma omp parallel for num_threads(opt.num_threads) ...
subq $0x10, %rsp movq %rdi, -0x10(%rsp) movq %rsi, -0x18(%rsp) movq %rdx, -0x20(%rsp) movq %rcx, -0x28(%rsp) movq -0x10(%rsp), %rax movl 0x30(%rax), %eax movl %eax, -0x2c(%rsp) movq -0x10(%rsp), %rax movl 0x2c(%rax), %eax movl %eax, -0x30(%rsp) movq -0x18(%rsp), %rax movl 0x30(%rax), %eax movl %eax, -0x34(%rsp) movq -0...
/Tencent[P]ncnn/src/layer/matmul.cpp
ncnn::MatMul_x86::create_pipeline(ncnn::Option const&)
int MatMul_x86::create_pipeline(const Option& opt) { gemm = ncnn::create_layer_cpu(ncnn::LayerType::Gemm); ncnn::ParamDict pd; pd.set(2, 0); // transA pd.set(3, transB); // transB pd.set(4, 0); // constantA pd.set(5, 0); // constantB pd.set(6, 1); // constantC pd.set...
subq $0x58, %rsp movq %rdi, 0x50(%rsp) movq %rsi, 0x48(%rsp) movq 0x50(%rsp), %rax movq %rax, 0x8(%rsp) movl $0x4a, %edi callq 0x89670 movq %rax, %rcx movq 0x8(%rsp), %rax movq %rcx, 0xd8(%rax) leaq 0x38(%rsp), %rdi movq %rdi, 0x10(%rsp) callq 0x78060 movq 0x10(%rsp), %rdi movl $0x2, %esi xorl %edx, %edx callq 0x79040 ...
/Tencent[P]ncnn/src/layer/x86/matmul_x86.cpp
ncnn::MatMul_x86_avx512::destroy_pipeline(ncnn::Option const&)
int MatMul_x86_avx512::destroy_pipeline(const Option& opt) { if (gemm) { gemm->destroy_pipeline(opt); delete gemm; gemm = 0; } return 0; }
subq $0x28, %rsp movq %rdi, 0x20(%rsp) movq %rsi, 0x18(%rsp) movq 0x20(%rsp), %rax movq %rax, 0x10(%rsp) cmpq $0x0, 0xd8(%rax) je 0x1c6e5eb movq 0x10(%rsp), %rax movq 0xd8(%rax), %rdi movq 0x18(%rsp), %rsi movq (%rdi), %rax callq *0x28(%rax) movq 0x10(%rsp), %rax movq 0xd8(%rax), %rax movq %rax, 0x8(%rsp) cmpq $0x0, %r...
/Tencent[P]ncnn/build_O0/src/layer/x86/matmul_x86_avx512.cpp
ncnn::MatMul_x86_avx512::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int MatMul_x86_avx512::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& A = bottom_blobs[0]; const Mat& B = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int Adims = A.dims; const int Bdims = B.dims; const int max_ABdims = s...
subq $0x26d8, %rsp # imm = 0x26D8 movq %rdi, 0x11e8(%rsp) movq %rsi, 0x11e0(%rsp) movq %rdx, 0x11d8(%rsp) movq %rcx, 0x11d0(%rsp) movq 0x11e8(%rsp), %rax movq %rax, 0x630(%rsp) movq 0x11e0(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq %rax, 0x11c8(%rsp) movq 0x11e0(%rsp), %rdi movl $0x1, %esi...
/Tencent[P]ncnn/build_O0/src/layer/x86/matmul_x86_avx512.cpp
ncnn::MatMul_x86_fma::destroy_pipeline(ncnn::Option const&)
int MatMul_x86_fma::destroy_pipeline(const Option& opt) { if (gemm) { gemm->destroy_pipeline(opt); delete gemm; gemm = 0; } return 0; }
subq $0x28, %rsp movq %rdi, 0x20(%rsp) movq %rsi, 0x18(%rsp) movq 0x20(%rsp), %rax movq %rax, 0x10(%rsp) cmpq $0x0, 0xd8(%rax) je 0x1c7af4b movq 0x10(%rsp), %rax movq 0xd8(%rax), %rdi movq 0x18(%rsp), %rsi movq (%rdi), %rax callq *0x28(%rax) movq 0x10(%rsp), %rax movq 0xd8(%rax), %rax movq %rax, 0x8(%rsp) cmpq $0x0, %r...
