system
stringclasses
7 values
input
stringlengths
10
94.5k
output
stringlengths
14
55.5k
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h>
#include "hip/hip_runtime.h" #include <stdio.h>
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include "hip/hip_runtime.h" #include <stdio.h>
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include "hip/hip_runtime.h" #include <stdio.h>
.text .file "kernel.hip" .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00155c98_00000000-6_kernel.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2059: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2059: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2082: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2082: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "kernel.hip" .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <cuda.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != cudaSuccess) { \ printf("cuda error with %s in line %d\n",cudaGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { cudaError_t err; cudaEvent_t start_event,stop_event; CHECK_CUDA_ERR ( cudaEventCreate(&start_event) ); CHECK_CUDA_ERR ( cudaEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( cudaMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( cudaEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( cudaMemcpy(dBuf,buf,sizeof(float) * N, cudaMemcpyHostToDevice)); CHECK_CUDA_ERR ( cudaEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( cudaDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( cudaEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( cudaEventDestroy(start_event) ); CHECK_CUDA_ERR ( cudaEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( cudaFree(dBuf) ); return 0; }
code for sm_80
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <cuda.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != cudaSuccess) { \ printf("cuda error with %s in line %d\n",cudaGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { cudaError_t err; cudaEvent_t start_event,stop_event; CHECK_CUDA_ERR ( cudaEventCreate(&start_event) ); CHECK_CUDA_ERR ( cudaEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( cudaMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( cudaEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( cudaMemcpy(dBuf,buf,sizeof(float) * N, cudaMemcpyHostToDevice)); CHECK_CUDA_ERR ( cudaEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( cudaDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( cudaEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( cudaEventDestroy(start_event) ); CHECK_CUDA_ERR ( cudaEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( cudaFree(dBuf) ); return 0; }
.file "tmpxft_00173363_00000000-6_elapsed.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2073: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2073: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "cuda error with %s in line %d\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "%f m second cost\n" .text .globl main .type main, @function main: .LFB2070: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 subq $48, %rsp .cfi_def_cfa_offset 64 movq %fs:40, %rax movq %rax, 40(%rsp) xorl %eax, %eax leaq 16(%rsp), %rdi call cudaEventCreate@PLT testl %eax, %eax jne .L17 leaq 24(%rsp), %rdi call cudaEventCreate@PLT testl %eax, %eax jne .L18 movl $4194304, %edi call malloc@PLT movq %rax, %rbx leaq 32(%rsp), %rdi movl $4194304, %esi call cudaMalloc@PLT testl %eax, %eax jne .L19 movl $0, %esi movq 16(%rsp), %rdi call cudaEventRecord@PLT testl %eax, %eax jne .L20 movl $1, %ecx movl $4194304, %edx movq %rbx, %rsi movq 32(%rsp), %rdi call cudaMemcpy@PLT testl %eax, %eax jne .L21 movl $0, %esi movq 24(%rsp), %rdi call cudaEventRecord@PLT testl %eax, %eax jne .L22 call cudaDeviceSynchronize@PLT testl %eax, %eax jne .L23 movl $0x42c80000, 12(%rsp) leaq 12(%rsp), %rdi movq 24(%rsp), %rdx movq 16(%rsp), %rsi call cudaEventElapsedTime@PLT testl %eax, %eax jne .L24 pxor %xmm0, %xmm0 cvtss2sd 12(%rsp), %xmm0 leaq .LC2(%rip), %rsi movl $2, %edi movl $1, %eax call __printf_chk@PLT movq 16(%rsp), %rdi call cudaEventDestroy@PLT testl %eax, %eax jne .L25 movq 24(%rsp), %rdi call cudaEventDestroy@PLT testl %eax, %eax jne .L26 movq %rbx, %rdi call free@PLT movq 32(%rsp), %rdi call cudaFree@PLT testl %eax, %eax jne .L27 movq 40(%rsp), %rax subq %fs:40, %rax jne .L28 movl $0, %eax addq $48, %rsp .cfi_remember_state .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 ret .L17: .cfi_restore_state movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $23, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L18: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $24, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L19: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $32, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L20: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $34, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L21: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $35, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L22: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $36, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L23: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $37, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L24: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $40, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L25: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $46, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L26: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $47, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L27: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $49, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L28: call __stack_chk_fail@PLT .cfi_endproc .LFE2070: .size main, .-main .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2096: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2096: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <cuda.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != cudaSuccess) { \ printf("cuda error with %s in line %d\n",cudaGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { cudaError_t err; cudaEvent_t start_event,stop_event; CHECK_CUDA_ERR ( cudaEventCreate(&start_event) ); CHECK_CUDA_ERR ( cudaEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( cudaMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( cudaEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( cudaMemcpy(dBuf,buf,sizeof(float) * N, cudaMemcpyHostToDevice)); CHECK_CUDA_ERR ( cudaEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( cudaDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( cudaEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( cudaEventDestroy(start_event) ); CHECK_CUDA_ERR ( cudaEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( cudaFree(dBuf) ); return 0; }
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <hip/hip_runtime.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != hipSuccess) { \ printf("cuda error with %s in line %d\n",hipGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { hipError_t err; hipEvent_t start_event,stop_event; CHECK_CUDA_ERR ( hipEventCreate(&start_event) ); CHECK_CUDA_ERR ( hipEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( hipMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( hipEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( hipMemcpy(dBuf,buf,sizeof(float) * N, hipMemcpyHostToDevice)); CHECK_CUDA_ERR ( hipEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( hipDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( hipEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( hipEventDestroy(start_event) ); CHECK_CUDA_ERR ( hipEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( hipFree(dBuf) ); return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <hip/hip_runtime.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != hipSuccess) { \ printf("cuda error with %s in line %d\n",hipGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { hipError_t err; hipEvent_t start_event,stop_event; CHECK_CUDA_ERR ( hipEventCreate(&start_event) ); CHECK_CUDA_ERR ( hipEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( hipMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( hipEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( hipMemcpy(dBuf,buf,sizeof(float) * N, hipMemcpyHostToDevice)); CHECK_CUDA_ERR ( hipEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( hipDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( hipEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( hipEventDestroy(start_event) ); CHECK_CUDA_ERR ( hipEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( hipFree(dBuf) ); return 0; }
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <hip/hip_runtime.h> #define N 1024 * 1024 #define CHECK_CUDA_ERR(x) { \ err = x; \ if (err != hipSuccess) { \ printf("cuda error with %s in line %d\n",hipGetErrorString(err),__LINE__); \ exit(1); \ } } int main() { hipError_t err; hipEvent_t start_event,stop_event; CHECK_CUDA_ERR ( hipEventCreate(&start_event) ); CHECK_CUDA_ERR ( hipEventCreate(&stop_event) ); void * buf = malloc(sizeof(float) * N); void * dBuf; CHECK_CUDA_ERR ( hipMalloc(&dBuf,sizeof(float) * N) ); CHECK_CUDA_ERR ( hipEventRecord(start_event,0) ) ; CHECK_CUDA_ERR ( hipMemcpy(dBuf,buf,sizeof(float) * N, hipMemcpyHostToDevice)); CHECK_CUDA_ERR ( hipEventRecord(stop_event,0) ); CHECK_CUDA_ERR ( hipDeviceSynchronize() ); float ms = 100.f; CHECK_CUDA_ERR ( hipEventElapsedTime(&ms,start_event,stop_event)); printf("%f m second cost\n",ms); CHECK_CUDA_ERR ( hipEventDestroy(start_event) ); CHECK_CUDA_ERR ( hipEventDestroy(stop_event) ); free(buf); CHECK_CUDA_ERR ( hipFree(dBuf) ); return 0; }
.text .file "elapsed.hip" .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 leaq 16(%rsp), %rdi callq hipEventCreate testl %eax, %eax jne .LBB0_1 # %bb.3: leaq 8(%rsp), %rdi callq hipEventCreate testl %eax, %eax jne .LBB0_4 # %bb.5: movl $4194304, %edi # imm = 0x400000 callq malloc movq %rax, %rbx leaq 24(%rsp), %rdi movl $4194304, %esi # imm = 0x400000 callq hipMalloc testl %eax, %eax jne .LBB0_6 # %bb.7: movq 16(%rsp), %rdi xorl %esi, %esi callq hipEventRecord testl %eax, %eax jne .LBB0_8 # %bb.9: movq 24(%rsp), %rdi movl $4194304, %edx # imm = 0x400000 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy testl %eax, %eax jne .LBB0_10 # %bb.11: movq 8(%rsp), %rdi xorl %esi, %esi callq hipEventRecord testl %eax, %eax jne .LBB0_12 # %bb.13: callq hipDeviceSynchronize testl %eax, %eax jne .LBB0_14 # %bb.15: movl $1120403456, 4(%rsp) # imm = 0x42C80000 movq 16(%rsp), %rsi movq 8(%rsp), %rdx leaq 4(%rsp), %rdi callq hipEventElapsedTime testl %eax, %eax jne .LBB0_16 # %bb.17: movss 4(%rsp), %xmm0 # xmm0 = mem[0],zero,zero,zero cvtss2sd %xmm0, %xmm0 movl $.L.str.1, %edi movb $1, %al callq printf movq 16(%rsp), %rdi callq hipEventDestroy testl %eax, %eax jne .LBB0_18 # %bb.19: movq 8(%rsp), %rdi callq hipEventDestroy testl %eax, %eax jne .LBB0_20 # %bb.21: movq %rbx, %rdi callq free movq 24(%rsp), %rdi callq hipFree testl %eax, %eax jne .LBB0_22 # %bb.23: xorl %eax, %eax addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 retq .LBB0_1: .cfi_def_cfa_offset 48 movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $23, %edx jmp .LBB0_2 .LBB0_4: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $24, %edx jmp .LBB0_2 .LBB0_6: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $32, %edx jmp .LBB0_2 .LBB0_8: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $34, %edx jmp .LBB0_2 .LBB0_10: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $35, %edx jmp .LBB0_2 .LBB0_12: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $36, %edx jmp .LBB0_2 .LBB0_14: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $37, %edx jmp .LBB0_2 .LBB0_16: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $40, %edx jmp .LBB0_2 .LBB0_18: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $46, %edx jmp .LBB0_2 .LBB0_20: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $47, %edx jmp .LBB0_2 .LBB0_22: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $49, %edx .LBB0_2: xorl %eax, %eax callq printf movl $1, %edi callq exit .Lfunc_end0: .size main, .Lfunc_end0-main .cfi_endproc # -- End function .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "cuda error with %s in line %d\n" .size .L.str, 31 .type .L.str.1,@object # @.str.1 .L.str.1: .asciz "%f m second cost\n" .size .L.str.1, 18 .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00173363_00000000-6_elapsed.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2073: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2073: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "cuda error with %s in line %d\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "%f m second cost\n" .text .globl main .type main, @function main: .LFB2070: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 subq $48, %rsp .cfi_def_cfa_offset 64 movq %fs:40, %rax movq %rax, 40(%rsp) xorl %eax, %eax leaq 16(%rsp), %rdi call cudaEventCreate@PLT testl %eax, %eax jne .L17 leaq 24(%rsp), %rdi call cudaEventCreate@PLT testl %eax, %eax jne .L18 movl $4194304, %edi call malloc@PLT movq %rax, %rbx leaq 32(%rsp), %rdi movl $4194304, %esi call cudaMalloc@PLT testl %eax, %eax jne .L19 movl $0, %esi movq 16(%rsp), %rdi call cudaEventRecord@PLT testl %eax, %eax jne .L20 movl $1, %ecx movl $4194304, %edx movq %rbx, %rsi movq 32(%rsp), %rdi call cudaMemcpy@PLT testl %eax, %eax jne .L21 movl $0, %esi movq 24(%rsp), %rdi call cudaEventRecord@PLT testl %eax, %eax jne .L22 call cudaDeviceSynchronize@PLT testl %eax, %eax jne .L23 movl $0x42c80000, 12(%rsp) leaq 12(%rsp), %rdi movq 24(%rsp), %rdx movq 16(%rsp), %rsi call cudaEventElapsedTime@PLT testl %eax, %eax jne .L24 pxor %xmm0, %xmm0 cvtss2sd 12(%rsp), %xmm0 leaq .LC2(%rip), %rsi movl $2, %edi movl $1, %eax call __printf_chk@PLT movq 16(%rsp), %rdi call cudaEventDestroy@PLT testl %eax, %eax jne .L25 movq 24(%rsp), %rdi call cudaEventDestroy@PLT testl %eax, %eax jne .L26 movq %rbx, %rdi call free@PLT movq 32(%rsp), %rdi call cudaFree@PLT testl %eax, %eax jne .L27 movq 40(%rsp), %rax subq %fs:40, %rax jne .L28 movl $0, %eax addq $48, %rsp .cfi_remember_state .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 ret .L17: .cfi_restore_state movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $23, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L18: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $24, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L19: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $32, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L20: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $34, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L21: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $35, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L22: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $36, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L23: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $37, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L24: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $40, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L25: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $46, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L26: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $47, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L27: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %rdx movl $49, %ecx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $1, %edi call exit@PLT .L28: call __stack_chk_fail@PLT .cfi_endproc .LFE2070: .size main, .-main .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2096: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2096: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "elapsed.hip" .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 leaq 16(%rsp), %rdi callq hipEventCreate testl %eax, %eax jne .LBB0_1 # %bb.3: leaq 8(%rsp), %rdi callq hipEventCreate testl %eax, %eax jne .LBB0_4 # %bb.5: movl $4194304, %edi # imm = 0x400000 callq malloc movq %rax, %rbx leaq 24(%rsp), %rdi movl $4194304, %esi # imm = 0x400000 callq hipMalloc testl %eax, %eax jne .LBB0_6 # %bb.7: movq 16(%rsp), %rdi xorl %esi, %esi callq hipEventRecord testl %eax, %eax jne .LBB0_8 # %bb.9: movq 24(%rsp), %rdi movl $4194304, %edx # imm = 0x400000 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy testl %eax, %eax jne .LBB0_10 # %bb.11: movq 8(%rsp), %rdi xorl %esi, %esi callq hipEventRecord testl %eax, %eax jne .LBB0_12 # %bb.13: callq hipDeviceSynchronize testl %eax, %eax jne .LBB0_14 # %bb.15: movl $1120403456, 4(%rsp) # imm = 0x42C80000 movq 16(%rsp), %rsi movq 8(%rsp), %rdx leaq 4(%rsp), %rdi callq hipEventElapsedTime testl %eax, %eax jne .LBB0_16 # %bb.17: movss 4(%rsp), %xmm0 # xmm0 = mem[0],zero,zero,zero cvtss2sd %xmm0, %xmm0 movl $.L.str.1, %edi movb $1, %al callq printf movq 16(%rsp), %rdi callq hipEventDestroy testl %eax, %eax jne .LBB0_18 # %bb.19: movq 8(%rsp), %rdi callq hipEventDestroy testl %eax, %eax jne .LBB0_20 # %bb.21: movq %rbx, %rdi callq free movq 24(%rsp), %rdi callq hipFree testl %eax, %eax jne .LBB0_22 # %bb.23: xorl %eax, %eax addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 retq .LBB0_1: .cfi_def_cfa_offset 48 movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $23, %edx jmp .LBB0_2 .LBB0_4: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $24, %edx jmp .LBB0_2 .LBB0_6: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $32, %edx jmp .LBB0_2 .LBB0_8: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $34, %edx jmp .LBB0_2 .LBB0_10: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $35, %edx jmp .LBB0_2 .LBB0_12: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $36, %edx jmp .LBB0_2 .LBB0_14: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $37, %edx jmp .LBB0_2 .LBB0_16: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $40, %edx jmp .LBB0_2 .LBB0_18: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $46, %edx jmp .LBB0_2 .LBB0_20: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $47, %edx jmp .LBB0_2 .LBB0_22: movl %eax, %edi callq hipGetErrorString movl $.L.str, %edi movq %rax, %rsi movl $49, %edx .LBB0_2: xorl %eax, %eax callq printf movl $1, %edi callq exit .Lfunc_end0: .size main, .Lfunc_end0-main .cfi_endproc # -- End function .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "cuda error with %s in line %d\n" .size .L.str, 31 .type .L.str.1,@object # @.str.1 .L.str.1: .asciz "%f m second cost\n" .size .L.str.1, 18 .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
//#include "shallow.h" #include <iostream> #include <time.h> #include <stdlib.h> #define RELU(a) ((a > 0) ? a : 0) #define KERNEL_SIZE(n_C) ((n_C * 2 > 8) ? 8 : n_C * 2) struct layer_param { int M; //Ilosc tablic podawanych na wejscie int pad; //Grubosc warstwy zer na krawedziach (zero-padding) int F; //Rozmiar 2D filtra (F x F) int F_MP; //Rozmiar 2D filtra do max poolingu (F_MP x F_MP) int stride; //Ilosc przeskakiwanych pikseli przy konwolucji na inkrementacje int stride_MP; //To samo, tylko przy max poolingu int n_Hprev; //Wysokosc tablicy wejsciowej podawanej na wejscie sieci int n_Wprev; //Szerokosc tablicy wejsciowej podawanej na wejscie sieci int n_Cprev; //Glebokosc tablicy wejsciowej, jednoczesnie musi to byc glebokosc filtra (F x F x C) int n_H; //Wysokosc tablicy uzyskanej po konwolucji kernela z wejsciem int n_W; int n_C; //Ilosc filtrow, jednoczesnie glebokosc wyjscia warstwy int n_Hout; //Wysokosc tablicy wyjsciowej warstwy int n_Wout; double alpha; //Predkosc uczenia }; struct cache_data { double** IN; //Tablica wejsciowa double** Z; //Wynik splotu double** A; //Wynik Aktywacji double** OUT; //Poprzedni wynik po max poolingu, jednoczescie wyjscie warstwy sieci double** kernel; //Filtr double** dW; //Gradient kosztu wzgledem kerneli double** dA; //Gradient kosztu wzgledem warstwy double** dAprev; //Gradient kosztu wzgledem wyjscia warstwy n_l - 1 double** dZ; //Gradient kosztu wzgledem wyniku konwolucji }; void set_random_IN(layer_param, double**&); //Ustawia losowe wejscie (do testowania) void set_new_IN(double**&, double**&, layer_param l); void show_results(layer_param, cache_data&); //Wyswietla zawartosc koncowych i posrednich wynikow w warstwie void brief_inf(layer_param, double**); //Krotka informacja o wyjsciu sieci void forw_prop(layer_param, cache_data&); //Najwazniejsza funkcja (konwolucja, aktywacja, maxpooling) void simple_del(double**&, int); //Usuwanie pamieci void update_param(layer_param&, layer_param&); //Ustawianie nowych parametrow warstwy void prep_new_arrays(layer_param, cache_data&); //Tworzenie nowych tablic wynikowych void prep_gradients(layer_param, cache_data&); //Tworzenie gradientow (narazie losowo, bez funkcji kosztu) void show_gradients(layer_param, cache_data&); void back_prop(layer_param, cache_data&); int main() { srand(time(NULL)); int number_of_layers = 2; layer_param* l = new layer_param[number_of_layers]; cache_data* cache = new cache_data[number_of_layers]; int n_l = 0; layer_param l_prev; l[n_l].M = 1; l[n_l].pad = 0; l[n_l].F = 3; l[n_l].F_MP = 2; l[n_l].stride = 1; l[n_l].stride_MP = 2; l[n_l].alpha = 0.1; int IN_size = 16; //Rzeczywisty rozmiar wejscia int IN_depth = 1; //Rzeczywista glebokosc wejscia l[n_l].n_Hprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Wprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Cprev = IN_depth; l[n_l].n_H = int((l[n_l].n_Hprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_W = int((l[n_l].n_Wprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_C = 1; l[n_l].n_Hout = int((l[n_l].n_H - l[n_l].F_MP) / l[n_l].stride_MP) + 1; l[n_l].n_Wout = int((l[n_l].n_W - l[n_l].F_MP) / l[n_l].stride_MP) + 1; for (n_l = 0; n_l < number_of_layers; n_l++) { std::cout << "\n\n#### WARSTWA: " << n_l + 1 << "#### \n"; if (n_l == 0) set_random_IN(l[n_l], cache[n_l].IN); else { l_prev = l[n_l - 1]; update_param(l_prev, l[n_l]); set_new_IN(cache[n_l].IN, cache[n_l - 1].OUT, l[n_l]); } prep_new_arrays(l[n_l], cache[n_l]); forw_prop(l[n_l], cache[n_l]); prep_gradients(l[n_l], cache[n_l]); if (l[n_l].n_H < 25) show_results(l[n_l], cache[n_l]); else brief_inf(l[n_l], cache[n_l].OUT); //back_prop(l[n_l], cache[n_l]); //show_gradients(l[n_l], cache[n_l]); //Funkcja wyswietla gradient } return 0; } void set_random_IN(layer_param l, double**& IN) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { //Dla kazdego badanego przypadku (np. pojedynczej mapy bajtowej- spektogram) for (int h = 0; h < l.n_Hprev; h++) { //Przejdz po kazdym wierszu for (int w = 0; w < l.n_Wprev; w++) { //Przejdz po kazdej kolumnie for (int c = 0; c < l.n_Cprev; c++) { //Przejdz po kazdym kanale (np. dla wejscia w postaci zdjecia rgb - 3 kanaly) if (h < l.pad || h > l.n_Hprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; //Ustawianie zer dla zero paddingu else if (w < l.pad || w > l.n_Wprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; else IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = (rand() % 10 + 1)/10.; //W tablicy wejsciowej beda same wartosci int } } } } } void set_new_IN(double**& IN, double**& OUT, layer_param l) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = OUT[m][w + l.n_Wprev * (h + l.n_Hprev * c)]; } } } } } void update_param(layer_param& l_prev, layer_param& l) { l.M = l_prev.M; l.pad = l_prev.pad; l.F = l_prev.F; l.F_MP = l_prev.F_MP; l.stride = l_prev.stride; l.stride_MP = l_prev.stride_MP; l.alpha = l_prev.alpha; l.n_Hprev = l_prev.n_Hout; l.n_Wprev = l_prev.n_Wout; l.n_Cprev = l_prev.n_C; l.n_H = int((l.n_Hprev - l.F) / l.stride) + 1; l.n_W = int((l.n_Wprev - l.F) / l.stride) + 1; l.n_C = KERNEL_SIZE(l_prev.n_C); l.n_Hout = int((l.n_H - l.F_MP) / l.stride_MP) + 1; l.n_Wout = int((l.n_W - l.F_MP) / l.stride_MP) + 1; } void prep_new_arrays(layer_param l, cache_data& cache) { cache.Z = new double* [l.M]; cache.A = new double* [l.M]; cache.OUT = new double* [l.M]; cache.kernel = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.Z[i] = new double[l.n_C * l.n_H * l.n_W]; cache.A[i] = new double[l.n_C * l.n_H * l.n_W]; cache.OUT[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.n_C; i++) { cache.kernel[i] = new double[l.n_Cprev * l.F * l.F]; } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.kernel[c][w + l.F * (h + l.F * d)] = (rand()%10000 - 5000) * 0.0001; //Ustawianie losowych wag filtra } } } } } void prep_gradients(layer_param l, cache_data& cache) { cache.dZ = new double* [l.M]; cache.dA = new double* [l.M]; cache.dAprev = new double* [l.M]; cache.dW = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.dZ[i] = new double[l.n_C * l.n_H * l.n_W]; } for (int i = 0; i < l.M; i++) { cache.dA[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.M; i++) { cache.dAprev[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int i = 0; i < l.n_C; i++) { cache.dW[i] = new double[l.n_Cprev * l.F * l.F]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] = (rand()%10000 - 5000) * 0.0001; } } } } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; } } } } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.dW[c][w + l.F * (h + l.F * d)] = 0; } } } } double maximum; int vert_start, vert_end; int horiz_start, horiz_end; for (int m = 0; m < l.M; m++) { //Dla kazdego przypadku for (int h = 0; h < l.n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < l.n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < l.n_C; c++) { //Dla kazdego kanalu wyjscia vert_start = h * l.stride_MP; vert_end = vert_start + l.F_MP; horiz_start = w * l.stride_MP; horiz_end = horiz_start + l.F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + l.n_W * (j + l.n_H * c)] > maximum) maximum = cache.A[m][k + l.n_W * (j + l.n_H * c)]; } } for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.A[m][k + l.n_W * (j + l.n_H * c)] != maximum || maximum == 0) cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = 0; else cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)]; } } } } } } for (int m = 0; m < l.M; m++) { for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { for (int c = 0; c < l.n_C; c++) { vert_start = h; vert_end = vert_start + l.F; horiz_start = w; horiz_end = horiz_start + l.F; for (int d = 0; d < l.n_Cprev; d++) { for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.dZ[m][w + l.n_W * (h + l.n_H * c)] < 0) cache.dZ[m][w + l.n_W * (h + l.n_H * c)] = 0; cache.dAprev[m][j + l.n_Wprev * (k + l.n_Hprev * d)] += cache.kernel[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; cache.dW[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] += cache.IN[m][j + l.n_Wprev * (k + l.n_Hprev * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; } } } } } } } } void brief_inf(layer_param l, double** OUT) { for (int m = 0; m < l.M; m++) std::cout << "Wyjscie: " << m + 1 << " Kanaly: " << l.n_C << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; } void forw_prop(layer_param l, cache_data& cache) { int M = l.M; int pad = l.pad; int F = l.F; int F_MP = l.F_MP; int stride = l.stride; int stride_MP = l.stride_MP; int n_Hprev = l.n_Hprev; int n_Wprev = l.n_Wprev; int n_Cprev = l.n_Cprev; int n_H = l.n_H; int n_W = l.n_W; int n_C = l.n_C; int n_Hout = l.n_Hout; int n_Wout = l.n_Wout; int vert_start = 0; int vert_end = 0; int horiz_start = 0; int horiz_end = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_H; h++) { //Dla kazdego wiersza for (int w = 0; w < n_W; w++) { //Dla kazdej kolumny for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu (kanalow bedzie tyle, ile chcemy kerneli) vert_start = h * stride; //Poczatek wycinka w pionie vert_end = vert_start + F; //Koniec wycinka w pionie horiz_start = w * stride; //Poczatek wycika w poziomie horiz_end = horiz_start + F; //Koniec wycinka w poziomie cache.Z[m][w + n_W * (h + n_H * c)] = 0; for (int d = 0; d < n_Cprev; d++) { //Dla kazdego kanalu w tablicy wejsciowej for (int j = vert_start; j < vert_end; j++) { //Dla wybranych wierszy for (int k = horiz_start; k < horiz_end; k++) { //Dla wybranych kolumn cache.Z[m][w + n_W * (h + n_H * c)] += cache.kernel[c][(k - horiz_start) + F * ((j - vert_start) + F * d)] * cache.IN[m][k + n_Wprev * (j + n_Hprev * d)]; //Pomnoz wartosc/piksel wycinka przez wage kernela i dodaj do wyniku konwolucji } } } cache.A[m][w + n_W * (h + n_H * c)] = RELU(cache.Z[m][w + n_W * (h + n_H * c)]); //Aktywowanie danej wartosci/neuronu } } } } double maximum = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu wyjscia if (n_Hout > 1) { vert_start = h * stride_MP; vert_end = vert_start + F_MP; horiz_start = w * stride_MP; horiz_end = horiz_start + F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + n_W * (j + n_H * c)] > maximum) maximum = cache.A[m][k + n_W * (j + n_H * c)]; //Wybierz maksimum z wycinka } } cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = maximum; } else cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = cache.A[m][0 + n_W * (0 + n_H * c)]; } } } } } void show_gradients(layer_param l, cache_data& cache) { for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "dAprev: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dW #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "dW: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dZ #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dZ: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.dZ[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### dA #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dA: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void simple_del(double**& arr, int len) { for (int i = 0; i < len; i++) { delete[] arr[i]; } delete[] arr; } void show_results(layer_param l, cache_data& cache) { std::cout.precision(4); for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "Wejscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### FILTRY #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Kernel: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### WYNIKI KONWOLUCJI #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Z: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.Z[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI AKTYWACJI (RELU) #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "A: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.A[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI MAX POOLINGU #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Wyjscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.OUT[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void back_prop(layer_param l, cache_data& cache) { std::cout << "#### FILTRY PO PO PROPAGACJI WSTECZNEJ #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Wagi: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] - l.alpha * cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } }
code for sm_80
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
//#include "shallow.h" #include <iostream> #include <time.h> #include <stdlib.h> #define RELU(a) ((a > 0) ? a : 0) #define KERNEL_SIZE(n_C) ((n_C * 2 > 8) ? 8 : n_C * 2) struct layer_param { int M; //Ilosc tablic podawanych na wejscie int pad; //Grubosc warstwy zer na krawedziach (zero-padding) int F; //Rozmiar 2D filtra (F x F) int F_MP; //Rozmiar 2D filtra do max poolingu (F_MP x F_MP) int stride; //Ilosc przeskakiwanych pikseli przy konwolucji na inkrementacje int stride_MP; //To samo, tylko przy max poolingu int n_Hprev; //Wysokosc tablicy wejsciowej podawanej na wejscie sieci int n_Wprev; //Szerokosc tablicy wejsciowej podawanej na wejscie sieci int n_Cprev; //Glebokosc tablicy wejsciowej, jednoczesnie musi to byc glebokosc filtra (F x F x C) int n_H; //Wysokosc tablicy uzyskanej po konwolucji kernela z wejsciem int n_W; int n_C; //Ilosc filtrow, jednoczesnie glebokosc wyjscia warstwy int n_Hout; //Wysokosc tablicy wyjsciowej warstwy int n_Wout; double alpha; //Predkosc uczenia }; struct cache_data { double** IN; //Tablica wejsciowa double** Z; //Wynik splotu double** A; //Wynik Aktywacji double** OUT; //Poprzedni wynik po max poolingu, jednoczescie wyjscie warstwy sieci double** kernel; //Filtr double** dW; //Gradient kosztu wzgledem kerneli double** dA; //Gradient kosztu wzgledem warstwy double** dAprev; //Gradient kosztu wzgledem wyjscia warstwy n_l - 1 double** dZ; //Gradient kosztu wzgledem wyniku konwolucji }; void set_random_IN(layer_param, double**&); //Ustawia losowe wejscie (do testowania) void set_new_IN(double**&, double**&, layer_param l); void show_results(layer_param, cache_data&); //Wyswietla zawartosc koncowych i posrednich wynikow w warstwie void brief_inf(layer_param, double**); //Krotka informacja o wyjsciu sieci void forw_prop(layer_param, cache_data&); //Najwazniejsza funkcja (konwolucja, aktywacja, maxpooling) void simple_del(double**&, int); //Usuwanie pamieci void update_param(layer_param&, layer_param&); //Ustawianie nowych parametrow warstwy void prep_new_arrays(layer_param, cache_data&); //Tworzenie nowych tablic wynikowych void prep_gradients(layer_param, cache_data&); //Tworzenie gradientow (narazie losowo, bez funkcji kosztu) void show_gradients(layer_param, cache_data&); void back_prop(layer_param, cache_data&); int main() { srand(time(NULL)); int number_of_layers = 2; layer_param* l = new layer_param[number_of_layers]; cache_data* cache = new cache_data[number_of_layers]; int n_l = 0; layer_param l_prev; l[n_l].M = 1; l[n_l].pad = 0; l[n_l].F = 3; l[n_l].F_MP = 2; l[n_l].stride = 1; l[n_l].stride_MP = 2; l[n_l].alpha = 0.1; int IN_size = 16; //Rzeczywisty rozmiar wejscia int IN_depth = 1; //Rzeczywista glebokosc wejscia l[n_l].n_Hprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Wprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Cprev = IN_depth; l[n_l].n_H = int((l[n_l].n_Hprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_W = int((l[n_l].n_Wprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_C = 1; l[n_l].n_Hout = int((l[n_l].n_H - l[n_l].F_MP) / l[n_l].stride_MP) + 1; l[n_l].n_Wout = int((l[n_l].n_W - l[n_l].F_MP) / l[n_l].stride_MP) + 1; for (n_l = 0; n_l < number_of_layers; n_l++) { std::cout << "\n\n#### WARSTWA: " << n_l + 1 << "#### \n"; if (n_l == 0) set_random_IN(l[n_l], cache[n_l].IN); else { l_prev = l[n_l - 1]; update_param(l_prev, l[n_l]); set_new_IN(cache[n_l].IN, cache[n_l - 1].OUT, l[n_l]); } prep_new_arrays(l[n_l], cache[n_l]); forw_prop(l[n_l], cache[n_l]); prep_gradients(l[n_l], cache[n_l]); if (l[n_l].n_H < 25) show_results(l[n_l], cache[n_l]); else brief_inf(l[n_l], cache[n_l].OUT); //back_prop(l[n_l], cache[n_l]); //show_gradients(l[n_l], cache[n_l]); //Funkcja wyswietla gradient } return 0; } void set_random_IN(layer_param l, double**& IN) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { //Dla kazdego badanego przypadku (np. pojedynczej mapy bajtowej- spektogram) for (int h = 0; h < l.n_Hprev; h++) { //Przejdz po kazdym wierszu for (int w = 0; w < l.n_Wprev; w++) { //Przejdz po kazdej kolumnie for (int c = 0; c < l.n_Cprev; c++) { //Przejdz po kazdym kanale (np. dla wejscia w postaci zdjecia rgb - 3 kanaly) if (h < l.pad || h > l.n_Hprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; //Ustawianie zer dla zero paddingu else if (w < l.pad || w > l.n_Wprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; else IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = (rand() % 10 + 1)/10.; //W tablicy wejsciowej beda same wartosci int } } } } } void set_new_IN(double**& IN, double**& OUT, layer_param l) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = OUT[m][w + l.n_Wprev * (h + l.n_Hprev * c)]; } } } } } void update_param(layer_param& l_prev, layer_param& l) { l.M = l_prev.M; l.pad = l_prev.pad; l.F = l_prev.F; l.F_MP = l_prev.F_MP; l.stride = l_prev.stride; l.stride_MP = l_prev.stride_MP; l.alpha = l_prev.alpha; l.n_Hprev = l_prev.n_Hout; l.n_Wprev = l_prev.n_Wout; l.n_Cprev = l_prev.n_C; l.n_H = int((l.n_Hprev - l.F) / l.stride) + 1; l.n_W = int((l.n_Wprev - l.F) / l.stride) + 1; l.n_C = KERNEL_SIZE(l_prev.n_C); l.n_Hout = int((l.n_H - l.F_MP) / l.stride_MP) + 1; l.n_Wout = int((l.n_W - l.F_MP) / l.stride_MP) + 1; } void prep_new_arrays(layer_param l, cache_data& cache) { cache.Z = new double* [l.M]; cache.A = new double* [l.M]; cache.OUT = new double* [l.M]; cache.kernel = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.Z[i] = new double[l.n_C * l.n_H * l.n_W]; cache.A[i] = new double[l.n_C * l.n_H * l.n_W]; cache.OUT[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.n_C; i++) { cache.kernel[i] = new double[l.n_Cprev * l.F * l.F]; } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.kernel[c][w + l.F * (h + l.F * d)] = (rand()%10000 - 5000) * 0.0001; //Ustawianie losowych wag filtra } } } } } void prep_gradients(layer_param l, cache_data& cache) { cache.dZ = new double* [l.M]; cache.dA = new double* [l.M]; cache.dAprev = new double* [l.M]; cache.dW = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.dZ[i] = new double[l.n_C * l.n_H * l.n_W]; } for (int i = 0; i < l.M; i++) { cache.dA[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.M; i++) { cache.dAprev[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int i = 0; i < l.n_C; i++) { cache.dW[i] = new double[l.n_Cprev * l.F * l.F]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] = (rand()%10000 - 5000) * 0.0001; } } } } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; } } } } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.dW[c][w + l.F * (h + l.F * d)] = 0; } } } } double maximum; int vert_start, vert_end; int horiz_start, horiz_end; for (int m = 0; m < l.M; m++) { //Dla kazdego przypadku for (int h = 0; h < l.n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < l.n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < l.n_C; c++) { //Dla kazdego kanalu wyjscia vert_start = h * l.stride_MP; vert_end = vert_start + l.F_MP; horiz_start = w * l.stride_MP; horiz_end = horiz_start + l.F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + l.n_W * (j + l.n_H * c)] > maximum) maximum = cache.A[m][k + l.n_W * (j + l.n_H * c)]; } } for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.A[m][k + l.n_W * (j + l.n_H * c)] != maximum || maximum == 0) cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = 0; else cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)]; } } } } } } for (int m = 0; m < l.M; m++) { for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { for (int c = 0; c < l.n_C; c++) { vert_start = h; vert_end = vert_start + l.F; horiz_start = w; horiz_end = horiz_start + l.F; for (int d = 0; d < l.n_Cprev; d++) { for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.dZ[m][w + l.n_W * (h + l.n_H * c)] < 0) cache.dZ[m][w + l.n_W * (h + l.n_H * c)] = 0; cache.dAprev[m][j + l.n_Wprev * (k + l.n_Hprev * d)] += cache.kernel[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; cache.dW[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] += cache.IN[m][j + l.n_Wprev * (k + l.n_Hprev * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; } } } } } } } } void brief_inf(layer_param l, double** OUT) { for (int m = 0; m < l.M; m++) std::cout << "Wyjscie: " << m + 1 << " Kanaly: " << l.n_C << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; } void forw_prop(layer_param l, cache_data& cache) { int M = l.M; int pad = l.pad; int F = l.F; int F_MP = l.F_MP; int stride = l.stride; int stride_MP = l.stride_MP; int n_Hprev = l.n_Hprev; int n_Wprev = l.n_Wprev; int n_Cprev = l.n_Cprev; int n_H = l.n_H; int n_W = l.n_W; int n_C = l.n_C; int n_Hout = l.n_Hout; int n_Wout = l.n_Wout; int vert_start = 0; int vert_end = 0; int horiz_start = 0; int horiz_end = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_H; h++) { //Dla kazdego wiersza for (int w = 0; w < n_W; w++) { //Dla kazdej kolumny for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu (kanalow bedzie tyle, ile chcemy kerneli) vert_start = h * stride; //Poczatek wycinka w pionie vert_end = vert_start + F; //Koniec wycinka w pionie horiz_start = w * stride; //Poczatek wycika w poziomie horiz_end = horiz_start + F; //Koniec wycinka w poziomie cache.Z[m][w + n_W * (h + n_H * c)] = 0; for (int d = 0; d < n_Cprev; d++) { //Dla kazdego kanalu w tablicy wejsciowej for (int j = vert_start; j < vert_end; j++) { //Dla wybranych wierszy for (int k = horiz_start; k < horiz_end; k++) { //Dla wybranych kolumn cache.Z[m][w + n_W * (h + n_H * c)] += cache.kernel[c][(k - horiz_start) + F * ((j - vert_start) + F * d)] * cache.IN[m][k + n_Wprev * (j + n_Hprev * d)]; //Pomnoz wartosc/piksel wycinka przez wage kernela i dodaj do wyniku konwolucji } } } cache.A[m][w + n_W * (h + n_H * c)] = RELU(cache.Z[m][w + n_W * (h + n_H * c)]); //Aktywowanie danej wartosci/neuronu } } } } double maximum = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu wyjscia if (n_Hout > 1) { vert_start = h * stride_MP; vert_end = vert_start + F_MP; horiz_start = w * stride_MP; horiz_end = horiz_start + F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + n_W * (j + n_H * c)] > maximum) maximum = cache.A[m][k + n_W * (j + n_H * c)]; //Wybierz maksimum z wycinka } } cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = maximum; } else cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = cache.A[m][0 + n_W * (0 + n_H * c)]; } } } } } void show_gradients(layer_param l, cache_data& cache) { for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "dAprev: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dW #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "dW: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dZ #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dZ: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.dZ[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### dA #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dA: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void simple_del(double**& arr, int len) { for (int i = 0; i < len; i++) { delete[] arr[i]; } delete[] arr; } void show_results(layer_param l, cache_data& cache) { std::cout.precision(4); for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "Wejscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### FILTRY #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Kernel: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### WYNIKI KONWOLUCJI #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Z: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.Z[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI AKTYWACJI (RELU) #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "A: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.A[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI MAX POOLINGU #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Wyjscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.OUT[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void back_prop(layer_param l, cache_data& cache) { std::cout << "#### FILTRY PO PO PROPAGACJI WSTECZNEJ #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Wagi: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] - l.alpha * cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } }
//#include "shallow.h" #include <hip/hip_runtime.h> #include <iostream> #include <time.h> #include <stdlib.h> #define RELU(a) ((a > 0) ? a : 0) #define KERNEL_SIZE(n_C) ((n_C * 2 > 8) ? 8 : n_C * 2) struct layer_param { int M; //Ilosc tablic podawanych na wejscie int pad; //Grubosc warstwy zer na krawedziach (zero-padding) int F; //Rozmiar 2D filtra (F x F) int F_MP; //Rozmiar 2D filtra do max poolingu (F_MP x F_MP) int stride; //Ilosc przeskakiwanych pikseli przy konwolucji na inkrementacje int stride_MP; //To samo, tylko przy max poolingu int n_Hprev; //Wysokosc tablicy wejsciowej podawanej na wejscie sieci int n_Wprev; //Szerokosc tablicy wejsciowej podawanej na wejscie sieci int n_Cprev; //Glebokosc tablicy wejsciowej, jednoczesnie musi to byc glebokosc filtra (F x F x C) int n_H; //Wysokosc tablicy uzyskanej po konwolucji kernela z wejsciem int n_W; int n_C; //Ilosc filtrow, jednoczesnie glebokosc wyjscia warstwy int n_Hout; //Wysokosc tablicy wyjsciowej warstwy int n_Wout; double alpha; //Predkosc uczenia }; struct cache_data { double** IN; //Tablica wejsciowa double** Z; //Wynik splotu double** A; //Wynik Aktywacji double** OUT; //Poprzedni wynik po max poolingu, jednoczescie wyjscie warstwy sieci double** kernel; //Filtr double** dW; //Gradient kosztu wzgledem kerneli double** dA; //Gradient kosztu wzgledem warstwy double** dAprev; //Gradient kosztu wzgledem wyjscia warstwy n_l - 1 double** dZ; //Gradient kosztu wzgledem wyniku konwolucji }; void set_random_IN(layer_param, double**&); //Ustawia losowe wejscie (do testowania) void set_new_IN(double**&, double**&, layer_param l); void show_results(layer_param, cache_data&); //Wyswietla zawartosc koncowych i posrednich wynikow w warstwie void brief_inf(layer_param, double**); //Krotka informacja o wyjsciu sieci void forw_prop(layer_param, cache_data&); //Najwazniejsza funkcja (konwolucja, aktywacja, maxpooling) void simple_del(double**&, int); //Usuwanie pamieci void update_param(layer_param&, layer_param&); //Ustawianie nowych parametrow warstwy void prep_new_arrays(layer_param, cache_data&); //Tworzenie nowych tablic wynikowych void prep_gradients(layer_param, cache_data&); //Tworzenie gradientow (narazie losowo, bez funkcji kosztu) void show_gradients(layer_param, cache_data&); void back_prop(layer_param, cache_data&); int main() { srand(time(NULL)); int number_of_layers = 2; layer_param* l = new layer_param[number_of_layers]; cache_data* cache = new cache_data[number_of_layers]; int n_l = 0; layer_param l_prev; l[n_l].M = 1; l[n_l].pad = 0; l[n_l].F = 3; l[n_l].F_MP = 2; l[n_l].stride = 1; l[n_l].stride_MP = 2; l[n_l].alpha = 0.1; int IN_size = 16; //Rzeczywisty rozmiar wejscia int IN_depth = 1; //Rzeczywista glebokosc wejscia l[n_l].n_Hprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Wprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Cprev = IN_depth; l[n_l].n_H = int((l[n_l].n_Hprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_W = int((l[n_l].n_Wprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_C = 1; l[n_l].n_Hout = int((l[n_l].n_H - l[n_l].F_MP) / l[n_l].stride_MP) + 1; l[n_l].n_Wout = int((l[n_l].n_W - l[n_l].F_MP) / l[n_l].stride_MP) + 1; for (n_l = 0; n_l < number_of_layers; n_l++) { std::cout << "\n\n#### WARSTWA: " << n_l + 1 << "#### \n"; if (n_l == 0) set_random_IN(l[n_l], cache[n_l].IN); else { l_prev = l[n_l - 1]; update_param(l_prev, l[n_l]); set_new_IN(cache[n_l].IN, cache[n_l - 1].OUT, l[n_l]); } prep_new_arrays(l[n_l], cache[n_l]); forw_prop(l[n_l], cache[n_l]); prep_gradients(l[n_l], cache[n_l]); if (l[n_l].n_H < 25) show_results(l[n_l], cache[n_l]); else brief_inf(l[n_l], cache[n_l].OUT); //back_prop(l[n_l], cache[n_l]); //show_gradients(l[n_l], cache[n_l]); //Funkcja wyswietla gradient } return 0; } void set_random_IN(layer_param l, double**& IN) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { //Dla kazdego badanego przypadku (np. pojedynczej mapy bajtowej- spektogram) for (int h = 0; h < l.n_Hprev; h++) { //Przejdz po kazdym wierszu for (int w = 0; w < l.n_Wprev; w++) { //Przejdz po kazdej kolumnie for (int c = 0; c < l.n_Cprev; c++) { //Przejdz po kazdym kanale (np. dla wejscia w postaci zdjecia rgb - 3 kanaly) if (h < l.pad || h > l.n_Hprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; //Ustawianie zer dla zero paddingu else if (w < l.pad || w > l.n_Wprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; else IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = (rand() % 10 + 1)/10.; //W tablicy wejsciowej beda same wartosci int } } } } } void set_new_IN(double**& IN, double**& OUT, layer_param l) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = OUT[m][w + l.n_Wprev * (h + l.n_Hprev * c)]; } } } } } void update_param(layer_param& l_prev, layer_param& l) { l.M = l_prev.M; l.pad = l_prev.pad; l.F = l_prev.F; l.F_MP = l_prev.F_MP; l.stride = l_prev.stride; l.stride_MP = l_prev.stride_MP; l.alpha = l_prev.alpha; l.n_Hprev = l_prev.n_Hout; l.n_Wprev = l_prev.n_Wout; l.n_Cprev = l_prev.n_C; l.n_H = int((l.n_Hprev - l.F) / l.stride) + 1; l.n_W = int((l.n_Wprev - l.F) / l.stride) + 1; l.n_C = KERNEL_SIZE(l_prev.n_C); l.n_Hout = int((l.n_H - l.F_MP) / l.stride_MP) + 1; l.n_Wout = int((l.n_W - l.F_MP) / l.stride_MP) + 1; } void prep_new_arrays(layer_param l, cache_data& cache) { cache.Z = new double* [l.M]; cache.A = new double* [l.M]; cache.OUT = new double* [l.M]; cache.kernel = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.Z[i] = new double[l.n_C * l.n_H * l.n_W]; cache.A[i] = new double[l.n_C * l.n_H * l.n_W]; cache.OUT[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.n_C; i++) { cache.kernel[i] = new double[l.n_Cprev * l.F * l.F]; } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.kernel[c][w + l.F * (h + l.F * d)] = (rand()%10000 - 5000) * 0.0001; //Ustawianie losowych wag filtra } } } } } void prep_gradients(layer_param l, cache_data& cache) { cache.dZ = new double* [l.M]; cache.dA = new double* [l.M]; cache.dAprev = new double* [l.M]; cache.dW = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.dZ[i] = new double[l.n_C * l.n_H * l.n_W]; } for (int i = 0; i < l.M; i++) { cache.dA[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.M; i++) { cache.dAprev[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int i = 0; i < l.n_C; i++) { cache.dW[i] = new double[l.n_Cprev * l.F * l.F]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] = (rand()%10000 - 5000) * 0.0001; } } } } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; } } } } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.dW[c][w + l.F * (h + l.F * d)] = 0; } } } } double maximum; int vert_start, vert_end; int horiz_start, horiz_end; for (int m = 0; m < l.M; m++) { //Dla kazdego przypadku for (int h = 0; h < l.n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < l.n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < l.n_C; c++) { //Dla kazdego kanalu wyjscia vert_start = h * l.stride_MP; vert_end = vert_start + l.F_MP; horiz_start = w * l.stride_MP; horiz_end = horiz_start + l.F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + l.n_W * (j + l.n_H * c)] > maximum) maximum = cache.A[m][k + l.n_W * (j + l.n_H * c)]; } } for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.A[m][k + l.n_W * (j + l.n_H * c)] != maximum || maximum == 0) cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = 0; else cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)]; } } } } } } for (int m = 0; m < l.M; m++) { for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { for (int c = 0; c < l.n_C; c++) { vert_start = h; vert_end = vert_start + l.F; horiz_start = w; horiz_end = horiz_start + l.F; for (int d = 0; d < l.n_Cprev; d++) { for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.dZ[m][w + l.n_W * (h + l.n_H * c)] < 0) cache.dZ[m][w + l.n_W * (h + l.n_H * c)] = 0; cache.dAprev[m][j + l.n_Wprev * (k + l.n_Hprev * d)] += cache.kernel[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; cache.dW[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] += cache.IN[m][j + l.n_Wprev * (k + l.n_Hprev * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; } } } } } } } } void brief_inf(layer_param l, double** OUT) { for (int m = 0; m < l.M; m++) std::cout << "Wyjscie: " << m + 1 << " Kanaly: " << l.n_C << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; } void forw_prop(layer_param l, cache_data& cache) { int M = l.M; int pad = l.pad; int F = l.F; int F_MP = l.F_MP; int stride = l.stride; int stride_MP = l.stride_MP; int n_Hprev = l.n_Hprev; int n_Wprev = l.n_Wprev; int n_Cprev = l.n_Cprev; int n_H = l.n_H; int n_W = l.n_W; int n_C = l.n_C; int n_Hout = l.n_Hout; int n_Wout = l.n_Wout; int vert_start = 0; int vert_end = 0; int horiz_start = 0; int horiz_end = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_H; h++) { //Dla kazdego wiersza for (int w = 0; w < n_W; w++) { //Dla kazdej kolumny for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu (kanalow bedzie tyle, ile chcemy kerneli) vert_start = h * stride; //Poczatek wycinka w pionie vert_end = vert_start + F; //Koniec wycinka w pionie horiz_start = w * stride; //Poczatek wycika w poziomie horiz_end = horiz_start + F; //Koniec wycinka w poziomie cache.Z[m][w + n_W * (h + n_H * c)] = 0; for (int d = 0; d < n_Cprev; d++) { //Dla kazdego kanalu w tablicy wejsciowej for (int j = vert_start; j < vert_end; j++) { //Dla wybranych wierszy for (int k = horiz_start; k < horiz_end; k++) { //Dla wybranych kolumn cache.Z[m][w + n_W * (h + n_H * c)] += cache.kernel[c][(k - horiz_start) + F * ((j - vert_start) + F * d)] * cache.IN[m][k + n_Wprev * (j + n_Hprev * d)]; //Pomnoz wartosc/piksel wycinka przez wage kernela i dodaj do wyniku konwolucji } } } cache.A[m][w + n_W * (h + n_H * c)] = RELU(cache.Z[m][w + n_W * (h + n_H * c)]); //Aktywowanie danej wartosci/neuronu } } } } double maximum = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu wyjscia if (n_Hout > 1) { vert_start = h * stride_MP; vert_end = vert_start + F_MP; horiz_start = w * stride_MP; horiz_end = horiz_start + F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + n_W * (j + n_H * c)] > maximum) maximum = cache.A[m][k + n_W * (j + n_H * c)]; //Wybierz maksimum z wycinka } } cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = maximum; } else cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = cache.A[m][0 + n_W * (0 + n_H * c)]; } } } } } void show_gradients(layer_param l, cache_data& cache) { for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "dAprev: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dW #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "dW: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dZ #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dZ: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.dZ[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### dA #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dA: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void simple_del(double**& arr, int len) { for (int i = 0; i < len; i++) { delete[] arr[i]; } delete[] arr; } void show_results(layer_param l, cache_data& cache) { std::cout.precision(4); for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "Wejscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### FILTRY #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Kernel: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### WYNIKI KONWOLUCJI #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Z: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.Z[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI AKTYWACJI (RELU) #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "A: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.A[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI MAX POOLINGU #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Wyjscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.OUT[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void back_prop(layer_param l, cache_data& cache) { std::cout << "#### FILTRY PO PO PROPAGACJI WSTECZNEJ #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Wagi: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] - l.alpha * cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
//#include "shallow.h" #include <hip/hip_runtime.h> #include <iostream> #include <time.h> #include <stdlib.h> #define RELU(a) ((a > 0) ? a : 0) #define KERNEL_SIZE(n_C) ((n_C * 2 > 8) ? 8 : n_C * 2) struct layer_param { int M; //Ilosc tablic podawanych na wejscie int pad; //Grubosc warstwy zer na krawedziach (zero-padding) int F; //Rozmiar 2D filtra (F x F) int F_MP; //Rozmiar 2D filtra do max poolingu (F_MP x F_MP) int stride; //Ilosc przeskakiwanych pikseli przy konwolucji na inkrementacje int stride_MP; //To samo, tylko przy max poolingu int n_Hprev; //Wysokosc tablicy wejsciowej podawanej na wejscie sieci int n_Wprev; //Szerokosc tablicy wejsciowej podawanej na wejscie sieci int n_Cprev; //Glebokosc tablicy wejsciowej, jednoczesnie musi to byc glebokosc filtra (F x F x C) int n_H; //Wysokosc tablicy uzyskanej po konwolucji kernela z wejsciem int n_W; int n_C; //Ilosc filtrow, jednoczesnie glebokosc wyjscia warstwy int n_Hout; //Wysokosc tablicy wyjsciowej warstwy int n_Wout; double alpha; //Predkosc uczenia }; struct cache_data { double** IN; //Tablica wejsciowa double** Z; //Wynik splotu double** A; //Wynik Aktywacji double** OUT; //Poprzedni wynik po max poolingu, jednoczescie wyjscie warstwy sieci double** kernel; //Filtr double** dW; //Gradient kosztu wzgledem kerneli double** dA; //Gradient kosztu wzgledem warstwy double** dAprev; //Gradient kosztu wzgledem wyjscia warstwy n_l - 1 double** dZ; //Gradient kosztu wzgledem wyniku konwolucji }; void set_random_IN(layer_param, double**&); //Ustawia losowe wejscie (do testowania) void set_new_IN(double**&, double**&, layer_param l); void show_results(layer_param, cache_data&); //Wyswietla zawartosc koncowych i posrednich wynikow w warstwie void brief_inf(layer_param, double**); //Krotka informacja o wyjsciu sieci void forw_prop(layer_param, cache_data&); //Najwazniejsza funkcja (konwolucja, aktywacja, maxpooling) void simple_del(double**&, int); //Usuwanie pamieci void update_param(layer_param&, layer_param&); //Ustawianie nowych parametrow warstwy void prep_new_arrays(layer_param, cache_data&); //Tworzenie nowych tablic wynikowych void prep_gradients(layer_param, cache_data&); //Tworzenie gradientow (narazie losowo, bez funkcji kosztu) void show_gradients(layer_param, cache_data&); void back_prop(layer_param, cache_data&); int main() { srand(time(NULL)); int number_of_layers = 2; layer_param* l = new layer_param[number_of_layers]; cache_data* cache = new cache_data[number_of_layers]; int n_l = 0; layer_param l_prev; l[n_l].M = 1; l[n_l].pad = 0; l[n_l].F = 3; l[n_l].F_MP = 2; l[n_l].stride = 1; l[n_l].stride_MP = 2; l[n_l].alpha = 0.1; int IN_size = 16; //Rzeczywisty rozmiar wejscia int IN_depth = 1; //Rzeczywista glebokosc wejscia l[n_l].n_Hprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Wprev = IN_size + 2 * l[n_l].pad; l[n_l].n_Cprev = IN_depth; l[n_l].n_H = int((l[n_l].n_Hprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_W = int((l[n_l].n_Wprev - l[n_l].F) / l[n_l].stride) + 1; l[n_l].n_C = 1; l[n_l].n_Hout = int((l[n_l].n_H - l[n_l].F_MP) / l[n_l].stride_MP) + 1; l[n_l].n_Wout = int((l[n_l].n_W - l[n_l].F_MP) / l[n_l].stride_MP) + 1; for (n_l = 0; n_l < number_of_layers; n_l++) { std::cout << "\n\n#### WARSTWA: " << n_l + 1 << "#### \n"; if (n_l == 0) set_random_IN(l[n_l], cache[n_l].IN); else { l_prev = l[n_l - 1]; update_param(l_prev, l[n_l]); set_new_IN(cache[n_l].IN, cache[n_l - 1].OUT, l[n_l]); } prep_new_arrays(l[n_l], cache[n_l]); forw_prop(l[n_l], cache[n_l]); prep_gradients(l[n_l], cache[n_l]); if (l[n_l].n_H < 25) show_results(l[n_l], cache[n_l]); else brief_inf(l[n_l], cache[n_l].OUT); //back_prop(l[n_l], cache[n_l]); //show_gradients(l[n_l], cache[n_l]); //Funkcja wyswietla gradient } return 0; } void set_random_IN(layer_param l, double**& IN) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { //Dla kazdego badanego przypadku (np. pojedynczej mapy bajtowej- spektogram) for (int h = 0; h < l.n_Hprev; h++) { //Przejdz po kazdym wierszu for (int w = 0; w < l.n_Wprev; w++) { //Przejdz po kazdej kolumnie for (int c = 0; c < l.n_Cprev; c++) { //Przejdz po kazdym kanale (np. dla wejscia w postaci zdjecia rgb - 3 kanaly) if (h < l.pad || h > l.n_Hprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; //Ustawianie zer dla zero paddingu else if (w < l.pad || w > l.n_Wprev - l.pad - 1) IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; else IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = (rand() % 10 + 1)/10.; //W tablicy wejsciowej beda same wartosci int } } } } } void set_new_IN(double**& IN, double**& OUT, layer_param l) { IN = new double* [l.M]; for (int i = 0; i < l.M; i++) { IN[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = OUT[m][w + l.n_Wprev * (h + l.n_Hprev * c)]; } } } } } void update_param(layer_param& l_prev, layer_param& l) { l.M = l_prev.M; l.pad = l_prev.pad; l.F = l_prev.F; l.F_MP = l_prev.F_MP; l.stride = l_prev.stride; l.stride_MP = l_prev.stride_MP; l.alpha = l_prev.alpha; l.n_Hprev = l_prev.n_Hout; l.n_Wprev = l_prev.n_Wout; l.n_Cprev = l_prev.n_C; l.n_H = int((l.n_Hprev - l.F) / l.stride) + 1; l.n_W = int((l.n_Wprev - l.F) / l.stride) + 1; l.n_C = KERNEL_SIZE(l_prev.n_C); l.n_Hout = int((l.n_H - l.F_MP) / l.stride_MP) + 1; l.n_Wout = int((l.n_W - l.F_MP) / l.stride_MP) + 1; } void prep_new_arrays(layer_param l, cache_data& cache) { cache.Z = new double* [l.M]; cache.A = new double* [l.M]; cache.OUT = new double* [l.M]; cache.kernel = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.Z[i] = new double[l.n_C * l.n_H * l.n_W]; cache.A[i] = new double[l.n_C * l.n_H * l.n_W]; cache.OUT[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.n_C; i++) { cache.kernel[i] = new double[l.n_Cprev * l.F * l.F]; } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.kernel[c][w + l.F * (h + l.F * d)] = (rand()%10000 - 5000) * 0.0001; //Ustawianie losowych wag filtra } } } } } void prep_gradients(layer_param l, cache_data& cache) { cache.dZ = new double* [l.M]; cache.dA = new double* [l.M]; cache.dAprev = new double* [l.M]; cache.dW = new double* [l.n_C]; for (int i = 0; i < l.M; i++) { cache.dZ[i] = new double[l.n_C * l.n_H * l.n_W]; } for (int i = 0; i < l.M; i++) { cache.dA[i] = new double[l.n_C * l.n_Hout * l.n_Wout]; } for (int i = 0; i < l.M; i++) { cache.dAprev[i] = new double[l.n_Cprev * l.n_Hprev * l.n_Wprev]; } for (int i = 0; i < l.n_C; i++) { cache.dW[i] = new double[l.n_Cprev * l.F * l.F]; } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] = (rand()%10000 - 5000) * 0.0001; } } } } for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] = 0; } } } } for (int c = 0; c < l.n_C; c++) { for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { for (int d = 0; d < l.n_Cprev; d++) { cache.dW[c][w + l.F * (h + l.F * d)] = 0; } } } } double maximum; int vert_start, vert_end; int horiz_start, horiz_end; for (int m = 0; m < l.M; m++) { //Dla kazdego przypadku for (int h = 0; h < l.n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < l.n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < l.n_C; c++) { //Dla kazdego kanalu wyjscia vert_start = h * l.stride_MP; vert_end = vert_start + l.F_MP; horiz_start = w * l.stride_MP; horiz_end = horiz_start + l.F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + l.n_W * (j + l.n_H * c)] > maximum) maximum = cache.A[m][k + l.n_W * (j + l.n_H * c)]; } } for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.A[m][k + l.n_W * (j + l.n_H * c)] != maximum || maximum == 0) cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = 0; else cache.dZ[m][k + l.n_W * (j + l.n_H * c)] = cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)]; } } } } } } for (int m = 0; m < l.M; m++) { for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { for (int c = 0; c < l.n_C; c++) { vert_start = h; vert_end = vert_start + l.F; horiz_start = w; horiz_end = horiz_start + l.F; for (int d = 0; d < l.n_Cprev; d++) { for (int j = vert_start; j < vert_end; j++) { for (int k = horiz_start; k < horiz_end; k++) { if (cache.dZ[m][w + l.n_W * (h + l.n_H * c)] < 0) cache.dZ[m][w + l.n_W * (h + l.n_H * c)] = 0; cache.dAprev[m][j + l.n_Wprev * (k + l.n_Hprev * d)] += cache.kernel[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; cache.dW[c][(k - horiz_start) + l.F * ((j - vert_start) + l.F * d)] += cache.IN[m][j + l.n_Wprev * (k + l.n_Hprev * d)] * cache.dZ[m][w + l.n_W * (h + l.n_H * c)]; } } } } } } } } void brief_inf(layer_param l, double** OUT) { for (int m = 0; m < l.M; m++) std::cout << "Wyjscie: " << m + 1 << " Kanaly: " << l.n_C << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; } void forw_prop(layer_param l, cache_data& cache) { int M = l.M; int pad = l.pad; int F = l.F; int F_MP = l.F_MP; int stride = l.stride; int stride_MP = l.stride_MP; int n_Hprev = l.n_Hprev; int n_Wprev = l.n_Wprev; int n_Cprev = l.n_Cprev; int n_H = l.n_H; int n_W = l.n_W; int n_C = l.n_C; int n_Hout = l.n_Hout; int n_Wout = l.n_Wout; int vert_start = 0; int vert_end = 0; int horiz_start = 0; int horiz_end = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_H; h++) { //Dla kazdego wiersza for (int w = 0; w < n_W; w++) { //Dla kazdej kolumny for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu (kanalow bedzie tyle, ile chcemy kerneli) vert_start = h * stride; //Poczatek wycinka w pionie vert_end = vert_start + F; //Koniec wycinka w pionie horiz_start = w * stride; //Poczatek wycika w poziomie horiz_end = horiz_start + F; //Koniec wycinka w poziomie cache.Z[m][w + n_W * (h + n_H * c)] = 0; for (int d = 0; d < n_Cprev; d++) { //Dla kazdego kanalu w tablicy wejsciowej for (int j = vert_start; j < vert_end; j++) { //Dla wybranych wierszy for (int k = horiz_start; k < horiz_end; k++) { //Dla wybranych kolumn cache.Z[m][w + n_W * (h + n_H * c)] += cache.kernel[c][(k - horiz_start) + F * ((j - vert_start) + F * d)] * cache.IN[m][k + n_Wprev * (j + n_Hprev * d)]; //Pomnoz wartosc/piksel wycinka przez wage kernela i dodaj do wyniku konwolucji } } } cache.A[m][w + n_W * (h + n_H * c)] = RELU(cache.Z[m][w + n_W * (h + n_H * c)]); //Aktywowanie danej wartosci/neuronu } } } } double maximum = 0; for (int m = 0; m < M; m++) { //Dla kazdego przypadku for (int h = 0; h < n_Hout; h++) { //Dla kazdego wiersza wyjscia (wyniku max poolingu) for (int w = 0; w < n_Wout; w++) { // Dla kazdej kolumny wyjscia for (int c = 0; c < n_C; c++) { //Dla kazdego kanalu wyjscia if (n_Hout > 1) { vert_start = h * stride_MP; vert_end = vert_start + F_MP; horiz_start = w * stride_MP; horiz_end = horiz_start + F_MP; maximum = 0; for (int j = vert_start; j < vert_end; j++) { //Dla kazdego wiersza wycinka wyniku aktywacji for (int k = horiz_start; k < horiz_end; k++) { //Dla kazdej kolumny wycinka wyniku aktywacji if (cache.A[m][k + n_W * (j + n_H * c)] > maximum) maximum = cache.A[m][k + n_W * (j + n_H * c)]; //Wybierz maksimum z wycinka } } cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = maximum; } else cache.OUT[m][w + n_Wout * (h + n_Hout * c)] = cache.A[m][0 + n_W * (0 + n_H * c)]; } } } } } void show_gradients(layer_param l, cache_data& cache) { for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "dAprev: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.dAprev[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dW #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "dW: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### dZ #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dZ: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.dZ[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### dA #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "dA: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.dA[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void simple_del(double**& arr, int len) { for (int i = 0; i < len; i++) { delete[] arr[i]; } delete[] arr; } void show_results(layer_param l, cache_data& cache) { std::cout.precision(4); for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_Cprev; c++) { std::cout << "Wejscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hprev << "x" << l.n_Wprev << "x" << l.n_Cprev << ")" << "\n"; for (int h = 0; h < l.n_Hprev; h++) { for (int w = 0; w < l.n_Wprev; w++) { std::cout << cache.IN[m][w + l.n_Wprev * (h + l.n_Hprev * c)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### FILTRY #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Kernel: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } std::cout << "#### WYNIKI KONWOLUCJI #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Z: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.Z[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI AKTYWACJI (RELU) #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "A: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_H << "x" << l.n_W << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_H; h++) { for (int w = 0; w < l.n_W; w++) { std::cout << cache.A[m][w + l.n_W * (h + l.n_H * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } std::cout << "#### WYNIKI MAX POOLINGU #### \n\n"; for (int m = 0; m < l.M; m++) { for (int c = 0; c < l.n_C; c++) { std::cout << "Wyjscie: " << m + 1 << " Kanal: " << c + 1 << " (" << l.n_Hout << "x" << l.n_Wout << "x" << l.n_C << ")" << "\n" << std::fixed; for (int h = 0; h < l.n_Hout; h++) { for (int w = 0; w < l.n_Wout; w++) { std::cout << cache.OUT[m][w + l.n_Wout * (h + l.n_Hout * c)] << " "; } std::cout << "\n"; } std::cout << "\n"; } } } void back_prop(layer_param l, cache_data& cache) { std::cout << "#### FILTRY PO PO PROPAGACJI WSTECZNEJ #### \n\n"; for (int c = 0; c < l.n_C; c++) { for (int d = 0; d < l.n_Cprev; d++) { std::cout << "Wagi: " << c + 1 << " Kanal: " << d + 1 << " (" << l.F << "x" << l.F << "x" << l.n_Cprev << ")" << "\n" << std::fixed; for (int h = 0; h < l.F; h++) { for (int w = 0; w < l.F; w++) { std::cout << cache.kernel[c][w + l.F * (h + l.F * d)] - l.alpha * cache.dW[c][w + l.F * (h + l.F * d)] << " "; } std::cout << "\n"; } } std::cout << "\n\n"; } }
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include <stdio.h> #define MIN(x, y) (((x) < (y)) ? (x) : (y)) __global__ void matmult_kernel1(int m, int n, int k, double *A, double *B, double *C){ // set C to zeros for (int i=0;i<m;i++){ for (int p=0;p<n;p++){ C[i*n+p]=0; //C[i][p] } } // do matmult with mkn loop order for (int i=0;i<m;i++) { for (int j=0;j<k;j++){ for (int p=0;p<n;p++) { C[i*n+p] += A[i*k+j] * B[j*n+p]; } } } } __global__ void matmult_kernel2(int m, int n, int k, double *A, double *B, double *C){ int j = blockIdx.x*blockDim.x+threadIdx.x; //looping through n int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp; if((j<n)&&(i<m)){ for(int p=0; p<k; p++){ //read row of A and col of B //row of A is A[mit*k+kit] //col of B is B[kit*n+nit] tmp += A[i*k+p] * B[p*n+j]; } //C is C[mit*n+nit] C[i*n+j] = tmp; } } __global__ void matmult_kernel3(int m, int n, int k, double *A, double *B, double *C){ //compute C(i,j) and C(i,j+1) int j = 2*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only half as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, either 1 or 0) int j_add = MIN(1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j]; if(j_add == 1) tmp2 += A[i*k+p] * B[p*n+j+1]; } C[i*n+j] = tmp1; if(j_add == 1) C[i*n+j+1] = tmp2; } } __global__ void matmult_kernel4(int m, int n, int k, double *A, double *B, double *C, int s){ //compute C(i,j), C(i,j+1), ... C(i,j+s) int j = s*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only 1/s as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; double tmp3 = 0.0; double tmp4 = 0.0; double tmp5 = 0.0; double tmp6 = 0.0; double tmp7 = 0.0; double tmp8 = 0.0; double tmp9 = 0.0; double tmp10 = 0.0; double tmp11 = 0.0; double tmp12 = 0.0; double tmp13 = 0.0; double tmp14 = 0.0; double tmp15 = 0.0; double tmp16 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, from 0 to s-1) int j_add = MIN(s-1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j+0]; if(j_add > 0) tmp2 += A[i*k+p] * B[p*n+j+1]; if(j_add > 1) tmp3 += A[i*k+p] * B[p*n+j+2]; if(j_add > 2) tmp4 += A[i*k+p] * B[p*n+j+3]; if(j_add > 3) tmp5 += A[i*k+p] * B[p*n+j+4]; if(j_add > 4) tmp6 += A[i*k+p] * B[p*n+j+5]; if(j_add > 5) tmp7 += A[i*k+p] * B[p*n+j+6]; if(j_add > 6) tmp8 += A[i*k+p] * B[p*n+j+7]; if(j_add > 7) tmp9 += A[i*k+p] * B[p*n+j+8]; if(j_add > 8) tmp10 += A[i*k+p] * B[p*n+j+9]; if(j_add > 9) tmp11 += A[i*k+p] * B[p*n+j+10]; if(j_add > 10) tmp12 += A[i*k+p] * B[p*n+j+11]; if(j_add > 11) tmp13 += A[i*k+p] * B[p*n+j+12]; if(j_add > 12) tmp14 += A[i*k+p] * B[p*n+j+13]; if(j_add > 13) tmp15 += A[i*k+p] * B[p*n+j+14]; if(j_add > 14) tmp16 += A[i*k+p] * B[p*n+j+15]; } C[i*n+j] = tmp1; if(j_add > 0) C[i*n+j+1] = tmp2; if(j_add > 1) C[i*n+j+2] = tmp3; if(j_add > 2) C[i*n+j+3] = tmp4; if(j_add > 3) C[i*n+j+4] = tmp5; if(j_add > 4) C[i*n+j+5] = tmp6; if(j_add > 5) C[i*n+j+6] = tmp7; if(j_add > 6) C[i*n+j+7] = tmp8; if(j_add > 7) C[i*n+j+8] = tmp9; if(j_add > 8) C[i*n+j+9] = tmp10; if(j_add > 9) C[i*n+j+10] = tmp11; if(j_add > 10) C[i*n+j+11] = tmp12; if(j_add > 11) C[i*n+j+12] = tmp13; if(j_add > 12) C[i*n+j+13] = tmp14; if(j_add > 13) C[i*n+j+14] = tmp15; if(j_add > 14) C[i*n+j+15] = tmp16; } } /* BEGIN GPU 5 ############################################################################## */ #define BLOCK_SIZE 16 typedef struct { int width; int height; int stride; double* elements; } Matrix; // Get a matrix element __device__ double GetElement(const Matrix A, int row, int col) { return A.elements[row * A.stride + col]; } // Set a matrix element __device__ void SetElement(Matrix A, int row, int col, double value) { A.elements[row * A.stride + col] = value; } // Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is // located col sub-matrices to the right and row sub-matrices down // from the upper-left corner of A __device__ Matrix GetSubMatrix(Matrix A, int row, int col) { Matrix Asub; Asub.width = BLOCK_SIZE; Asub.height = BLOCK_SIZE; Asub.stride = A.stride; Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col]; return Asub; } // Matrix multiplication kernel called by MatMul() __global__ void gpu5_kernel(const Matrix A, const Matrix B, Matrix C) { // Block row and column int blockRow = blockIdx.y; int blockCol = blockIdx.x; // Each thread block computes one sub-matrix Csub of C Matrix Csub = GetSubMatrix(C, blockRow, blockCol); // Each thread computes one element of Csub // by accumulating results into Cvalue double Cvalue = 0; // Thread row and column within Csub int row = threadIdx.y; int col = threadIdx.x; // Loop over all the sub-matrices of A and B that are // required to compute Csub // Multiply each pair of sub-matrices together // and accumulate the results for (int m = 0; m < (A.width / BLOCK_SIZE); ++m) { // Get sub-matrix Asub of A Matrix Asub = GetSubMatrix(A, blockRow, m); // Get sub-matrix Bsub of B Matrix Bsub = GetSubMatrix(B, m, blockCol); // Shared memory used to store Asub and Bsub respectively __shared__ double As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ double Bs[BLOCK_SIZE][BLOCK_SIZE]; // Load Asub and Bsub from device memory to shared memory // Each thread loads one element of each sub-matrix As[row][col] = GetElement(Asub, row, col); Bs[row][col] = GetElement(Bsub, row, col); // Synchronize to make sure the sub-matrices are loaded // before starting the computation __syncthreads(); // Multiply Asub and Bsub together for (int e = 0; e < BLOCK_SIZE; ++e) Cvalue += As[row][e] * Bs[e][col]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } // Write Csub to device memory // Each thread writes one element SetElement(Csub, row, col, Cvalue); }