/Tencent[P]ncnn/build_O0/src/layer/x86/matmul_x86_fma.cpp
ncnn::MatMul_x86_avx::create_pipeline(ncnn::Option const&)
int MatMul_x86_avx::create_pipeline(const Option& opt) { gemm = ncnn::create_layer_cpu(ncnn::LayerType::Gemm); ncnn::ParamDict pd; pd.set(2, 0); // transA pd.set(3, transB); // transB pd.set(4, 0); // constantA pd.set(5, 0); // constantB pd.set(6, 1); // constantC pd...
subq $0x58, %rsp movq %rdi, 0x50(%rsp) movq %rsi, 0x48(%rsp) movq 0x50(%rsp), %rax movq %rax, 0x8(%rsp) movl $0x4a, %edi callq 0x89670 movq %rax, %rcx movq 0x8(%rsp), %rax movq %rcx, 0xd8(%rax) leaq 0x38(%rsp), %rdi movq %rdi, 0x10(%rsp) callq 0x78060 movq 0x10(%rsp), %rdi movl $0x2, %esi xorl %edx, %edx callq 0x79040 ...
/Tencent[P]ncnn/build_O0/src/layer/x86/matmul_x86_avx.cpp
ncnn::Deconvolution1D::Deconvolution1D()
Deconvolution1D::Deconvolution1D() { one_blob_only = true; support_inplace = false; }
subq $0x98, %rsp movq %rdi, 0x38(%rsp) movq 0x38(%rsp), %rdi movq %rdi, 0x18(%rsp) callq 0x88680 movq 0x18(%rsp), %rax leaq 0x291f51(%rip), %rcx # 0x1f25ee8 addq $0x10, %rcx movq %rcx, (%rax) addq $0x100, %rax # imm = 0x100 movq %rax, 0x40(%rsp) movq 0x40(%rsp), %rax movq %rax, 0x20(%rsp) movq $0x0, (%rax...
/Tencent[P]ncnn/src/layer/deconvolution1d.cpp
ncnn::Deconvolution1D::load_param(ncnn::ParamDict const&)
int Deconvolution1D::load_param(const ParamDict& pd) { num_output = pd.get(0, 0); kernel_w = pd.get(1, 0); dilation_w = pd.get(2, 1); stride_w = pd.get(3, 1); pad_left = pd.get(4, 0); pad_right = pd.get(15, pad_left); output_pad_right = pd.get(18, 0); output_w = pd.get(20, 0); bias_t...
subq $0x1b8, %rsp # imm = 0x1B8 movq %rdi, 0xf0(%rsp) movq %rsi, 0xe8(%rsp) movq 0xf0(%rsp), %rax movq %rax, 0x38(%rsp) movq 0xe8(%rsp), %rdi xorl %edx, %edx movl %edx, 0x34(%rsp) movl %edx, %esi callq 0x78d90 movl 0x34(%rsp), %edx movl %eax, %ecx movq 0x38(%rsp), %rax movl %ecx, 0xd0(%rax) movq 0xe8(%rsp), ...
/Tencent[P]ncnn/src/layer/deconvolution1d.cpp
ncnn::Deconvolution1D::load_model(ncnn::ModelBin const&)
int Deconvolution1D::load_model(const ModelBin& mb) { if (dynamic_weight) return 0; weight_data = mb.load(weight_data_size, 0); if (weight_data.empty()) return -100; if (bias_term) { bias_data = mb.load(num_output, 1); if (bias_data.empty()) return -100;...
subq $0x238, %rsp # imm = 0x238 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq 0x118(%rsp), %rax movq %rax, 0x68(%rsp) cmpl $0x0, 0x148(%rax) je 0x1c94a3d movl $0x0, 0x124(%rsp) jmp 0x1c95444 movq 0x68(%rsp), %rax movq 0x110(%rsp), %rsi movl 0xf4(%rax), %edx movq (%rsi), %rax leaq 0xc8(%rsp), %rdi xorl %...
/Tencent[P]ncnn/src/layer/deconvolution1d.cpp
ncnn::DeconvolutionDepthWise1D::DeconvolutionDepthWise1D()
DeconvolutionDepthWise1D::DeconvolutionDepthWise1D() { one_blob_only = true; support_inplace = false; }
subq $0x98, %rsp movq %rdi, 0x38(%rsp) movq 0x38(%rsp), %rdi movq %rdi, 0x18(%rsp) callq 0x88680 movq 0x18(%rsp), %rax leaq 0x28dbd9(%rip), %rcx # 0x1f25f60 addq $0x10, %rcx movq %rcx, (%rax) addq $0x100, %rax # imm = 0x100 movq %rax, 0x40(%rsp) movq 0x40(%rsp), %rax movq %rax, 0x20(%rsp) movq $0x0, (%rax...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise1d.cpp
ncnn::DeconvolutionDepthWise1D::load_model(ncnn::ModelBin const&)
int DeconvolutionDepthWise1D::load_model(const ModelBin& mb) { if (dynamic_weight) return 0; weight_data = mb.load(weight_data_size, 0); if (weight_data.empty()) return -100; if (bias_term) { bias_data = mb.load(num_output, 1); if (bias_data.empty()) ret...