.file "tmpxft_00152ae4_00000000-6_matmult_kernels.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2062: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2062: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z10GetElement6Matrixii .type _Z10GetElement6Matrixii, @function _Z10GetElement6Matrixii: .LFB2057: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2057: .size _Z10GetElement6Matrixii, .-_Z10GetElement6Matrixii .globl _Z10SetElement6Matrixiid .type _Z10SetElement6Matrixiid, @function _Z10SetElement6Matrixiid: .LFB2058: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2058: .size _Z10SetElement6Matrixiid, .-_Z10SetElement6Matrixiid .globl _Z12GetSubMatrix6Matrixii .type _Z12GetSubMatrix6Matrixii, @function _Z12GetSubMatrix6Matrixii: .LFB2059: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2059: .size _Z12GetSubMatrix6Matrixii, .-_Z12GetSubMatrix6Matrixii .globl _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_: .LFB2084: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L13 .L9: movq 168(%rsp), %rax subq %fs:40, %rax jne .L14 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L13: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel1iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L9 .L14: call __stack_chk_fail@PLT .cfi_endproc .LFE2084: .size _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel1iiiPdS_S_ .type _Z15matmult_kernel1iiiPdS_S_, @function _Z15matmult_kernel1iiiPdS_S_: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _Z15matmult_kernel1iiiPdS_S_, .-_Z15matmult_kernel1iiiPdS_S_ .globl _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_: .LFB2086: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L21 .L17: movq 168(%rsp), %rax subq %fs:40, %rax jne .L22 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L21: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel2iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L17 .L22: call __stack_chk_fail@PLT .cfi_endproc .LFE2086: .size _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel2iiiPdS_S_ .type _Z15matmult_kernel2iiiPdS_S_, @function _Z15matmult_kernel2iiiPdS_S_: .LFB2087: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2087: .size _Z15matmult_kernel2iiiPdS_S_, .-_Z15matmult_kernel2iiiPdS_S_ .globl _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_: .LFB2088: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L29 .L25: movq 168(%rsp), %rax subq %fs:40, %rax jne .L30 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L29: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel3iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L25 .L30: call __stack_chk_fail@PLT .cfi_endproc .LFE2088: .size _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel3iiiPdS_S_ .type _Z15matmult_kernel3iiiPdS_S_, @function _Z15matmult_kernel3iiiPdS_S_: .LFB2089: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2089: .size _Z15matmult_kernel3iiiPdS_S_, .-_Z15matmult_kernel3iiiPdS_S_ .globl _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i .type _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i, @function _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i: .LFB2090: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) leaq 192(%rsp), %rax movq %rax, 160(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L37 .L33: movq 168(%rsp), %rax subq %fs:40, %rax jne .L38 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L37: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel4iiiPdS_S_i(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L33 .L38: call __stack_chk_fail@PLT .cfi_endproc .LFE2090: .size _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i, .-_Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i .globl _Z15matmult_kernel4iiiPdS_S_i .type _Z15matmult_kernel4iiiPdS_S_i, @function _Z15matmult_kernel4iiiPdS_S_i: .LFB2091: .cfi_startproc endbr64 subq $16, %rsp .cfi_def_cfa_offset 24 movl 24(%rsp), %eax pushq %rax .cfi_def_cfa_offset 32 call _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i addq $24, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2091: .size _Z15matmult_kernel4iiiPdS_S_i, .-_Z15matmult_kernel4iiiPdS_S_i .globl _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ .type _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_, @function _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_: .LFB2092: .cfi_startproc endbr64 subq $104, %rsp .cfi_def_cfa_offset 112 movq %fs:40, %rax movq %rax, 88(%rsp) xorl %eax, %eax movq %rdi, 64(%rsp) movq %rsi, 72(%rsp) movq %rdx, 80(%rsp) movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L45 .L41: movq 88(%rsp), %rax subq %fs:40, %rax jne .L46 addq $104, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L45: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 120 pushq 8(%rsp) .cfi_def_cfa_offset 128 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z11gpu5_kernel6MatrixS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 112 jmp .L41 .L46: call __stack_chk_fail@PLT .cfi_endproc .LFE2092: .size _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_, .-_Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ .globl _Z11gpu5_kernel6MatrixS_S_ .type _Z11gpu5_kernel6MatrixS_S_, @function _Z11gpu5_kernel6MatrixS_S_: .LFB2093: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq 64(%rsp), %rdx leaq 40(%rsp), %rsi leaq 16(%rsp), %rdi call _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2093: .size _Z11gpu5_kernel6MatrixS_S_, .-_Z11gpu5_kernel6MatrixS_S_ .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "_Z11gpu5_kernel6MatrixS_S_" .LC1: .string "_Z15matmult_kernel4iiiPdS_S_i" .LC2: .string "_Z15matmult_kernel3iiiPdS_S_" .LC3: .string "_Z15matmult_kernel2iiiPdS_S_" .LC4: .string "_Z15matmult_kernel1iiiPdS_S_" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2095: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rbx movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z11gpu5_kernel6MatrixS_S_(%rip), %rsi movq %rax, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC1(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel4iiiPdS_S_i(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC2(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel3iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC3(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel2iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel1iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2095: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include <stdio.h> #define MIN(x, y) (((x) < (y)) ? (x) : (y)) __global__ void matmult_kernel1(int m, int n, int k, double *A, double *B, double *C){ // set C to zeros for (int i=0;i<m;i++){ for (int p=0;p<n;p++){ C[i*n+p]=0; //C[i][p] } } // do matmult with mkn loop order for (int i=0;i<m;i++) { for (int j=0;j<k;j++){ for (int p=0;p<n;p++) { C[i*n+p] += A[i*k+j] * B[j*n+p]; } } } } __global__ void matmult_kernel2(int m, int n, int k, double *A, double *B, double *C){ int j = blockIdx.x*blockDim.x+threadIdx.x; //looping through n int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp; if((j<n)&&(i<m)){ for(int p=0; p<k; p++){ //read row of A and col of B //row of A is A[mit*k+kit] //col of B is B[kit*n+nit] tmp += A[i*k+p] * B[p*n+j]; } //C is C[mit*n+nit] C[i*n+j] = tmp; } } __global__ void matmult_kernel3(int m, int n, int k, double *A, double *B, double *C){ //compute C(i,j) and C(i,j+1) int j = 2*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only half as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, either 1 or 0) int j_add = MIN(1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j]; if(j_add == 1) tmp2 += A[i*k+p] * B[p*n+j+1]; } C[i*n+j] = tmp1; if(j_add == 1) C[i*n+j+1] = tmp2; } } __global__ void matmult_kernel4(int m, int n, int k, double *A, double *B, double *C, int s){ //compute C(i,j), C(i,j+1), ... C(i,j+s) int j = s*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only 1/s as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; double tmp3 = 0.0; double tmp4 = 0.0; double tmp5 = 0.0; double tmp6 = 0.0; double tmp7 = 0.0; double tmp8 = 0.0; double tmp9 = 0.0; double tmp10 = 0.0; double tmp11 = 0.0; double tmp12 = 0.0; double tmp13 = 0.0; double tmp14 = 0.0; double tmp15 = 0.0; double tmp16 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, from 0 to s-1) int j_add = MIN(s-1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j+0]; if(j_add > 0) tmp2 += A[i*k+p] * B[p*n+j+1]; if(j_add > 1) tmp3 += A[i*k+p] * B[p*n+j+2]; if(j_add > 2) tmp4 += A[i*k+p] * B[p*n+j+3]; if(j_add > 3) tmp5 += A[i*k+p] * B[p*n+j+4]; if(j_add > 4) tmp6 += A[i*k+p] * B[p*n+j+5]; if(j_add > 5) tmp7 += A[i*k+p] * B[p*n+j+6]; if(j_add > 6) tmp8 += A[i*k+p] * B[p*n+j+7]; if(j_add > 7) tmp9 += A[i*k+p] * B[p*n+j+8]; if(j_add > 8) tmp10 += A[i*k+p] * B[p*n+j+9]; if(j_add > 9) tmp11 += A[i*k+p] * B[p*n+j+10]; if(j_add > 10) tmp12 += A[i*k+p] * B[p*n+j+11]; if(j_add > 11) tmp13 += A[i*k+p] * B[p*n+j+12]; if(j_add > 12) tmp14 += A[i*k+p] * B[p*n+j+13]; if(j_add > 13) tmp15 += A[i*k+p] * B[p*n+j+14]; if(j_add > 14) tmp16 += A[i*k+p] * B[p*n+j+15]; } C[i*n+j] = tmp1; if(j_add > 0) C[i*n+j+1] = tmp2; if(j_add > 1) C[i*n+j+2] = tmp3; if(j_add > 2) C[i*n+j+3] = tmp4; if(j_add > 3) C[i*n+j+4] = tmp5; if(j_add > 4) C[i*n+j+5] = tmp6; if(j_add > 5) C[i*n+j+6] = tmp7; if(j_add > 6) C[i*n+j+7] = tmp8; if(j_add > 7) C[i*n+j+8] = tmp9; if(j_add > 8) C[i*n+j+9] = tmp10; if(j_add > 9) C[i*n+j+10] = tmp11; if(j_add > 10) C[i*n+j+11] = tmp12; if(j_add > 11) C[i*n+j+12] = tmp13; if(j_add > 12) C[i*n+j+13] = tmp14; if(j_add > 13) C[i*n+j+14] = tmp15; if(j_add > 14) C[i*n+j+15] = tmp16; } } /* BEGIN GPU 5 ############################################################################## */ #define BLOCK_SIZE 16 typedef struct { int width; int height; int stride; double* elements; } Matrix; // Get a matrix element __device__ double GetElement(const Matrix A, int row, int col) { return A.elements[row * A.stride + col]; } // Set a matrix element __device__ void SetElement(Matrix A, int row, int col, double value) { A.elements[row * A.stride + col] = value; } // Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is // located col sub-matrices to the right and row sub-matrices down // from the upper-left corner of A __device__ Matrix GetSubMatrix(Matrix A, int row, int col) { Matrix Asub; Asub.width = BLOCK_SIZE; Asub.height = BLOCK_SIZE; Asub.stride = A.stride; Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col]; return Asub; } // Matrix multiplication kernel called by MatMul() __global__ void gpu5_kernel(const Matrix A, const Matrix B, Matrix C) { // Block row and column int blockRow = blockIdx.y; int blockCol = blockIdx.x; // Each thread block computes one sub-matrix Csub of C Matrix Csub = GetSubMatrix(C, blockRow, blockCol); // Each thread computes one element of Csub // by accumulating results into Cvalue double Cvalue = 0; // Thread row and column within Csub int row = threadIdx.y; int col = threadIdx.x; // Loop over all the sub-matrices of A and B that are // required to compute Csub // Multiply each pair of sub-matrices together // and accumulate the results for (int m = 0; m < (A.width / BLOCK_SIZE); ++m) { // Get sub-matrix Asub of A Matrix Asub = GetSubMatrix(A, blockRow, m); // Get sub-matrix Bsub of B Matrix Bsub = GetSubMatrix(B, m, blockCol); // Shared memory used to store Asub and Bsub respectively __shared__ double As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ double Bs[BLOCK_SIZE][BLOCK_SIZE]; // Load Asub and Bsub from device memory to shared memory // Each thread loads one element of each sub-matrix As[row][col] = GetElement(Asub, row, col); Bs[row][col] = GetElement(Bsub, row, col); // Synchronize to make sure the sub-matrices are loaded // before starting the computation __syncthreads(); // Multiply Asub and Bsub together for (int e = 0; e < BLOCK_SIZE; ++e) Cvalue += As[row][e] * Bs[e][col]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } // Write Csub to device memory // Each thread writes one element SetElement(Csub, row, col, Cvalue); }
#include <hip/hip_runtime.h> #include <stdio.h> #define MIN(x, y) (((x) < (y)) ? (x) : (y)) __global__ void matmult_kernel1(int m, int n, int k, double *A, double *B, double *C){ // set C to zeros for (int i=0;i<m;i++){ for (int p=0;p<n;p++){ C[i*n+p]=0; //C[i][p] } } // do matmult with mkn loop order for (int i=0;i<m;i++) { for (int j=0;j<k;j++){ for (int p=0;p<n;p++) { C[i*n+p] += A[i*k+j] * B[j*n+p]; } } } } __global__ void matmult_kernel2(int m, int n, int k, double *A, double *B, double *C){ int j = blockIdx.x*blockDim.x+threadIdx.x; //looping through n int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp; if((j<n)&&(i<m)){ for(int p=0; p<k; p++){ //read row of A and col of B //row of A is A[mit*k+kit] //col of B is B[kit*n+nit] tmp += A[i*k+p] * B[p*n+j]; } //C is C[mit*n+nit] C[i*n+j] = tmp; } } __global__ void matmult_kernel3(int m, int n, int k, double *A, double *B, double *C){ //compute C(i,j) and C(i,j+1) int j = 2*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only half as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, either 1 or 0) int j_add = MIN(1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j]; if(j_add == 1) tmp2 += A[i*k+p] * B[p*n+j+1]; } C[i*n+j] = tmp1; if(j_add == 1) C[i*n+j+1] = tmp2; } } __global__ void matmult_kernel4(int m, int n, int k, double *A, double *B, double *C, int s){ //compute C(i,j), C(i,j+1), ... C(i,j+s) int j = s*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only 1/s as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; double tmp3 = 0.0; double tmp4 = 0.0; double tmp5 = 0.0; double tmp6 = 0.0; double tmp7 = 0.0; double tmp8 = 0.0; double tmp9 = 0.0; double tmp10 = 0.0; double tmp11 = 0.0; double tmp12 = 0.0; double tmp13 = 0.0; double tmp14 = 0.0; double tmp15 = 0.0; double tmp16 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, from 0 to s-1) int j_add = MIN(s-1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j+0]; if(j_add > 0) tmp2 += A[i*k+p] * B[p*n+j+1]; if(j_add > 1) tmp3 += A[i*k+p] * B[p*n+j+2]; if(j_add > 2) tmp4 += A[i*k+p] * B[p*n+j+3]; if(j_add > 3) tmp5 += A[i*k+p] * B[p*n+j+4]; if(j_add > 4) tmp6 += A[i*k+p] * B[p*n+j+5]; if(j_add > 5) tmp7 += A[i*k+p] * B[p*n+j+6]; if(j_add > 6) tmp8 += A[i*k+p] * B[p*n+j+7]; if(j_add > 7) tmp9 += A[i*k+p] * B[p*n+j+8]; if(j_add > 8) tmp10 += A[i*k+p] * B[p*n+j+9]; if(j_add > 9) tmp11 += A[i*k+p] * B[p*n+j+10]; if(j_add > 10) tmp12 += A[i*k+p] * B[p*n+j+11]; if(j_add > 11) tmp13 += A[i*k+p] * B[p*n+j+12]; if(j_add > 12) tmp14 += A[i*k+p] * B[p*n+j+13]; if(j_add > 13) tmp15 += A[i*k+p] * B[p*n+j+14]; if(j_add > 14) tmp16 += A[i*k+p] * B[p*n+j+15]; } C[i*n+j] = tmp1; if(j_add > 0) C[i*n+j+1] = tmp2; if(j_add > 1) C[i*n+j+2] = tmp3; if(j_add > 2) C[i*n+j+3] = tmp4; if(j_add > 3) C[i*n+j+4] = tmp5; if(j_add > 4) C[i*n+j+5] = tmp6; if(j_add > 5) C[i*n+j+6] = tmp7; if(j_add > 6) C[i*n+j+7] = tmp8; if(j_add > 7) C[i*n+j+8] = tmp9; if(j_add > 8) C[i*n+j+9] = tmp10; if(j_add > 9) C[i*n+j+10] = tmp11; if(j_add > 10) C[i*n+j+11] = tmp12; if(j_add > 11) C[i*n+j+12] = tmp13; if(j_add > 12) C[i*n+j+13] = tmp14; if(j_add > 13) C[i*n+j+14] = tmp15; if(j_add > 14) C[i*n+j+15] = tmp16; } } /* BEGIN GPU 5 ############################################################################## */ #define BLOCK_SIZE 16 typedef struct { int width; int height; int stride; double* elements; } Matrix; // Get a matrix element __device__ double GetElement(const Matrix A, int row, int col) { return A.elements[row * A.stride + col]; } // Set a matrix element __device__ void SetElement(Matrix A, int row, int col, double value) { A.elements[row * A.stride + col] = value; } // Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is // located col sub-matrices to the right and row sub-matrices down // from the upper-left corner of A __device__ Matrix GetSubMatrix(Matrix A, int row, int col) { Matrix Asub; Asub.width = BLOCK_SIZE; Asub.height = BLOCK_SIZE; Asub.stride = A.stride; Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col]; return Asub; } // Matrix multiplication kernel called by MatMul() __global__ void gpu5_kernel(const Matrix A, const Matrix B, Matrix C) { // Block row and column int blockRow = blockIdx.y; int blockCol = blockIdx.x; // Each thread block computes one sub-matrix Csub of C Matrix Csub = GetSubMatrix(C, blockRow, blockCol); // Each thread computes one element of Csub // by accumulating results into Cvalue double Cvalue = 0; // Thread row and column within Csub int row = threadIdx.y; int col = threadIdx.x; // Loop over all the sub-matrices of A and B that are // required to compute Csub // Multiply each pair of sub-matrices together // and accumulate the results for (int m = 0; m < (A.width / BLOCK_SIZE); ++m) { // Get sub-matrix Asub of A Matrix Asub = GetSubMatrix(A, blockRow, m); // Get sub-matrix Bsub of B Matrix Bsub = GetSubMatrix(B, m, blockCol); // Shared memory used to store Asub and Bsub respectively __shared__ double As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ double Bs[BLOCK_SIZE][BLOCK_SIZE]; // Load Asub and Bsub from device memory to shared memory // Each thread loads one element of each sub-matrix As[row][col] = GetElement(Asub, row, col); Bs[row][col] = GetElement(Bsub, row, col); // Synchronize to make sure the sub-matrices are loaded // before starting the computation __syncthreads(); // Multiply Asub and Bsub together for (int e = 0; e < BLOCK_SIZE; ++e) Cvalue += As[row][e] * Bs[e][col]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } // Write Csub to device memory // Each thread writes one element SetElement(Csub, row, col, Cvalue); }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include <stdio.h> #define MIN(x, y) (((x) < (y)) ? (x) : (y)) __global__ void matmult_kernel1(int m, int n, int k, double *A, double *B, double *C){ // set C to zeros for (int i=0;i<m;i++){ for (int p=0;p<n;p++){ C[i*n+p]=0; //C[i][p] } } // do matmult with mkn loop order for (int i=0;i<m;i++) { for (int j=0;j<k;j++){ for (int p=0;p<n;p++) { C[i*n+p] += A[i*k+j] * B[j*n+p]; } } } } __global__ void matmult_kernel2(int m, int n, int k, double *A, double *B, double *C){ int j = blockIdx.x*blockDim.x+threadIdx.x; //looping through n int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp; if((j<n)&&(i<m)){ for(int p=0; p<k; p++){ //read row of A and col of B //row of A is A[mit*k+kit] //col of B is B[kit*n+nit] tmp += A[i*k+p] * B[p*n+j]; } //C is C[mit*n+nit] C[i*n+j] = tmp; } } __global__ void matmult_kernel3(int m, int n, int k, double *A, double *B, double *C){ //compute C(i,j) and C(i,j+1) int j = 2*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only half as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, either 1 or 0) int j_add = MIN(1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j]; if(j_add == 1) tmp2 += A[i*k+p] * B[p*n+j+1]; } C[i*n+j] = tmp1; if(j_add == 1) C[i*n+j+1] = tmp2; } } __global__ void matmult_kernel4(int m, int n, int k, double *A, double *B, double *C, int s){ //compute C(i,j), C(i,j+1), ... C(i,j+s) int j = s*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only 1/s as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; double tmp3 = 0.0; double tmp4 = 0.0; double tmp5 = 0.0; double tmp6 = 0.0; double tmp7 = 0.0; double tmp8 = 0.0; double tmp9 = 0.0; double tmp10 = 0.0; double tmp11 = 0.0; double tmp12 = 0.0; double tmp13 = 0.0; double tmp14 = 0.0; double tmp15 = 0.0; double tmp16 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, from 0 to s-1) int j_add = MIN(s-1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j+0]; if(j_add > 0) tmp2 += A[i*k+p] * B[p*n+j+1]; if(j_add > 1) tmp3 += A[i*k+p] * B[p*n+j+2]; if(j_add > 2) tmp4 += A[i*k+p] * B[p*n+j+3]; if(j_add > 3) tmp5 += A[i*k+p] * B[p*n+j+4]; if(j_add > 4) tmp6 += A[i*k+p] * B[p*n+j+5]; if(j_add > 5) tmp7 += A[i*k+p] * B[p*n+j+6]; if(j_add > 6) tmp8 += A[i*k+p] * B[p*n+j+7]; if(j_add > 7) tmp9 += A[i*k+p] * B[p*n+j+8]; if(j_add > 8) tmp10 += A[i*k+p] * B[p*n+j+9]; if(j_add > 9) tmp11 += A[i*k+p] * B[p*n+j+10]; if(j_add > 10) tmp12 += A[i*k+p] * B[p*n+j+11]; if(j_add > 11) tmp13 += A[i*k+p] * B[p*n+j+12]; if(j_add > 12) tmp14 += A[i*k+p] * B[p*n+j+13]; if(j_add > 13) tmp15 += A[i*k+p] * B[p*n+j+14]; if(j_add > 14) tmp16 += A[i*k+p] * B[p*n+j+15]; } C[i*n+j] = tmp1; if(j_add > 0) C[i*n+j+1] = tmp2; if(j_add > 1) C[i*n+j+2] = tmp3; if(j_add > 2) C[i*n+j+3] = tmp4; if(j_add > 3) C[i*n+j+4] = tmp5; if(j_add > 4) C[i*n+j+5] = tmp6; if(j_add > 5) C[i*n+j+6] = tmp7; if(j_add > 6) C[i*n+j+7] = tmp8; if(j_add > 7) C[i*n+j+8] = tmp9; if(j_add > 8) C[i*n+j+9] = tmp10; if(j_add > 9) C[i*n+j+10] = tmp11; if(j_add > 10) C[i*n+j+11] = tmp12; if(j_add > 11) C[i*n+j+12] = tmp13; if(j_add > 12) C[i*n+j+13] = tmp14; if(j_add > 13) C[i*n+j+14] = tmp15; if(j_add > 14) C[i*n+j+15] = tmp16; } } /* BEGIN GPU 5 ############################################################################## */ #define BLOCK_SIZE 16 typedef struct { int width; int height; int stride; double* elements; } Matrix; // Get a matrix element __device__ double GetElement(const Matrix A, int row, int col) { return A.elements[row * A.stride + col]; } // Set a matrix element __device__ void SetElement(Matrix A, int row, int col, double value) { A.elements[row * A.stride + col] = value; } // Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is // located col sub-matrices to the right and row sub-matrices down // from the upper-left corner of A __device__ Matrix GetSubMatrix(Matrix A, int row, int col) { Matrix Asub; Asub.width = BLOCK_SIZE; Asub.height = BLOCK_SIZE; Asub.stride = A.stride; Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col]; return Asub; } // Matrix multiplication kernel called by MatMul() __global__ void gpu5_kernel(const Matrix A, const Matrix B, Matrix C) { // Block row and column int blockRow = blockIdx.y; int blockCol = blockIdx.x; // Each thread block computes one sub-matrix Csub of C Matrix Csub = GetSubMatrix(C, blockRow, blockCol); // Each thread computes one element of Csub // by accumulating results into Cvalue double Cvalue = 0; // Thread row and column within Csub int row = threadIdx.y; int col = threadIdx.x; // Loop over all the sub-matrices of A and B that are // required to compute Csub // Multiply each pair of sub-matrices together // and accumulate the results for (int m = 0; m < (A.width / BLOCK_SIZE); ++m) { // Get sub-matrix Asub of A Matrix Asub = GetSubMatrix(A, blockRow, m); // Get sub-matrix Bsub of B Matrix Bsub = GetSubMatrix(B, m, blockCol); // Shared memory used to store Asub and Bsub respectively __shared__ double As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ double Bs[BLOCK_SIZE][BLOCK_SIZE]; // Load Asub and Bsub from device memory to shared memory // Each thread loads one element of each sub-matrix As[row][col] = GetElement(Asub, row, col); Bs[row][col] = GetElement(Bsub, row, col); // Synchronize to make sure the sub-matrices are loaded // before starting the computation __syncthreads(); // Multiply Asub and Bsub together for (int e = 0; e < BLOCK_SIZE; ++e) Cvalue += As[row][e] * Bs[e][col]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } // Write Csub to device memory // Each thread writes one element SetElement(Csub, row, col, Cvalue); }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z15matmult_kernel1iiiPdS_S_ .globl _Z15matmult_kernel1iiiPdS_S_ .p2align 8 .type _Z15matmult_kernel1iiiPdS_S_,@function _Z15matmult_kernel1iiiPdS_S_: s_clause 0x1 s_load_b64 s[8:9], s[0:1], 0x0 s_load_b64 s[10:11], s[0:1], 0x20 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s8, 1 s_cbranch_scc1 .LBB0_6 s_cmp_gt_i32 s9, 0 s_mov_b32 s5, 0 s_cselect_b32 s2, -1, 0 s_mov_b32 s4, s5 v_cndmask_b32_e64 v0, 0, 1, s2 s_mov_b32 s3, s5 v_mov_b32_e32 v1, 0 s_delay_alu instid0(VALU_DEP_2) v_cmp_ne_u32_e64 s2, 1, v0 v_mov_b32_e32 v0, 0 s_branch .LBB0_3 .p2align 6 .LBB0_2: s_add_i32 s3, s3, 1 s_add_i32 s4, s4, s9 s_cmp_eq_u32 s3, s8 s_cbranch_scc1 .LBB0_6 .LBB0_3: s_delay_alu instid0(VALU_DEP_2) s_and_b32 vcc_lo, exec_lo, s2 s_cbranch_vccnz .LBB0_2 s_lshl_b64 s[6:7], s[4:5], 3 s_mov_b32 s12, s9 s_add_u32 s6, s10, s6 s_addc_u32 s7, s11, s7 .LBB0_5: v_mov_b32_e32 v2, v1 s_add_i32 s12, s12, -1 global_store_b64 v0, v[1:2], s[6:7] s_add_u32 s6, s6, 8 s_addc_u32 s7, s7, 0 s_cmp_eq_u32 s12, 0 s_cbranch_scc0 .LBB0_5 s_branch .LBB0_2 .LBB0_6: s_cmp_lt_i32 s8, 1 s_cbranch_scc1 .LBB0_15 s_clause 0x1 s_load_b32 s22, s[0:1], 0x8 s_load_b128 s[4:7], s[0:1], 0x10 s_mov_b32 s3, 0 s_mov_b32 s23, 0 s_mov_b32 s12, s3 s_waitcnt lgkmcnt(0) s_cmp_gt_i32 s22, 0 s_cselect_b32 s0, -1, 0 s_cmp_gt_i32 s9, 0 v_cndmask_b32_e64 v0, 0, 1, s0 s_cselect_b32 s0, -1, 0 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_cndmask_b32_e64 v1, 0, 1, s0 v_cmp_ne_u32_e64 s0, 1, v0 v_mov_b32_e32 v0, 0 s_delay_alu instid0(VALU_DEP_3) v_cmp_ne_u32_e64 s1, 1, v1 s_set_inst_prefetch_distance 0x1 s_branch .LBB0_9 .p2align 6 .LBB0_8: s_add_i32 s23, s23, 1 s_add_i32 s12, s12, s9 s_cmp_eq_u32 s23, s8 s_cbranch_scc1 .LBB0_15 .LBB0_9: s_and_b32 vcc_lo, exec_lo, s0 s_cbranch_vccnz .LBB0_8 s_mov_b32 s13, s3 s_mov_b32 s2, 0 s_lshl_b64 s[14:15], s[12:13], 3 s_mul_i32 s24, s23, s22 s_add_u32 s14, s10, s14 s_addc_u32 s15, s11, s15 s_mov_b32 s13, s2 s_branch .LBB0_12 .p2align 6 .LBB0_11: s_add_i32 s13, s13, 1 s_add_i32 s2, s2, s9 s_cmp_eq_u32 s13, s22 s_cbranch_scc1 .LBB0_8 .LBB0_12: s_and_b32 vcc_lo, exec_lo, s1 s_cbranch_vccnz .LBB0_11 s_lshl_b64 s[16:17], s[2:3], 3 s_mov_b32 s19, s3 s_add_u32 s16, s6, s16 s_addc_u32 s17, s7, s17 s_add_i32 s18, s13, s24 s_mov_b32 s25, s9 s_lshl_b64 s[20:21], s[18:19], 3 s_mov_b64 s[18:19], s[14:15] s_add_u32 s20, s4, s20 s_addc_u32 s21, s5, s21 .p2align 6 .LBB0_14: s_clause 0x2 global_load_b64 v[1:2], v0, s[20:21] global_load_b64 v[3:4], v0, s[16:17] global_load_b64 v[5:6], v0, s[18:19] s_add_i32 s25, s25, -1 s_waitcnt vmcnt(0) v_fma_f64 v[1:2], v[1:2], v[3:4], v[5:6] global_store_b64 v0, v[1:2], s[18:19] s_add_u32 s18, s18, 8 s_addc_u32 s19, s19, 0 s_add_u32 s16, s16, 8 s_addc_u32 s17, s17, 0 s_cmp_eq_u32 s25, 0 s_cbranch_scc0 .LBB0_14 s_branch .LBB0_11 .LBB0_15: s_set_inst_prefetch_distance 0x2 s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z15matmult_kernel1iiiPdS_S_ .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 40 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 7 .amdhsa_next_free_sgpr 26 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z15matmult_kernel1iiiPdS_S_, .Lfunc_end0-_Z15matmult_kernel1iiiPdS_S_ .section .AMDGPU.csdata,"",@progbits .text .protected _Z15matmult_kernel2iiiPdS_S_ .globl _Z15matmult_kernel2iiiPdS_S_ .p2align 8 .type _Z15matmult_kernel2iiiPdS_S_,@function _Z15matmult_kernel2iiiPdS_S_: s_clause 0x1 s_load_b32 s4, s[0:1], 0x34 s_load_b64 s[2:3], s[0:1], 0x0 v_and_b32_e32 v2, 0x3ff, v0 v_bfe_u32 v3, v0, 10, 10 s_waitcnt lgkmcnt(0) s_and_b32 s5, s4, 0xffff s_lshr_b32 s4, s4, 16 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_mad_u64_u32 v[0:1], null, s14, s5, v[2:3] v_mad_u64_u32 v[1:2], null, s15, s4, v[3:4] v_cmp_gt_i32_e32 vcc_lo, s3, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_cmp_gt_i32_e64 s2, s2, v1 s_and_b32 s2, vcc_lo, s2 s_delay_alu instid0(SALU_CYCLE_1) s_and_saveexec_b32 s4, s2 s_cbranch_execz .LBB1_6 s_load_b32 s2, s[0:1], 0x8 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s2, 1 s_cbranch_scc1 .LBB1_4 s_load_b128 s[4:7], s[0:1], 0x10 v_mul_lo_u32 v2, v1, s2 v_mov_b32_e32 v6, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_ashrrev_i32_e32 v3, 31, v2 v_lshlrev_b64 v[2:3], 3, v[2:3] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v4, vcc_lo, s4, v2 v_add_co_ci_u32_e32 v5, vcc_lo, s5, v3, vcc_lo .p2align 6 .LBB1_3: v_ashrrev_i32_e32 v7, 31, v6 s_add_i32 s2, s2, -1 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) s_cmp_eq_u32 s2, 0 v_lshlrev_b64 v[7:8], 3, v[6:7] v_add_nc_u32_e32 v6, s3, v6 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) v_add_co_u32 v7, vcc_lo, s6, v7 v_add_co_ci_u32_e32 v8, vcc_lo, s7, v8, vcc_lo global_load_b64 v[9:10], v[4:5], off global_load_b64 v[7:8], v[7:8], off v_add_co_u32 v4, vcc_lo, v4, 8 v_add_co_ci_u32_e32 v5, vcc_lo, 0, v5, vcc_lo s_waitcnt vmcnt(0) v_fma_f64 v[2:3], v[9:10], v[7:8], v[2:3] s_cbranch_scc0 .LBB1_3 s_branch .LBB1_5 .LBB1_4: .LBB1_5: s_load_b64 s[0:1], s[0:1], 0x20 v_mad_u64_u32 v[4:5], null, v1, s3, v[0:1] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_ashrrev_i32_e32 v5, 31, v4 v_lshlrev_b64 v[0:1], 3, v[4:5] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v0, vcc_lo, s0, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo global_store_b64 v[0:1], v[2:3], off .LBB1_6: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z15matmult_kernel2iiiPdS_S_ .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 296 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 11 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end1: .size _Z15matmult_kernel2iiiPdS_S_, .Lfunc_end1-_Z15matmult_kernel2iiiPdS_S_ .section .AMDGPU.csdata,"",@progbits .text .protected _Z15matmult_kernel3iiiPdS_S_ .globl _Z15matmult_kernel3iiiPdS_S_ .p2align 8 .type _Z15matmult_kernel3iiiPdS_S_,@function _Z15matmult_kernel3iiiPdS_S_: s_clause 0x1 s_load_b32 s4, s[0:1], 0x34 s_load_b64 s[2:3], s[0:1], 0x0 v_and_b32_e32 v3, 0x3ff, v0 v_bfe_u32 v2, v0, 10, 10 s_waitcnt lgkmcnt(0) s_and_b32 s5, s4, 0xffff s_lshr_b32 s4, s4, 16 s_mul_i32 s14, s14, s5 v_mad_u64_u32 v[0:1], null, s15, s4, v[2:3] v_add_lshl_u32 v1, s14, v3, 1 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_cmp_gt_i32_e32 vcc_lo, s3, v1 v_cmp_gt_i32_e64 s2, s2, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_and_b32 s2, vcc_lo, s2 s_and_saveexec_b32 s4, s2 s_cbranch_execz .LBB2_8 s_load_b32 s6, s[0:1], 0x8 v_mov_b32_e32 v2, 0 v_mov_b32_e32 v3, 0 v_xad_u32 v14, v1, -1, s3 s_delay_alu instid0(VALU_DEP_2) v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s6, 1 s_cbranch_scc1 .LBB2_6 s_load_b128 s[8:11], s[0:1], 0x10 v_mul_lo_u32 v4, v0, s6 v_ashrrev_i32_e32 v2, 31, v1 v_cmp_lt_i32_e32 vcc_lo, 0, v14 s_ashr_i32 s5, s3, 31 s_mov_b32 s4, s3 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2) v_lshlrev_b64 v[6:7], 3, v[1:2] v_mov_b32_e32 v2, 0 v_ashrrev_i32_e32 v5, 31, v4 v_mov_b32_e32 v3, 0 s_lshl_b64 s[4:5], s[4:5], 3 v_lshlrev_b64 v[4:5], 3, v[4:5] s_waitcnt lgkmcnt(0) v_add_co_u32 v8, s2, v6, s10 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_add_co_ci_u32_e64 v9, s2, s11, v7, s2 v_add_co_u32 v6, s2, s8, v4 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4) v_add_co_ci_u32_e64 v7, s2, s9, v5, s2 v_add_co_u32 v8, s2, v8, 8 s_delay_alu instid0(VALU_DEP_1) v_add_co_ci_u32_e64 v9, s2, 0, v9, s2 v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2 s_set_inst_prefetch_distance 0x1 s_branch .LBB2_4 .p2align 6 .LBB2_3: s_or_b32 exec_lo, exec_lo, s2 s_waitcnt vmcnt(0) v_fma_f64 v[2:3], v[10:11], v[12:13], v[2:3] v_add_co_u32 v8, s2, v8, s4 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_add_co_ci_u32_e64 v9, s2, s5, v9, s2 v_add_co_u32 v6, s2, v6, 8 v_add_co_ci_u32_e64 v7, s2, 0, v7, s2 s_add_i32 s6, s6, -1 s_delay_alu instid0(SALU_CYCLE_1) s_cmp_eq_u32 s6, 0 s_cbranch_scc1 .LBB2_6 .LBB2_4: global_load_b64 v[10:11], v[6:7], off global_load_b64 v[12:13], v[8:9], off offset:-8 s_and_saveexec_b32 s2, vcc_lo s_cbranch_execz .LBB2_3 global_load_b64 v[15:16], v[8:9], off s_waitcnt vmcnt(0) v_fma_f64 v[4:5], v[10:11], v[15:16], v[4:5] s_branch .LBB2_3 .LBB2_6: s_set_inst_prefetch_distance 0x2 s_load_b64 s[0:1], s[0:1], 0x20 v_mad_u64_u32 v[6:7], null, v0, s3, v[1:2] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_ashrrev_i32_e32 v7, 31, v6 v_lshlrev_b64 v[0:1], 3, v[6:7] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v0, vcc_lo, s0, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo v_cmp_lt_i32_e32 vcc_lo, 0, v14 global_store_b64 v[0:1], v[2:3], off s_and_b32 exec_lo, exec_lo, vcc_lo s_cbranch_execz .LBB2_8 global_store_b64 v[0:1], v[4:5], off offset:8 .LBB2_8: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z15matmult_kernel3iiiPdS_S_ .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 296 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 17 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end2: .size _Z15matmult_kernel3iiiPdS_S_, .Lfunc_end2-_Z15matmult_kernel3iiiPdS_S_ .section .AMDGPU.csdata,"",@progbits .text .protected _Z15matmult_kernel4iiiPdS_S_i .globl _Z15matmult_kernel4iiiPdS_S_i .p2align 8 .type _Z15matmult_kernel4iiiPdS_S_i,@function _Z15matmult_kernel4iiiPdS_S_i: s_clause 0x1 s_load_b32 s2, s[0:1], 0x3c s_load_b32 s3, s[0:1], 0x28 v_and_b32_e32 v1, 0x3ff, v0 s_load_b64 s[16:17], s[0:1], 0x0 v_bfe_u32 v0, v0, 10, 10 s_waitcnt lgkmcnt(0) s_and_b32 s4, s2, 0xffff s_lshr_b32 s2, s2, 16 v_mad_u64_u32 v[2:3], null, s14, s4, v[1:2] v_mad_u64_u32 v[15:16], null, s15, s2, v[0:1] s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_mul_lo_u32 v12, v2, s3 v_cmp_gt_i32_e64 s2, s16, v15 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_cmp_gt_i32_e32 vcc_lo, s17, v12 s_and_b32 s2, vcc_lo, s2 s_delay_alu instid0(SALU_CYCLE_1) s_and_saveexec_b32 s4, s2 s_cbranch_execz .LBB3_64 s_load_b32 s24, s[0:1], 0x8 v_xad_u32 v0, v12, -1, s17 v_mov_b32_e32 v26, 0 s_add_i32 s2, s3, -1 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_cmp_gt_i32_e32 vcc_lo, s3, v0 v_dual_mov_b32 v27, 0 :: v_dual_cndmask_b32 v42, s2, v0 v_dual_mov_b32 v0, v26 :: v_dual_mov_b32 v1, v27 v_dual_mov_b32 v2, v26 :: v_dual_mov_b32 v3, v27 v_dual_mov_b32 v4, v26 :: v_dual_mov_b32 v5, v27 v_dual_mov_b32 v6, v26 :: v_dual_mov_b32 v7, v27 v_dual_mov_b32 v8, v26 :: v_dual_mov_b32 v9, v27 v_dual_mov_b32 v10, v26 :: v_dual_mov_b32 v11, v27 v_dual_mov_b32 v13, v26 :: v_dual_mov_b32 v14, v27 v_dual_mov_b32 v16, v26 :: v_dual_mov_b32 v17, v27 v_dual_mov_b32 v18, v26 :: v_dual_mov_b32 v19, v27 v_dual_mov_b32 v20, v26 :: v_dual_mov_b32 v21, v27 v_dual_mov_b32 v22, v26 :: v_dual_mov_b32 v23, v27 v_dual_mov_b32 v24, v26 :: v_dual_mov_b32 v25, v27 v_dual_mov_b32 v29, v27 :: v_dual_mov_b32 v28, v26 v_dual_mov_b32 v31, v27 :: v_dual_mov_b32 v30, v26 v_dual_mov_b32 v33, v27 :: v_dual_mov_b32 v32, v26 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s24, 1 s_cbranch_scc1 .LBB3_34 s_load_b128 s[20:23], s[0:1], 0x10 v_mul_lo_u32 v0, v15, s24 v_ashrrev_i32_e32 v13, 31, v12 v_mov_b32_e32 v26, 0 v_mov_b32_e32 v27, 0 v_cmp_lt_i32_e32 vcc_lo, 0, v42 v_cmp_lt_i32_e64 s2, 1, v42 v_lshlrev_b64 v[2:3], 3, v[12:13] v_cmp_lt_i32_e64 s3, 2, v42 v_ashrrev_i32_e32 v1, 31, v0 v_cmp_lt_i32_e64 s4, 3, v42 v_cmp_lt_i32_e64 s5, 4, v42 v_cmp_lt_i32_e64 s6, 5, v42 v_cmp_lt_i32_e64 s7, 6, v42 v_lshlrev_b64 v[0:1], 3, v[0:1] v_cmp_lt_i32_e64 s8, 7, v42 v_cmp_lt_i32_e64 s9, 8, v42 v_cmp_lt_i32_e64 s10, 9, v42 v_cmp_lt_i32_e64 s11, 10, v42 v_cmp_lt_i32_e64 s12, 11, v42 s_waitcnt lgkmcnt(0) v_add_co_u32 v2, s16, v2, s22 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_add_co_ci_u32_e64 v3, s16, s23, v3, s16 v_cmp_lt_i32_e64 s13, 12, v42 v_add_co_u32 v34, s16, v2, 0x78 s_delay_alu instid0(VALU_DEP_1) v_add_co_ci_u32_e64 v35, s16, 0, v3, s16 v_add_co_u32 v36, s16, s20, v0 v_cmp_lt_i32_e64 s14, 13, v42 v_cmp_lt_i32_e64 s15, 14, v42 v_add_co_ci_u32_e64 v37, s16, s21, v1, s16 v_dual_mov_b32 v0, v26 :: v_dual_mov_b32 v1, v27 v_dual_mov_b32 v2, v26 :: v_dual_mov_b32 v3, v27 v_dual_mov_b32 v4, v26 :: v_dual_mov_b32 v5, v27 v_dual_mov_b32 v6, v26 :: v_dual_mov_b32 v7, v27 v_dual_mov_b32 v8, v26 :: v_dual_mov_b32 v9, v27 v_dual_mov_b32 v10, v26 :: v_dual_mov_b32 v11, v27 v_dual_mov_b32 v13, v26 :: v_dual_mov_b32 v14, v27 v_dual_mov_b32 v16, v26 :: v_dual_mov_b32 v17, v27 v_dual_mov_b32 v18, v26 :: v_dual_mov_b32 v19, v27 v_dual_mov_b32 v20, v26 :: v_dual_mov_b32 v21, v27 v_dual_mov_b32 v22, v26 :: v_dual_mov_b32 v23, v27 v_dual_mov_b32 v24, v26 :: v_dual_mov_b32 v25, v27 v_dual_mov_b32 v29, v27 :: v_dual_mov_b32 v28, v26 v_dual_mov_b32 v31, v27 :: v_dual_mov_b32 v30, v26 v_dual_mov_b32 v33, v27 :: v_dual_mov_b32 v32, v26 s_ashr_i32 s19, s17, 31 s_mov_b32 s18, s17 s_delay_alu instid0(SALU_CYCLE_1) s_lshl_b64 s[18:19], s[18:19], 3 s_branch .LBB3_4 .LBB3_3: s_or_b32 exec_lo, exec_lo, s16 s_waitcnt vmcnt(0) v_fma_f64 v[26:27], v[38:39], v[40:41], v[26:27] v_add_co_u32 v34, s16, v34, s18 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_add_co_ci_u32_e64 v35, s16, s19, v35, s16 v_add_co_u32 v36, s16, v36, 8 v_add_co_ci_u32_e64 v37, s16, 0, v37, s16 s_add_i32 s24, s24, -1 s_delay_alu instid0(SALU_CYCLE_1) s_cmp_eq_u32 s24, 0 s_cbranch_scc1 .LBB3_34 .LBB3_4: global_load_b64 v[38:39], v[36:37], off global_load_b64 v[40:41], v[34:35], off offset:-120 s_and_saveexec_b32 s16, vcc_lo s_cbranch_execz .LBB3_6 global_load_b64 v[43:44], v[34:35], off offset:-112 s_waitcnt vmcnt(0) v_fma_f64 v[32:33], v[38:39], v[43:44], v[32:33] .LBB3_6: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s2 s_cbranch_execz .LBB3_8 global_load_b64 v[43:44], v[34:35], off offset:-104 s_waitcnt vmcnt(0) v_fma_f64 v[30:31], v[38:39], v[43:44], v[30:31] .LBB3_8: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s3 s_cbranch_execz .LBB3_10 global_load_b64 v[43:44], v[34:35], off offset:-96 s_waitcnt vmcnt(0) v_fma_f64 v[28:29], v[38:39], v[43:44], v[28:29] .LBB3_10: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s4 s_cbranch_execz .LBB3_12 global_load_b64 v[43:44], v[34:35], off offset:-88 s_waitcnt vmcnt(0) v_fma_f64 v[24:25], v[38:39], v[43:44], v[24:25] .LBB3_12: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s5 s_cbranch_execz .LBB3_14 global_load_b64 v[43:44], v[34:35], off offset:-80 s_waitcnt vmcnt(0) v_fma_f64 v[22:23], v[38:39], v[43:44], v[22:23] .LBB3_14: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s6 s_cbranch_execz .LBB3_16 global_load_b64 v[43:44], v[34:35], off offset:-72 s_waitcnt vmcnt(0) v_fma_f64 v[20:21], v[38:39], v[43:44], v[20:21] .LBB3_16: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s7 s_cbranch_execz .LBB3_18 global_load_b64 v[43:44], v[34:35], off offset:-64 s_waitcnt vmcnt(0) v_fma_f64 v[18:19], v[38:39], v[43:44], v[18:19] .LBB3_18: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s8 s_cbranch_execz .LBB3_20 global_load_b64 v[43:44], v[34:35], off offset:-56 s_waitcnt vmcnt(0) v_fma_f64 v[16:17], v[38:39], v[43:44], v[16:17] .LBB3_20: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s9 s_cbranch_execz .LBB3_22 global_load_b64 v[43:44], v[34:35], off offset:-48 s_waitcnt vmcnt(0) v_fma_f64 v[13:14], v[38:39], v[43:44], v[13:14] .LBB3_22: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s10 s_cbranch_execz .LBB3_24 global_load_b64 v[43:44], v[34:35], off offset:-40 s_waitcnt vmcnt(0) v_fma_f64 v[10:11], v[38:39], v[43:44], v[10:11] .LBB3_24: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s11 s_cbranch_execz .LBB3_26 global_load_b64 v[43:44], v[34:35], off offset:-32 s_waitcnt vmcnt(0) v_fma_f64 v[8:9], v[38:39], v[43:44], v[8:9] .LBB3_26: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s12 s_cbranch_execz .LBB3_28 global_load_b64 v[43:44], v[34:35], off offset:-24 s_waitcnt vmcnt(0) v_fma_f64 v[6:7], v[38:39], v[43:44], v[6:7] .LBB3_28: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s13 s_cbranch_execz .LBB3_30 global_load_b64 v[43:44], v[34:35], off offset:-16 s_waitcnt vmcnt(0) v_fma_f64 v[4:5], v[38:39], v[43:44], v[4:5] .LBB3_30: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s14 s_cbranch_execz .LBB3_32 global_load_b64 v[43:44], v[34:35], off offset:-8 s_waitcnt vmcnt(0) v_fma_f64 v[2:3], v[38:39], v[43:44], v[2:3] .LBB3_32: s_or_b32 exec_lo, exec_lo, s16 s_and_saveexec_b32 s16, s15 s_cbranch_execz .LBB3_3 global_load_b64 v[43:44], v[34:35], off s_waitcnt vmcnt(0) v_fma_f64 v[0:1], v[38:39], v[43:44], v[0:1] s_branch .LBB3_3 .LBB3_34: s_load_b64 s[0:1], s[0:1], 0x20 v_mad_u64_u32 v[34:35], null, v15, s17, v[12:13] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_ashrrev_i32_e32 v35, 31, v34 v_lshlrev_b64 v[34:35], 3, v[34:35] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v34, vcc_lo, s0, v34 v_add_co_ci_u32_e32 v35, vcc_lo, s1, v35, vcc_lo s_mov_b32 s0, exec_lo global_store_b64 v[34:35], v[26:27], off v_cmpx_lt_i32_e32 0, v42 s_cbranch_execz .LBB3_36 global_store_b64 v[34:35], v[32:33], off offset:8 .LBB3_36: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 1, v42 s_cbranch_execz .LBB3_38 global_store_b64 v[34:35], v[30:31], off offset:16 .LBB3_38: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 2, v42 s_cbranch_execz .LBB3_40 global_store_b64 v[34:35], v[28:29], off offset:24 .LBB3_40: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 3, v42 s_cbranch_execz .LBB3_42 global_store_b64 v[34:35], v[24:25], off offset:32 .LBB3_42: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 4, v42 s_cbranch_execz .LBB3_44 global_store_b64 v[34:35], v[22:23], off offset:40 .LBB3_44: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 5, v42 s_cbranch_execz .LBB3_46 global_store_b64 v[34:35], v[20:21], off offset:48 .LBB3_46: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 6, v42 s_cbranch_execz .LBB3_48 global_store_b64 v[34:35], v[18:19], off offset:56 .LBB3_48: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 7, v42 s_cbranch_execz .LBB3_50 global_store_b64 v[34:35], v[16:17], off offset:64 .LBB3_50: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 8, v42 s_cbranch_execz .LBB3_52 global_store_b64 v[34:35], v[13:14], off offset:72 .LBB3_52: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 9, v42 s_cbranch_execz .LBB3_54 global_store_b64 v[34:35], v[10:11], off offset:80 .LBB3_54: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 10, v42 s_cbranch_execz .LBB3_56 global_store_b64 v[34:35], v[8:9], off offset:88 .LBB3_56: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 11, v42 s_cbranch_execz .LBB3_58 global_store_b64 v[34:35], v[6:7], off offset:96 .LBB3_58: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 12, v42 s_cbranch_execz .LBB3_60 global_store_b64 v[34:35], v[4:5], off offset:104 .LBB3_60: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(SALU_CYCLE_1) s_mov_b32 s0, exec_lo v_cmpx_lt_i32_e32 13, v42 s_cbranch_execz .LBB3_62 global_store_b64 v[34:35], v[2:3], off offset:112 .LBB3_62: s_or_b32 exec_lo, exec_lo, s0 v_cmp_lt_i32_e32 vcc_lo, 14, v42 s_and_b32 exec_lo, exec_lo, vcc_lo s_cbranch_execz .LBB3_64 global_store_b64 v[34:35], v[0:1], off offset:120 .LBB3_64: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z15matmult_kernel4iiiPdS_S_i .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 304 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 45 .amdhsa_next_free_sgpr 25 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end3: .size _Z15matmult_kernel4iiiPdS_S_i, .Lfunc_end3-_Z15matmult_kernel4iiiPdS_S_i .section .AMDGPU.csdata,"",@progbits .text .protected _Z11gpu5_kernel6MatrixS_S_ .globl _Z11gpu5_kernel6MatrixS_S_ .p2align 8 .type _Z11gpu5_kernel6MatrixS_S_,@function _Z11gpu5_kernel6MatrixS_S_: s_clause 0x1 s_load_b32 s5, s[0:1], 0x0 s_load_b64 s[2:3], s[0:1], 0x40 v_mov_b32_e32 v1, 0 v_bfe_u32 v3, v0, 10, 10 v_and_b32_e32 v0, 0x3ff, v0 v_mov_b32_e32 v2, 0 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s5, 16 s_cbranch_scc1 .LBB4_5 s_clause 0x3 s_load_b32 s10, s[0:1], 0x8 s_load_b32 s4, s[0:1], 0x20 s_load_b64 s[6:7], s[0:1], 0x10 s_load_b64 s[8:9], s[0:1], 0x28 v_lshlrev_b32_e32 v11, 3, v0 v_lshlrev_b32_e32 v4, 7, v3 s_ashr_i32 s11, s5, 31 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_lshr_b32 s11, s11, 28 s_add_i32 s5, s5, s11 s_delay_alu instid0(SALU_CYCLE_1) s_ashr_i32 s5, s5, 4 s_waitcnt lgkmcnt(0) v_mad_u64_u32 v[5:6], null, s10, v3, v[0:1] v_mad_u64_u32 v[7:8], null, s4, v3, v[0:1] v_mov_b32_e32 v1, 0 v_mov_b32_e32 v2, 0 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_ashrrev_i32_e32 v6, 31, v5 v_ashrrev_i32_e32 v8, 31, v7 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4) v_lshlrev_b64 v[9:10], 3, v[5:6] v_add_nc_u32_e32 v5, v4, v11 v_add_nc_u32_e32 v6, 0x800, v11 v_lshlrev_b64 v[11:12], 3, v[7:8] s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_add_co_u32 v7, vcc_lo, s6, v9 v_add_co_ci_u32_e32 v8, vcc_lo, s7, v10, vcc_lo v_add_co_u32 v9, vcc_lo, s8, v11 s_delay_alu instid0(VALU_DEP_4) v_add_co_ci_u32_e32 v10, vcc_lo, s9, v12, vcc_lo v_add_nc_u32_e32 v11, v6, v4 s_mul_i32 s6, s10, s15 s_mov_b32 s7, 0 s_set_inst_prefetch_distance 0x1 .p2align 6 .LBB4_2: s_mul_i32 s8, s7, s4 s_add_i32 s9, s7, s6 s_add_i32 s10, s8, s14 s_lshl_b32 s8, s9, 4 s_lshl_b32 s10, s10, 4 s_ashr_i32 s9, s8, 31 s_ashr_i32 s11, s10, 31 s_lshl_b64 s[8:9], s[8:9], 3 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) v_add_co_u32 v12, vcc_lo, v7, s8 v_add_co_ci_u32_e32 v13, vcc_lo, s9, v8, vcc_lo s_lshl_b64 s[8:9], s[10:11], 3 v_add_co_u32 v14, vcc_lo, v9, s8 v_add_co_ci_u32_e32 v15, vcc_lo, s9, v10, vcc_lo global_load_b64 v[16:17], v[12:13], off global_load_b64 v[13:14], v[14:15], off v_mov_b32_e32 v12, v6 s_mov_b32 s8, 0 s_waitcnt vmcnt(1) ds_store_b64 v5, v[16:17] s_waitcnt vmcnt(0) ds_store_b64 v11, v[13:14] s_waitcnt lgkmcnt(0) s_barrier buffer_gl0_inv .LBB4_3: v_add_nc_u32_e32 v15, s8, v4 s_add_i32 s8, s8, 8 ds_load_b64 v[13:14], v12 ds_load_b64 v[15:16], v15 v_add_nc_u32_e32 v12, 0x80, v12 s_cmpk_eq_i32 s8, 0x80 s_waitcnt lgkmcnt(0) v_fma_f64 v[1:2], v[15:16], v[13:14], v[1:2] s_cbranch_scc0 .LBB4_3 s_add_i32 s7, s7, 1 s_delay_alu instid0(SALU_CYCLE_1) s_cmp_eq_u32 s7, s5 s_barrier buffer_gl0_inv s_cbranch_scc0 .LBB4_2 .LBB4_5: s_set_inst_prefetch_distance 0x2 s_load_b32 s0, s[0:1], 0x38 s_waitcnt lgkmcnt(0) v_mad_u64_u32 v[4:5], null, s0, v3, v[0:1] s_mul_i32 s0, s0, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_add_i32 s0, s0, s14 s_lshl_b32 s0, s0, 4 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) s_ashr_i32 s1, s0, 31 v_ashrrev_i32_e32 v5, 31, v4 s_lshl_b64 s[0:1], s[0:1], 3 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) s_add_u32 s0, s2, s0 s_addc_u32 s1, s3, s1 v_lshlrev_b64 v[3:4], 3, v[4:5] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v3, vcc_lo, s0, v3 v_add_co_ci_u32_e32 v4, vcc_lo, s1, v4, vcc_lo global_store_b64 v[3:4], v[1:2], off s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z11gpu5_kernel6MatrixS_S_ .amdhsa_group_segment_fixed_size 4096 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 72 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 18 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end4: .size _Z11gpu5_kernel6MatrixS_S_, .Lfunc_end4-_Z11gpu5_kernel6MatrixS_S_ .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .offset: 8 .size: 4 .value_kind: by_value - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 32 .size: 8 .value_kind: global_buffer .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 40 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z15matmult_kernel1iiiPdS_S_ .private_segment_fixed_size: 0 .sgpr_count: 28 .sgpr_spill_count: 0 .symbol: _Z15matmult_kernel1iiiPdS_S_.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 7 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .offset: 8 .size: 4 .value_kind: by_value - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 32 .size: 8 .value_kind: global_buffer - .offset: 40 .size: 4 .value_kind: hidden_block_count_x - .offset: 44 .size: 4 .value_kind: hidden_block_count_y - .offset: 48 .size: 4 .value_kind: hidden_block_count_z - .offset: 52 .size: 2 .value_kind: hidden_group_size_x - .offset: 54 .size: 2 .value_kind: hidden_group_size_y - .offset: 56 .size: 2 .value_kind: hidden_group_size_z - .offset: 58 .size: 2 .value_kind: hidden_remainder_x - .offset: 60 .size: 2 .value_kind: hidden_remainder_y - .offset: 62 .size: 2 .value_kind: hidden_remainder_z - .offset: 80 .size: 8 .value_kind: hidden_global_offset_x - .offset: 88 .size: 8 .value_kind: hidden_global_offset_y - .offset: 96 .size: 8 .value_kind: hidden_global_offset_z - .offset: 104 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 296 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z15matmult_kernel2iiiPdS_S_ .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z15matmult_kernel2iiiPdS_S_.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 11 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .offset: 8 .size: 4 .value_kind: by_value - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 32 .size: 8 .value_kind: global_buffer - .offset: 40 .size: 4 .value_kind: hidden_block_count_x - .offset: 44 .size: 4 .value_kind: hidden_block_count_y - .offset: 48 .size: 4 .value_kind: hidden_block_count_z - .offset: 52 .size: 2 .value_kind: hidden_group_size_x - .offset: 54 .size: 2 .value_kind: hidden_group_size_y - .offset: 56 .size: 2 .value_kind: hidden_group_size_z - .offset: 58 .size: 2 .value_kind: hidden_remainder_x - .offset: 60 .size: 2 .value_kind: hidden_remainder_y - .offset: 62 .size: 2 .value_kind: hidden_remainder_z - .offset: 80 .size: 8 .value_kind: hidden_global_offset_x - .offset: 88 .size: 8 .value_kind: hidden_global_offset_y - .offset: 96 .size: 8 .value_kind: hidden_global_offset_z - .offset: 104 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 296 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z15matmult_kernel3iiiPdS_S_ .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z15matmult_kernel3iiiPdS_S_.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 17 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .offset: 8 .size: 4 .value_kind: by_value - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 32 .size: 8 .value_kind: global_buffer - .offset: 40 .size: 4 .value_kind: by_value - .offset: 48 .size: 4 .value_kind: hidden_block_count_x - .offset: 52 .size: 4 .value_kind: hidden_block_count_y - .offset: 56 .size: 4 .value_kind: hidden_block_count_z - .offset: 60 .size: 2 .value_kind: hidden_group_size_x - .offset: 62 .size: 2 .value_kind: hidden_group_size_y - .offset: 64 .size: 2 .value_kind: hidden_group_size_z - .offset: 66 .size: 2 .value_kind: hidden_remainder_x - .offset: 68 .size: 2 .value_kind: hidden_remainder_y - .offset: 70 .size: 2 .value_kind: hidden_remainder_z - .offset: 88 .size: 8 .value_kind: hidden_global_offset_x - .offset: 96 .size: 8 .value_kind: hidden_global_offset_y - .offset: 104 .size: 8 .value_kind: hidden_global_offset_z - .offset: 112 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 304 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z15matmult_kernel4iiiPdS_S_i .private_segment_fixed_size: 0 .sgpr_count: 27 .sgpr_spill_count: 0 .symbol: _Z15matmult_kernel4iiiPdS_S_i.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 45 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 24 .value_kind: by_value - .offset: 24 .size: 24 .value_kind: by_value - .offset: 48 .size: 24 .value_kind: by_value .group_segment_fixed_size: 4096 .kernarg_segment_align: 8 .kernarg_segment_size: 72 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z11gpu5_kernel6MatrixS_S_ .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z11gpu5_kernel6MatrixS_S_.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 18 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include <stdio.h> #define MIN(x, y) (((x) < (y)) ? (x) : (y)) __global__ void matmult_kernel1(int m, int n, int k, double *A, double *B, double *C){ // set C to zeros for (int i=0;i<m;i++){ for (int p=0;p<n;p++){ C[i*n+p]=0; //C[i][p] } } // do matmult with mkn loop order for (int i=0;i<m;i++) { for (int j=0;j<k;j++){ for (int p=0;p<n;p++) { C[i*n+p] += A[i*k+j] * B[j*n+p]; } } } } __global__ void matmult_kernel2(int m, int n, int k, double *A, double *B, double *C){ int j = blockIdx.x*blockDim.x+threadIdx.x; //looping through n int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp; if((j<n)&&(i<m)){ for(int p=0; p<k; p++){ //read row of A and col of B //row of A is A[mit*k+kit] //col of B is B[kit*n+nit] tmp += A[i*k+p] * B[p*n+j]; } //C is C[mit*n+nit] C[i*n+j] = tmp; } } __global__ void matmult_kernel3(int m, int n, int k, double *A, double *B, double *C){ //compute C(i,j) and C(i,j+1) int j = 2*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only half as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, either 1 or 0) int j_add = MIN(1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j]; if(j_add == 1) tmp2 += A[i*k+p] * B[p*n+j+1]; } C[i*n+j] = tmp1; if(j_add == 1) C[i*n+j+1] = tmp2; } } __global__ void matmult_kernel4(int m, int n, int k, double *A, double *B, double *C, int s){ //compute C(i,j), C(i,j+1), ... C(i,j+s) int j = s*(blockIdx.x*blockDim.x+threadIdx.x); //looping through n (only 1/s as many threads/blocks) int i = blockIdx.y*blockDim.y+threadIdx.y; //looping through m double tmp1 = 0.0; double tmp2 = 0.0; double tmp3 = 0.0; double tmp4 = 0.0; double tmp5 = 0.0; double tmp6 = 0.0; double tmp7 = 0.0; double tmp8 = 0.0; double tmp9 = 0.0; double tmp10 = 0.0; double tmp11 = 0.0; double tmp12 = 0.0; double tmp13 = 0.0; double tmp14 = 0.0; double tmp15 = 0.0; double tmp16 = 0.0; if((j<n)&&(i<m)){ //additional j to compute (here, from 0 to s-1) int j_add = MIN(s-1,n-1-j); for(int p=0; p<k; p++){ //row of A and col of B tmp1 += A[i*k+p] * B[p*n+j+0]; if(j_add > 0) tmp2 += A[i*k+p] * B[p*n+j+1]; if(j_add > 1) tmp3 += A[i*k+p] * B[p*n+j+2]; if(j_add > 2) tmp4 += A[i*k+p] * B[p*n+j+3]; if(j_add > 3) tmp5 += A[i*k+p] * B[p*n+j+4]; if(j_add > 4) tmp6 += A[i*k+p] * B[p*n+j+5]; if(j_add > 5) tmp7 += A[i*k+p] * B[p*n+j+6]; if(j_add > 6) tmp8 += A[i*k+p] * B[p*n+j+7]; if(j_add > 7) tmp9 += A[i*k+p] * B[p*n+j+8]; if(j_add > 8) tmp10 += A[i*k+p] * B[p*n+j+9]; if(j_add > 9) tmp11 += A[i*k+p] * B[p*n+j+10]; if(j_add > 10) tmp12 += A[i*k+p] * B[p*n+j+11]; if(j_add > 11) tmp13 += A[i*k+p] * B[p*n+j+12]; if(j_add > 12) tmp14 += A[i*k+p] * B[p*n+j+13]; if(j_add > 13) tmp15 += A[i*k+p] * B[p*n+j+14]; if(j_add > 14) tmp16 += A[i*k+p] * B[p*n+j+15]; } C[i*n+j] = tmp1; if(j_add > 0) C[i*n+j+1] = tmp2; if(j_add > 1) C[i*n+j+2] = tmp3; if(j_add > 2) C[i*n+j+3] = tmp4; if(j_add > 3) C[i*n+j+4] = tmp5; if(j_add > 4) C[i*n+j+5] = tmp6; if(j_add > 5) C[i*n+j+6] = tmp7; if(j_add > 6) C[i*n+j+7] = tmp8; if(j_add > 7) C[i*n+j+8] = tmp9; if(j_add > 8) C[i*n+j+9] = tmp10; if(j_add > 9) C[i*n+j+10] = tmp11; if(j_add > 10) C[i*n+j+11] = tmp12; if(j_add > 11) C[i*n+j+12] = tmp13; if(j_add > 12) C[i*n+j+13] = tmp14; if(j_add > 13) C[i*n+j+14] = tmp15; if(j_add > 14) C[i*n+j+15] = tmp16; } } /* BEGIN GPU 5 ############################################################################## */ #define BLOCK_SIZE 16 typedef struct { int width; int height; int stride; double* elements; } Matrix; // Get a matrix element __device__ double GetElement(const Matrix A, int row, int col) { return A.elements[row * A.stride + col]; } // Set a matrix element __device__ void SetElement(Matrix A, int row, int col, double value) { A.elements[row * A.stride + col] = value; } // Get the BLOCK_SIZExBLOCK_SIZE sub-matrix Asub of A that is // located col sub-matrices to the right and row sub-matrices down // from the upper-left corner of A __device__ Matrix GetSubMatrix(Matrix A, int row, int col) { Matrix Asub; Asub.width = BLOCK_SIZE; Asub.height = BLOCK_SIZE; Asub.stride = A.stride; Asub.elements = &A.elements[A.stride * BLOCK_SIZE * row + BLOCK_SIZE * col]; return Asub; } // Matrix multiplication kernel called by MatMul() __global__ void gpu5_kernel(const Matrix A, const Matrix B, Matrix C) { // Block row and column int blockRow = blockIdx.y; int blockCol = blockIdx.x; // Each thread block computes one sub-matrix Csub of C Matrix Csub = GetSubMatrix(C, blockRow, blockCol); // Each thread computes one element of Csub // by accumulating results into Cvalue double Cvalue = 0; // Thread row and column within Csub int row = threadIdx.y; int col = threadIdx.x; // Loop over all the sub-matrices of A and B that are // required to compute Csub // Multiply each pair of sub-matrices together // and accumulate the results for (int m = 0; m < (A.width / BLOCK_SIZE); ++m) { // Get sub-matrix Asub of A Matrix Asub = GetSubMatrix(A, blockRow, m); // Get sub-matrix Bsub of B Matrix Bsub = GetSubMatrix(B, m, blockCol); // Shared memory used to store Asub and Bsub respectively __shared__ double As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ double Bs[BLOCK_SIZE][BLOCK_SIZE]; // Load Asub and Bsub from device memory to shared memory // Each thread loads one element of each sub-matrix As[row][col] = GetElement(Asub, row, col); Bs[row][col] = GetElement(Bsub, row, col); // Synchronize to make sure the sub-matrices are loaded // before starting the computation __syncthreads(); // Multiply Asub and Bsub together for (int e = 0; e < BLOCK_SIZE; ++e) Cvalue += As[row][e] * Bs[e][col]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } // Write Csub to device memory // Each thread writes one element SetElement(Csub, row, col, Cvalue); }
.text .file "matmult_kernels.hip" .globl _Z30__device_stub__matmult_kernel1iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel1iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel1iiiPdS_S_,@function _Z30__device_stub__matmult_kernel1iiiPdS_S_: # @_Z30__device_stub__matmult_kernel1iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel1iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end0: .size _Z30__device_stub__matmult_kernel1iiiPdS_S_, .Lfunc_end0-_Z30__device_stub__matmult_kernel1iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel2iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel2iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel2iiiPdS_S_,@function _Z30__device_stub__matmult_kernel2iiiPdS_S_: # @_Z30__device_stub__matmult_kernel2iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel2iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end1: .size _Z30__device_stub__matmult_kernel2iiiPdS_S_, .Lfunc_end1-_Z30__device_stub__matmult_kernel2iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel3iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel3iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel3iiiPdS_S_,@function _Z30__device_stub__matmult_kernel3iiiPdS_S_: # @_Z30__device_stub__matmult_kernel3iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel3iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end2: .size _Z30__device_stub__matmult_kernel3iiiPdS_S_, .Lfunc_end2-_Z30__device_stub__matmult_kernel3iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel4iiiPdS_S_i # -- Begin function _Z30__device_stub__matmult_kernel4iiiPdS_S_i .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel4iiiPdS_S_i,@function _Z30__device_stub__matmult_kernel4iiiPdS_S_i: # @_Z30__device_stub__matmult_kernel4iiiPdS_S_i .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 160(%rsp), %rax movq %rax, 144(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel4iiiPdS_S_i, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end3: .size _Z30__device_stub__matmult_kernel4iiiPdS_S_i, .Lfunc_end3-_Z30__device_stub__matmult_kernel4iiiPdS_S_i .cfi_endproc # -- End function .globl _Z26__device_stub__gpu5_kernel6MatrixS_S_ # -- Begin function _Z26__device_stub__gpu5_kernel6MatrixS_S_ .p2align 4, 0x90 .type _Z26__device_stub__gpu5_kernel6MatrixS_S_,@function _Z26__device_stub__gpu5_kernel6MatrixS_S_: # @_Z26__device_stub__gpu5_kernel6MatrixS_S_ .cfi_startproc # %bb.0: subq $72, %rsp .cfi_def_cfa_offset 80 leaq 80(%rsp), %rax movq %rax, 48(%rsp) leaq 104(%rsp), %rax movq %rax, 56(%rsp) leaq 128(%rsp), %rax movq %rax, 64(%rsp) leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z11gpu5_kernel6MatrixS_S_, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $88, %rsp .cfi_adjust_cfa_offset -88 retq .Lfunc_end4: .size _Z26__device_stub__gpu5_kernel6MatrixS_S_, .Lfunc_end4-_Z26__device_stub__gpu5_kernel6MatrixS_S_ .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB5_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB5_2: movq __hip_gpubin_handle(%rip), %rbx xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel1iiiPdS_S_, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel2iiiPdS_S_, %esi movl $.L__unnamed_2, %edx movl $.L__unnamed_2, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel3iiiPdS_S_, %esi movl $.L__unnamed_3, %edx movl $.L__unnamed_3, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel4iiiPdS_S_i, %esi movl $.L__unnamed_4, %edx movl $.L__unnamed_4, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z11gpu5_kernel6MatrixS_S_, %esi movl $.L__unnamed_5, %edx movl $.L__unnamed_5, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end5: .size __hip_module_ctor, .Lfunc_end5-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB6_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB6_2: retq .Lfunc_end6: .size __hip_module_dtor, .Lfunc_end6-__hip_module_dtor .cfi_endproc # -- End function .type _Z15matmult_kernel1iiiPdS_S_,@object # @_Z15matmult_kernel1iiiPdS_S_ .section .rodata,"a",@progbits .globl _Z15matmult_kernel1iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel1iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel1iiiPdS_S_ .size _Z15matmult_kernel1iiiPdS_S_, 8 .type _Z15matmult_kernel2iiiPdS_S_,@object # @_Z15matmult_kernel2iiiPdS_S_ .globl _Z15matmult_kernel2iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel2iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel2iiiPdS_S_ .size _Z15matmult_kernel2iiiPdS_S_, 8 .type _Z15matmult_kernel3iiiPdS_S_,@object # @_Z15matmult_kernel3iiiPdS_S_ .globl _Z15matmult_kernel3iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel3iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel3iiiPdS_S_ .size _Z15matmult_kernel3iiiPdS_S_, 8 .type _Z15matmult_kernel4iiiPdS_S_i,@object # @_Z15matmult_kernel4iiiPdS_S_i .globl _Z15matmult_kernel4iiiPdS_S_i .p2align 3, 0x0 _Z15matmult_kernel4iiiPdS_S_i: .quad _Z30__device_stub__matmult_kernel4iiiPdS_S_i .size _Z15matmult_kernel4iiiPdS_S_i, 8 .type _Z11gpu5_kernel6MatrixS_S_,@object # @_Z11gpu5_kernel6MatrixS_S_ .globl _Z11gpu5_kernel6MatrixS_S_ .p2align 3, 0x0 _Z11gpu5_kernel6MatrixS_S_: .quad _Z26__device_stub__gpu5_kernel6MatrixS_S_ .size _Z11gpu5_kernel6MatrixS_S_, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z15matmult_kernel1iiiPdS_S_" .size .L__unnamed_1, 29 .type .L__unnamed_2,@object # @1 .L__unnamed_2: .asciz "_Z15matmult_kernel2iiiPdS_S_" .size .L__unnamed_2, 29 .type .L__unnamed_3,@object # @2 .L__unnamed_3: .asciz "_Z15matmult_kernel3iiiPdS_S_" .size .L__unnamed_3, 29 .type .L__unnamed_4,@object # @3 .L__unnamed_4: .asciz "_Z15matmult_kernel4iiiPdS_S_i" .size .L__unnamed_4, 30 .type .L__unnamed_5,@object # @4 .L__unnamed_5: .asciz "_Z11gpu5_kernel6MatrixS_S_" .size .L__unnamed_5, 27 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z30__device_stub__matmult_kernel1iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel2iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel3iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel4iiiPdS_S_i .addrsig_sym _Z26__device_stub__gpu5_kernel6MatrixS_S_ .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z15matmult_kernel1iiiPdS_S_ .addrsig_sym _Z15matmult_kernel2iiiPdS_S_ .addrsig_sym _Z15matmult_kernel3iiiPdS_S_ .addrsig_sym _Z15matmult_kernel4iiiPdS_S_i .addrsig_sym _Z11gpu5_kernel6MatrixS_S_ .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00152ae4_00000000-6_matmult_kernels.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2062: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2062: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z10GetElement6Matrixii .type _Z10GetElement6Matrixii, @function _Z10GetElement6Matrixii: .LFB2057: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2057: .size _Z10GetElement6Matrixii, .-_Z10GetElement6Matrixii .globl _Z10SetElement6Matrixiid .type _Z10SetElement6Matrixiid, @function _Z10SetElement6Matrixiid: .LFB2058: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2058: .size _Z10SetElement6Matrixiid, .-_Z10SetElement6Matrixiid .globl _Z12GetSubMatrix6Matrixii .type _Z12GetSubMatrix6Matrixii, @function _Z12GetSubMatrix6Matrixii: .LFB2059: .cfi_startproc endbr64 pushq %rax .cfi_def_cfa_offset 16 popq %rax .cfi_def_cfa_offset 8 subq $24, %rsp .cfi_def_cfa_offset 32 movl $1, 12(%rsp) movl 12(%rsp), %edi call exit@PLT .cfi_endproc .LFE2059: .size _Z12GetSubMatrix6Matrixii, .-_Z12GetSubMatrix6Matrixii .globl _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_: .LFB2084: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L13 .L9: movq 168(%rsp), %rax subq %fs:40, %rax jne .L14 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L13: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel1iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L9 .L14: call __stack_chk_fail@PLT .cfi_endproc .LFE2084: .size _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel1iiiPdS_S_ .type _Z15matmult_kernel1iiiPdS_S_, @function _Z15matmult_kernel1iiiPdS_S_: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel1iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _Z15matmult_kernel1iiiPdS_S_, .-_Z15matmult_kernel1iiiPdS_S_ .globl _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_: .LFB2086: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L21 .L17: movq 168(%rsp), %rax subq %fs:40, %rax jne .L22 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L21: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel2iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L17 .L22: call __stack_chk_fail@PLT .cfi_endproc .LFE2086: .size _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel2iiiPdS_S_ .type _Z15matmult_kernel2iiiPdS_S_, @function _Z15matmult_kernel2iiiPdS_S_: .LFB2087: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel2iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2087: .size _Z15matmult_kernel2iiiPdS_S_, .-_Z15matmult_kernel2iiiPdS_S_ .globl _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ .type _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_, @function _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_: .LFB2088: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L29 .L25: movq 168(%rsp), %rax subq %fs:40, %rax jne .L30 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L29: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel3iiiPdS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L25 .L30: call __stack_chk_fail@PLT .cfi_endproc .LFE2088: .size _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_, .-_Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ .globl _Z15matmult_kernel3iiiPdS_S_ .type _Z15matmult_kernel3iiiPdS_S_, @function _Z15matmult_kernel3iiiPdS_S_: .LFB2089: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z42__device_stub__Z15matmult_kernel3iiiPdS_S_iiiPdS_S_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2089: .size _Z15matmult_kernel3iiiPdS_S_, .-_Z15matmult_kernel3iiiPdS_S_ .globl _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i .type _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i, @function _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i: .LFB2090: .cfi_startproc endbr64 subq $184, %rsp .cfi_def_cfa_offset 192 movl %edi, 44(%rsp) movl %esi, 40(%rsp) movl %edx, 36(%rsp) movq %rcx, 24(%rsp) movq %r8, 16(%rsp) movq %r9, 8(%rsp) movq %fs:40, %rax movq %rax, 168(%rsp) xorl %eax, %eax leaq 44(%rsp), %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rax movq %rax, 120(%rsp) leaq 36(%rsp), %rax movq %rax, 128(%rsp) leaq 24(%rsp), %rax movq %rax, 136(%rsp) leaq 16(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) leaq 192(%rsp), %rax movq %rax, 160(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L37 .L33: movq 168(%rsp), %rax subq %fs:40, %rax jne .L38 addq $184, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L37: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 200 pushq 56(%rsp) .cfi_def_cfa_offset 208 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z15matmult_kernel4iiiPdS_S_i(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 192 jmp .L33 .L38: call __stack_chk_fail@PLT .cfi_endproc .LFE2090: .size _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i, .-_Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i .globl _Z15matmult_kernel4iiiPdS_S_i .type _Z15matmult_kernel4iiiPdS_S_i, @function _Z15matmult_kernel4iiiPdS_S_i: .LFB2091: .cfi_startproc endbr64 subq $16, %rsp .cfi_def_cfa_offset 24 movl 24(%rsp), %eax pushq %rax .cfi_def_cfa_offset 32 call _Z43__device_stub__Z15matmult_kernel4iiiPdS_S_iiiiPdS_S_i addq $24, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2091: .size _Z15matmult_kernel4iiiPdS_S_i, .-_Z15matmult_kernel4iiiPdS_S_i .globl _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ .type _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_, @function _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_: .LFB2092: .cfi_startproc endbr64 subq $104, %rsp .cfi_def_cfa_offset 112 movq %fs:40, %rax movq %rax, 88(%rsp) xorl %eax, %eax movq %rdi, 64(%rsp) movq %rsi, 72(%rsp) movq %rdx, 80(%rsp) movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L45 .L41: movq 88(%rsp), %rax subq %fs:40, %rax jne .L46 addq $104, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L45: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 120 pushq 8(%rsp) .cfi_def_cfa_offset 128 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z11gpu5_kernel6MatrixS_S_(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 112 jmp .L41 .L46: call __stack_chk_fail@PLT .cfi_endproc .LFE2092: .size _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_, .-_Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ .globl _Z11gpu5_kernel6MatrixS_S_ .type _Z11gpu5_kernel6MatrixS_S_, @function _Z11gpu5_kernel6MatrixS_S_: .LFB2093: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq 64(%rsp), %rdx leaq 40(%rsp), %rsi leaq 16(%rsp), %rdi call _Z40__device_stub__Z11gpu5_kernel6MatrixS_S_RK6MatrixS1_RS_ addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2093: .size _Z11gpu5_kernel6MatrixS_S_, .-_Z11gpu5_kernel6MatrixS_S_ .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "_Z11gpu5_kernel6MatrixS_S_" .LC1: .string "_Z15matmult_kernel4iiiPdS_S_i" .LC2: .string "_Z15matmult_kernel3iiiPdS_S_" .LC3: .string "_Z15matmult_kernel2iiiPdS_S_" .LC4: .string "_Z15matmult_kernel1iiiPdS_S_" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2095: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rbx movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z11gpu5_kernel6MatrixS_S_(%rip), %rsi movq %rax, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC1(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel4iiiPdS_S_i(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC2(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel3iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC3(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel2iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z15matmult_kernel1iiiPdS_S_(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2095: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "matmult_kernels.hip" .globl _Z30__device_stub__matmult_kernel1iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel1iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel1iiiPdS_S_,@function _Z30__device_stub__matmult_kernel1iiiPdS_S_: # @_Z30__device_stub__matmult_kernel1iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel1iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end0: .size _Z30__device_stub__matmult_kernel1iiiPdS_S_, .Lfunc_end0-_Z30__device_stub__matmult_kernel1iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel2iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel2iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel2iiiPdS_S_,@function _Z30__device_stub__matmult_kernel2iiiPdS_S_: # @_Z30__device_stub__matmult_kernel2iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel2iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end1: .size _Z30__device_stub__matmult_kernel2iiiPdS_S_, .Lfunc_end1-_Z30__device_stub__matmult_kernel2iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel3iiiPdS_S_ # -- Begin function _Z30__device_stub__matmult_kernel3iiiPdS_S_ .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel3iiiPdS_S_,@function _Z30__device_stub__matmult_kernel3iiiPdS_S_: # @_Z30__device_stub__matmult_kernel3iiiPdS_S_ .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel3iiiPdS_S_, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end2: .size _Z30__device_stub__matmult_kernel3iiiPdS_S_, .Lfunc_end2-_Z30__device_stub__matmult_kernel3iiiPdS_S_ .cfi_endproc # -- End function .globl _Z30__device_stub__matmult_kernel4iiiPdS_S_i # -- Begin function _Z30__device_stub__matmult_kernel4iiiPdS_S_i .p2align 4, 0x90 .type _Z30__device_stub__matmult_kernel4iiiPdS_S_i,@function _Z30__device_stub__matmult_kernel4iiiPdS_S_i: # @_Z30__device_stub__matmult_kernel4iiiPdS_S_i .cfi_startproc # %bb.0: subq $152, %rsp .cfi_def_cfa_offset 160 movl %edi, 20(%rsp) movl %esi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, 88(%rsp) movq %r8, 80(%rsp) movq %r9, 72(%rsp) leaq 20(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) leaq 88(%rsp), %rax movq %rax, 120(%rsp) leaq 80(%rsp), %rax movq %rax, 128(%rsp) leaq 72(%rsp), %rax movq %rax, 136(%rsp) leaq 160(%rsp), %rax movq %rax, 144(%rsp) leaq 56(%rsp), %rdi leaq 40(%rsp), %rsi leaq 32(%rsp), %rdx leaq 24(%rsp), %rcx callq __hipPopCallConfiguration movq 56(%rsp), %rsi movl 64(%rsp), %edx movq 40(%rsp), %rcx movl 48(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z15matmult_kernel4iiiPdS_S_i, %edi pushq 24(%rsp) .cfi_adjust_cfa_offset 8 pushq 40(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $168, %rsp .cfi_adjust_cfa_offset -168 retq .Lfunc_end3: .size _Z30__device_stub__matmult_kernel4iiiPdS_S_i, .Lfunc_end3-_Z30__device_stub__matmult_kernel4iiiPdS_S_i .cfi_endproc # -- End function .globl _Z26__device_stub__gpu5_kernel6MatrixS_S_ # -- Begin function _Z26__device_stub__gpu5_kernel6MatrixS_S_ .p2align 4, 0x90 .type _Z26__device_stub__gpu5_kernel6MatrixS_S_,@function _Z26__device_stub__gpu5_kernel6MatrixS_S_: # @_Z26__device_stub__gpu5_kernel6MatrixS_S_ .cfi_startproc # %bb.0: subq $72, %rsp .cfi_def_cfa_offset 80 leaq 80(%rsp), %rax movq %rax, 48(%rsp) leaq 104(%rsp), %rax movq %rax, 56(%rsp) leaq 128(%rsp), %rax movq %rax, 64(%rsp) leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z11gpu5_kernel6MatrixS_S_, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $88, %rsp .cfi_adjust_cfa_offset -88 retq .Lfunc_end4: .size _Z26__device_stub__gpu5_kernel6MatrixS_S_, .Lfunc_end4-_Z26__device_stub__gpu5_kernel6MatrixS_S_ .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB5_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB5_2: movq __hip_gpubin_handle(%rip), %rbx xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel1iiiPdS_S_, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel2iiiPdS_S_, %esi movl $.L__unnamed_2, %edx movl $.L__unnamed_2, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel3iiiPdS_S_, %esi movl $.L__unnamed_3, %edx movl $.L__unnamed_3, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z15matmult_kernel4iiiPdS_S_i, %esi movl $.L__unnamed_4, %edx movl $.L__unnamed_4, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z11gpu5_kernel6MatrixS_S_, %esi movl $.L__unnamed_5, %edx movl $.L__unnamed_5, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end5: .size __hip_module_ctor, .Lfunc_end5-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB6_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB6_2: retq .Lfunc_end6: .size __hip_module_dtor, .Lfunc_end6-__hip_module_dtor .cfi_endproc # -- End function .type _Z15matmult_kernel1iiiPdS_S_,@object # @_Z15matmult_kernel1iiiPdS_S_ .section .rodata,"a",@progbits .globl _Z15matmult_kernel1iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel1iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel1iiiPdS_S_ .size _Z15matmult_kernel1iiiPdS_S_, 8 .type _Z15matmult_kernel2iiiPdS_S_,@object # @_Z15matmult_kernel2iiiPdS_S_ .globl _Z15matmult_kernel2iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel2iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel2iiiPdS_S_ .size _Z15matmult_kernel2iiiPdS_S_, 8 .type _Z15matmult_kernel3iiiPdS_S_,@object # @_Z15matmult_kernel3iiiPdS_S_ .globl _Z15matmult_kernel3iiiPdS_S_ .p2align 3, 0x0 _Z15matmult_kernel3iiiPdS_S_: .quad _Z30__device_stub__matmult_kernel3iiiPdS_S_ .size _Z15matmult_kernel3iiiPdS_S_, 8 .type _Z15matmult_kernel4iiiPdS_S_i,@object # @_Z15matmult_kernel4iiiPdS_S_i .globl _Z15matmult_kernel4iiiPdS_S_i .p2align 3, 0x0 _Z15matmult_kernel4iiiPdS_S_i: .quad _Z30__device_stub__matmult_kernel4iiiPdS_S_i .size _Z15matmult_kernel4iiiPdS_S_i, 8 .type _Z11gpu5_kernel6MatrixS_S_,@object # @_Z11gpu5_kernel6MatrixS_S_ .globl _Z11gpu5_kernel6MatrixS_S_ .p2align 3, 0x0 _Z11gpu5_kernel6MatrixS_S_: .quad _Z26__device_stub__gpu5_kernel6MatrixS_S_ .size _Z11gpu5_kernel6MatrixS_S_, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z15matmult_kernel1iiiPdS_S_" .size .L__unnamed_1, 29 .type .L__unnamed_2,@object # @1 .L__unnamed_2: .asciz "_Z15matmult_kernel2iiiPdS_S_" .size .L__unnamed_2, 29 .type .L__unnamed_3,@object # @2 .L__unnamed_3: .asciz "_Z15matmult_kernel3iiiPdS_S_" .size .L__unnamed_3, 29 .type .L__unnamed_4,@object # @3 .L__unnamed_4: .asciz "_Z15matmult_kernel4iiiPdS_S_i" .size .L__unnamed_4, 30 .type .L__unnamed_5,@object # @4 .L__unnamed_5: .asciz "_Z11gpu5_kernel6MatrixS_S_" .size .L__unnamed_5, 27 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z30__device_stub__matmult_kernel1iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel2iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel3iiiPdS_S_ .addrsig_sym _Z30__device_stub__matmult_kernel4iiiPdS_S_i .addrsig_sym _Z26__device_stub__gpu5_kernel6MatrixS_S_ .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z15matmult_kernel1iiiPdS_S_ .addrsig_sym _Z15matmult_kernel2iiiPdS_S_ .addrsig_sym _Z15matmult_kernel3iiiPdS_S_ .addrsig_sym _Z15matmult_kernel4iiiPdS_S_i .addrsig_sym _Z11gpu5_kernel6MatrixS_S_ .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); cudaMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); cudaMemcpy(host_array, device_array, num_bytes, cudaMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); cudaFree(device_array); return 0; }
code for sm_80 Function : _Z6kernelPi .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R2, SR_CTAID.Y ; /* 0x0000000000027919 */ /* 0x000e220000002600 */ /*0020*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fc60000000a00 */ /*0030*/ S2R R7, SR_TID.Y ; /* 0x0000000000077919 */ /* 0x000e280000002200 */ /*0040*/ S2R R5, SR_CTAID.X ; /* 0x0000000000057919 */ /* 0x000e680000002500 */ /*0050*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000ea20000002100 */ /*0060*/ IMAD R0, R2, c[0x0][0x4], R7 ; /* 0x0000010002007a24 */ /* 0x001fe200078e0207 */ /*0070*/ HFMA2.MMA R7, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff077435 */ /* 0x000fc600000001ff */ /*0080*/ IMAD R0, R0, c[0x0][0xc], R5.reuse ; /* 0x0000030000007a24 */ /* 0x102fe400078e0205 */ /*0090*/ IMAD R5, R2, c[0x0][0xc], R5 ; /* 0x0000030002057a24 */ /* 0x000fe400078e0205 */ /*00a0*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x004fc800078e0203 */ /*00b0*/ IMAD.WIDE R2, R0, R7, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fca00078e0207 */ /*00c0*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x000fe2000c101904 */ /*00d0*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*00e0*/ BRA 0xe0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0100*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0110*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0120*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0130*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); cudaMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); cudaMemcpy(host_array, device_array, num_bytes, cudaMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); cudaFree(device_array); return 0; }
.file "tmpxft_000ff563_00000000-6_two_d_kernel.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z25__device_stub__Z6kernelPiPi .type _Z25__device_stub__Z6kernelPiPi, @function _Z25__device_stub__Z6kernelPiPi: .LFB2082: .cfi_startproc endbr64 subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 8(%rsp) movq %fs:40, %rax movq %rax, 88(%rsp) xorl %eax, %eax leaq 8(%rsp), %rax movq %rax, 80(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 88(%rsp), %rax subq %fs:40, %rax jne .L8 addq $104, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 120 pushq 24(%rsp) .cfi_def_cfa_offset 128 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z6kernelPi(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 112 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z25__device_stub__Z6kernelPiPi, .-_Z25__device_stub__Z6kernelPiPi .globl _Z6kernelPi .type _Z6kernelPi, @function _Z6kernelPi: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z25__device_stub__Z6kernelPiPi addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z6kernelPi, .-_Z6kernelPi .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "%2d " .LC1: .string "\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $56, %rsp .cfi_def_cfa_offset 112 movq %fs:40, %rax movq %rax, 40(%rsp) xorl %eax, %eax movq $0, 8(%rsp) movl $1024, %edi call malloc@PLT movq %rax, %r15 leaq 8(%rsp), %rdi movl $1024, %esi call cudaMalloc@PLT movl $4, 28(%rsp) movl $4, 32(%rsp) movl $4, 16(%rsp) movl $4, 20(%rsp) movl $0, %r9d movl $0, %r8d movq 16(%rsp), %rdx movl $1, %ecx movq 28(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L19 .L12: movl $2, %ecx movl $1024, %edx movq 8(%rsp), %rsi movq %r15, %rdi call cudaMemcpy@PLT leaq 64(%r15), %rbp leaq 1088(%r15), %r14 leaq .LC0(%rip), %r12 leaq .LC1(%rip), %r13 .L13: leaq -64(%rbp), %rbx .L14: movl (%rbx), %edx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addq $4, %rbx cmpq %rbp, %rbx jne .L14 movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addq $64, %rbp cmpq %r14, %rbp jne .L13 movq %r15, %rdi call free@PLT movq 8(%rsp), %rdi call cudaFree@PLT movq 40(%rsp), %rax subq %fs:40, %rax jne .L20 movl $0, %eax addq $56, %rsp .cfi_remember_state .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .L19: .cfi_restore_state movq 8(%rsp), %rdi call _Z25__device_stub__Z6kernelPiPi jmp .L12 .L20: call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC2: .string "_Z6kernelPi" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC2(%rip), %rdx movq %rdx, %rcx leaq _Z6kernelPi(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); cudaMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); cudaMemcpy(host_array, device_array, num_bytes, cudaMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); cudaFree(device_array); return 0; }