subq $0x238, %rsp # imm = 0x238 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq 0x118(%rsp), %rax movq %rax, 0x68(%rsp) cmpl $0x0, 0x148(%rax) je 0x1c98e4d movl $0x0, 0x124(%rsp) jmp 0x1c99854 movq 0x68(%rsp), %rax movq 0x110(%rsp), %rsi movl 0xf4(%rax), %edx movq (%rsi), %rax leaq 0xc8(%rsp), %rdi xorl %...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise1d.cpp
ncnn::DeconvolutionDepthWise1D::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int DeconvolutionDepthWise1D::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; size_t elemsize = bottom_blob.elemsize; const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1; int outw = (w - 1) * stride_w + kernel_extent_w + output_pad_right; M...
pushq %r14 pushq %rbx subq $0x1c8, %rsp # imm = 0x1C8 movq %rdi, 0x118(%rsp) movq %rsi, 0x110(%rsp) movq %rdx, 0x108(%rsp) movq %rcx, 0x100(%rsp) movq 0x118(%rsp), %rax movq %rax, 0x78(%rsp) movq 0x110(%rsp), %rcx movl 0x2c(%rcx), %ecx movl %ecx, 0xfc(%rsp) movq 0x110(%rsp), %rcx movq 0x10(%rcx), %rcx movq %...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise1d.cpp
ncnn::deconvolutiondepthwise1d(ncnn::Mat const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static int deconvolutiondepthwise1d(const Mat& bottom_blob, Mat& top_blob, const Mat& weight_data, const Mat& bias_data, int kernel_w, int stride_w, int dilation_w, int group, int activation_type, const Mat& activation_params, const Option& opt) { const int w = bottom_blob.w; const int h = bottom_blob.h; c...
subq $0x518, %rsp # imm = 0x518 movq 0x540(%rsp), %rax movq 0x538(%rsp), %rax movl 0x530(%rsp), %eax movl 0x528(%rsp), %eax movl 0x520(%rsp), %eax movq %rdi, 0x220(%rsp) movq %rsi, 0x218(%rsp) movq %rdx, 0x210(%rsp) movq %rcx, 0x208(%rsp) movl %r8d, 0x204(%rsp) movl %r9d, 0x200(%rsp) movq 0x220(%rsp), %rax m...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise1d.cpp
ncnn::deconvolution3d(ncnn::Mat const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static int deconvolution3d(const Mat& bottom_blob, Mat& top_blob, const Mat& weight_data, const Mat& bias_data, int kernel_w, int kernel_h, int kernel_d, int stride_w, int stride_h, int stride_d, int dilation_w, int dilation_h, int dilation_d, int activation_type, const Mat& activation_params, const Option& opt) { ...
subq $0x688, %rsp # imm = 0x688 movq 0x6d8(%rsp), %rax movq 0x6d0(%rsp), %rax movl 0x6c8(%rsp), %eax movl 0x6c0(%rsp), %eax movl 0x6b8(%rsp), %eax movl 0x6b0(%rsp), %eax movl 0x6a8(%rsp), %eax movl 0x6a0(%rsp), %eax movl 0x698(%rsp), %eax movl 0x690(%rsp), %eax movq %rdi, 0x2f0(%rsp) movq %rsi, 0x2e8(%rsp) m...
/Tencent[P]ncnn/src/layer/deconvolution3d.cpp
ncnn::DeconvolutionDepthWise3D::DeconvolutionDepthWise3D()
DeconvolutionDepthWise3D::DeconvolutionDepthWise3D() { one_blob_only = true; support_inplace = false; }
subq $0x98, %rsp movq %rdi, 0x38(%rsp) movq 0x38(%rsp), %rdi movq %rdi, 0x18(%rsp) callq 0x88680 movq 0x18(%rsp), %rax leaq 0x284a89(%rip), %rcx # 0x1f26050 addq $0x10, %rcx movq %rcx, (%rax) addq $0x138, %rax # imm = 0x138 movq %rax, 0x40(%rsp) movq 0x40(%rsp), %rax movq %rax, 0x20(%rsp) movq $0x0, (%rax...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise3d.cpp
ncnn::DeconvolutionDepthWise3D::load_param(ncnn::ParamDict const&)
int DeconvolutionDepthWise3D::load_param(const ParamDict& pd) { num_output = pd.get(0, 0); kernel_w = pd.get(1, 0); kernel_h = pd.get(11, kernel_w); kernel_d = pd.get(21, kernel_w); dilation_w = pd.get(2, 1); dilation_h = pd.get(12, dilation_w); dilation_d = pd.get(22, dilation_w); strid...
subq $0x1b8, %rsp # imm = 0x1B8 movq %rdi, 0xf0(%rsp) movq %rsi, 0xe8(%rsp) movq 0xf0(%rsp), %rax movq %rax, 0x38(%rsp) movq 0xe8(%rsp), %rdi xorl %edx, %edx movl %edx, 0x34(%rsp) movl %edx, %esi callq 0x78d90 movl 0x34(%rsp), %edx movl %eax, %ecx movq 0x38(%rsp), %rax movl %ecx, 0xd0(%rax) movq 0xe8(%rsp), ...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise3d.cpp
ncnn::deconvolutiondepthwise3d(ncnn::Mat const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, int, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static int deconvolutiondepthwise3d(const Mat& bottom_blob, Mat& top_blob, const Mat& weight_data, const Mat& bias_data, int kernel_w, int kernel_h, int kernel_d, int stride_w, int stride_h, int stride_d, int dilation_w, int dilation_h, int dilation_d, int group, int activation_type, const Mat& activation_params, const...
subq $0xb98, %rsp # imm = 0xB98 movq 0xbf0(%rsp), %rax movq 0xbe8(%rsp), %rax movl 0xbe0(%rsp), %eax movl 0xbd8(%rsp), %eax movl 0xbd0(%rsp), %eax movl 0xbc8(%rsp), %eax movl 0xbc0(%rsp), %eax movl 0xbb8(%rsp), %eax movl 0xbb0(%rsp), %eax movl 0xba8(%rsp), %eax movl 0xba0(%rsp), %eax movq %rdi, 0x500(%rsp) m...
/Tencent[P]ncnn/src/layer/deconvolutiondepthwise3d.cpp
ncnn::Einsum::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int Einsum::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { // assert bottom_blobs.size() == lhs_tokens.size() // assert top_blobs.size() == 1 size_t elemsize = bottom_blobs[0].elemsize; if (lhs_tokens.empty() && rhs_token == "ii") { //...
subq $0x6d8, %rsp # imm = 0x6D8 movq %rdi, 0x410(%rsp) movq %rsi, 0x408(%rsp) movq %rdx, 0x400(%rsp) movq %rcx, 0x3f8(%rsp) movq 0x410(%rsp), %rax movq %rax, 0x210(%rsp) movq 0x408(%rsp), %rdi xorl %eax, %eax movl %eax, %esi callq 0x89920 movq 0x210(%rsp), %rdi movq 0x10(%rax), %rax movq %rax, 0x3f0(%rsp) ad...
/Tencent[P]ncnn/src/layer/einsum.cpp
ncnn::sum_dim(std::vector<int, std::allocator<int>> const&, int, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>>> const&, s...
static float sum_dim(const std::vector<int>& dim_sizes, int d, const std::vector<Mat>& bottom_blobs, const std::vector<std::string>& tokens, std::vector<int>& indexes) { if (d == (int)dim_sizes.size()) { float v = 1.f; for (size_t b = 0; b < bottom_blobs.size(); b++) { v *= g...
subq $0x68, %rsp movq %rdi, 0x58(%rsp) movl %esi, 0x54(%rsp) movq %rdx, 0x48(%rsp) movq %rcx, 0x40(%rsp) movq %r8, 0x38(%rsp) movl 0x54(%rsp), %eax movl %eax, 0x1c(%rsp) movq 0x58(%rsp), %rdi callq 0x2fb30 movq %rax, %rcx movl 0x1c(%rsp), %eax cmpl %ecx, %eax jne 0x1ca9807 movss 0x1ee8af(%rip), %xmm0 # 0x1e98024 movs...
/Tencent[P]ncnn/src/layer/einsum.cpp
ncnn::DeformableConv2D::DeformableConv2D()
DeformableConv2D::DeformableConv2D() { one_blob_only = false; support_inplace = false; }
subq $0x98, %rsp movq %rdi, 0x38(%rsp) movq 0x38(%rsp), %rdi movq %rdi, 0x18(%rsp) callq 0x88680 movq 0x18(%rsp), %rax leaq 0x27aad9(%rip), %rcx # 0x1f26140 addq $0x10, %rcx movq %rcx, (%rax) addq $0x108, %rax # imm = 0x108 movq %rax, 0x40(%rsp) movq 0x40(%rsp), %rax movq %rax, 0x20(%rsp) movq $0x0, (%rax...