#include <hip/hip_runtime.h> #include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); hipMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); hipMemcpy(host_array, device_array, num_bytes, hipMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); hipFree(device_array); return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); hipMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); hipMemcpy(host_array, device_array, num_bytes, hipMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); hipFree(device_array); return 0; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6kernelPi .globl _Z6kernelPi .p2align 8 .type _Z6kernelPi,@function _Z6kernelPi: s_clause 0x1 s_load_b32 s2, s[0:1], 0x14 s_load_b32 s3, s[0:1], 0x8 v_bfe_u32 v1, v0, 10, 10 v_and_b32_e32 v0, 0x3ff, v0 s_load_b64 s[0:1], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_lshr_b32 s4, s2, 16 s_and_b32 s2, s2, 0xffff v_mad_u64_u32 v[2:3], null, s15, s4, v[1:2] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mad_u64_u32 v[3:4], null, v2, s3, s[14:15] v_mad_u64_u32 v[1:2], null, v3, s2, v[0:1] s_mul_i32 s2, s3, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) s_add_i32 s2, s2, s14 v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_lshlrev_b64 v[0:1], 2, v[1:2] v_mov_b32_e32 v2, s2 v_add_co_u32 v0, vcc_lo, s0, v0 s_delay_alu instid0(VALU_DEP_3) v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo global_store_b32 v[0:1], v2, off s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6kernelPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 264 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 5 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6kernelPi, .Lfunc_end0-_Z6kernelPi .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .offset: 8 .size: 4 .value_kind: hidden_block_count_x - .offset: 12 .size: 4 .value_kind: hidden_block_count_y - .offset: 16 .size: 4 .value_kind: hidden_block_count_z - .offset: 20 .size: 2 .value_kind: hidden_group_size_x - .offset: 22 .size: 2 .value_kind: hidden_group_size_y - .offset: 24 .size: 2 .value_kind: hidden_group_size_z - .offset: 26 .size: 2 .value_kind: hidden_remainder_x - .offset: 28 .size: 2 .value_kind: hidden_remainder_y - .offset: 30 .size: 2 .value_kind: hidden_remainder_z - .offset: 48 .size: 8 .value_kind: hidden_global_offset_x - .offset: 56 .size: 8 .value_kind: hidden_global_offset_y - .offset: 64 .size: 8 .value_kind: hidden_global_offset_z - .offset: 72 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 264 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6kernelPi .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z6kernelPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 5 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include <stdio.h> #include <stdlib.h> __global__ void kernel(int *array){ int index_x = blockIdx.x * blockDim.x + threadIdx.x; int index_y = blockIdx.y * blockDim.y + threadIdx.y; int grid_width = gridDim.x * blockDim.x; int index = index_y * grid_width + index_x; int result = blockIdx.y * gridDim.x + blockIdx.x; array[index] = result; } int main(){ int num_elements_x = 16; int num_elements_y = 16; int num_bytes = num_elements_x * num_elements_y * sizeof(int); int *device_array = NULL; int *host_array = 0; host_array = (int *) malloc(num_bytes); hipMalloc((void **) &device_array, num_bytes); dim3 block_size; block_size.x = 4; block_size.y = 4; dim3 grid_size; grid_size.x = num_elements_x / block_size.x; grid_size.y = num_elements_y / block_size.y; kernel<<<grid_size, block_size>>>(device_array); hipMemcpy(host_array, device_array, num_bytes, hipMemcpyDeviceToHost); int row, col; for(row = 0; row <num_elements_y; ++row){ for(col = 0; col<num_elements_x; ++col) printf("%2d ", host_array[row*num_elements_x + col]); printf("\n"); } free(host_array); hipFree(device_array); return 0; }
.text .file "two_d_kernel.hip" .globl _Z21__device_stub__kernelPi # -- Begin function _Z21__device_stub__kernelPi .p2align 4, 0x90 .type _Z21__device_stub__kernelPi,@function _Z21__device_stub__kernelPi: # @_Z21__device_stub__kernelPi .cfi_startproc # %bb.0: subq $72, %rsp .cfi_def_cfa_offset 80 movq %rdi, 64(%rsp) leaq 64(%rsp), %rax movq %rax, (%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d movq %rsp, %r9 movl $_Z6kernelPi, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $88, %rsp .cfi_adjust_cfa_offset -88 retq .Lfunc_end0: .size _Z21__device_stub__kernelPi, .Lfunc_end0-_Z21__device_stub__kernelPi .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %r15 .cfi_def_cfa_offset 16 pushq %r14 .cfi_def_cfa_offset 24 pushq %r12 .cfi_def_cfa_offset 32 pushq %rbx .cfi_def_cfa_offset 40 subq $88, %rsp .cfi_def_cfa_offset 128 .cfi_offset %rbx, -40 .cfi_offset %r12, -32 .cfi_offset %r14, -24 .cfi_offset %r15, -16 movq $0, 8(%rsp) movl $1024, %edi # imm = 0x400 callq malloc movq %rax, %rbx leaq 8(%rsp), %rdi movl $1024, %esi # imm = 0x400 callq hipMalloc movabsq $17179869188, %rdi # imm = 0x400000004 movl $1, %esi movq %rdi, %rdx movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_2 # %bb.1: movq 8(%rsp), %rax movq %rax, 80(%rsp) leaq 80(%rsp), %rax movq %rax, 16(%rsp) leaq 64(%rsp), %rdi leaq 48(%rsp), %rsi leaq 40(%rsp), %rdx leaq 32(%rsp), %rcx callq __hipPopCallConfiguration movq 64(%rsp), %rsi movl 72(%rsp), %edx movq 48(%rsp), %rcx movl 56(%rsp), %r8d leaq 16(%rsp), %r9 movl $_Z6kernelPi, %edi pushq 32(%rsp) .cfi_adjust_cfa_offset 8 pushq 48(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_2: movq 8(%rsp), %rsi movl $1024, %edx # imm = 0x400 movq %rbx, %rdi movl $2, %ecx callq hipMemcpy movq %rbx, %r14 xorl %r15d, %r15d .p2align 4, 0x90 .LBB1_3: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB1_4 Depth 2 xorl %r12d, %r12d .p2align 4, 0x90 .LBB1_4: # Parent Loop BB1_3 Depth=1 # => This Inner Loop Header: Depth=2 movl (%r14,%r12,4), %esi movl $.L.str, %edi xorl %eax, %eax callq printf incq %r12 cmpq $16, %r12 jne .LBB1_4 # %bb.5: # in Loop: Header=BB1_3 Depth=1 movl $10, %edi callq putchar@PLT incq %r15 addq $64, %r14 cmpq $16, %r15 jne .LBB1_3 # %bb.6: movq %rbx, %rdi callq free movq 8(%rsp), %rdi callq hipFree xorl %eax, %eax addq $88, %rsp .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %r12 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6kernelPi, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z6kernelPi,@object # @_Z6kernelPi .section .rodata,"a",@progbits .globl _Z6kernelPi .p2align 3, 0x0 _Z6kernelPi: .quad _Z21__device_stub__kernelPi .size _Z6kernelPi, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "%2d " .size .L.str, 5 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6kernelPi" .size .L__unnamed_1, 12 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__kernelPi .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6kernelPi .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z6kernelPi .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R2, SR_CTAID.Y ; /* 0x0000000000027919 */ /* 0x000e220000002600 */ /*0020*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fc60000000a00 */ /*0030*/ S2R R7, SR_TID.Y ; /* 0x0000000000077919 */ /* 0x000e280000002200 */ /*0040*/ S2R R5, SR_CTAID.X ; /* 0x0000000000057919 */ /* 0x000e680000002500 */ /*0050*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000ea20000002100 */ /*0060*/ IMAD R0, R2, c[0x0][0x4], R7 ; /* 0x0000010002007a24 */ /* 0x001fe200078e0207 */ /*0070*/ HFMA2.MMA R7, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff077435 */ /* 0x000fc600000001ff */ /*0080*/ IMAD R0, R0, c[0x0][0xc], R5.reuse ; /* 0x0000030000007a24 */ /* 0x102fe400078e0205 */ /*0090*/ IMAD R5, R2, c[0x0][0xc], R5 ; /* 0x0000030002057a24 */ /* 0x000fe400078e0205 */ /*00a0*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x004fc800078e0203 */ /*00b0*/ IMAD.WIDE R2, R0, R7, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fca00078e0207 */ /*00c0*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x000fe2000c101904 */ /*00d0*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*00e0*/ BRA 0xe0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0100*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0110*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0120*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0130*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6kernelPi .globl _Z6kernelPi .p2align 8 .type _Z6kernelPi,@function _Z6kernelPi: s_clause 0x1 s_load_b32 s2, s[0:1], 0x14 s_load_b32 s3, s[0:1], 0x8 v_bfe_u32 v1, v0, 10, 10 v_and_b32_e32 v0, 0x3ff, v0 s_load_b64 s[0:1], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_lshr_b32 s4, s2, 16 s_and_b32 s2, s2, 0xffff v_mad_u64_u32 v[2:3], null, s15, s4, v[1:2] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mad_u64_u32 v[3:4], null, v2, s3, s[14:15] v_mad_u64_u32 v[1:2], null, v3, s2, v[0:1] s_mul_i32 s2, s3, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) s_add_i32 s2, s2, s14 v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_lshlrev_b64 v[0:1], 2, v[1:2] v_mov_b32_e32 v2, s2 v_add_co_u32 v0, vcc_lo, s0, v0 s_delay_alu instid0(VALU_DEP_3) v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo global_store_b32 v[0:1], v2, off s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6kernelPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 264 .amdhsa_user_sgpr_count 14 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 1 .amdhsa_next_free_vgpr 5 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6kernelPi, .Lfunc_end0-_Z6kernelPi .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .offset: 8 .size: 4 .value_kind: hidden_block_count_x - .offset: 12 .size: 4 .value_kind: hidden_block_count_y - .offset: 16 .size: 4 .value_kind: hidden_block_count_z - .offset: 20 .size: 2 .value_kind: hidden_group_size_x - .offset: 22 .size: 2 .value_kind: hidden_group_size_y - .offset: 24 .size: 2 .value_kind: hidden_group_size_z - .offset: 26 .size: 2 .value_kind: hidden_remainder_x - .offset: 28 .size: 2 .value_kind: hidden_remainder_y - .offset: 30 .size: 2 .value_kind: hidden_remainder_z - .offset: 48 .size: 8 .value_kind: hidden_global_offset_x - .offset: 56 .size: 8 .value_kind: hidden_global_offset_y - .offset: 64 .size: 8 .value_kind: hidden_global_offset_z - .offset: 72 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 264 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6kernelPi .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z6kernelPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 5 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_000ff563_00000000-6_two_d_kernel.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z25__device_stub__Z6kernelPiPi .type _Z25__device_stub__Z6kernelPiPi, @function _Z25__device_stub__Z6kernelPiPi: .LFB2082: .cfi_startproc endbr64 subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 8(%rsp) movq %fs:40, %rax movq %rax, 88(%rsp) xorl %eax, %eax leaq 8(%rsp), %rax movq %rax, 80(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 88(%rsp), %rax subq %fs:40, %rax jne .L8 addq $104, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 120 pushq 24(%rsp) .cfi_def_cfa_offset 128 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z6kernelPi(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 112 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z25__device_stub__Z6kernelPiPi, .-_Z25__device_stub__Z6kernelPiPi .globl _Z6kernelPi .type _Z6kernelPi, @function _Z6kernelPi: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z25__device_stub__Z6kernelPiPi addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z6kernelPi, .-_Z6kernelPi .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "%2d " .LC1: .string "\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $56, %rsp .cfi_def_cfa_offset 112 movq %fs:40, %rax movq %rax, 40(%rsp) xorl %eax, %eax movq $0, 8(%rsp) movl $1024, %edi call malloc@PLT movq %rax, %r15 leaq 8(%rsp), %rdi movl $1024, %esi call cudaMalloc@PLT movl $4, 28(%rsp) movl $4, 32(%rsp) movl $4, 16(%rsp) movl $4, 20(%rsp) movl $0, %r9d movl $0, %r8d movq 16(%rsp), %rdx movl $1, %ecx movq 28(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L19 .L12: movl $2, %ecx movl $1024, %edx movq 8(%rsp), %rsi movq %r15, %rdi call cudaMemcpy@PLT leaq 64(%r15), %rbp leaq 1088(%r15), %r14 leaq .LC0(%rip), %r12 leaq .LC1(%rip), %r13 .L13: leaq -64(%rbp), %rbx .L14: movl (%rbx), %edx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addq $4, %rbx cmpq %rbp, %rbx jne .L14 movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addq $64, %rbp cmpq %r14, %rbp jne .L13 movq %r15, %rdi call free@PLT movq 8(%rsp), %rdi call cudaFree@PLT movq 40(%rsp), %rax subq %fs:40, %rax jne .L20 movl $0, %eax addq $56, %rsp .cfi_remember_state .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .L19: .cfi_restore_state movq 8(%rsp), %rdi call _Z25__device_stub__Z6kernelPiPi jmp .L12 .L20: call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC2: .string "_Z6kernelPi" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC2(%rip), %rdx movq %rdx, %rcx leaq _Z6kernelPi(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "two_d_kernel.hip" .globl _Z21__device_stub__kernelPi # -- Begin function _Z21__device_stub__kernelPi .p2align 4, 0x90 .type _Z21__device_stub__kernelPi,@function _Z21__device_stub__kernelPi: # @_Z21__device_stub__kernelPi .cfi_startproc # %bb.0: subq $72, %rsp .cfi_def_cfa_offset 80 movq %rdi, 64(%rsp) leaq 64(%rsp), %rax movq %rax, (%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d movq %rsp, %r9 movl $_Z6kernelPi, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $88, %rsp .cfi_adjust_cfa_offset -88 retq .Lfunc_end0: .size _Z21__device_stub__kernelPi, .Lfunc_end0-_Z21__device_stub__kernelPi .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %r15 .cfi_def_cfa_offset 16 pushq %r14 .cfi_def_cfa_offset 24 pushq %r12 .cfi_def_cfa_offset 32 pushq %rbx .cfi_def_cfa_offset 40 subq $88, %rsp .cfi_def_cfa_offset 128 .cfi_offset %rbx, -40 .cfi_offset %r12, -32 .cfi_offset %r14, -24 .cfi_offset %r15, -16 movq $0, 8(%rsp) movl $1024, %edi # imm = 0x400 callq malloc movq %rax, %rbx leaq 8(%rsp), %rdi movl $1024, %esi # imm = 0x400 callq hipMalloc movabsq $17179869188, %rdi # imm = 0x400000004 movl $1, %esi movq %rdi, %rdx movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_2 # %bb.1: movq 8(%rsp), %rax movq %rax, 80(%rsp) leaq 80(%rsp), %rax movq %rax, 16(%rsp) leaq 64(%rsp), %rdi leaq 48(%rsp), %rsi leaq 40(%rsp), %rdx leaq 32(%rsp), %rcx callq __hipPopCallConfiguration movq 64(%rsp), %rsi movl 72(%rsp), %edx movq 48(%rsp), %rcx movl 56(%rsp), %r8d leaq 16(%rsp), %r9 movl $_Z6kernelPi, %edi pushq 32(%rsp) .cfi_adjust_cfa_offset 8 pushq 48(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_2: movq 8(%rsp), %rsi movl $1024, %edx # imm = 0x400 movq %rbx, %rdi movl $2, %ecx callq hipMemcpy movq %rbx, %r14 xorl %r15d, %r15d .p2align 4, 0x90 .LBB1_3: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB1_4 Depth 2 xorl %r12d, %r12d .p2align 4, 0x90 .LBB1_4: # Parent Loop BB1_3 Depth=1 # => This Inner Loop Header: Depth=2 movl (%r14,%r12,4), %esi movl $.L.str, %edi xorl %eax, %eax callq printf incq %r12 cmpq $16, %r12 jne .LBB1_4 # %bb.5: # in Loop: Header=BB1_3 Depth=1 movl $10, %edi callq putchar@PLT incq %r15 addq $64, %r14 cmpq $16, %r15 jne .LBB1_3 # %bb.6: movq %rbx, %rdi callq free movq 8(%rsp), %rdi callq hipFree xorl %eax, %eax addq $88, %rsp .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %r12 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6kernelPi, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z6kernelPi,@object # @_Z6kernelPi .section .rodata,"a",@progbits .globl _Z6kernelPi .p2align 3, 0x0 _Z6kernelPi: .quad _Z21__device_stub__kernelPi .size _Z6kernelPi, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "%2d " .size .L.str, 5 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6kernelPi" .size .L__unnamed_1, 12 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__kernelPi .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6kernelPi .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
code for sm_80 Function : _Z8mykernelv .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0020*/ BRA 0x20; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0030*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0040*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0050*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0060*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0070*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0080*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
.file "tmpxft_0010205b_00000000-6_cuda-hello1.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z26__device_stub__Z8mykernelvv .type _Z26__device_stub__Z8mykernelvv, @function _Z26__device_stub__Z8mykernelvv: .LFB2082: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 72(%rsp), %rax subq %fs:40, %rax jne .L8 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z8mykernelv(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z26__device_stub__Z8mykernelvv, .-_Z26__device_stub__Z8mykernelvv .globl _Z8mykernelv .type _Z8mykernelv, @function _Z8mykernelv: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z26__device_stub__Z8mykernelvv addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z8mykernelv, .-_Z8mykernelv .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "Hello, world!\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 subq $40, %rsp .cfi_def_cfa_offset 48 movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 8(%rsp) movl $1, 12(%rsp) movl $0, %r9d movl $0, %r8d movq 20(%rsp), %rdx movl $1, %ecx movq 8(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L14 .L12: leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $0, %eax addq $40, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L14: .cfi_restore_state call _Z26__device_stub__Z8mykernelvv jmp .L12 .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC1: .string "_Z8mykernelv" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC1(%rip), %rdx movq %rdx, %rcx leaq _Z8mykernelv(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <hip/hip_runtime.h> #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <hip/hip_runtime.h> #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z8mykernelv .globl _Z8mykernelv .p2align 8 .type _Z8mykernelv,@function _Z8mykernelv: s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z8mykernelv .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 0 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 0 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 1 .amdhsa_next_free_sgpr 1 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z8mykernelv, .Lfunc_end0-_Z8mykernelv .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: [] .group_segment_fixed_size: 0 .kernarg_segment_align: 4 .kernarg_segment_size: 0 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z8mykernelv .private_segment_fixed_size: 0 .sgpr_count: 0 .sgpr_spill_count: 0 .symbol: _Z8mykernelv.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 0 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
/**************************************************************************** * * cuda-hello1.cu - Hello world with CUDA (with dummy device code) * * Written in 2017 by Moreno Marzolla <moreno.marzolla(at)unibo.it> * * To the extent possible under law, the author(s) have dedicated all * copyright and related and neighboring rights to this software to the * public domain worldwide. This software is distributed without any warranty. * * You should have received a copy of the CC0 Public Domain Dedication * along with this software. If not, see * <http://creativecommons.org/publicdomain/zero/1.0/>. * * --------------------------------------------------------------------------- * * Based on the example shown in the CUDA toolkit documentation * http://docs.nvidia.com/cuda/cuda-c-programming-guide/ * * Compile with: * nvcc cuda-hello1.cu -o cuda-hello1 * * Run with: * ./cuda-hello1 * ****************************************************************************/ #include <hip/hip_runtime.h> #include <stdio.h> __global__ void mykernel( void ) { } int main( void ) { mykernel<<<1,1>>>( ); printf("Hello, world!\n"); return 0; }
.text .file "cuda-hello1.hip" .globl _Z23__device_stub__mykernelv # -- Begin function _Z23__device_stub__mykernelv .p2align 4, 0x90 .type _Z23__device_stub__mykernelv,@function _Z23__device_stub__mykernelv: # @_Z23__device_stub__mykernelv .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z8mykernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z23__device_stub__mykernelv, .Lfunc_end0-_Z23__device_stub__mykernelv .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 movabsq $4294967297, %rdi # imm = 0x100000001 movl $1, %esi movq %rdi, %rdx movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_2 # %bb.1: leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z8mykernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_2: movl $.Lstr, %edi callq puts@PLT xorl %eax, %eax addq $56, %rsp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z8mykernelv, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z8mykernelv,@object # @_Z8mykernelv .section .rodata,"a",@progbits .globl _Z8mykernelv .p2align 3, 0x0 _Z8mykernelv: .quad _Z23__device_stub__mykernelv .size _Z8mykernelv, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z8mykernelv" .size .L__unnamed_1, 13 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "Hello, world!" .size .Lstr, 14 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z23__device_stub__mykernelv .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z8mykernelv .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z8mykernelv .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0020*/ BRA 0x20; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0030*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0040*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0050*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0060*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0070*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0080*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z8mykernelv .globl _Z8mykernelv .p2align 8 .type _Z8mykernelv,@function _Z8mykernelv: s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z8mykernelv .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 0 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 0 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 1 .amdhsa_next_free_sgpr 1 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z8mykernelv, .Lfunc_end0-_Z8mykernelv .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: [] .group_segment_fixed_size: 0 .kernarg_segment_align: 4 .kernarg_segment_size: 0 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z8mykernelv .private_segment_fixed_size: 0 .sgpr_count: 0 .sgpr_spill_count: 0 .symbol: _Z8mykernelv.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 0 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_0010205b_00000000-6_cuda-hello1.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z26__device_stub__Z8mykernelvv .type _Z26__device_stub__Z8mykernelvv, @function _Z26__device_stub__Z8mykernelvv: .LFB2082: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 72(%rsp), %rax subq %fs:40, %rax jne .L8 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z8mykernelv(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z26__device_stub__Z8mykernelvv, .-_Z26__device_stub__Z8mykernelvv .globl _Z8mykernelv .type _Z8mykernelv, @function _Z8mykernelv: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z26__device_stub__Z8mykernelvv addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z8mykernelv, .-_Z8mykernelv .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "Hello, world!\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 subq $40, %rsp .cfi_def_cfa_offset 48 movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 8(%rsp) movl $1, 12(%rsp) movl $0, %r9d movl $0, %r8d movq 20(%rsp), %rdx movl $1, %ecx movq 8(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L14 .L12: leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $0, %eax addq $40, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L14: .cfi_restore_state call _Z26__device_stub__Z8mykernelvv jmp .L12 .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC1: .string "_Z8mykernelv" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC1(%rip), %rdx movq %rdx, %rcx leaq _Z8mykernelv(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "cuda-hello1.hip" .globl _Z23__device_stub__mykernelv # -- Begin function _Z23__device_stub__mykernelv .p2align 4, 0x90 .type _Z23__device_stub__mykernelv,@function _Z23__device_stub__mykernelv: # @_Z23__device_stub__mykernelv .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z8mykernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z23__device_stub__mykernelv, .Lfunc_end0-_Z23__device_stub__mykernelv .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 movabsq $4294967297, %rdi # imm = 0x100000001 movl $1, %esi movq %rdi, %rdx movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_2 # %bb.1: leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z8mykernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_2: movl $.Lstr, %edi callq puts@PLT xorl %eax, %eax addq $56, %rsp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z8mykernelv, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z8mykernelv,@object # @_Z8mykernelv .section .rodata,"a",@progbits .globl _Z8mykernelv .p2align 3, 0x0 _Z8mykernelv: .quad _Z23__device_stub__mykernelv .size _Z8mykernelv, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z8mykernelv" .size .L__unnamed_1, 13 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "Hello, world!" .size .Lstr, 14 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z23__device_stub__mykernelv .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z8mykernelv .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
/* * * Authors: Steven Faulkner, Blaine Oakley, Felipe Gutierrez * * Final Project for CIS 4930, Implementation of K-means clustering * optimized with shared memory and reduction methods. * * To compile nvcc kmeans.cu * To run ./a.out "input.txt" "K" "iterations" * * @data file: is the specified input file * @k: is the number of centroids tro be determined * @iterations: total number of iterations to be performed * */ #include<cuda_runtime.h> #include <cstdlib> #include <ctime> #include <algorithm> #include <fstream> #include <iostream> #include <sstream> #include <vector> #include <string> void ErrorCheck( cudaError_t err, const char op[]){ /* * @cudaError_t err: cuda api call catcher. * * all cuda api's usually return * cudaSuccess * @const char op[]: error string will tell where api call * failed * * Error Catch Function, will wrap all malloc, memset and * memcopy calls * */ if( err != cudaSuccess ) { printf("CUDA Error: %s, %s ", op, cudaGetErrorString(err)); exit(EXIT_FAILURE); } } struct Point { /* * * Struct for the imported data points * * Follows CADRe stanadards, moved cuadaMalloc and cudaFree * to the constructors, destrucor. Just makes the code look * a little cleaner * * * @double x: x data point * @double y: y data point * @int size: size ofdata point * @int bytes: # of bytes allocated for storage */ Point(long dataSize) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(cudaMalloc(&x, num_of_bytes),"Allocate x data\n"); ErrorCheck(cudaMalloc(&y, num_of_bytes), "Allocate y data\n"); ErrorCheck(cudaMemset(x, 0, num_of_bytes), "Set x data to '0'\n"); ErrorCheck(cudaMemset(y, 0, num_of_bytes), "Set y data to '0'\n"); } Point(long dataSize, std::vector<double>& x_data, std::vector<double>& y_data) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(cudaMalloc(&x, num_of_bytes),"Allocate x array\n"); ErrorCheck(cudaMalloc(&y, num_of_bytes), "Allocate y array\n");; ErrorCheck(cudaMemcpy(x, x_data.data(), num_of_bytes, cudaMemcpyHostToDevice),"Copy x array to device\n"); ErrorCheck(cudaMemcpy(y, y_data.data(), num_of_bytes, cudaMemcpyHostToDevice), "Copy y array to device\n"); } ~Point() { ErrorCheck(cudaFree(x),"Freeing x \n"); ErrorCheck(cudaFree(y),"Freeing y \n"); } double* x; double* y; long dataSize; int num_of_bytes; }; __device__ double euclidean_distance(double x_1, double y_1, double x_2, double y_2) { /* * * @double x_1, y_1, x_2, y_2: x and y coordinates from Point struct * * * Standard Euclidean Distance function returns a straight line distance * Point A to Point B. * * //If I Have Time\\ * We can exapnd this for higher dimensional data(add more x_n - x_m) or preprocess our data * with PCA(prinicpal component analysis) to reduce to 2 dimensions * */ return sqrt((x_1 - x_2) * (x_1 - x_2) + (y_1 - y_2) * (y_1 - y_2)); } __global__ void Assignment(double * data_x, double * data_y, int data_size, double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * @double* data_x: array of x data points * @double* data_y: array of y data points * @int data_size: size of data array * @centroid_x: array of x centroids * @centroid_y: array of y centroids * @device_new_x: updated array for x * @device_new_y: updated array for y * @int k: # of centroids * @int* device_counts: int array, holds count * for total points among all centodsi * * K-Means Algorithm : each x,y cluster is assigned to its closest centroid * then each centroid is averaged over all the points * assigned to it and then updated with this new value */ extern __shared__ double shared_mem[]; int reg = threadIdx.x; int unique_index = blockIdx.x * blockDim.x + threadIdx.x; //out of range if (unique_index >= data_size) return; //loading in centroids if (reg < k) { //1D array seperated by K values for each point shared_mem[reg] = centroids_x[reg]; shared_mem[k + reg] = centroids_y[reg]; } __syncthreads(); // load to registers double x_value = data_x[unique_index]; double y_value = data_y[unique_index]; //none of our distance values will be large enough to use stl::infinity or FLT_MAX, // arbitrary sentinal values will suffice for these two variables double min_distance = 1000; int best_cluster = -1; //iterate over the all centroids keeping track of closest one for storage for (int cluster = 0; cluster < k; ++cluster) { double distance = euclidean_distance(x_value, y_value, shared_mem[cluster], shared_mem[k + cluster]); if (distance < min_distance) { min_distance = distance; best_cluster = cluster; } } __syncthreads(); // tree-reduction start int x = reg; int y = reg + blockDim.x; int count = reg + blockDim.x + blockDim.x; //check if thread is assigned to centroid, writing to local memory if true or 0 if false for (int cluster = 0; cluster < k; ++cluster) { shared_mem[x] = (best_cluster == cluster) ? x_value : 0; shared_mem[y] = (best_cluster == cluster) ? y_value : 0; shared_mem[count] = (best_cluster == cluster) ? 1 : 0; __syncthreads(); // Reduction for local memory. for (int stride = blockDim.x / 2; stride > 0; stride /= 2) { if (reg < stride) { shared_mem[x] += shared_mem[x + stride]; shared_mem[y] += shared_mem[y + stride]; shared_mem[count] += shared_mem[count + stride]; } __syncthreads(); } //push_back from shared mem to update array if (reg == 0) { int cluster_index = blockIdx.x * k + cluster; device_new_x[cluster_index] = shared_mem[x]; device_new_y[cluster_index] = shared_mem[y]; device_counts[cluster_index] = shared_mem[count]; } __syncthreads(); } } __global__ void centroid_recompute(double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * * @double * centroids_x: array for x centroids * @double * centroids_y: array for y centroids * @double * new_sums_x: updated x array * @double * new_sums_y: updated y array * @int k: # of centroids * @int * device_counts: int array,holds count * for total points among all centodsi * * * * centroid Recompute: Recomputes the centroids from all * points assigned to it. * * */ //local memory declaration extern __shared__ double shared_mem[]; int reg = threadIdx.x; int b_Dim = blockDim.x; //load into local memory shared_mem[reg] = device_new_x[reg]; shared_mem[b_Dim + reg] = device_new_y[reg]; __syncthreads(); //summination of every stride length block for (int stride = blockDim.x / 2; stride >= k; stride /= 2) { if (reg < stride) { shared_mem[reg] += shared_mem[reg + stride]; shared_mem[b_Dim + reg] += shared_mem[b_Dim + reg + stride]; } __syncthreads(); } //recomputing centroid centers if (reg < k) { int count = max(1, device_counts[reg]); centroids_x[reg] = device_new_x[reg] / count; centroids_y[reg] = device_new_y[reg] / count; device_new_y[reg] = 0; device_new_x[reg] = 0; device_counts[reg] = 0; } } int main(int argc, const char * argv[]) { if (argc < 4) { std::cout << "Incorrect startup execution: <./a.out 'input.txt' 'K' 'iterations' " << std::endl; std::exit(EXIT_FAILURE); } int k = std::atoi(argv[2]); int number_of_iterations = std::atoi(argv[3]); std::vector<double> x_data; std::vector<double> y_data; std::ifstream stream_in(argv[1]); std::string line; if(stream_in){ while (std::getline(stream_in, line)) { std::istringstream line_stream(line); double x, y; line_stream >> x >> y; x_data.push_back(x); y_data.push_back(y); } } else{ std::cout << "Error Opening File" << std::endl; return(EXIT_FAILURE); } // dinput data up to 1,000,000 points long number_of_elements = x_data.size(); // centroids are initalized to first k points of array // in order to chose 'randomly' we shuffle the array // input array after we initilize the devize point array // and before we initilize the centroid array Point device_data(number_of_elements, x_data, y_data); std::srand(std::time(0)); random_shuffle(x_data.begin(),x_data.end()); random_shuffle(y_data.begin(),y_data.end()); Point device_centroids(k, x_data, y_data); int threads = 1024; int blocks = (number_of_elements + threads - 1) / threads; std::cout << "\nProcessing " << number_of_elements << " points\n" << std::endl; int kmeans_shared_memory = 3 * threads * sizeof(double); int centroid_reduction_memory = 2 * k * blocks * sizeof(double); Point device_sum(k * blocks); int * device_count; ErrorCheck(cudaMalloc(&device_count, k * blocks * sizeof(int)), "Allocate size for device_count\n"); ErrorCheck(cudaMemset(device_count, 0, k * blocks * sizeof(int)),"Set device_count to '0' \n"); // cuda api time start cudaEvent_t start,stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start,0); // start iteration loop, assigning and updating centroid on each iteration for (size_t iteration = 0; iteration < number_of_iterations; ++iteration) { Assignment<<<blocks, threads, kmeans_shared_memory>>>(device_data.x,device_data.y, device_data.dataSize, device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); cudaDeviceSynchronize(); centroid_recompute<<<1, k * blocks, centroid_reduction_memory>>>(device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); cudaDeviceSynchronize(); } // cuda api time stop cudaEventRecord(stop,0); cudaEventSynchronize(stop); float elapsedTime; cudaEventElapsedTime(&elapsedTime,start,stop); cudaEventDestroy(start); cudaEventDestroy(stop); std::string unit = ""; unit = (elapsedTime > 999) ? "seconds" : "milliseconds"; elapsedTime = (elapsedTime > 999 ) ? elapsedTime/1000 : elapsedTime; std::cout << "Elapsed time of kernal calls: " << elapsedTime << " " << unit << "\n" << std::endl; ErrorCheck(cudaFree(device_count),"Freeing Device Memory"); std::vector<double> centroid_x(k, 0); std::vector<double> centroid_y(k, 0); ErrorCheck(cudaMemcpy(centroid_x.data(), device_centroids.x, device_centroids.num_of_bytes, cudaMemcpyDeviceToHost), "Moving Array back to host\n"); ErrorCheck(cudaMemcpy(centroid_y.data(), device_centroids.y, device_centroids.num_of_bytes, cudaMemcpyDeviceToHost), "Moving Array back to host\n"); std::cout << "centroids:" << std::endl; for (size_t cluster = 0; cluster < k; ++cluster) { std::cout << centroid_x[cluster] << " " << centroid_y[cluster] << std::endl; } std::cout << "\n" << std::endl; return(EXIT_SUCCESS); }