/Tencent[P]ncnn/src/layer/deformableconv2d.cpp
ncnn::DeformableConv2D::load_param(ncnn::ParamDict const&)
int DeformableConv2D::load_param(const ParamDict& pd) { num_output = pd.get(0, 0); kernel_w = pd.get(1, 0); kernel_h = pd.get(11, kernel_w); dilation_w = pd.get(2, 1); dilation_h = pd.get(12, dilation_w); stride_w = pd.get(3, 1); stride_h = pd.get(13, stride_w); pad_left = pd.get(4, 0); ...
subq $0x1b8, %rsp # imm = 0x1B8 movq %rdi, 0xf0(%rsp) movq %rsi, 0xe8(%rsp) movq 0xf0(%rsp), %rax movq %rax, 0x38(%rsp) movq 0xe8(%rsp), %rdi xorl %edx, %edx movl %edx, 0x34(%rsp) movl %edx, %esi callq 0x78d90 movl 0x34(%rsp), %edx movl %eax, %ecx movq 0x38(%rsp), %rax movl %ecx, 0xd0(%rax) movq 0xe8(%rsp), ...
/Tencent[P]ncnn/src/layer/deformableconv2d.cpp
ncnn::deformableconv2d_pack1to4_sse(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static void deformableconv2d_pack1to4_sse(const std::vector<Mat>& bottom_blobs, Mat& top_blob, const Mat& weight_data_packed, const Mat& bias_data, int kernel_w, int kernel_h, int dilation_w, int dilation_h, int stride_w, int stride_h, int pad_left, int pad_top, int activation_type, const Mat& activation_params, const ...
pushq %r14 pushq %rbx subq $0x2c98, %rsp # imm = 0x2C98 movq 0x2cf0(%rsp), %rax movq 0x2ce8(%rsp), %rax movl 0x2ce0(%rsp), %eax movl 0x2cd8(%rsp), %eax movl 0x2cd0(%rsp), %eax movl 0x2cc8(%rsp), %eax movl 0x2cc0(%rsp), %eax movl 0x2cb8(%rsp), %eax movl 0x2cb0(%rsp), %eax movq %rdi, 0x6c0(%rsp) movq %rsi, 0x6b...
/Tencent[P]ncnn/src/layer/x86/deformableconv2d_pack1to4.h
ncnn::deformableconv2d_transform_kernel_packed_sse(ncnn::Mat const&, ncnn::Mat&, int, int, int, int, int, int)
static void deformableconv2d_transform_kernel_packed_sse(const Mat& weight_data, Mat& weight_data_tm, int num_input, int num_output, int kernel_w, int kernel_h, int elempack, int out_elempack) { const int maxk = kernel_w * kernel_h; // src = kw-kh-inch-outch // dst = pb-pa-inch/pa-kw-kh-outch/pb { ...
pushq %rbp pushq %rbx subq $0x198, %rsp # imm = 0x198 movl 0x1b8(%rsp), %eax movl 0x1b0(%rsp), %eax movq %rdi, 0x108(%rsp) movq %rsi, 0x100(%rsp) movl %edx, 0xfc(%rsp) movl %ecx, 0xf8(%rsp) movl %r8d, 0xf4(%rsp) movl %r9d, 0xf0(%rsp) movl 0xf4(%rsp), %eax imull 0xf0(%rsp), %eax movl %eax, 0xec(%rsp) movq 0x1...
/Tencent[P]ncnn/build_O0/src/layer/x86/deformableconv2d_x86_avx512.cpp
ncnn::deformableconv2d_pack1to16_avx512(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static void deformableconv2d_pack1to16_avx512(const std::vector<Mat>& bottom_blobs, Mat& top_blob, const Mat& weight_data_packed, const Mat& bias_data, int kernel_w, int kernel_h, int dilation_w, int dilation_h, int stride_w, int stride_h, int pad_left, int pad_top, int activation_type, const Mat& activation_params, co...
pushq %rbp movq %rsp, %rbp andq $-0x40, %rsp subq $0x7000, %rsp # imm = 0x7000 movq 0x50(%rbp), %rax movq 0x48(%rbp), %rax movl 0x40(%rbp), %eax movl 0x38(%rbp), %eax movl 0x30(%rbp), %eax movl 0x28(%rbp), %eax movl 0x20(%rbp), %eax movl 0x18(%rbp), %eax movl 0x10(%rbp), %eax movq %rdi, 0xa60(%rsp) movq %rsi,...
/Tencent[P]ncnn/src/layer/x86/deformableconv2d_pack1to16.h