/* * * Authors: Steven Faulkner, Blaine Oakley, Felipe Gutierrez * * Final Project for CIS 4930, Implementation of K-means clustering * optimized with shared memory and reduction methods. * * To compile nvcc kmeans.cu * To run ./a.out "input.txt" "K" "iterations" * * @data file: is the specified input file * @k: is the number of centroids tro be determined * @iterations: total number of iterations to be performed * */ #include<hip/hip_runtime.h> #include <cstdlib> #include <ctime> #include <algorithm> #include <fstream> #include <iostream> #include <sstream> #include <vector> #include <string> void ErrorCheck( hipError_t err, const char op[]){ /* * @cudaError_t err: cuda api call catcher. * * all cuda api's usually return * cudaSuccess * @const char op[]: error string will tell where api call * failed * * Error Catch Function, will wrap all malloc, memset and * memcopy calls * */ if( err != hipSuccess ) { printf("CUDA Error: %s, %s ", op, hipGetErrorString(err)); exit(EXIT_FAILURE); } } struct Point { /* * * Struct for the imported data points * * Follows CADRe stanadards, moved cuadaMalloc and cudaFree * to the constructors, destrucor. Just makes the code look * a little cleaner * * * @double x: x data point * @double y: y data point * @int size: size ofdata point * @int bytes: # of bytes allocated for storage */ Point(long dataSize) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(hipMalloc(&x, num_of_bytes),"Allocate x data\n"); ErrorCheck(hipMalloc(&y, num_of_bytes), "Allocate y data\n"); ErrorCheck(hipMemset(x, 0, num_of_bytes), "Set x data to '0'\n"); ErrorCheck(hipMemset(y, 0, num_of_bytes), "Set y data to '0'\n"); } Point(long dataSize, std::vector<double>& x_data, std::vector<double>& y_data) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(hipMalloc(&x, num_of_bytes),"Allocate x array\n"); ErrorCheck(hipMalloc(&y, num_of_bytes), "Allocate y array\n");; ErrorCheck(hipMemcpy(x, x_data.data(), num_of_bytes, hipMemcpyHostToDevice),"Copy x array to device\n"); ErrorCheck(hipMemcpy(y, y_data.data(), num_of_bytes, hipMemcpyHostToDevice), "Copy y array to device\n"); } ~Point() { ErrorCheck(hipFree(x),"Freeing x \n"); ErrorCheck(hipFree(y),"Freeing y \n"); } double* x; double* y; long dataSize; int num_of_bytes; }; __device__ double euclidean_distance(double x_1, double y_1, double x_2, double y_2) { /* * * @double x_1, y_1, x_2, y_2: x and y coordinates from Point struct * * * Standard Euclidean Distance function returns a straight line distance * Point A to Point B. * * //If I Have Time\\ * We can exapnd this for higher dimensional data(add more x_n - x_m) or preprocess our data * with PCA(prinicpal component analysis) to reduce to 2 dimensions * */ return sqrt((x_1 - x_2) * (x_1 - x_2) + (y_1 - y_2) * (y_1 - y_2)); } __global__ void Assignment(double * data_x, double * data_y, int data_size, double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * @double* data_x: array of x data points * @double* data_y: array of y data points * @int data_size: size of data array * @centroid_x: array of x centroids * @centroid_y: array of y centroids * @device_new_x: updated array for x * @device_new_y: updated array for y * @int k: # of centroids * @int* device_counts: int array, holds count * for total points among all centodsi * * K-Means Algorithm : each x,y cluster is assigned to its closest centroid * then each centroid is averaged over all the points * assigned to it and then updated with this new value */ extern __shared__ double shared_mem[]; int reg = threadIdx.x; int unique_index = blockIdx.x * blockDim.x + threadIdx.x; //out of range if (unique_index >= data_size) return; //loading in centroids if (reg < k) { //1D array seperated by K values for each point shared_mem[reg] = centroids_x[reg]; shared_mem[k + reg] = centroids_y[reg]; } __syncthreads(); // load to registers double x_value = data_x[unique_index]; double y_value = data_y[unique_index]; //none of our distance values will be large enough to use stl::infinity or FLT_MAX, // arbitrary sentinal values will suffice for these two variables double min_distance = 1000; int best_cluster = -1; //iterate over the all centroids keeping track of closest one for storage for (int cluster = 0; cluster < k; ++cluster) { double distance = euclidean_distance(x_value, y_value, shared_mem[cluster], shared_mem[k + cluster]); if (distance < min_distance) { min_distance = distance; best_cluster = cluster; } } __syncthreads(); // tree-reduction start int x = reg; int y = reg + blockDim.x; int count = reg + blockDim.x + blockDim.x; //check if thread is assigned to centroid, writing to local memory if true or 0 if false for (int cluster = 0; cluster < k; ++cluster) { shared_mem[x] = (best_cluster == cluster) ? x_value : 0; shared_mem[y] = (best_cluster == cluster) ? y_value : 0; shared_mem[count] = (best_cluster == cluster) ? 1 : 0; __syncthreads(); // Reduction for local memory. for (int stride = blockDim.x / 2; stride > 0; stride /= 2) { if (reg < stride) { shared_mem[x] += shared_mem[x + stride]; shared_mem[y] += shared_mem[y + stride]; shared_mem[count] += shared_mem[count + stride]; } __syncthreads(); } //push_back from shared mem to update array if (reg == 0) { int cluster_index = blockIdx.x * k + cluster; device_new_x[cluster_index] = shared_mem[x]; device_new_y[cluster_index] = shared_mem[y]; device_counts[cluster_index] = shared_mem[count]; } __syncthreads(); } } __global__ void centroid_recompute(double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * * @double * centroids_x: array for x centroids * @double * centroids_y: array for y centroids * @double * new_sums_x: updated x array * @double * new_sums_y: updated y array * @int k: # of centroids * @int * device_counts: int array,holds count * for total points among all centodsi * * * * centroid Recompute: Recomputes the centroids from all * points assigned to it. * * */ //local memory declaration extern __shared__ double shared_mem[]; int reg = threadIdx.x; int b_Dim = blockDim.x; //load into local memory shared_mem[reg] = device_new_x[reg]; shared_mem[b_Dim + reg] = device_new_y[reg]; __syncthreads(); //summination of every stride length block for (int stride = blockDim.x / 2; stride >= k; stride /= 2) { if (reg < stride) { shared_mem[reg] += shared_mem[reg + stride]; shared_mem[b_Dim + reg] += shared_mem[b_Dim + reg + stride]; } __syncthreads(); } //recomputing centroid centers if (reg < k) { int count = max(1, device_counts[reg]); centroids_x[reg] = device_new_x[reg] / count; centroids_y[reg] = device_new_y[reg] / count; device_new_y[reg] = 0; device_new_x[reg] = 0; device_counts[reg] = 0; } } int main(int argc, const char * argv[]) { if (argc < 4) { std::cout << "Incorrect startup execution: <./a.out 'input.txt' 'K' 'iterations' " << std::endl; std::exit(EXIT_FAILURE); } int k = std::atoi(argv[2]); int number_of_iterations = std::atoi(argv[3]); std::vector<double> x_data; std::vector<double> y_data; std::ifstream stream_in(argv[1]); std::string line; if(stream_in){ while (std::getline(stream_in, line)) { std::istringstream line_stream(line); double x, y; line_stream >> x >> y; x_data.push_back(x); y_data.push_back(y); } } else{ std::cout << "Error Opening File" << std::endl; return(EXIT_FAILURE); } // dinput data up to 1,000,000 points long number_of_elements = x_data.size(); // centroids are initalized to first k points of array // in order to chose 'randomly' we shuffle the array // input array after we initilize the devize point array // and before we initilize the centroid array Point device_data(number_of_elements, x_data, y_data); std::srand(std::time(0)); random_shuffle(x_data.begin(),x_data.end()); random_shuffle(y_data.begin(),y_data.end()); Point device_centroids(k, x_data, y_data); int threads = 1024; int blocks = (number_of_elements + threads - 1) / threads; std::cout << "\nProcessing " << number_of_elements << " points\n" << std::endl; int kmeans_shared_memory = 3 * threads * sizeof(double); int centroid_reduction_memory = 2 * k * blocks * sizeof(double); Point device_sum(k * blocks); int * device_count; ErrorCheck(hipMalloc(&device_count, k * blocks * sizeof(int)), "Allocate size for device_count\n"); ErrorCheck(hipMemset(device_count, 0, k * blocks * sizeof(int)),"Set device_count to '0' \n"); // cuda api time start hipEvent_t start,stop; hipEventCreate(&start); hipEventCreate(&stop); hipEventRecord(start,0); // start iteration loop, assigning and updating centroid on each iteration for (size_t iteration = 0; iteration < number_of_iterations; ++iteration) { Assignment<<<blocks, threads, kmeans_shared_memory>>>(device_data.x,device_data.y, device_data.dataSize, device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); hipDeviceSynchronize(); centroid_recompute<<<1, k * blocks, centroid_reduction_memory>>>(device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); hipDeviceSynchronize(); } // cuda api time stop hipEventRecord(stop,0); hipEventSynchronize(stop); float elapsedTime; hipEventElapsedTime(&elapsedTime,start,stop); hipEventDestroy(start); hipEventDestroy(stop); std::string unit = ""; unit = (elapsedTime > 999) ? "seconds" : "milliseconds"; elapsedTime = (elapsedTime > 999 ) ? elapsedTime/1000 : elapsedTime; std::cout << "Elapsed time of kernal calls: " << elapsedTime << " " << unit << "\n" << std::endl; ErrorCheck(hipFree(device_count),"Freeing Device Memory"); std::vector<double> centroid_x(k, 0); std::vector<double> centroid_y(k, 0); ErrorCheck(hipMemcpy(centroid_x.data(), device_centroids.x, device_centroids.num_of_bytes, hipMemcpyDeviceToHost), "Moving Array back to host\n"); ErrorCheck(hipMemcpy(centroid_y.data(), device_centroids.y, device_centroids.num_of_bytes, hipMemcpyDeviceToHost), "Moving Array back to host\n"); std::cout << "centroids:" << std::endl; for (size_t cluster = 0; cluster < k; ++cluster) { std::cout << centroid_x[cluster] << " " << centroid_y[cluster] << std::endl; } std::cout << "\n" << std::endl; return(EXIT_SUCCESS); }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
/* * * Authors: Steven Faulkner, Blaine Oakley, Felipe Gutierrez * * Final Project for CIS 4930, Implementation of K-means clustering * optimized with shared memory and reduction methods. * * To compile nvcc kmeans.cu * To run ./a.out "input.txt" "K" "iterations" * * @data file: is the specified input file * @k: is the number of centroids tro be determined * @iterations: total number of iterations to be performed * */ #include<hip/hip_runtime.h> #include <cstdlib> #include <ctime> #include <algorithm> #include <fstream> #include <iostream> #include <sstream> #include <vector> #include <string> void ErrorCheck( hipError_t err, const char op[]){ /* * @cudaError_t err: cuda api call catcher. * * all cuda api's usually return * cudaSuccess * @const char op[]: error string will tell where api call * failed * * Error Catch Function, will wrap all malloc, memset and * memcopy calls * */ if( err != hipSuccess ) { printf("CUDA Error: %s, %s ", op, hipGetErrorString(err)); exit(EXIT_FAILURE); } } struct Point { /* * * Struct for the imported data points * * Follows CADRe stanadards, moved cuadaMalloc and cudaFree * to the constructors, destrucor. Just makes the code look * a little cleaner * * * @double x: x data point * @double y: y data point * @int size: size ofdata point * @int bytes: # of bytes allocated for storage */ Point(long dataSize) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(hipMalloc(&x, num_of_bytes),"Allocate x data\n"); ErrorCheck(hipMalloc(&y, num_of_bytes), "Allocate y data\n"); ErrorCheck(hipMemset(x, 0, num_of_bytes), "Set x data to '0'\n"); ErrorCheck(hipMemset(y, 0, num_of_bytes), "Set y data to '0'\n"); } Point(long dataSize, std::vector<double>& x_data, std::vector<double>& y_data) : dataSize(dataSize), num_of_bytes(dataSize * sizeof(double)) { ErrorCheck(hipMalloc(&x, num_of_bytes),"Allocate x array\n"); ErrorCheck(hipMalloc(&y, num_of_bytes), "Allocate y array\n");; ErrorCheck(hipMemcpy(x, x_data.data(), num_of_bytes, hipMemcpyHostToDevice),"Copy x array to device\n"); ErrorCheck(hipMemcpy(y, y_data.data(), num_of_bytes, hipMemcpyHostToDevice), "Copy y array to device\n"); } ~Point() { ErrorCheck(hipFree(x),"Freeing x \n"); ErrorCheck(hipFree(y),"Freeing y \n"); } double* x; double* y; long dataSize; int num_of_bytes; }; __device__ double euclidean_distance(double x_1, double y_1, double x_2, double y_2) { /* * * @double x_1, y_1, x_2, y_2: x and y coordinates from Point struct * * * Standard Euclidean Distance function returns a straight line distance * Point A to Point B. * * //If I Have Time\\ * We can exapnd this for higher dimensional data(add more x_n - x_m) or preprocess our data * with PCA(prinicpal component analysis) to reduce to 2 dimensions * */ return sqrt((x_1 - x_2) * (x_1 - x_2) + (y_1 - y_2) * (y_1 - y_2)); } __global__ void Assignment(double * data_x, double * data_y, int data_size, double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * @double* data_x: array of x data points * @double* data_y: array of y data points * @int data_size: size of data array * @centroid_x: array of x centroids * @centroid_y: array of y centroids * @device_new_x: updated array for x * @device_new_y: updated array for y * @int k: # of centroids * @int* device_counts: int array, holds count * for total points among all centodsi * * K-Means Algorithm : each x,y cluster is assigned to its closest centroid * then each centroid is averaged over all the points * assigned to it and then updated with this new value */ extern __shared__ double shared_mem[]; int reg = threadIdx.x; int unique_index = blockIdx.x * blockDim.x + threadIdx.x; //out of range if (unique_index >= data_size) return; //loading in centroids if (reg < k) { //1D array seperated by K values for each point shared_mem[reg] = centroids_x[reg]; shared_mem[k + reg] = centroids_y[reg]; } __syncthreads(); // load to registers double x_value = data_x[unique_index]; double y_value = data_y[unique_index]; //none of our distance values will be large enough to use stl::infinity or FLT_MAX, // arbitrary sentinal values will suffice for these two variables double min_distance = 1000; int best_cluster = -1; //iterate over the all centroids keeping track of closest one for storage for (int cluster = 0; cluster < k; ++cluster) { double distance = euclidean_distance(x_value, y_value, shared_mem[cluster], shared_mem[k + cluster]); if (distance < min_distance) { min_distance = distance; best_cluster = cluster; } } __syncthreads(); // tree-reduction start int x = reg; int y = reg + blockDim.x; int count = reg + blockDim.x + blockDim.x; //check if thread is assigned to centroid, writing to local memory if true or 0 if false for (int cluster = 0; cluster < k; ++cluster) { shared_mem[x] = (best_cluster == cluster) ? x_value : 0; shared_mem[y] = (best_cluster == cluster) ? y_value : 0; shared_mem[count] = (best_cluster == cluster) ? 1 : 0; __syncthreads(); // Reduction for local memory. for (int stride = blockDim.x / 2; stride > 0; stride /= 2) { if (reg < stride) { shared_mem[x] += shared_mem[x + stride]; shared_mem[y] += shared_mem[y + stride]; shared_mem[count] += shared_mem[count + stride]; } __syncthreads(); } //push_back from shared mem to update array if (reg == 0) { int cluster_index = blockIdx.x * k + cluster; device_new_x[cluster_index] = shared_mem[x]; device_new_y[cluster_index] = shared_mem[y]; device_counts[cluster_index] = shared_mem[count]; } __syncthreads(); } } __global__ void centroid_recompute(double * centroids_x, double * centroids_y, double * device_new_x, double * device_new_y, int k, int * device_counts) { /* * * * @double * centroids_x: array for x centroids * @double * centroids_y: array for y centroids * @double * new_sums_x: updated x array * @double * new_sums_y: updated y array * @int k: # of centroids * @int * device_counts: int array,holds count * for total points among all centodsi * * * * centroid Recompute: Recomputes the centroids from all * points assigned to it. * * */ //local memory declaration extern __shared__ double shared_mem[]; int reg = threadIdx.x; int b_Dim = blockDim.x; //load into local memory shared_mem[reg] = device_new_x[reg]; shared_mem[b_Dim + reg] = device_new_y[reg]; __syncthreads(); //summination of every stride length block for (int stride = blockDim.x / 2; stride >= k; stride /= 2) { if (reg < stride) { shared_mem[reg] += shared_mem[reg + stride]; shared_mem[b_Dim + reg] += shared_mem[b_Dim + reg + stride]; } __syncthreads(); } //recomputing centroid centers if (reg < k) { int count = max(1, device_counts[reg]); centroids_x[reg] = device_new_x[reg] / count; centroids_y[reg] = device_new_y[reg] / count; device_new_y[reg] = 0; device_new_x[reg] = 0; device_counts[reg] = 0; } } int main(int argc, const char * argv[]) { if (argc < 4) { std::cout << "Incorrect startup execution: <./a.out 'input.txt' 'K' 'iterations' " << std::endl; std::exit(EXIT_FAILURE); } int k = std::atoi(argv[2]); int number_of_iterations = std::atoi(argv[3]); std::vector<double> x_data; std::vector<double> y_data; std::ifstream stream_in(argv[1]); std::string line; if(stream_in){ while (std::getline(stream_in, line)) { std::istringstream line_stream(line); double x, y; line_stream >> x >> y; x_data.push_back(x); y_data.push_back(y); } } else{ std::cout << "Error Opening File" << std::endl; return(EXIT_FAILURE); } // dinput data up to 1,000,000 points long number_of_elements = x_data.size(); // centroids are initalized to first k points of array // in order to chose 'randomly' we shuffle the array // input array after we initilize the devize point array // and before we initilize the centroid array Point device_data(number_of_elements, x_data, y_data); std::srand(std::time(0)); random_shuffle(x_data.begin(),x_data.end()); random_shuffle(y_data.begin(),y_data.end()); Point device_centroids(k, x_data, y_data); int threads = 1024; int blocks = (number_of_elements + threads - 1) / threads; std::cout << "\nProcessing " << number_of_elements << " points\n" << std::endl; int kmeans_shared_memory = 3 * threads * sizeof(double); int centroid_reduction_memory = 2 * k * blocks * sizeof(double); Point device_sum(k * blocks); int * device_count; ErrorCheck(hipMalloc(&device_count, k * blocks * sizeof(int)), "Allocate size for device_count\n"); ErrorCheck(hipMemset(device_count, 0, k * blocks * sizeof(int)),"Set device_count to '0' \n"); // cuda api time start hipEvent_t start,stop; hipEventCreate(&start); hipEventCreate(&stop); hipEventRecord(start,0); // start iteration loop, assigning and updating centroid on each iteration for (size_t iteration = 0; iteration < number_of_iterations; ++iteration) { Assignment<<<blocks, threads, kmeans_shared_memory>>>(device_data.x,device_data.y, device_data.dataSize, device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); hipDeviceSynchronize(); centroid_recompute<<<1, k * blocks, centroid_reduction_memory>>>(device_centroids.x, device_centroids.y, device_sum.x, device_sum.y, k, device_count); hipDeviceSynchronize(); } // cuda api time stop hipEventRecord(stop,0); hipEventSynchronize(stop); float elapsedTime; hipEventElapsedTime(&elapsedTime,start,stop); hipEventDestroy(start); hipEventDestroy(stop); std::string unit = ""; unit = (elapsedTime > 999) ? "seconds" : "milliseconds"; elapsedTime = (elapsedTime > 999 ) ? elapsedTime/1000 : elapsedTime; std::cout << "Elapsed time of kernal calls: " << elapsedTime << " " << unit << "\n" << std::endl; ErrorCheck(hipFree(device_count),"Freeing Device Memory"); std::vector<double> centroid_x(k, 0); std::vector<double> centroid_y(k, 0); ErrorCheck(hipMemcpy(centroid_x.data(), device_centroids.x, device_centroids.num_of_bytes, hipMemcpyDeviceToHost), "Moving Array back to host\n"); ErrorCheck(hipMemcpy(centroid_y.data(), device_centroids.y, device_centroids.num_of_bytes, hipMemcpyDeviceToHost), "Moving Array back to host\n"); std::cout << "centroids:" << std::endl; for (size_t cluster = 0; cluster < k; ++cluster) { std::cout << centroid_x[cluster] << " " << centroid_y[cluster] << std::endl; } std::cout << "\n" << std::endl; return(EXIT_SUCCESS); }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z10AssignmentPdS_iS_S_S_S_iPi .globl _Z10AssignmentPdS_iS_S_S_S_iPi .p2align 8 .type _Z10AssignmentPdS_iS_S_S_S_iPi,@function _Z10AssignmentPdS_iS_S_S_S_iPi: s_clause 0x1 s_load_b32 s2, s[0:1], 0x54 s_load_b32 s3, s[0:1], 0x10 s_waitcnt lgkmcnt(0) s_and_b32 s8, s2, 0xffff s_mov_b32 s2, exec_lo v_mad_u64_u32 v[1:2], null, s15, s8, v[0:1] s_delay_alu instid0(VALU_DEP_1) v_cmpx_gt_i32_e64 s3, v1 s_cbranch_execz .LBB0_16 s_load_b32 s9, s[0:1], 0x38 s_mov_b32 s2, exec_lo s_waitcnt lgkmcnt(0) v_cmpx_gt_i32_e64 s9, v0 s_cbranch_execz .LBB0_3 s_load_b128 s[4:7], s[0:1], 0x18 v_lshlrev_b32_e32 v6, 3, v0 v_add_nc_u32_e32 v7, s9, v0 s_waitcnt lgkmcnt(0) s_clause 0x1 global_load_b64 v[2:3], v6, s[4:5] global_load_b64 v[4:5], v6, s[6:7] v_add_nc_u32_e32 v6, 0, v6 v_lshl_add_u32 v7, v7, 3, 0 s_waitcnt vmcnt(1) ds_store_b64 v6, v[2:3] s_waitcnt vmcnt(0) ds_store_b64 v7, v[4:5] .LBB0_3: s_or_b32 exec_lo, exec_lo, s2 s_load_b128 s[4:7], s[0:1], 0x0 v_ashrrev_i32_e32 v2, 31, v1 s_waitcnt lgkmcnt(0) s_barrier buffer_gl0_inv s_cmp_lt_i32 s9, 1 v_lshlrev_b64 v[1:2], 3, v[1:2] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v3, vcc_lo, s4, v1 v_add_co_ci_u32_e32 v4, vcc_lo, s5, v2, vcc_lo v_add_co_u32 v5, vcc_lo, s6, v1 v_add_co_ci_u32_e32 v6, vcc_lo, s7, v2, vcc_lo global_load_b64 v[1:2], v[3:4], off global_load_b64 v[3:4], v[5:6], off s_cbranch_scc1 .LBB0_6 v_mov_b32_e32 v5, 0 v_dual_mov_b32 v6, 0x408f4000 :: v_dual_mov_b32 v7, -1 s_lshl_b32 s2, s9, 3 s_mov_b32 s3, 0 s_mov_b32 s4, 0 .LBB0_5: s_delay_alu instid0(SALU_CYCLE_1) s_add_i32 s5, s4, s2 v_mov_b32_e32 v8, s4 v_mov_b32_e32 v10, s5 s_add_i32 s4, s4, 8 ds_load_b64 v[8:9], v8 ds_load_b64 v[10:11], v10 s_waitcnt vmcnt(1) lgkmcnt(1) v_add_f64 v[8:9], v[1:2], -v[8:9] s_waitcnt vmcnt(0) lgkmcnt(0) v_add_f64 v[10:11], v[3:4], -v[10:11] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_f64 v[10:11], v[10:11], v[10:11] v_fma_f64 v[8:9], v[8:9], v[8:9], v[10:11] s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_cmp_gt_f64_e32 vcc_lo, 0x10000000, v[8:9] v_cndmask_b32_e64 v10, 0, 1, vcc_lo v_lshlrev_b32_e32 v10, 8, v10 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_ldexp_f64 v[8:9], v[8:9], v10 v_rsq_f64_e32 v[10:11], v[8:9] s_waitcnt_depctr 0xfff v_mul_f64 v[12:13], v[8:9], v[10:11] v_mul_f64 v[10:11], v[10:11], 0.5 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[14:15], -v[10:11], v[12:13], 0.5 v_fma_f64 v[12:13], v[12:13], v[14:15], v[12:13] v_fma_f64 v[10:11], v[10:11], v[14:15], v[10:11] s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[14:15], -v[12:13], v[12:13], v[8:9] v_fma_f64 v[12:13], v[14:15], v[10:11], v[12:13] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[14:15], -v[12:13], v[12:13], v[8:9] v_fma_f64 v[10:11], v[14:15], v[10:11], v[12:13] v_cndmask_b32_e64 v12, 0, 0xffffff80, vcc_lo v_cmp_class_f64_e64 vcc_lo, v[8:9], 0x260 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_ldexp_f64 v[10:11], v[10:11], v12 v_dual_cndmask_b32 v9, v11, v9 :: v_dual_cndmask_b32 v8, v10, v8 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) v_cmp_lt_f64_e32 vcc_lo, v[8:9], v[5:6] v_cndmask_b32_e32 v5, v5, v8, vcc_lo v_cndmask_b32_e64 v7, v7, s3, vcc_lo v_cndmask_b32_e32 v6, v6, v9, vcc_lo s_add_i32 s3, s3, 1 s_cmp_eq_u32 s9, s3 s_cbranch_scc0 .LBB0_5 s_branch .LBB0_7 .LBB0_6: v_mov_b32_e32 v7, -1 .LBB0_7: s_cmp_lt_i32 s9, 1 s_waitcnt vmcnt(0) s_barrier buffer_gl0_inv s_cbranch_scc1 .LBB0_16 s_clause 0x1 s_load_b128 s[4:7], s[0:1], 0x28 s_load_b64 s[2:3], s[0:1], 0x40 v_add_nc_u32_e32 v5, s8, v0 v_lshl_add_u32 v8, v0, 3, 0 v_cmp_eq_u32_e64 s0, 0, v0 s_cmp_gt_u32 s8, 1 s_delay_alu instid0(VALU_DEP_3) v_dual_mov_b32 v11, 0 :: v_dual_add_nc_u32 v6, s8, v5 v_lshl_add_u32 v9, v5, 3, 0 s_cselect_b32 s1, -1, 0 s_mov_b32 s10, 0 s_mul_i32 s15, s15, s9 v_lshl_add_u32 v10, v6, 3, 0 s_branch .LBB0_10 .LBB0_9: s_or_b32 exec_lo, exec_lo, s11 s_add_i32 s10, s10, 1 s_waitcnt_vscnt null, 0x0 s_cmp_lg_u32 s10, s9 s_barrier buffer_gl0_inv s_cbranch_scc0 .LBB0_16 .LBB0_10: v_cmp_eq_u32_e32 vcc_lo, s10, v7 s_mov_b32 s11, s8 v_dual_mov_b32 v14, 0 :: v_dual_cndmask_b32 v13, 0, v2 v_dual_cndmask_b32 v12, 0, v1 :: v_dual_cndmask_b32 v17, 0, v4 v_cndmask_b32_e32 v16, 0, v3, vcc_lo v_cndmask_b32_e64 v15, 0, 0x3ff00000, vcc_lo s_and_not1_b32 vcc_lo, exec_lo, s1 ds_store_b64 v8, v[12:13] ds_store_b64 v9, v[16:17] ds_store_b64 v10, v[14:15] s_waitcnt lgkmcnt(0) s_barrier buffer_gl0_inv s_cbranch_vccz .LBB0_14 .LBB0_11: s_and_saveexec_b32 s11, s0 s_cbranch_execz .LBB0_9 ds_load_b64 v[12:13], v10 s_add_i32 s12, s10, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_ashr_i32 s13, s12, 31 s_lshl_b64 s[16:17], s[12:13], 3 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) s_add_u32 s18, s4, s16 s_addc_u32 s19, s5, s17 s_add_u32 s16, s6, s16 s_addc_u32 s17, s7, s17 s_lshl_b64 s[12:13], s[12:13], 2 s_add_u32 s12, s2, s12 s_addc_u32 s13, s3, s13 s_waitcnt lgkmcnt(0) v_cvt_i32_f64_e32 v16, v[12:13] ds_load_b64 v[12:13], v8 ds_load_b64 v[14:15], v9 s_waitcnt lgkmcnt(1) global_store_b64 v11, v[12:13], s[18:19] s_waitcnt lgkmcnt(0) s_clause 0x1 global_store_b64 v11, v[14:15], s[16:17] global_store_b32 v11, v16, s[12:13] s_branch .LBB0_9 .p2align 6 .LBB0_13: s_or_b32 exec_lo, exec_lo, s13 s_cmp_lt_u32 s11, 4 s_mov_b32 s11, s12 s_waitcnt lgkmcnt(0) s_barrier buffer_gl0_inv s_cbranch_scc1 .LBB0_11 .LBB0_14: s_lshr_b32 s12, s11, 1 s_mov_b32 s13, exec_lo v_cmpx_gt_u32_e64 s12, v0 s_cbranch_execz .LBB0_13 v_add_nc_u32_e32 v12, s12, v0 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v12, v12, 3, 0 ds_load_b64 v[12:13], v12 ds_load_b64 v[14:15], v8 s_waitcnt lgkmcnt(0) v_add_f64 v[12:13], v[12:13], v[14:15] v_add_nc_u32_e32 v14, s12, v5 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v14, v14, 3, 0 ds_store_b64 v8, v[12:13] ds_load_b64 v[12:13], v14 ds_load_b64 v[14:15], v9 s_waitcnt lgkmcnt(0) v_add_f64 v[12:13], v[12:13], v[14:15] v_add_nc_u32_e32 v14, s12, v6 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v14, v14, 3, 0 ds_store_b64 v9, v[12:13] ds_load_b64 v[12:13], v14 ds_load_b64 v[14:15], v10 s_waitcnt lgkmcnt(0) v_add_f64 v[12:13], v[12:13], v[14:15] ds_store_b64 v10, v[12:13] s_branch .LBB0_13 .LBB0_16: s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z10AssignmentPdS_iS_S_S_S_iPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 328 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 18 .amdhsa_next_free_sgpr 20 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z10AssignmentPdS_iS_S_S_S_iPi, .Lfunc_end0-_Z10AssignmentPdS_iS_S_S_S_iPi .section .AMDGPU.csdata,"",@progbits .text .protected _Z18centroid_recomputePdS_S_S_iPi .globl _Z18centroid_recomputePdS_S_S_iPi .p2align 8 .type _Z18centroid_recomputePdS_S_S_iPi,@function _Z18centroid_recomputePdS_S_S_iPi: s_load_b128 s[4:7], s[0:1], 0x10 v_lshlrev_b32_e32 v1, 3, v0 s_clause 0x1 s_load_b32 s2, s[0:1], 0x20 s_load_b32 s3, s[0:1], 0x3c s_waitcnt lgkmcnt(0) s_clause 0x1 global_load_b64 v[5:6], v1, s[4:5] global_load_b64 v[7:8], v1, s[6:7] v_add_nc_u32_e32 v3, 0, v1 s_and_b32 s8, s3, 0xffff s_bfe_u32 s3, s3, 0xf0001 v_add_nc_u32_e32 v2, s8, v0 s_cmp_lt_i32 s3, s2 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v4, v2, 3, 0 s_waitcnt vmcnt(1) ds_store_b64 v3, v[5:6] s_waitcnt vmcnt(0) ds_store_b64 v4, v[7:8] s_waitcnt lgkmcnt(0) s_set_inst_prefetch_distance 0x1 s_branch .LBB1_2 .p2align 6 .LBB1_1: s_or_b32 exec_lo, exec_lo, s8 s_lshr_b32 s3, s3, 1 s_waitcnt lgkmcnt(0) s_cmp_lt_i32 s3, s2 .LBB1_2: s_barrier buffer_gl0_inv s_cbranch_scc1 .LBB1_5 s_mov_b32 s8, exec_lo v_cmpx_gt_u32_e64 s3, v0 s_cbranch_execz .LBB1_1 v_add_nc_u32_e32 v5, s3, v0 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v5, v5, 3, 0 ds_load_b64 v[5:6], v5 ds_load_b64 v[7:8], v3 s_waitcnt lgkmcnt(0) v_add_f64 v[5:6], v[5:6], v[7:8] v_add_nc_u32_e32 v7, s3, v2 s_delay_alu instid0(VALU_DEP_1) v_lshl_add_u32 v7, v7, 3, 0 ds_store_b64 v3, v[5:6] ds_load_b64 v[5:6], v7 ds_load_b64 v[7:8], v4 s_waitcnt lgkmcnt(0) v_add_f64 v[5:6], v[5:6], v[7:8] ds_store_b64 v4, v[5:6] s_branch .LBB1_1 .LBB1_5: s_set_inst_prefetch_distance 0x2 v_cmp_gt_i32_e32 vcc_lo, s2, v0 s_and_saveexec_b32 s2, vcc_lo s_cbranch_execz .LBB1_7 s_load_b64 s[8:9], s[0:1], 0x28 v_lshlrev_b32_e32 v16, 2, v0 v_add_co_u32 v2, s2, s4, v1 s_delay_alu instid0(VALU_DEP_1) v_add_co_ci_u32_e64 v3, null, s5, 0, s2 s_load_b128 s[0:3], s[0:1], 0x0 v_lshlrev_b32_e32 v17, 3, v0 v_add_co_u32 v0, s4, s6, v1 global_load_b64 v[4:5], v[2:3], off v_add_co_ci_u32_e64 v1, null, s7, 0, s4 s_waitcnt lgkmcnt(0) global_load_b32 v6, v16, s[8:9] s_waitcnt vmcnt(0) v_max_i32_e32 v6, 1, v6 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_cvt_f64_i32_e32 v[6:7], v6 v_div_scale_f64 v[8:9], null, v[6:7], v[6:7], v[4:5] v_div_scale_f64 v[14:15], vcc_lo, v[4:5], v[6:7], v[4:5] s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) v_rcp_f64_e32 v[10:11], v[8:9] s_waitcnt_depctr 0xfff v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0 v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0 v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_f64 v[12:13], v[14:15], v[10:11] v_fma_f64 v[8:9], -v[8:9], v[12:13], v[14:15] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[12:13] v_div_fixup_f64 v[4:5], v[8:9], v[6:7], v[4:5] global_store_b64 v17, v[4:5], s[0:1] global_load_b64 v[4:5], v[0:1], off s_mov_b32 s0, 0 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2) s_mov_b32 s1, s0 s_waitcnt vmcnt(0) v_div_scale_f64 v[8:9], null, v[6:7], v[6:7], v[4:5] v_div_scale_f64 v[14:15], vcc_lo, v[4:5], v[6:7], v[4:5] v_rcp_f64_e32 v[10:11], v[8:9] s_waitcnt_depctr 0xfff v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11] v_fma_f64 v[12:13], -v[8:9], v[10:11], 1.0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[10:11], v[10:11], v[12:13], v[10:11] v_mul_f64 v[12:13], v[14:15], v[10:11] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fma_f64 v[8:9], -v[8:9], v[12:13], v[14:15] v_div_fmas_f64 v[8:9], v[8:9], v[10:11], v[12:13] s_delay_alu instid0(VALU_DEP_1) v_div_fixup_f64 v[4:5], v[8:9], v[6:7], v[4:5] v_dual_mov_b32 v7, s1 :: v_dual_mov_b32 v6, s0 v_mov_b32_e32 v8, 0 global_store_b64 v17, v[4:5], s[2:3] global_store_b64 v[0:1], v[6:7], off global_store_b64 v[2:3], v[6:7], off global_store_b32 v16, v8, s[8:9] .LBB1_7: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z18centroid_recomputePdS_S_S_iPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 304 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 18 .amdhsa_next_free_sgpr 10 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end1: .size _Z18centroid_recomputePdS_S_S_iPi, .Lfunc_end1-_Z18centroid_recomputePdS_S_S_iPi .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 32 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 40 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 48 .size: 8 .value_kind: global_buffer - .offset: 56 .size: 4 .value_kind: by_value - .address_space: global .offset: 64 .size: 8 .value_kind: global_buffer - .offset: 72 .size: 4 .value_kind: hidden_block_count_x - .offset: 76 .size: 4 .value_kind: hidden_block_count_y - .offset: 80 .size: 4 .value_kind: hidden_block_count_z - .offset: 84 .size: 2 .value_kind: hidden_group_size_x - .offset: 86 .size: 2 .value_kind: hidden_group_size_y - .offset: 88 .size: 2 .value_kind: hidden_group_size_z - .offset: 90 .size: 2 .value_kind: hidden_remainder_x - .offset: 92 .size: 2 .value_kind: hidden_remainder_y - .offset: 94 .size: 2 .value_kind: hidden_remainder_z - .offset: 112 .size: 8 .value_kind: hidden_global_offset_x - .offset: 120 .size: 8 .value_kind: hidden_global_offset_y - .offset: 128 .size: 8 .value_kind: hidden_global_offset_z - .offset: 136 .size: 2 .value_kind: hidden_grid_dims - .offset: 192 .size: 4 .value_kind: hidden_dynamic_lds_size .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 328 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z10AssignmentPdS_iS_S_S_S_iPi .private_segment_fixed_size: 0 .sgpr_count: 22 .sgpr_spill_count: 0 .symbol: _Z10AssignmentPdS_iS_S_S_S_iPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 18 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .offset: 32 .size: 4 .value_kind: by_value - .address_space: global .offset: 40 .size: 8 .value_kind: global_buffer - .offset: 48 .size: 4 .value_kind: hidden_block_count_x - .offset: 52 .size: 4 .value_kind: hidden_block_count_y - .offset: 56 .size: 4 .value_kind: hidden_block_count_z - .offset: 60 .size: 2 .value_kind: hidden_group_size_x - .offset: 62 .size: 2 .value_kind: hidden_group_size_y - .offset: 64 .size: 2 .value_kind: hidden_group_size_z - .offset: 66 .size: 2 .value_kind: hidden_remainder_x - .offset: 68 .size: 2 .value_kind: hidden_remainder_y - .offset: 70 .size: 2 .value_kind: hidden_remainder_z - .offset: 88 .size: 8 .value_kind: hidden_global_offset_x - .offset: 96 .size: 8 .value_kind: hidden_global_offset_y - .offset: 104 .size: 8 .value_kind: hidden_global_offset_z - .offset: 112 .size: 2 .value_kind: hidden_grid_dims - .offset: 168 .size: 4 .value_kind: hidden_dynamic_lds_size .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 304 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z18centroid_recomputePdS_S_S_iPi .private_segment_fixed_size: 0 .sgpr_count: 12 .sgpr_spill_count: 0 .symbol: _Z18centroid_recomputePdS_S_S_iPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 18 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
code for sm_80 Function : _Z25callOperationSharedStaticPiS_iS_i .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.X ; /* 0x0000000000007919 */ /* 0x000e280000002500 */ /*0020*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000e240000002100 */ /*0030*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x001fca00078e0203 */ /*0040*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x180], PT ; /* 0x0000600000007a0c */ /* 0x000fda0003f06270 */ /*0050*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0060*/ HFMA2.MMA R7, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff077435 */ /* 0x000fe200000001ff */ /*0070*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fd20000000a00 */ /*0080*/ IMAD.WIDE R2, R0, R7, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fc800078e0207 */ /*0090*/ IMAD.WIDE R4, R0.reuse, R7.reuse, c[0x0][0x168] ; /* 0x00005a0000047625 */ /* 0x0c0fe400078e0207 */ /*00a0*/ LDG.E R3, [R2.64] ; /* 0x0000000402037981 */ /* 0x000ea8000c1e1900 */ /*00b0*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ee2000c1e1900 */ /*00c0*/ MOV R8, c[0x0][0x170] ; /* 0x00005c0000087a02 */ /* 0x000fe20000000f00 */ /*00d0*/ IMAD.WIDE R6, R0, R7, c[0x0][0x178] ; /* 0x00005e0000067625 */ /* 0x000fc800078e0207 */ /*00e0*/ STS [0x14], R8 ; /* 0x00001408ff007388 */ /* 0x000fe80000000800 */ /*00f0*/ STS [R0.X4], R3 ; /* 0x0000000300007388 */ /* 0x004fe20000004800 */ /*0100*/ IMAD R9, R3, c[0x0][0x170], R4 ; /* 0x00005c0003097a24 */ /* 0x008fca00078e0204 */ /*0110*/ STG.E [R6.64], R9 ; /* 0x0000000906007986 */ /* 0x000fe2000c101904 */ /*0120*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0130*/ BRA 0x130; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
.file "tmpxft_0016f500_00000000-6_callOperationSharedStatic.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i .type _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i, @function _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i: .LFB2051: .cfi_startproc endbr64 subq $152, %rsp .cfi_def_cfa_offset 160 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, (%rsp) movl %r8d, 8(%rsp) movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movq %rsp, %rax movq %rax, 120(%rsp) leaq 8(%rsp), %rax movq %rax, 128(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 136(%rsp), %rax subq %fs:40, %rax jne .L8 addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 168 pushq 40(%rsp) .cfi_def_cfa_offset 176 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z25callOperationSharedStaticPiS_iS_i(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 160 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i, .-_Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i .globl _Z25callOperationSharedStaticPiS_iS_i .type _Z25callOperationSharedStaticPiS_iS_i, @function _Z25callOperationSharedStaticPiS_iS_i: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z25callOperationSharedStaticPiS_iS_i, .-_Z25callOperationSharedStaticPiS_iS_i .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "_Z25callOperationSharedStaticPiS_iS_i" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z25callOperationSharedStaticPiS_iS_i(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
#include <hip/hip_runtime.h> #include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z25callOperationSharedStaticPiS_iS_i .globl _Z25callOperationSharedStaticPiS_iS_i .p2align 8 .type _Z25callOperationSharedStaticPiS_iS_i,@function _Z25callOperationSharedStaticPiS_iS_i: s_clause 0x1 s_load_b32 s2, s[0:1], 0x34 s_load_b32 s3, s[0:1], 0x20 s_waitcnt lgkmcnt(0) s_and_b32 s2, s2, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_mad_u64_u32 v[1:2], null, s15, s2, v[0:1] s_mov_b32 s2, exec_lo v_cmpx_gt_i32_e64 s3, v1 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b128 s[4:7], s[0:1], 0x0 s_load_b32 s2, s[0:1], 0x10 v_ashrrev_i32_e32 v2, 31, v1 s_load_b64 s[0:1], s[0:1], 0x18 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[1:2] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s4, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s5, v1, vcc_lo v_add_co_u32 v4, vcc_lo, s6, v0 v_add_co_ci_u32_e32 v5, vcc_lo, s7, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s0, v0 global_load_b32 v6, v[2:3], off global_load_b32 v2, v[4:5], off v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo s_waitcnt vmcnt(0) v_mad_u64_u32 v[3:4], null, v6, s2, v[2:3] global_store_b32 v[0:1], v3, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z25callOperationSharedStaticPiS_iS_i .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 296 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 7 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z25callOperationSharedStaticPiS_iS_i, .Lfunc_end0-_Z25callOperationSharedStaticPiS_iS_i .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .offset: 32 .size: 4 .value_kind: by_value - .offset: 40 .size: 4 .value_kind: hidden_block_count_x - .offset: 44 .size: 4 .value_kind: hidden_block_count_y - .offset: 48 .size: 4 .value_kind: hidden_block_count_z - .offset: 52 .size: 2 .value_kind: hidden_group_size_x - .offset: 54 .size: 2 .value_kind: hidden_group_size_y - .offset: 56 .size: 2 .value_kind: hidden_group_size_z - .offset: 58 .size: 2 .value_kind: hidden_remainder_x - .offset: 60 .size: 2 .value_kind: hidden_remainder_y - .offset: 62 .size: 2 .value_kind: hidden_remainder_z - .offset: 80 .size: 8 .value_kind: hidden_global_offset_x - .offset: 88 .size: 8 .value_kind: hidden_global_offset_y - .offset: 96 .size: 8 .value_kind: hidden_global_offset_z - .offset: 104 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 296 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z25callOperationSharedStaticPiS_iS_i .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z25callOperationSharedStaticPiS_iS_i.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 7 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include "includes.h" __global__ void callOperationSharedStatic(int *a, int *b, int x, int *res, int n) { int tid = blockDim.x * blockIdx.x + threadIdx.x; if (tid >= n) { return; } __shared__ int s_a[size], s_b[size], s_res[size]; __shared__ int s_x; s_x = x; s_a[tid] = a[tid]; s_b[tid] = b[tid]; s_res[tid] = ((s_a[tid] * s_x) + s_b[tid]); res[tid] = s_res[tid]; }
.text .file "callOperationSharedStatic.hip" .globl _Z40__device_stub__callOperationSharedStaticPiS_iS_i # -- Begin function _Z40__device_stub__callOperationSharedStaticPiS_iS_i .p2align 4, 0x90 .type _Z40__device_stub__callOperationSharedStaticPiS_iS_i,@function _Z40__device_stub__callOperationSharedStaticPiS_iS_i: # @_Z40__device_stub__callOperationSharedStaticPiS_iS_i .cfi_startproc # %bb.0: subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 4(%rsp) movq %rcx, 56(%rsp) movl %r8d, (%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 4(%rsp), %rax movq %rax, 96(%rsp) leaq 56(%rsp), %rax movq %rax, 104(%rsp) movq %rsp, %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z25callOperationSharedStaticPiS_iS_i, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $136, %rsp .cfi_adjust_cfa_offset -136 retq .Lfunc_end0: .size _Z40__device_stub__callOperationSharedStaticPiS_iS_i, .Lfunc_end0-_Z40__device_stub__callOperationSharedStaticPiS_iS_i .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z25callOperationSharedStaticPiS_iS_i, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z25callOperationSharedStaticPiS_iS_i,@object # @_Z25callOperationSharedStaticPiS_iS_i .section .rodata,"a",@progbits .globl _Z25callOperationSharedStaticPiS_iS_i .p2align 3, 0x0 _Z25callOperationSharedStaticPiS_iS_i: .quad _Z40__device_stub__callOperationSharedStaticPiS_iS_i .size _Z25callOperationSharedStaticPiS_iS_i, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z25callOperationSharedStaticPiS_iS_i" .size .L__unnamed_1, 38 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z40__device_stub__callOperationSharedStaticPiS_iS_i .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z25callOperationSharedStaticPiS_iS_i .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z25callOperationSharedStaticPiS_iS_i .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.X ; /* 0x0000000000007919 */ /* 0x000e280000002500 */ /*0020*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000e240000002100 */ /*0030*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x001fca00078e0203 */ /*0040*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x180], PT ; /* 0x0000600000007a0c */ /* 0x000fda0003f06270 */ /*0050*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0060*/ HFMA2.MMA R7, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff077435 */ /* 0x000fe200000001ff */ /*0070*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fd20000000a00 */ /*0080*/ IMAD.WIDE R2, R0, R7, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fc800078e0207 */ /*0090*/ IMAD.WIDE R4, R0.reuse, R7.reuse, c[0x0][0x168] ; /* 0x00005a0000047625 */ /* 0x0c0fe400078e0207 */ /*00a0*/ LDG.E R3, [R2.64] ; /* 0x0000000402037981 */ /* 0x000ea8000c1e1900 */ /*00b0*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ee2000c1e1900 */ /*00c0*/ MOV R8, c[0x0][0x170] ; /* 0x00005c0000087a02 */ /* 0x000fe20000000f00 */ /*00d0*/ IMAD.WIDE R6, R0, R7, c[0x0][0x178] ; /* 0x00005e0000067625 */ /* 0x000fc800078e0207 */ /*00e0*/ STS [0x14], R8 ; /* 0x00001408ff007388 */ /* 0x000fe80000000800 */ /*00f0*/ STS [R0.X4], R3 ; /* 0x0000000300007388 */ /* 0x004fe20000004800 */ /*0100*/ IMAD R9, R3, c[0x0][0x170], R4 ; /* 0x00005c0003097a24 */ /* 0x008fca00078e0204 */ /*0110*/ STG.E [R6.64], R9 ; /* 0x0000000906007986 */ /* 0x000fe2000c101904 */ /*0120*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0130*/ BRA 0x130; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z25callOperationSharedStaticPiS_iS_i .globl _Z25callOperationSharedStaticPiS_iS_i .p2align 8 .type _Z25callOperationSharedStaticPiS_iS_i,@function _Z25callOperationSharedStaticPiS_iS_i: s_clause 0x1 s_load_b32 s2, s[0:1], 0x34 s_load_b32 s3, s[0:1], 0x20 s_waitcnt lgkmcnt(0) s_and_b32 s2, s2, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_mad_u64_u32 v[1:2], null, s15, s2, v[0:1] s_mov_b32 s2, exec_lo v_cmpx_gt_i32_e64 s3, v1 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b128 s[4:7], s[0:1], 0x0 s_load_b32 s2, s[0:1], 0x10 v_ashrrev_i32_e32 v2, 31, v1 s_load_b64 s[0:1], s[0:1], 0x18 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[1:2] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s4, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s5, v1, vcc_lo v_add_co_u32 v4, vcc_lo, s6, v0 v_add_co_ci_u32_e32 v5, vcc_lo, s7, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s0, v0 global_load_b32 v6, v[2:3], off global_load_b32 v2, v[4:5], off v_add_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo s_waitcnt vmcnt(0) v_mad_u64_u32 v[3:4], null, v6, s2, v[2:3] global_store_b32 v[0:1], v3, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z25callOperationSharedStaticPiS_iS_i .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 296 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 7 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z25callOperationSharedStaticPiS_iS_i, .Lfunc_end0-_Z25callOperationSharedStaticPiS_iS_i .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .offset: 32 .size: 4 .value_kind: by_value - .offset: 40 .size: 4 .value_kind: hidden_block_count_x - .offset: 44 .size: 4 .value_kind: hidden_block_count_y - .offset: 48 .size: 4 .value_kind: hidden_block_count_z - .offset: 52 .size: 2 .value_kind: hidden_group_size_x - .offset: 54 .size: 2 .value_kind: hidden_group_size_y - .offset: 56 .size: 2 .value_kind: hidden_group_size_z - .offset: 58 .size: 2 .value_kind: hidden_remainder_x - .offset: 60 .size: 2 .value_kind: hidden_remainder_y - .offset: 62 .size: 2 .value_kind: hidden_remainder_z - .offset: 80 .size: 8 .value_kind: hidden_global_offset_x - .offset: 88 .size: 8 .value_kind: hidden_global_offset_y - .offset: 96 .size: 8 .value_kind: hidden_global_offset_z - .offset: 104 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 296 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z25callOperationSharedStaticPiS_iS_i .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z25callOperationSharedStaticPiS_iS_i.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 7 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_0016f500_00000000-6_callOperationSharedStatic.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i .type _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i, @function _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i: .LFB2051: .cfi_startproc endbr64 subq $152, %rsp .cfi_def_cfa_offset 160 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %rcx, (%rsp) movl %r8d, 8(%rsp) movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movq %rsp, %rax movq %rax, 120(%rsp) leaq 8(%rsp), %rax movq %rax, 128(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 136(%rsp), %rax subq %fs:40, %rax jne .L8 addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 168 pushq 40(%rsp) .cfi_def_cfa_offset 176 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z25callOperationSharedStaticPiS_iS_i(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 160 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i, .-_Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i .globl _Z25callOperationSharedStaticPiS_iS_i .type _Z25callOperationSharedStaticPiS_iS_i, @function _Z25callOperationSharedStaticPiS_iS_i: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z51__device_stub__Z25callOperationSharedStaticPiS_iS_iPiS_iS_i addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z25callOperationSharedStaticPiS_iS_i, .-_Z25callOperationSharedStaticPiS_iS_i .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "_Z25callOperationSharedStaticPiS_iS_i" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z25callOperationSharedStaticPiS_iS_i(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "callOperationSharedStatic.hip" .globl _Z40__device_stub__callOperationSharedStaticPiS_iS_i # -- Begin function _Z40__device_stub__callOperationSharedStaticPiS_iS_i .p2align 4, 0x90 .type _Z40__device_stub__callOperationSharedStaticPiS_iS_i,@function _Z40__device_stub__callOperationSharedStaticPiS_iS_i: # @_Z40__device_stub__callOperationSharedStaticPiS_iS_i .cfi_startproc # %bb.0: subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 4(%rsp) movq %rcx, 56(%rsp) movl %r8d, (%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 4(%rsp), %rax movq %rax, 96(%rsp) leaq 56(%rsp), %rax movq %rax, 104(%rsp) movq %rsp, %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z25callOperationSharedStaticPiS_iS_i, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $136, %rsp .cfi_adjust_cfa_offset -136 retq .Lfunc_end0: .size _Z40__device_stub__callOperationSharedStaticPiS_iS_i, .Lfunc_end0-_Z40__device_stub__callOperationSharedStaticPiS_iS_i .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z25callOperationSharedStaticPiS_iS_i, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z25callOperationSharedStaticPiS_iS_i,@object # @_Z25callOperationSharedStaticPiS_iS_i .section .rodata,"a",@progbits .globl _Z25callOperationSharedStaticPiS_iS_i .p2align 3, 0x0 _Z25callOperationSharedStaticPiS_iS_i: .quad _Z40__device_stub__callOperationSharedStaticPiS_iS_i .size _Z25callOperationSharedStaticPiS_iS_i, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z25callOperationSharedStaticPiS_iS_i" .size .L__unnamed_1, 38 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z40__device_stub__callOperationSharedStaticPiS_iS_i .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z25callOperationSharedStaticPiS_iS_i .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
code for sm_80 Function : add32 .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.Z ; /* 0x0000000000007919 */ /* 0x000e280000002700 */ /*0020*/ S2R R5, SR_CTAID.Y ; /* 0x0000000000057919 */ /* 0x000e280000002600 */ /*0030*/ S2R R3, SR_CTAID.X ; /* 0x0000000000037919 */ /* 0x000e680000002500 */ /*0040*/ S2R R7, SR_TID.Z ; /* 0x0000000000077919 */ /* 0x000ea80000002300 */ /*0050*/ S2R R9, SR_TID.Y ; /* 0x0000000000097919 */ /* 0x000ee80000002200 */ /*0060*/ S2R R11, SR_TID.X ; /* 0x00000000000b7919 */ /* 0x000f220000002100 */ /*0070*/ IMAD R0, R0, c[0x0][0x10], R5 ; /* 0x0000040000007a24 */ /* 0x001fc800078e0205 */ /*0080*/ IMAD R0, R0, c[0x0][0xc], R3 ; /* 0x0000030000007a24 */ /* 0x002fc800078e0203 */ /*0090*/ IMAD R0, R0, c[0x0][0x8], R7 ; /* 0x0000020000007a24 */ /* 0x004fc800078e0207 */ /*00a0*/ IMAD R0, R0, c[0x0][0x4], R9 ; /* 0x0000010000007a24 */ /* 0x008fc800078e0209 */ /*00b0*/ IMAD R0, R0, c[0x0][0x0], R11 ; /* 0x0000000000007a24 */ /* 0x010fca00078e020b */ /*00c0*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x000fda0003f06270 */ /*00d0*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*00e0*/ HFMA2.MMA R3, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff037435 */ /* 0x000fe200000001ff */ /*00f0*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fd20000000a00 */ /*0100*/ IMAD.WIDE R4, R0, R3, c[0x0][0x168] ; /* 0x00005a0000047625 */ /* 0x000fc800078e0203 */ /*0110*/ IMAD.WIDE R2, R0, R3, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fe400078e0203 */ /*0120*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ea8000c1e1900 */ /*0130*/ LDG.E R7, [R2.64] ; /* 0x0000000402077981 */ /* 0x000ea4000c1e1900 */ /*0140*/ FADD R7, R4, R7 ; /* 0x0000000704077221 */ /* 0x004fca0000000000 */ /*0150*/ STG.E [R2.64], R7 ; /* 0x0000000702007986 */ /* 0x000fe2000c101904 */ /*0160*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0170*/ BRA 0x170; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
.file "tmpxft_001263e2_00000000-6_add32.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z27__device_stub__Z5add32PfS_iPfS_i .type _Z27__device_stub__Z5add32PfS_iPfS_i, @function _Z27__device_stub__Z5add32PfS_iPfS_i: .LFB2051: .cfi_startproc endbr64 subq $136, %rsp .cfi_def_cfa_offset 144 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %fs:40, %rax movq %rax, 120(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 120(%rsp), %rax subq %fs:40, %rax jne .L8 addq $136, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 152 pushq 40(%rsp) .cfi_def_cfa_offset 160 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq add32(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 144 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z27__device_stub__Z5add32PfS_iPfS_i, .-_Z27__device_stub__Z5add32PfS_iPfS_i .globl add32 .type add32, @function add32: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z27__device_stub__Z5add32PfS_iPfS_i addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size add32, .-add32 .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "add32" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq add32(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
#include <hip/hip_runtime.h> #include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected add32 .globl add32 .p2align 8 .type add32,@function add32: s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b64 s[4:5], s[0:1], 0x24 v_bfe_u32 v1, v0, 20, 10 s_waitcnt lgkmcnt(0) s_mul_i32 s3, s3, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_add_i32 s3, s3, s14 s_mul_i32 s2, s3, s2 s_and_b32 s3, s5, 0xffff s_add_i32 s2, s2, s13 s_delay_alu instid0(SALU_CYCLE_1) v_mad_u64_u32 v[2:3], null, s2, s3, v[1:2] v_bfe_u32 v1, v0, 10, 10 s_load_b32 s2, s[0:1], 0x10 s_lshr_b32 s3, s4, 16 s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[3:4], null, v2, s3, v[1:2] v_and_b32_e32 v2, 0x3ff, v0 s_and_b32 s3, s4, 0xffff s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[0:1], null, v3, s3, v[2:3] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) v_cmp_gt_i32_e32 vcc_lo, s2, v0 s_and_saveexec_b32 s2, vcc_lo s_cbranch_execz .LBB0_2 s_load_b128 s[0:3], s[0:1], 0x0 v_ashrrev_i32_e32 v1, 31, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[0:1] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s0, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s1, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s2, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s3, v1, vcc_lo global_load_b32 v4, v[2:3], off global_load_b32 v0, v[0:1], off s_waitcnt vmcnt(0) v_add_f32_e32 v0, v4, v0 global_store_b32 v[2:3], v0, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel add32 .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 280 .amdhsa_user_sgpr_count 13 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 1 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 2 .amdhsa_next_free_vgpr 5 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size add32, .Lfunc_end0-add32 .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .offset: 24 .size: 4 .value_kind: hidden_block_count_x - .offset: 28 .size: 4 .value_kind: hidden_block_count_y - .offset: 32 .size: 4 .value_kind: hidden_block_count_z - .offset: 36 .size: 2 .value_kind: hidden_group_size_x - .offset: 38 .size: 2 .value_kind: hidden_group_size_y - .offset: 40 .size: 2 .value_kind: hidden_group_size_z - .offset: 42 .size: 2 .value_kind: hidden_remainder_x - .offset: 44 .size: 2 .value_kind: hidden_remainder_y - .offset: 46 .size: 2 .value_kind: hidden_remainder_z - .offset: 64 .size: 8 .value_kind: hidden_global_offset_x - .offset: 72 .size: 8 .value_kind: hidden_global_offset_y - .offset: 80 .size: 8 .value_kind: hidden_global_offset_z - .offset: 88 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 280 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: add32 .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: add32.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 5 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include "includes.h" extern "C" __global__ void add32(float* A, float *B, int size) { int block = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int index = block * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x; if(index >= size) return; A[index] = A[index] + B[index]; }
.text .file "add32.hip" .globl __device_stub__add32 # -- Begin function __device_stub__add32 .p2align 4, 0x90 .type __device_stub__add32,@function __device_stub__add32: # @__device_stub__add32 .cfi_startproc # %bb.0: subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 12(%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 12(%rsp), %rax movq %rax, 96(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 80(%rsp), %r9 movl $add32, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $120, %rsp .cfi_adjust_cfa_offset -120 retq .Lfunc_end0: .size __device_stub__add32, .Lfunc_end0-__device_stub__add32 .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $add32, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type add32,@object # @add32 .section .rodata,"a",@progbits .globl add32 .p2align 3, 0x0 add32: .quad __device_stub__add32 .size add32, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "add32" .size .L__unnamed_1, 6 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __device_stub__add32 .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym add32 .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : add32 .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.Z ; /* 0x0000000000007919 */ /* 0x000e280000002700 */ /*0020*/ S2R R5, SR_CTAID.Y ; /* 0x0000000000057919 */ /* 0x000e280000002600 */ /*0030*/ S2R R3, SR_CTAID.X ; /* 0x0000000000037919 */ /* 0x000e680000002500 */ /*0040*/ S2R R7, SR_TID.Z ; /* 0x0000000000077919 */ /* 0x000ea80000002300 */ /*0050*/ S2R R9, SR_TID.Y ; /* 0x0000000000097919 */ /* 0x000ee80000002200 */ /*0060*/ S2R R11, SR_TID.X ; /* 0x00000000000b7919 */ /* 0x000f220000002100 */ /*0070*/ IMAD R0, R0, c[0x0][0x10], R5 ; /* 0x0000040000007a24 */ /* 0x001fc800078e0205 */ /*0080*/ IMAD R0, R0, c[0x0][0xc], R3 ; /* 0x0000030000007a24 */ /* 0x002fc800078e0203 */ /*0090*/ IMAD R0, R0, c[0x0][0x8], R7 ; /* 0x0000020000007a24 */ /* 0x004fc800078e0207 */ /*00a0*/ IMAD R0, R0, c[0x0][0x4], R9 ; /* 0x0000010000007a24 */ /* 0x008fc800078e0209 */ /*00b0*/ IMAD R0, R0, c[0x0][0x0], R11 ; /* 0x0000000000007a24 */ /* 0x010fca00078e020b */ /*00c0*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x000fda0003f06270 */ /*00d0*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*00e0*/ HFMA2.MMA R3, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff037435 */ /* 0x000fe200000001ff */ /*00f0*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fd20000000a00 */ /*0100*/ IMAD.WIDE R4, R0, R3, c[0x0][0x168] ; /* 0x00005a0000047625 */ /* 0x000fc800078e0203 */ /*0110*/ IMAD.WIDE R2, R0, R3, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fe400078e0203 */ /*0120*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ea8000c1e1900 */ /*0130*/ LDG.E R7, [R2.64] ; /* 0x0000000402077981 */ /* 0x000ea4000c1e1900 */ /*0140*/ FADD R7, R4, R7 ; /* 0x0000000704077221 */ /* 0x004fca0000000000 */ /*0150*/ STG.E [R2.64], R7 ; /* 0x0000000702007986 */ /* 0x000fe2000c101904 */ /*0160*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0170*/ BRA 0x170; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected add32 .globl add32 .p2align 8 .type add32,@function add32: s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b64 s[4:5], s[0:1], 0x24 v_bfe_u32 v1, v0, 20, 10 s_waitcnt lgkmcnt(0) s_mul_i32 s3, s3, s15 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_add_i32 s3, s3, s14 s_mul_i32 s2, s3, s2 s_and_b32 s3, s5, 0xffff s_add_i32 s2, s2, s13 s_delay_alu instid0(SALU_CYCLE_1) v_mad_u64_u32 v[2:3], null, s2, s3, v[1:2] v_bfe_u32 v1, v0, 10, 10 s_load_b32 s2, s[0:1], 0x10 s_lshr_b32 s3, s4, 16 s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[3:4], null, v2, s3, v[1:2] v_and_b32_e32 v2, 0x3ff, v0 s_and_b32 s3, s4, 0xffff s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[0:1], null, v3, s3, v[2:3] s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) v_cmp_gt_i32_e32 vcc_lo, s2, v0 s_and_saveexec_b32 s2, vcc_lo s_cbranch_execz .LBB0_2 s_load_b128 s[0:3], s[0:1], 0x0 v_ashrrev_i32_e32 v1, 31, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[0:1] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s0, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s1, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s2, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s3, v1, vcc_lo global_load_b32 v4, v[2:3], off global_load_b32 v0, v[0:1], off s_waitcnt vmcnt(0) v_add_f32_e32 v0, v4, v0 global_store_b32 v[2:3], v0, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel add32 .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 280 .amdhsa_user_sgpr_count 13 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 1 .amdhsa_system_sgpr_workgroup_id_z 1 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 2 .amdhsa_next_free_vgpr 5 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size add32, .Lfunc_end0-add32 .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .offset: 24 .size: 4 .value_kind: hidden_block_count_x - .offset: 28 .size: 4 .value_kind: hidden_block_count_y - .offset: 32 .size: 4 .value_kind: hidden_block_count_z - .offset: 36 .size: 2 .value_kind: hidden_group_size_x - .offset: 38 .size: 2 .value_kind: hidden_group_size_y - .offset: 40 .size: 2 .value_kind: hidden_group_size_z - .offset: 42 .size: 2 .value_kind: hidden_remainder_x - .offset: 44 .size: 2 .value_kind: hidden_remainder_y - .offset: 46 .size: 2 .value_kind: hidden_remainder_z - .offset: 64 .size: 8 .value_kind: hidden_global_offset_x - .offset: 72 .size: 8 .value_kind: hidden_global_offset_y - .offset: 80 .size: 8 .value_kind: hidden_global_offset_z - .offset: 88 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 280 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: add32 .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: add32.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 5 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_001263e2_00000000-6_add32.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z27__device_stub__Z5add32PfS_iPfS_i .type _Z27__device_stub__Z5add32PfS_iPfS_i, @function _Z27__device_stub__Z5add32PfS_iPfS_i: .LFB2051: .cfi_startproc endbr64 subq $136, %rsp .cfi_def_cfa_offset 144 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %fs:40, %rax movq %rax, 120(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 120(%rsp), %rax subq %fs:40, %rax jne .L8 addq $136, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 152 pushq 40(%rsp) .cfi_def_cfa_offset 160 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq add32(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 144 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z27__device_stub__Z5add32PfS_iPfS_i, .-_Z27__device_stub__Z5add32PfS_iPfS_i .globl add32 .type add32, @function add32: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z27__device_stub__Z5add32PfS_iPfS_i addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size add32, .-add32 .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "add32" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq add32(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "add32.hip" .globl __device_stub__add32 # -- Begin function __device_stub__add32 .p2align 4, 0x90 .type __device_stub__add32,@function __device_stub__add32: # @__device_stub__add32 .cfi_startproc # %bb.0: subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 12(%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 12(%rsp), %rax movq %rax, 96(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 80(%rsp), %r9 movl $add32, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $120, %rsp .cfi_adjust_cfa_offset -120 retq .Lfunc_end0: .size __device_stub__add32, .Lfunc_end0-__device_stub__add32 .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $add32, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type add32,@object # @add32 .section .rodata,"a",@progbits .globl add32 .p2align 3, 0x0 add32: .quad __device_stub__add32 .size add32, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "add32" .size .L__unnamed_1, 6 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __device_stub__add32 .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym add32 .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include <stdio.h> #include <sys/time.h> #include <cuda.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( cudaError err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( cudaSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { cudaStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamCreate(&streams[i])); CudaSafeCall(cudaMallocHost((void**)&A[i], sizeof(int) * N)); CudaSafeCall(cudaMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(dA[i], A[i], N * sizeof(int), cudaMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(A[i], dA[i], N * sizeof(int), cudaMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
code for sm_80 Function : _Z6assignPii .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.X ; /* 0x0000000000007919 */ /* 0x000e280000002500 */ /*0020*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000e240000002100 */ /*0030*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x001fca00078e0203 */ /*0040*/ ISETP.GT.AND P0, PT, R0, 0xfffff, PT ; /* 0x000fffff0000780c */ /* 0x000fda0003f04270 */ /*0050*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0060*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fe40000000a00 */ /*0070*/ HFMA2.MMA R3, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff037435 */ /* 0x001fd400000001ff */ /*0080*/ IMAD.WIDE R2, R0, R3, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fca00078e0203 */ /*0090*/ LDG.E R4, [R2.64] ; /* 0x0000000402047981 */ /* 0x000ea2000c1e1900 */ /*00a0*/ MOV R5, 0x100000 ; /* 0x0010000000057802 */ /* 0x000fe40000000f00 */ /*00b0*/ MOV R7, c[0x0][0x0] ; /* 0x0000000000077a02 */ /* 0x000fc60000000f00 */ /*00c0*/ IMAD R5, R5, c[0x0][0x168], R0.reuse ; /* 0x00005a0005057a24 */ /* 0x100fe400078e0200 */ /*00d0*/ IMAD R0, R7, c[0x0][0xc], R0 ; /* 0x0000030007007a24 */ /* 0x000fca00078e0200 */ /*00e0*/ ISETP.GE.AND P0, PT, R0, 0x100000, PT ; /* 0x001000000000780c */ /* 0x000fe40003f06270 */ /*00f0*/ IADD3 R5, R4, R5, RZ ; /* 0x0000000504057210 */ /* 0x004fca0007ffe0ff */ /*0100*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x0001ec000c101904 */ /*0110*/ @!P0 BRA 0x70 ; /* 0xffffff5000008947 */ /* 0x000fea000383ffff */ /*0120*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0130*/ BRA 0x130; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include <stdio.h> #include <sys/time.h> #include <cuda.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( cudaError err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( cudaSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { cudaStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamCreate(&streams[i])); CudaSafeCall(cudaMallocHost((void**)&A[i], sizeof(int) * N)); CudaSafeCall(cudaMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(dA[i], A[i], N * sizeof(int), cudaMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(A[i], dA[i], N * sizeof(int), cudaMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
.file "tmpxft_00098435_00000000-6_streams.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2062: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2062: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z14getCurrentTimev .type _Z14getCurrentTimev, @function _Z14getCurrentTimev: .LFB2057: .cfi_startproc endbr64 subq $40, %rsp .cfi_def_cfa_offset 48 movq %fs:40, %rax movq %rax, 24(%rsp) xorl %eax, %eax movq %rsp, %rdi movl $0, %esi call gettimeofday@PLT imulq $1000000, (%rsp), %rax addq 8(%rsp), %rax movq 24(%rsp), %rdx subq %fs:40, %rdx jne .L6 addq $40, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L6: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size _Z14getCurrentTimev, .-_Z14getCurrentTimev .globl _Z26__device_stub__Z6assignPiiPii .type _Z26__device_stub__Z6assignPiiPii, @function _Z26__device_stub__Z6assignPiiPii: .LFB2084: .cfi_startproc endbr64 subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 8(%rsp) movl %esi, 4(%rsp) movq %fs:40, %rax movq %rax, 104(%rsp) xorl %eax, %eax leaq 8(%rsp), %rax movq %rax, 80(%rsp) leaq 4(%rsp), %rax movq %rax, 88(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L11 .L7: movq 104(%rsp), %rax subq %fs:40, %rax jne .L12 addq $120, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L11: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 136 pushq 24(%rsp) .cfi_def_cfa_offset 144 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z6assignPii(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 128 jmp .L7 .L12: call __stack_chk_fail@PLT .cfi_endproc .LFE2084: .size _Z26__device_stub__Z6assignPiiPii, .-_Z26__device_stub__Z6assignPiiPii .globl _Z6assignPii .type _Z6assignPii, @function _Z6assignPii: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z26__device_stub__Z6assignPiiPii addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _Z6assignPii, .-_Z6assignPii .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "/home/ubuntu/Datasets/stackv2/train-structured/ryuji0123/cuda-tutorial/master/waseda2018/examples/streams/streams.cu" .align 8 .LC1: .string "cudaSafeCall() failed at %s:%i : %s\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "VERIFILED\n" .LC3: .string "NOT VERIFIED" .text .globl main .type main, @function main: .LFB2059: .cfi_startproc endbr64 pushq %r13 .cfi_def_cfa_offset 16 .cfi_offset 13, -16 pushq %r12 .cfi_def_cfa_offset 24 .cfi_offset 12, -24 pushq %rbp .cfi_def_cfa_offset 32 .cfi_offset 6, -32 pushq %rbx .cfi_def_cfa_offset 40 .cfi_offset 3, -40 subq $152, %rsp .cfi_def_cfa_offset 192 movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax movl $0, %r12d leaq 96(%rsp), %r13 .L19: leaq 32(%rsp), %rbp leaq (%r12,%rbp), %rdi call cudaStreamCreate@PLT testl %eax, %eax jne .L42 leaq 64(%rsp), %rbx leaq (%r12,%rbx), %rdi movl $4194304, %esi call cudaMallocHost@PLT testl %eax, %eax jne .L43 leaq 0(%r13,%r12), %rdi movl $4194304, %esi call cudaMalloc@PLT testl %eax, %eax jne .L44 addq $8, %r12 cmpq $32, %r12 jne .L19 movl $0, %r12d .L21: movq 64(%rsp,%r12), %rsi movq 96(%rsp,%r12), %rdi movq 32(%rsp,%r12), %r8 movl $1, %ecx movl $4194304, %edx call cudaMemcpyAsync@PLT testl %eax, %eax jne .L45 addq $8, %r12 cmpq $32, %r12 jne .L21 movl $0, %r12d jmp .L23 .L42: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $45, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L43: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $46, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L44: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $47, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L45: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $52, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L22: addq $1, %r12 cmpq $4, %r12 je .L46 .L23: movl $1024, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 8(%rsp) movl $1, 12(%rsp) movl $1, 16(%rsp) movq 32(%rsp,%r12,8), %r9 movl $0, %r8d movq 20(%rsp), %rdx movl $1, %ecx movq 8(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax jne .L22 movq 96(%rsp,%r12,8), %rdi movl %r12d, %esi call _Z26__device_stub__Z6assignPiiPii jmp .L22 .L46: movl $0, %r12d .L25: movq 96(%rsp,%r12), %rsi movq 64(%rsp,%r12), %rdi movq 32(%rsp,%r12), %r8 movl $2, %ecx movl $4194304, %edx call cudaMemcpyAsync@PLT testl %eax, %eax jne .L47 addq $8, %r12 cmpq $32, %r12 jne .L25 leaq 32(%rbp), %r12 .L27: movq 0(%rbp), %rdi call cudaStreamSynchronize@PLT testl %eax, %eax jne .L48 addq $8, %rbp cmpq %r12, %rbp jne .L27 movl $0, %esi jmp .L28 .L47: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $63, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L48: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $69, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L29: addq $1, %rdx cmpq $1048576, %rdx je .L49 .L30: leal (%rax,%rdx), %ecx cmpl %ecx, (%rdi,%rdx,4) je .L29 addl $1, %esi jmp .L29 .L49: addq $8, %rbx addl $1048576, %eax cmpl $4194304, %eax je .L31 .L28: movq (%rbx), %rdi movl $0, %edx jmp .L30 .L31: testl %esi, %esi jne .L32 leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L33: movq 136(%rsp), %rax subq %fs:40, %rax jne .L50 movl $0, %eax addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %rbp .cfi_def_cfa_offset 24 popq %r12 .cfi_def_cfa_offset 16 popq %r13 .cfi_def_cfa_offset 8 ret .L32: .cfi_restore_state leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L33 .L50: call __stack_chk_fail@PLT .cfi_endproc .LFE2059: .size main, .-main .section .rodata.str1.1 .LC4: .string "_Z6assignPii" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2087: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z6assignPii(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2087: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include <stdio.h> #include <sys/time.h> #include <cuda.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( cudaError err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( cudaSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { cudaStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamCreate(&streams[i])); CudaSafeCall(cudaMallocHost((void**)&A[i], sizeof(int) * N)); CudaSafeCall(cudaMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(dA[i], A[i], N * sizeof(int), cudaMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(cudaMemcpyAsync(A[i], dA[i], N * sizeof(int), cudaMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(cudaStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
#include <stdio.h> #include <sys/time.h> #include <hip/hip_runtime.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( hipError_t err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( hipSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, hipGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { hipStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamCreate(&streams[i])); CudaSafeCall(hipHostMalloc((void**)&A[i], sizeof(int) * N, hipHostMallocDefault)); CudaSafeCall(hipMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(dA[i], A[i], N * sizeof(int), hipMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(A[i], dA[i], N * sizeof(int), hipMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <stdio.h> #include <sys/time.h> #include <hip/hip_runtime.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( hipError_t err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( hipSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, hipGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { hipStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamCreate(&streams[i])); CudaSafeCall(hipHostMalloc((void**)&A[i], sizeof(int) * N, hipHostMallocDefault)); CudaSafeCall(hipMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(dA[i], A[i], N * sizeof(int), hipMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(A[i], dA[i], N * sizeof(int), hipMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6assignPii .globl _Z6assignPii .p2align 8 .type _Z6assignPii,@function _Z6assignPii: s_load_b32 s4, s[0:1], 0x1c s_add_u32 s2, s0, 16 s_addc_u32 s3, s1, 0 s_mov_b32 s5, exec_lo s_waitcnt lgkmcnt(0) s_and_b32 s4, s4, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mad_u64_u32 v[1:2], null, s15, s4, v[0:1] v_cmpx_gt_i32_e32 0x100000, v1 s_cbranch_execz .LBB0_3 s_load_b32 s5, s[0:1], 0x8 s_load_b32 s3, s[2:3], 0x0 s_load_b64 s[0:1], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_lshl_b32 s2, s5, 20 s_mul_i32 s3, s3, s4 s_mov_b32 s4, 0 .p2align 6 .LBB0_2: v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[2:3], 2, v[1:2] v_add_co_u32 v2, vcc_lo, s0, v2 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s1, v3, vcc_lo global_load_b32 v0, v[2:3], off s_waitcnt vmcnt(0) v_add3_u32 v0, s2, v1, v0 v_add_nc_u32_e32 v1, s3, v1 global_store_b32 v[2:3], v0, off v_cmp_lt_i32_e32 vcc_lo, 0xfffff, v1 s_or_b32 s4, vcc_lo, s4 s_delay_alu instid0(SALU_CYCLE_1) s_and_not1_b32 exec_lo, exec_lo, s4 s_cbranch_execnz .LBB0_2 .LBB0_3: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6assignPii .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 272 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 4 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6assignPii, .Lfunc_end0-_Z6assignPii .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .offset: 8 .size: 4 .value_kind: by_value - .offset: 16 .size: 4 .value_kind: hidden_block_count_x - .offset: 20 .size: 4 .value_kind: hidden_block_count_y - .offset: 24 .size: 4 .value_kind: hidden_block_count_z - .offset: 28 .size: 2 .value_kind: hidden_group_size_x - .offset: 30 .size: 2 .value_kind: hidden_group_size_y - .offset: 32 .size: 2 .value_kind: hidden_group_size_z - .offset: 34 .size: 2 .value_kind: hidden_remainder_x - .offset: 36 .size: 2 .value_kind: hidden_remainder_y - .offset: 38 .size: 2 .value_kind: hidden_remainder_z - .offset: 56 .size: 8 .value_kind: hidden_global_offset_x - .offset: 64 .size: 8 .value_kind: hidden_global_offset_y - .offset: 72 .size: 8 .value_kind: hidden_global_offset_z - .offset: 80 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 272 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6assignPii .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z6assignPii.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 4 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <stdio.h> #include <sys/time.h> #include <hip/hip_runtime.h> long long getCurrentTime() { struct timeval te; gettimeofday(&te, NULL); // get current time long long microseconds = te.tv_sec*1000000LL + te.tv_usec; return microseconds; } #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) inline void __cudaSafeCall( hipError_t err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( hipSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, hipGetErrorString( err ) ); exit( -1 ); } #endif return; } __global__ void assign(int *A, int streamId) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < 1024 * 1024; i += blockDim.x * gridDim.x) { A[i] += 1024*1024*streamId + i; } } int main() { hipStream_t streams[4]; int *A[4], *dA[4]; int N = 1024*1024; // Initialization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamCreate(&streams[i])); CudaSafeCall(hipHostMalloc((void**)&A[i], sizeof(int) * N, hipHostMallocDefault)); CudaSafeCall(hipMalloc(&dA[i], N * sizeof(int))); } // Asynchronous H2D copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(dA[i], A[i], N * sizeof(int), hipMemcpyHostToDevice, streams[i])); } // Asynchronous kernel launches for (int i = 0; i < 4; i++) { assign<<<1, 1024, 0, streams[i]>>>(dA[i], i); } // Asynchronous D2H copies for (int i = 0; i < 4; i++) { CudaSafeCall(hipMemcpyAsync(A[i], dA[i], N * sizeof(int), hipMemcpyDeviceToHost, streams[i])); } // Synchronization for (int i = 0; i < 4; i++) { CudaSafeCall(hipStreamSynchronize(streams[i])); } // Verification int error = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < N; j++) { if (A[i][j] != i * N + j) { error++; } } } if (!error) { printf("VERIFILED\n"); } else { printf("NOT VERIFIED"); } return 0; }
.text .file "streams.hip" .globl _Z14getCurrentTimev # -- Begin function _Z14getCurrentTimev .p2align 4, 0x90 .type _Z14getCurrentTimev,@function _Z14getCurrentTimev: # @_Z14getCurrentTimev .cfi_startproc # %bb.0: subq $24, %rsp .cfi_def_cfa_offset 32 leaq 8(%rsp), %rdi xorl %esi, %esi callq gettimeofday imulq $1000000, 8(%rsp), %rax # imm = 0xF4240 addq 16(%rsp), %rax addq $24, %rsp .cfi_def_cfa_offset 8 retq .Lfunc_end0: .size _Z14getCurrentTimev, .Lfunc_end0-_Z14getCurrentTimev .cfi_endproc # -- End function .globl _Z21__device_stub__assignPii # -- Begin function _Z21__device_stub__assignPii .p2align 4, 0x90 .type _Z21__device_stub__assignPii,@function _Z21__device_stub__assignPii: # @_Z21__device_stub__assignPii .cfi_startproc # %bb.0: subq $88, %rsp .cfi_def_cfa_offset 96 movq %rdi, 56(%rsp) movl %esi, 4(%rsp) leaq 56(%rsp), %rax movq %rax, 64(%rsp) leaq 4(%rsp), %rax movq %rax, 72(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 64(%rsp), %r9 movl $_Z6assignPii, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $104, %rsp .cfi_adjust_cfa_offset -104 retq .Lfunc_end1: .size _Z21__device_stub__assignPii, .Lfunc_end1-_Z21__device_stub__assignPii .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $184, %rsp .cfi_def_cfa_offset 240 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_1: # =>This Inner Loop Header: Depth=1 leaq (%rsp,%rbx), %rdi addq $80, %rdi callq hipStreamCreate testl %eax, %eax jne .LBB2_2 # %bb.8: # %_Z14__cudaSafeCall10hipError_tPKci.exit # in Loop: Header=BB2_1 Depth=1 leaq (%rsp,%rbx), %rdi addq $144, %rdi movl $4194304, %esi # imm = 0x400000 xorl %edx, %edx callq hipHostMalloc testl %eax, %eax jne .LBB2_9 # %bb.10: # %_Z14__cudaSafeCall10hipError_tPKci.exit52 # in Loop: Header=BB2_1 Depth=1 leaq (%rsp,%rbx), %rdi addq $112, %rdi movl $4194304, %esi # imm = 0x400000 callq hipMalloc testl %eax, %eax jne .LBB2_11 # %bb.4: # in Loop: Header=BB2_1 Depth=1 addq $8, %rbx cmpq $32, %rbx jne .LBB2_1 # %bb.5: # %_Z14__cudaSafeCall10hipError_tPKci.exit56.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_6: # %_Z14__cudaSafeCall10hipError_tPKci.exit56 # =>This Inner Loop Header: Depth=1 movq 112(%rsp,%rbx,8), %rdi movq 144(%rsp,%rbx,8), %rsi movq 80(%rsp,%rbx,8), %r8 movl $4194304, %edx # imm = 0x400000 movl $1, %ecx callq hipMemcpyAsync testl %eax, %eax jne .LBB2_7 # %bb.12: # in Loop: Header=BB2_6 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_6 # %bb.13: # %.preheader67 movabsq $4294967297, %rbx # imm = 0x100000001 xorl %r12d, %r12d leaq 1023(%rbx), %r14 leaq 16(%rsp), %r13 leaq 8(%rsp), %rbp leaq 64(%rsp), %r15 jmp .LBB2_14 .p2align 4, 0x90 .LBB2_16: # in Loop: Header=BB2_14 Depth=1 incq %r12 cmpq $4, %r12 je .LBB2_17 .LBB2_14: # =>This Inner Loop Header: Depth=1 movq 80(%rsp,%r12,8), %r9 movq %rbx, %rdi movl $1, %esi movq %r14, %rdx movl $1, %ecx xorl %r8d, %r8d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB2_16 # %bb.15: # in Loop: Header=BB2_14 Depth=1 movq 112(%rsp,%r12,8), %rax movq %rax, 56(%rsp) movl %r12d, 4(%rsp) leaq 56(%rsp), %rax movq %rax, 64(%rsp) leaq 4(%rsp), %rax movq %rax, 72(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi movq %r13, %rdx movq %rbp, %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d movl $_Z6assignPii, %edi movq %r15, %r9 pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 jmp .LBB2_16 .LBB2_17: # %_Z14__cudaSafeCall10hipError_tPKci.exit58.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_18: # %_Z14__cudaSafeCall10hipError_tPKci.exit58 # =>This Inner Loop Header: Depth=1 movq 144(%rsp,%rbx,8), %rdi movq 112(%rsp,%rbx,8), %rsi movq 80(%rsp,%rbx,8), %r8 movl $4194304, %edx # imm = 0x400000 movl $2, %ecx callq hipMemcpyAsync testl %eax, %eax jne .LBB2_23 # %bb.19: # in Loop: Header=BB2_18 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_18 # %bb.20: # %_Z14__cudaSafeCall10hipError_tPKci.exit60.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_21: # %_Z14__cudaSafeCall10hipError_tPKci.exit60 # =>This Inner Loop Header: Depth=1 movq 80(%rsp,%rbx,8), %rdi callq hipStreamSynchronize testl %eax, %eax jne .LBB2_22 # %bb.24: # in Loop: Header=BB2_21 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_21 # %bb.25: # %.preheader.preheader xorl %eax, %eax xorl %edx, %edx xorl %ecx, %ecx .p2align 4, 0x90 .LBB2_26: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB2_27 Depth 2 movq 144(%rsp,%rdx,8), %rsi xorl %edi, %edi .p2align 4, 0x90 .LBB2_27: # Parent Loop BB2_26 Depth=1 # => This Inner Loop Header: Depth=2 movl (%rsi,%rdi,4), %r8d leaq (%rax,%rdi), %r9 xorl %r10d, %r10d cmpq %r8, %r9 setne %r10b addl %r10d, %ecx incq %rdi cmpq $1048576, %rdi # imm = 0x100000 jne .LBB2_27 # %bb.28: # in Loop: Header=BB2_26 Depth=1 incq %rdx addq $1048576, %rax # imm = 0x100000 cmpq $4, %rdx jne .LBB2_26 # %bb.29: testl %ecx, %ecx je .LBB2_30 # %bb.31: movl $.L.str.2, %edi xorl %eax, %eax callq printf jmp .LBB2_32 .LBB2_30: movl $.Lstr, %edi callq puts@PLT .LBB2_32: xorl %eax, %eax addq $184, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .LBB2_11: .cfi_def_cfa_offset 240 movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $47, %ecx jmp .LBB2_3 .LBB2_9: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $46, %ecx jmp .LBB2_3 .LBB2_2: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $45, %ecx jmp .LBB2_3 .LBB2_7: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $53, %ecx jmp .LBB2_3 .LBB2_23: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $64, %ecx jmp .LBB2_3 .LBB2_22: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $69, %ecx .LBB2_3: movq %rax, %r8 xorl %eax, %eax callq fprintf movl $-1, %edi callq exit .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6assignPii, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z6assignPii,@object # @_Z6assignPii .section .rodata,"a",@progbits .globl _Z6assignPii .p2align 3, 0x0 _Z6assignPii: .quad _Z21__device_stub__assignPii .size _Z6assignPii, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "/home/ubuntu/Datasets/stackv2/train-structured-repos-hip/ryuji0123/cuda-tutorial/master/waseda2018/examples/streams/streams.hip" .size .L.str, 128 .type .L.str.2,@object # @.str.2 .L.str.2: .asciz "NOT VERIFIED" .size .L.str.2, 13 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "cudaSafeCall() failed at %s:%i : %s\n" .size .L.str.3, 37 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6assignPii" .size .L__unnamed_1, 13 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "VERIFILED" .size .Lstr, 10 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__assignPii .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6assignPii .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z6assignPii .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ S2R R0, SR_CTAID.X ; /* 0x0000000000007919 */ /* 0x000e280000002500 */ /*0020*/ S2R R3, SR_TID.X ; /* 0x0000000000037919 */ /* 0x000e240000002100 */ /*0030*/ IMAD R0, R0, c[0x0][0x0], R3 ; /* 0x0000000000007a24 */ /* 0x001fca00078e0203 */ /*0040*/ ISETP.GT.AND P0, PT, R0, 0xfffff, PT ; /* 0x000fffff0000780c */ /* 0x000fda0003f04270 */ /*0050*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0060*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fe40000000a00 */ /*0070*/ HFMA2.MMA R3, -RZ, RZ, 0, 2.384185791015625e-07 ; /* 0x00000004ff037435 */ /* 0x001fd400000001ff */ /*0080*/ IMAD.WIDE R2, R0, R3, c[0x0][0x160] ; /* 0x0000580000027625 */ /* 0x000fca00078e0203 */ /*0090*/ LDG.E R4, [R2.64] ; /* 0x0000000402047981 */ /* 0x000ea2000c1e1900 */ /*00a0*/ MOV R5, 0x100000 ; /* 0x0010000000057802 */ /* 0x000fe40000000f00 */ /*00b0*/ MOV R7, c[0x0][0x0] ; /* 0x0000000000077a02 */ /* 0x000fc60000000f00 */ /*00c0*/ IMAD R5, R5, c[0x0][0x168], R0.reuse ; /* 0x00005a0005057a24 */ /* 0x100fe400078e0200 */ /*00d0*/ IMAD R0, R7, c[0x0][0xc], R0 ; /* 0x0000030007007a24 */ /* 0x000fca00078e0200 */ /*00e0*/ ISETP.GE.AND P0, PT, R0, 0x100000, PT ; /* 0x001000000000780c */ /* 0x000fe40003f06270 */ /*00f0*/ IADD3 R5, R4, R5, RZ ; /* 0x0000000504057210 */ /* 0x004fca0007ffe0ff */ /*0100*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x0001ec000c101904 */ /*0110*/ @!P0 BRA 0x70 ; /* 0xffffff5000008947 */ /* 0x000fea000383ffff */ /*0120*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0130*/ BRA 0x130; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0180*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0190*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*01f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6assignPii .globl _Z6assignPii .p2align 8 .type _Z6assignPii,@function _Z6assignPii: s_load_b32 s4, s[0:1], 0x1c s_add_u32 s2, s0, 16 s_addc_u32 s3, s1, 0 s_mov_b32 s5, exec_lo s_waitcnt lgkmcnt(0) s_and_b32 s4, s4, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mad_u64_u32 v[1:2], null, s15, s4, v[0:1] v_cmpx_gt_i32_e32 0x100000, v1 s_cbranch_execz .LBB0_3 s_load_b32 s5, s[0:1], 0x8 s_load_b32 s3, s[2:3], 0x0 s_load_b64 s[0:1], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_lshl_b32 s2, s5, 20 s_mul_i32 s3, s3, s4 s_mov_b32 s4, 0 .p2align 6 .LBB0_2: v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[2:3], 2, v[1:2] v_add_co_u32 v2, vcc_lo, s0, v2 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s1, v3, vcc_lo global_load_b32 v0, v[2:3], off s_waitcnt vmcnt(0) v_add3_u32 v0, s2, v1, v0 v_add_nc_u32_e32 v1, s3, v1 global_store_b32 v[2:3], v0, off v_cmp_lt_i32_e32 vcc_lo, 0xfffff, v1 s_or_b32 s4, vcc_lo, s4 s_delay_alu instid0(SALU_CYCLE_1) s_and_not1_b32 exec_lo, exec_lo, s4 s_cbranch_execnz .LBB0_2 .LBB0_3: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6assignPii .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 272 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 4 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6assignPii, .Lfunc_end0-_Z6assignPii .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .offset: 8 .size: 4 .value_kind: by_value - .offset: 16 .size: 4 .value_kind: hidden_block_count_x - .offset: 20 .size: 4 .value_kind: hidden_block_count_y - .offset: 24 .size: 4 .value_kind: hidden_block_count_z - .offset: 28 .size: 2 .value_kind: hidden_group_size_x - .offset: 30 .size: 2 .value_kind: hidden_group_size_y - .offset: 32 .size: 2 .value_kind: hidden_group_size_z - .offset: 34 .size: 2 .value_kind: hidden_remainder_x - .offset: 36 .size: 2 .value_kind: hidden_remainder_y - .offset: 38 .size: 2 .value_kind: hidden_remainder_z - .offset: 56 .size: 8 .value_kind: hidden_global_offset_x - .offset: 64 .size: 8 .value_kind: hidden_global_offset_y - .offset: 72 .size: 8 .value_kind: hidden_global_offset_z - .offset: 80 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 272 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6assignPii .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z6assignPii.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 4 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00098435_00000000-6_streams.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2062: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2062: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z14getCurrentTimev .type _Z14getCurrentTimev, @function _Z14getCurrentTimev: .LFB2057: .cfi_startproc endbr64 subq $40, %rsp .cfi_def_cfa_offset 48 movq %fs:40, %rax movq %rax, 24(%rsp) xorl %eax, %eax movq %rsp, %rdi movl $0, %esi call gettimeofday@PLT imulq $1000000, (%rsp), %rax addq 8(%rsp), %rax movq 24(%rsp), %rdx subq %fs:40, %rdx jne .L6 addq $40, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L6: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size _Z14getCurrentTimev, .-_Z14getCurrentTimev .globl _Z26__device_stub__Z6assignPiiPii .type _Z26__device_stub__Z6assignPiiPii, @function _Z26__device_stub__Z6assignPiiPii: .LFB2084: .cfi_startproc endbr64 subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 8(%rsp) movl %esi, 4(%rsp) movq %fs:40, %rax movq %rax, 104(%rsp) xorl %eax, %eax leaq 8(%rsp), %rax movq %rax, 80(%rsp) leaq 4(%rsp), %rax movq %rax, 88(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L11 .L7: movq 104(%rsp), %rax subq %fs:40, %rax jne .L12 addq $120, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L11: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 136 pushq 24(%rsp) .cfi_def_cfa_offset 144 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z6assignPii(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 128 jmp .L7 .L12: call __stack_chk_fail@PLT .cfi_endproc .LFE2084: .size _Z26__device_stub__Z6assignPiiPii, .-_Z26__device_stub__Z6assignPiiPii .globl _Z6assignPii .type _Z6assignPii, @function _Z6assignPii: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z26__device_stub__Z6assignPiiPii addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _Z6assignPii, .-_Z6assignPii .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "/home/ubuntu/Datasets/stackv2/train-structured/ryuji0123/cuda-tutorial/master/waseda2018/examples/streams/streams.cu" .align 8 .LC1: .string "cudaSafeCall() failed at %s:%i : %s\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "VERIFILED\n" .LC3: .string "NOT VERIFIED" .text .globl main .type main, @function main: .LFB2059: .cfi_startproc endbr64 pushq %r13 .cfi_def_cfa_offset 16 .cfi_offset 13, -16 pushq %r12 .cfi_def_cfa_offset 24 .cfi_offset 12, -24 pushq %rbp .cfi_def_cfa_offset 32 .cfi_offset 6, -32 pushq %rbx .cfi_def_cfa_offset 40 .cfi_offset 3, -40 subq $152, %rsp .cfi_def_cfa_offset 192 movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax movl $0, %r12d leaq 96(%rsp), %r13 .L19: leaq 32(%rsp), %rbp leaq (%r12,%rbp), %rdi call cudaStreamCreate@PLT testl %eax, %eax jne .L42 leaq 64(%rsp), %rbx leaq (%r12,%rbx), %rdi movl $4194304, %esi call cudaMallocHost@PLT testl %eax, %eax jne .L43 leaq 0(%r13,%r12), %rdi movl $4194304, %esi call cudaMalloc@PLT testl %eax, %eax jne .L44 addq $8, %r12 cmpq $32, %r12 jne .L19 movl $0, %r12d .L21: movq 64(%rsp,%r12), %rsi movq 96(%rsp,%r12), %rdi movq 32(%rsp,%r12), %r8 movl $1, %ecx movl $4194304, %edx call cudaMemcpyAsync@PLT testl %eax, %eax jne .L45 addq $8, %r12 cmpq $32, %r12 jne .L21 movl $0, %r12d jmp .L23 .L42: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $45, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L43: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $46, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L44: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $47, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L45: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $52, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L22: addq $1, %r12 cmpq $4, %r12 je .L46 .L23: movl $1024, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 8(%rsp) movl $1, 12(%rsp) movl $1, 16(%rsp) movq 32(%rsp,%r12,8), %r9 movl $0, %r8d movq 20(%rsp), %rdx movl $1, %ecx movq 8(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax jne .L22 movq 96(%rsp,%r12,8), %rdi movl %r12d, %esi call _Z26__device_stub__Z6assignPiiPii jmp .L22 .L46: movl $0, %r12d .L25: movq 96(%rsp,%r12), %rsi movq 64(%rsp,%r12), %rdi movq 32(%rsp,%r12), %r8 movl $2, %ecx movl $4194304, %edx call cudaMemcpyAsync@PLT testl %eax, %eax jne .L47 addq $8, %r12 cmpq $32, %r12 jne .L25 leaq 32(%rbp), %r12 .L27: movq 0(%rbp), %rdi call cudaStreamSynchronize@PLT testl %eax, %eax jne .L48 addq $8, %rbp cmpq %r12, %rbp jne .L27 movl $0, %esi jmp .L28 .L47: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $63, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L48: movl %eax, %edi call cudaGetErrorString@PLT movq %rax, %r9 movl $69, %r8d leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $2, %esi movq stderr(%rip), %rdi movl $0, %eax call __fprintf_chk@PLT movl $-1, %edi call exit@PLT .L29: addq $1, %rdx cmpq $1048576, %rdx je .L49 .L30: leal (%rax,%rdx), %ecx cmpl %ecx, (%rdi,%rdx,4) je .L29 addl $1, %esi jmp .L29 .L49: addq $8, %rbx addl $1048576, %eax cmpl $4194304, %eax je .L31 .L28: movq (%rbx), %rdi movl $0, %edx jmp .L30 .L31: testl %esi, %esi jne .L32 leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L33: movq 136(%rsp), %rax subq %fs:40, %rax jne .L50 movl $0, %eax addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %rbp .cfi_def_cfa_offset 24 popq %r12 .cfi_def_cfa_offset 16 popq %r13 .cfi_def_cfa_offset 8 ret .L32: .cfi_restore_state leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L33 .L50: call __stack_chk_fail@PLT .cfi_endproc .LFE2059: .size main, .-main .section .rodata.str1.1 .LC4: .string "_Z6assignPii" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2087: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z6assignPii(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2087: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "streams.hip" .globl _Z14getCurrentTimev # -- Begin function _Z14getCurrentTimev .p2align 4, 0x90 .type _Z14getCurrentTimev,@function _Z14getCurrentTimev: # @_Z14getCurrentTimev .cfi_startproc # %bb.0: subq $24, %rsp .cfi_def_cfa_offset 32 leaq 8(%rsp), %rdi xorl %esi, %esi callq gettimeofday imulq $1000000, 8(%rsp), %rax # imm = 0xF4240 addq 16(%rsp), %rax addq $24, %rsp .cfi_def_cfa_offset 8 retq .Lfunc_end0: .size _Z14getCurrentTimev, .Lfunc_end0-_Z14getCurrentTimev .cfi_endproc # -- End function .globl _Z21__device_stub__assignPii # -- Begin function _Z21__device_stub__assignPii .p2align 4, 0x90 .type _Z21__device_stub__assignPii,@function _Z21__device_stub__assignPii: # @_Z21__device_stub__assignPii .cfi_startproc # %bb.0: subq $88, %rsp .cfi_def_cfa_offset 96 movq %rdi, 56(%rsp) movl %esi, 4(%rsp) leaq 56(%rsp), %rax movq %rax, 64(%rsp) leaq 4(%rsp), %rax movq %rax, 72(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 64(%rsp), %r9 movl $_Z6assignPii, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $104, %rsp .cfi_adjust_cfa_offset -104 retq .Lfunc_end1: .size _Z21__device_stub__assignPii, .Lfunc_end1-_Z21__device_stub__assignPii .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $184, %rsp .cfi_def_cfa_offset 240 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_1: # =>This Inner Loop Header: Depth=1 leaq (%rsp,%rbx), %rdi addq $80, %rdi callq hipStreamCreate testl %eax, %eax jne .LBB2_2 # %bb.8: # %_Z14__cudaSafeCall10hipError_tPKci.exit # in Loop: Header=BB2_1 Depth=1 leaq (%rsp,%rbx), %rdi addq $144, %rdi movl $4194304, %esi # imm = 0x400000 xorl %edx, %edx callq hipHostMalloc testl %eax, %eax jne .LBB2_9 # %bb.10: # %_Z14__cudaSafeCall10hipError_tPKci.exit52 # in Loop: Header=BB2_1 Depth=1 leaq (%rsp,%rbx), %rdi addq $112, %rdi movl $4194304, %esi # imm = 0x400000 callq hipMalloc testl %eax, %eax jne .LBB2_11 # %bb.4: # in Loop: Header=BB2_1 Depth=1 addq $8, %rbx cmpq $32, %rbx jne .LBB2_1 # %bb.5: # %_Z14__cudaSafeCall10hipError_tPKci.exit56.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_6: # %_Z14__cudaSafeCall10hipError_tPKci.exit56 # =>This Inner Loop Header: Depth=1 movq 112(%rsp,%rbx,8), %rdi movq 144(%rsp,%rbx,8), %rsi movq 80(%rsp,%rbx,8), %r8 movl $4194304, %edx # imm = 0x400000 movl $1, %ecx callq hipMemcpyAsync testl %eax, %eax jne .LBB2_7 # %bb.12: # in Loop: Header=BB2_6 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_6 # %bb.13: # %.preheader67 movabsq $4294967297, %rbx # imm = 0x100000001 xorl %r12d, %r12d leaq 1023(%rbx), %r14 leaq 16(%rsp), %r13 leaq 8(%rsp), %rbp leaq 64(%rsp), %r15 jmp .LBB2_14 .p2align 4, 0x90 .LBB2_16: # in Loop: Header=BB2_14 Depth=1 incq %r12 cmpq $4, %r12 je .LBB2_17 .LBB2_14: # =>This Inner Loop Header: Depth=1 movq 80(%rsp,%r12,8), %r9 movq %rbx, %rdi movl $1, %esi movq %r14, %rdx movl $1, %ecx xorl %r8d, %r8d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB2_16 # %bb.15: # in Loop: Header=BB2_14 Depth=1 movq 112(%rsp,%r12,8), %rax movq %rax, 56(%rsp) movl %r12d, 4(%rsp) leaq 56(%rsp), %rax movq %rax, 64(%rsp) leaq 4(%rsp), %rax movq %rax, 72(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi movq %r13, %rdx movq %rbp, %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d movl $_Z6assignPii, %edi movq %r15, %r9 pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 jmp .LBB2_16 .LBB2_17: # %_Z14__cudaSafeCall10hipError_tPKci.exit58.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_18: # %_Z14__cudaSafeCall10hipError_tPKci.exit58 # =>This Inner Loop Header: Depth=1 movq 144(%rsp,%rbx,8), %rdi movq 112(%rsp,%rbx,8), %rsi movq 80(%rsp,%rbx,8), %r8 movl $4194304, %edx # imm = 0x400000 movl $2, %ecx callq hipMemcpyAsync testl %eax, %eax jne .LBB2_23 # %bb.19: # in Loop: Header=BB2_18 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_18 # %bb.20: # %_Z14__cudaSafeCall10hipError_tPKci.exit60.preheader xorl %ebx, %ebx .p2align 4, 0x90 .LBB2_21: # %_Z14__cudaSafeCall10hipError_tPKci.exit60 # =>This Inner Loop Header: Depth=1 movq 80(%rsp,%rbx,8), %rdi callq hipStreamSynchronize testl %eax, %eax jne .LBB2_22 # %bb.24: # in Loop: Header=BB2_21 Depth=1 incq %rbx cmpq $4, %rbx jne .LBB2_21 # %bb.25: # %.preheader.preheader xorl %eax, %eax xorl %edx, %edx xorl %ecx, %ecx .p2align 4, 0x90 .LBB2_26: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB2_27 Depth 2 movq 144(%rsp,%rdx,8), %rsi xorl %edi, %edi .p2align 4, 0x90 .LBB2_27: # Parent Loop BB2_26 Depth=1 # => This Inner Loop Header: Depth=2 movl (%rsi,%rdi,4), %r8d leaq (%rax,%rdi), %r9 xorl %r10d, %r10d cmpq %r8, %r9 setne %r10b addl %r10d, %ecx incq %rdi cmpq $1048576, %rdi # imm = 0x100000 jne .LBB2_27 # %bb.28: # in Loop: Header=BB2_26 Depth=1 incq %rdx addq $1048576, %rax # imm = 0x100000 cmpq $4, %rdx jne .LBB2_26 # %bb.29: testl %ecx, %ecx je .LBB2_30 # %bb.31: movl $.L.str.2, %edi xorl %eax, %eax callq printf jmp .LBB2_32 .LBB2_30: movl $.Lstr, %edi callq puts@PLT .LBB2_32: xorl %eax, %eax addq $184, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .LBB2_11: .cfi_def_cfa_offset 240 movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $47, %ecx jmp .LBB2_3 .LBB2_9: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $46, %ecx jmp .LBB2_3 .LBB2_2: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $45, %ecx jmp .LBB2_3 .LBB2_7: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $53, %ecx jmp .LBB2_3 .LBB2_23: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $64, %ecx jmp .LBB2_3 .LBB2_22: movq stderr(%rip), %rbx movl %eax, %edi callq hipGetErrorString movl $.L.str.3, %esi movl $.L.str, %edx movq %rbx, %rdi movl $69, %ecx .LBB2_3: movq %rax, %r8 xorl %eax, %eax callq fprintf movl $-1, %edi callq exit .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6assignPii, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z6assignPii,@object # @_Z6assignPii .section .rodata,"a",@progbits .globl _Z6assignPii .p2align 3, 0x0 _Z6assignPii: .quad _Z21__device_stub__assignPii .size _Z6assignPii, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "/home/ubuntu/Datasets/stackv2/train-structured-repos-hip/ryuji0123/cuda-tutorial/master/waseda2018/examples/streams/streams.hip" .size .L.str, 128 .type .L.str.2,@object # @.str.2 .L.str.2: .asciz "NOT VERIFIED" .size .L.str.2, 13 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "cudaSafeCall() failed at %s:%i : %s\n" .size .L.str.3, 37 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6assignPii" .size .L__unnamed_1, 13 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "VERIFILED" .size .Lstr, 10 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__assignPii .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6assignPii .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
code for sm_80 Function : _Z5totalPfS_j .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R7, SR_TID.X ; /* 0x0000000000077919 */ /* 0x000e220000002100 */ /*0020*/ ULDC UR4, c[0x0][0x0] ; /* 0x0000000000047ab9 */ /* 0x000fe20000000800 */ /*0030*/ IMAD.MOV.U32 R0, RZ, RZ, RZ ; /* 0x000000ffff007224 */ /* 0x000fe200078e00ff */ /*0040*/ USHF.L.U32 UR5, UR4, 0x1, URZ ; /* 0x0000000104057899 */ /* 0x000fe2000800063f */ /*0050*/ S2R R6, SR_CTAID.X ; /* 0x0000000000067919 */ /* 0x000e620000002500 */ /*0060*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fe20000000a00 */ /*0070*/ ISETP.GE.U32.AND P1, PT, R7, c[0x0][0x170], PT ; /* 0x00005c0007007a0c */ /* 0x001fc60003f26070 */ /*0080*/ IMAD R2, R6, UR5, R7 ; /* 0x0000000506027c24 */ /* 0x002fca000f8e0207 */ /*0090*/ IADD3 R4, R2, c[0x0][0x0], RZ ; /* 0x0000000002047a10 */ /* 0x000fc80007ffe0ff */ /*00a0*/ ISETP.GE.U32.AND P0, PT, R4, c[0x0][0x170], PT ; /* 0x00005c0004007a0c */ /* 0x000fe20003f06070 */ /*00b0*/ @!P1 IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff039424 */ /* 0x000fc800078e00ff */ /*00c0*/ @!P1 IMAD.WIDE.U32 R2, R2, R3, c[0x0][0x160] ; /* 0x0000580002029625 */ /* 0x000fca00078e0003 */ /*00d0*/ @!P1 LDG.E R0, [R2.64] ; /* 0x0000000602009981 */ /* 0x000ea6000c1e1900 */ /*00e0*/ @!P0 MOV R5, 0x4 ; /* 0x0000000400058802 */ /* 0x000fca0000000f00 */ /*00f0*/ @!P0 IMAD.WIDE.U32 R4, R4, R5, c[0x0][0x160] ; /* 0x0000580004048625 */ /* 0x000fcc00078e0005 */ /*0100*/ @!P0 LDG.E R5, [R4.64] ; /* 0x0000000604058981 */ /* 0x000ea2000c1e1900 */ /*0110*/ USHF.R.U32.HI UR4, URZ, 0x1, UR4 ; /* 0x000000013f047899 */ /* 0x000fcc0008011604 */ /*0120*/ ISETP.NE.AND P2, PT, RZ, UR4, PT ; /* 0x00000004ff007c0c */ /* 0x000fe4000bf45270 */ /*0130*/ ISETP.NE.AND P1, PT, R7, RZ, PT ; /* 0x000000ff0700720c */ /* 0x000fe20003f25270 */ /*0140*/ @!P0 FADD R0, R0, R5 ; /* 0x0000000500008221 */ /* 0x004fca0000000000 */ /*0150*/ STS [R7.X4], R0 ; /* 0x0000000007007388 */ /* 0x0001e80000004800 */ /*0160*/ BAR.SYNC.DEFER_BLOCKING 0x0 ; /* 0x0000000000007b1d */ /* 0x000fec0000010000 */ /*0170*/ @!P2 BRA 0x230 ; /* 0x000000b00000a947 */ /* 0x000fea0003800000 */ /*0180*/ SHF.L.U32 R2, R7, 0x2, RZ ; /* 0x0000000207027819 */ /* 0x001fe200000006ff */ /*0190*/ IMAD.U32 R3, RZ, RZ, UR4 ; /* 0x00000004ff037e24 */ /* 0x000fca000f8e00ff */ /*01a0*/ ISETP.GE.U32.AND P0, PT, R7, R3, PT ; /* 0x000000030700720c */ /* 0x000fda0003f06070 */ /*01b0*/ @!P0 LEA R4, R3, R2, 0x2 ; /* 0x0000000203048211 */ /* 0x000fe400078e10ff */ /*01c0*/ SHF.R.U32.HI R3, RZ, 0x1, R3 ; /* 0x00000001ff037819 */ /* 0x000fc60000011603 */ /*01d0*/ @!P0 LDS R5, [R4] ; /* 0x0000000004058984 */ /* 0x000e240000000800 */ /*01e0*/ @!P0 FADD R0, R0, R5 ; /* 0x0000000500008221 */ /* 0x001fca0000000000 */ /*01f0*/ @!P0 STS [R7.X4], R0 ; /* 0x0000000007008388 */ /* 0x0001e80000004800 */ /*0200*/ BAR.SYNC.DEFER_BLOCKING 0x0 ; /* 0x0000000000007b1d */ /* 0x000fe20000010000 */ /*0210*/ ISETP.NE.AND P0, PT, R3, RZ, PT ; /* 0x000000ff0300720c */ /* 0x000fda0003f05270 */ /*0220*/ @P0 BRA 0x1a0 ; /* 0xffffff7000000947 */ /* 0x001fea000383ffff */ /*0230*/ @P1 EXIT ; /* 0x000000000000194d */ /* 0x001fea0003800000 */ /*0240*/ LDS R5, [RZ] ; /* 0x00000000ff057984 */ /* 0x000e220000000800 */ /*0250*/ IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff037424 */ /* 0x000fc800078e00ff */ /*0260*/ IMAD.WIDE.U32 R2, R6, R3, c[0x0][0x168] ; /* 0x00005a0006027625 */ /* 0x000fca00078e0003 */ /*0270*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x001fe2000c101906 */ /*0280*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0290*/ BRA 0x290; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*02a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0300*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0310*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0320*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0330*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0340*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0350*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0360*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0370*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
.file "tmpxft_00015a86_00000000-6_total.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z27__device_stub__Z5totalPfS_jPfS_j .type _Z27__device_stub__Z5totalPfS_jPfS_j, @function _Z27__device_stub__Z5totalPfS_jPfS_j: .LFB2051: .cfi_startproc endbr64 subq $136, %rsp .cfi_def_cfa_offset 144 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %fs:40, %rax movq %rax, 120(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 120(%rsp), %rax subq %fs:40, %rax jne .L8 addq $136, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 152 pushq 40(%rsp) .cfi_def_cfa_offset 160 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z5totalPfS_j(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 144 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z27__device_stub__Z5totalPfS_jPfS_j, .-_Z27__device_stub__Z5totalPfS_jPfS_j .globl _Z5totalPfS_j .type _Z5totalPfS_j, @function _Z5totalPfS_j: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z27__device_stub__Z5totalPfS_jPfS_j addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z5totalPfS_j, .-_Z5totalPfS_j .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "_Z5totalPfS_j" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z5totalPfS_j(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
#include <hip/hip_runtime.h> #include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z5totalPfS_j .globl _Z5totalPfS_j .p2align 8 .type _Z5totalPfS_j,@function _Z5totalPfS_j: s_clause 0x2 s_load_b32 s3, s[0:1], 0x24 s_load_b32 s6, s[0:1], 0x10 s_load_b64 s[4:5], s[0:1], 0x0 v_mov_b32_e32 v3, 0 s_mov_b32 s2, s15 s_waitcnt lgkmcnt(0) s_and_b32 s3, s3, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_mul_i32 s7, s15, s3 v_lshl_add_u32 v1, s7, 1, v0 s_mov_b32 s7, exec_lo v_cmpx_gt_u32_e64 s6, v0 s_cbranch_execz .LBB0_2 v_mov_b32_e32 v2, 0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[2:3], 2, v[1:2] v_add_co_u32 v2, vcc_lo, s4, v2 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s5, v3, vcc_lo global_load_b32 v3, v[2:3], off .LBB0_2: s_or_b32 exec_lo, exec_lo, s7 v_add_nc_u32_e32 v1, s3, v1 s_delay_alu instid0(VALU_DEP_1) v_cmp_gt_u32_e32 vcc_lo, s6, v1 s_and_saveexec_b32 s6, vcc_lo s_cbranch_execz .LBB0_4 v_mov_b32_e32 v2, 0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[1:2], 2, v[1:2] v_add_co_u32 v1, vcc_lo, s4, v1 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v2, vcc_lo, s5, v2, vcc_lo global_load_b32 v1, v[1:2], off s_waitcnt vmcnt(0) v_add_f32_e32 v3, v3, v1 .LBB0_4: s_or_b32 exec_lo, exec_lo, s6 v_lshlrev_b32_e32 v1, 2, v0 s_cmp_lt_u32 s3, 2 s_waitcnt vmcnt(0) ds_store_b32 v1, v3 s_waitcnt lgkmcnt(0) s_barrier s_branch .LBB0_6 .p2align 6 .LBB0_5: s_or_b32 exec_lo, exec_lo, s5 s_waitcnt lgkmcnt(0) s_barrier s_cmp_lt_u32 s3, 4 s_mov_b32 s3, s4 .LBB0_6: buffer_gl0_inv s_cbranch_scc1 .LBB0_9 s_lshr_b32 s4, s3, 1 s_mov_b32 s5, exec_lo v_cmpx_gt_u32_e64 s4, v0 s_cbranch_execz .LBB0_5 v_add_lshl_u32 v2, s4, v0, 2 ds_load_b32 v2, v2 s_waitcnt lgkmcnt(0) v_add_f32_e32 v3, v3, v2 ds_store_b32 v1, v3 s_branch .LBB0_5 .LBB0_9: s_mov_b32 s3, 0 s_mov_b32 s4, exec_lo v_cmpx_eq_u32_e32 0, v0 s_cbranch_execz .LBB0_11 v_mov_b32_e32 v0, 0 s_load_b64 s[0:1], s[0:1], 0x8 s_lshl_b64 s[2:3], s[2:3], 2 ds_load_b32 v1, v0 s_waitcnt lgkmcnt(0) s_add_u32 s0, s0, s2 s_addc_u32 s1, s1, s3 global_store_b32 v0, v1, s[0:1] .LBB0_11: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z5totalPfS_j .amdhsa_group_segment_fixed_size 8192 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 280 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 4 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z5totalPfS_j, .Lfunc_end0-_Z5totalPfS_j .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .offset: 24 .size: 4 .value_kind: hidden_block_count_x - .offset: 28 .size: 4 .value_kind: hidden_block_count_y - .offset: 32 .size: 4 .value_kind: hidden_block_count_z - .offset: 36 .size: 2 .value_kind: hidden_group_size_x - .offset: 38 .size: 2 .value_kind: hidden_group_size_y - .offset: 40 .size: 2 .value_kind: hidden_group_size_z - .offset: 42 .size: 2 .value_kind: hidden_remainder_x - .offset: 44 .size: 2 .value_kind: hidden_remainder_y - .offset: 46 .size: 2 .value_kind: hidden_remainder_z - .offset: 64 .size: 8 .value_kind: hidden_global_offset_x - .offset: 72 .size: 8 .value_kind: hidden_global_offset_y - .offset: 80 .size: 8 .value_kind: hidden_global_offset_z - .offset: 88 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 8192 .kernarg_segment_align: 8 .kernarg_segment_size: 280 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z5totalPfS_j .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z5totalPfS_j.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 4 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include "includes.h" #define BLOCK_SIZE 1024 #ifndef THREADS # define THREADS 1024 #endif __global__ void total(float * input, float * output, unsigned int len) { __shared__ float sum[2*BLOCK_SIZE]; unsigned int i = threadIdx.x; unsigned int j = blockIdx.x * (blockDim.x * 2) + threadIdx.x; float localSum = (i < len) ? input[j] : 0; if (j + blockDim.x < len) localSum += input[j + blockDim.x]; sum[i] = localSum; __syncthreads(); for (unsigned int step = blockDim.x / 2; step >= 1; step >>= 1) { if (i < step) sum[i] = localSum = localSum + sum[i + step]; __syncthreads(); } if(i == 0) output[blockIdx.x] = sum[0]; }
.text .file "total.hip" .globl _Z20__device_stub__totalPfS_j # -- Begin function _Z20__device_stub__totalPfS_j .p2align 4, 0x90 .type _Z20__device_stub__totalPfS_j,@function _Z20__device_stub__totalPfS_j: # @_Z20__device_stub__totalPfS_j .cfi_startproc # %bb.0: subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 12(%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 12(%rsp), %rax movq %rax, 96(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z5totalPfS_j, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $120, %rsp .cfi_adjust_cfa_offset -120 retq .Lfunc_end0: .size _Z20__device_stub__totalPfS_j, .Lfunc_end0-_Z20__device_stub__totalPfS_j .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z5totalPfS_j, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z5totalPfS_j,@object # @_Z5totalPfS_j .section .rodata,"a",@progbits .globl _Z5totalPfS_j .p2align 3, 0x0 _Z5totalPfS_j: .quad _Z20__device_stub__totalPfS_j .size _Z5totalPfS_j, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z5totalPfS_j" .size .L__unnamed_1, 14 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z20__device_stub__totalPfS_j .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z5totalPfS_j .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z5totalPfS_j .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R7, SR_TID.X ; /* 0x0000000000077919 */ /* 0x000e220000002100 */ /*0020*/ ULDC UR4, c[0x0][0x0] ; /* 0x0000000000047ab9 */ /* 0x000fe20000000800 */ /*0030*/ IMAD.MOV.U32 R0, RZ, RZ, RZ ; /* 0x000000ffff007224 */ /* 0x000fe200078e00ff */ /*0040*/ USHF.L.U32 UR5, UR4, 0x1, URZ ; /* 0x0000000104057899 */ /* 0x000fe2000800063f */ /*0050*/ S2R R6, SR_CTAID.X ; /* 0x0000000000067919 */ /* 0x000e620000002500 */ /*0060*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fe20000000a00 */ /*0070*/ ISETP.GE.U32.AND P1, PT, R7, c[0x0][0x170], PT ; /* 0x00005c0007007a0c */ /* 0x001fc60003f26070 */ /*0080*/ IMAD R2, R6, UR5, R7 ; /* 0x0000000506027c24 */ /* 0x002fca000f8e0207 */ /*0090*/ IADD3 R4, R2, c[0x0][0x0], RZ ; /* 0x0000000002047a10 */ /* 0x000fc80007ffe0ff */ /*00a0*/ ISETP.GE.U32.AND P0, PT, R4, c[0x0][0x170], PT ; /* 0x00005c0004007a0c */ /* 0x000fe20003f06070 */ /*00b0*/ @!P1 IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff039424 */ /* 0x000fc800078e00ff */ /*00c0*/ @!P1 IMAD.WIDE.U32 R2, R2, R3, c[0x0][0x160] ; /* 0x0000580002029625 */ /* 0x000fca00078e0003 */ /*00d0*/ @!P1 LDG.E R0, [R2.64] ; /* 0x0000000602009981 */ /* 0x000ea6000c1e1900 */ /*00e0*/ @!P0 MOV R5, 0x4 ; /* 0x0000000400058802 */ /* 0x000fca0000000f00 */ /*00f0*/ @!P0 IMAD.WIDE.U32 R4, R4, R5, c[0x0][0x160] ; /* 0x0000580004048625 */ /* 0x000fcc00078e0005 */ /*0100*/ @!P0 LDG.E R5, [R4.64] ; /* 0x0000000604058981 */ /* 0x000ea2000c1e1900 */ /*0110*/ USHF.R.U32.HI UR4, URZ, 0x1, UR4 ; /* 0x000000013f047899 */ /* 0x000fcc0008011604 */ /*0120*/ ISETP.NE.AND P2, PT, RZ, UR4, PT ; /* 0x00000004ff007c0c */ /* 0x000fe4000bf45270 */ /*0130*/ ISETP.NE.AND P1, PT, R7, RZ, PT ; /* 0x000000ff0700720c */ /* 0x000fe20003f25270 */ /*0140*/ @!P0 FADD R0, R0, R5 ; /* 0x0000000500008221 */ /* 0x004fca0000000000 */ /*0150*/ STS [R7.X4], R0 ; /* 0x0000000007007388 */ /* 0x0001e80000004800 */ /*0160*/ BAR.SYNC.DEFER_BLOCKING 0x0 ; /* 0x0000000000007b1d */ /* 0x000fec0000010000 */ /*0170*/ @!P2 BRA 0x230 ; /* 0x000000b00000a947 */ /* 0x000fea0003800000 */ /*0180*/ SHF.L.U32 R2, R7, 0x2, RZ ; /* 0x0000000207027819 */ /* 0x001fe200000006ff */ /*0190*/ IMAD.U32 R3, RZ, RZ, UR4 ; /* 0x00000004ff037e24 */ /* 0x000fca000f8e00ff */ /*01a0*/ ISETP.GE.U32.AND P0, PT, R7, R3, PT ; /* 0x000000030700720c */ /* 0x000fda0003f06070 */ /*01b0*/ @!P0 LEA R4, R3, R2, 0x2 ; /* 0x0000000203048211 */ /* 0x000fe400078e10ff */ /*01c0*/ SHF.R.U32.HI R3, RZ, 0x1, R3 ; /* 0x00000001ff037819 */ /* 0x000fc60000011603 */ /*01d0*/ @!P0 LDS R5, [R4] ; /* 0x0000000004058984 */ /* 0x000e240000000800 */ /*01e0*/ @!P0 FADD R0, R0, R5 ; /* 0x0000000500008221 */ /* 0x001fca0000000000 */ /*01f0*/ @!P0 STS [R7.X4], R0 ; /* 0x0000000007008388 */ /* 0x0001e80000004800 */ /*0200*/ BAR.SYNC.DEFER_BLOCKING 0x0 ; /* 0x0000000000007b1d */ /* 0x000fe20000010000 */ /*0210*/ ISETP.NE.AND P0, PT, R3, RZ, PT ; /* 0x000000ff0300720c */ /* 0x000fda0003f05270 */ /*0220*/ @P0 BRA 0x1a0 ; /* 0xffffff7000000947 */ /* 0x001fea000383ffff */ /*0230*/ @P1 EXIT ; /* 0x000000000000194d */ /* 0x001fea0003800000 */ /*0240*/ LDS R5, [RZ] ; /* 0x00000000ff057984 */ /* 0x000e220000000800 */ /*0250*/ IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff037424 */ /* 0x000fc800078e00ff */ /*0260*/ IMAD.WIDE.U32 R2, R6, R3, c[0x0][0x168] ; /* 0x00005a0006027625 */ /* 0x000fca00078e0003 */ /*0270*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x001fe2000c101906 */ /*0280*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0290*/ BRA 0x290; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*02a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0300*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0310*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0320*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0330*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0340*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0350*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0360*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0370*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z5totalPfS_j .globl _Z5totalPfS_j .p2align 8 .type _Z5totalPfS_j,@function _Z5totalPfS_j: s_clause 0x2 s_load_b32 s3, s[0:1], 0x24 s_load_b32 s6, s[0:1], 0x10 s_load_b64 s[4:5], s[0:1], 0x0 v_mov_b32_e32 v3, 0 s_mov_b32 s2, s15 s_waitcnt lgkmcnt(0) s_and_b32 s3, s3, 0xffff s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_mul_i32 s7, s15, s3 v_lshl_add_u32 v1, s7, 1, v0 s_mov_b32 s7, exec_lo v_cmpx_gt_u32_e64 s6, v0 s_cbranch_execz .LBB0_2 v_mov_b32_e32 v2, 0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[2:3], 2, v[1:2] v_add_co_u32 v2, vcc_lo, s4, v2 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s5, v3, vcc_lo global_load_b32 v3, v[2:3], off .LBB0_2: s_or_b32 exec_lo, exec_lo, s7 v_add_nc_u32_e32 v1, s3, v1 s_delay_alu instid0(VALU_DEP_1) v_cmp_gt_u32_e32 vcc_lo, s6, v1 s_and_saveexec_b32 s6, vcc_lo s_cbranch_execz .LBB0_4 v_mov_b32_e32 v2, 0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_lshlrev_b64 v[1:2], 2, v[1:2] v_add_co_u32 v1, vcc_lo, s4, v1 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v2, vcc_lo, s5, v2, vcc_lo global_load_b32 v1, v[1:2], off s_waitcnt vmcnt(0) v_add_f32_e32 v3, v3, v1 .LBB0_4: s_or_b32 exec_lo, exec_lo, s6 v_lshlrev_b32_e32 v1, 2, v0 s_cmp_lt_u32 s3, 2 s_waitcnt vmcnt(0) ds_store_b32 v1, v3 s_waitcnt lgkmcnt(0) s_barrier s_branch .LBB0_6 .p2align 6 .LBB0_5: s_or_b32 exec_lo, exec_lo, s5 s_waitcnt lgkmcnt(0) s_barrier s_cmp_lt_u32 s3, 4 s_mov_b32 s3, s4 .LBB0_6: buffer_gl0_inv s_cbranch_scc1 .LBB0_9 s_lshr_b32 s4, s3, 1 s_mov_b32 s5, exec_lo v_cmpx_gt_u32_e64 s4, v0 s_cbranch_execz .LBB0_5 v_add_lshl_u32 v2, s4, v0, 2 ds_load_b32 v2, v2 s_waitcnt lgkmcnt(0) v_add_f32_e32 v3, v3, v2 ds_store_b32 v1, v3 s_branch .LBB0_5 .LBB0_9: s_mov_b32 s3, 0 s_mov_b32 s4, exec_lo v_cmpx_eq_u32_e32 0, v0 s_cbranch_execz .LBB0_11 v_mov_b32_e32 v0, 0 s_load_b64 s[0:1], s[0:1], 0x8 s_lshl_b64 s[2:3], s[2:3], 2 ds_load_b32 v1, v0 s_waitcnt lgkmcnt(0) s_add_u32 s0, s0, s2 s_addc_u32 s1, s1, s3 global_store_b32 v0, v1, s[0:1] .LBB0_11: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z5totalPfS_j .amdhsa_group_segment_fixed_size 8192 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 280 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 4 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z5totalPfS_j, .Lfunc_end0-_Z5totalPfS_j .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .offset: 16 .size: 4 .value_kind: by_value - .offset: 24 .size: 4 .value_kind: hidden_block_count_x - .offset: 28 .size: 4 .value_kind: hidden_block_count_y - .offset: 32 .size: 4 .value_kind: hidden_block_count_z - .offset: 36 .size: 2 .value_kind: hidden_group_size_x - .offset: 38 .size: 2 .value_kind: hidden_group_size_y - .offset: 40 .size: 2 .value_kind: hidden_group_size_z - .offset: 42 .size: 2 .value_kind: hidden_remainder_x - .offset: 44 .size: 2 .value_kind: hidden_remainder_y - .offset: 46 .size: 2 .value_kind: hidden_remainder_z - .offset: 64 .size: 8 .value_kind: hidden_global_offset_x - .offset: 72 .size: 8 .value_kind: hidden_global_offset_y - .offset: 80 .size: 8 .value_kind: hidden_global_offset_z - .offset: 88 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 8192 .kernarg_segment_align: 8 .kernarg_segment_size: 280 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z5totalPfS_j .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z5totalPfS_j.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 4 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00015a86_00000000-6_total.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z27__device_stub__Z5totalPfS_jPfS_j .type _Z27__device_stub__Z5totalPfS_jPfS_j, @function _Z27__device_stub__Z5totalPfS_jPfS_j: .LFB2051: .cfi_startproc endbr64 subq $136, %rsp .cfi_def_cfa_offset 144 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movl %edx, 12(%rsp) movq %fs:40, %rax movq %rax, 120(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 12(%rsp), %rax movq %rax, 112(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 120(%rsp), %rax subq %fs:40, %rax jne .L8 addq $136, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 152 pushq 40(%rsp) .cfi_def_cfa_offset 160 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z5totalPfS_j(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 144 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z27__device_stub__Z5totalPfS_jPfS_j, .-_Z27__device_stub__Z5totalPfS_jPfS_j .globl _Z5totalPfS_j .type _Z5totalPfS_j, @function _Z5totalPfS_j: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z27__device_stub__Z5totalPfS_jPfS_j addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z5totalPfS_j, .-_Z5totalPfS_j .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "_Z5totalPfS_j" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z5totalPfS_j(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "total.hip" .globl _Z20__device_stub__totalPfS_j # -- Begin function _Z20__device_stub__totalPfS_j .p2align 4, 0x90 .type _Z20__device_stub__totalPfS_j,@function _Z20__device_stub__totalPfS_j: # @_Z20__device_stub__totalPfS_j .cfi_startproc # %bb.0: subq $104, %rsp .cfi_def_cfa_offset 112 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movl %edx, 12(%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 12(%rsp), %rax movq %rax, 96(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z5totalPfS_j, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $120, %rsp .cfi_adjust_cfa_offset -120 retq .Lfunc_end0: .size _Z20__device_stub__totalPfS_j, .Lfunc_end0-_Z20__device_stub__totalPfS_j .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z5totalPfS_j, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z5totalPfS_j,@object # @_Z5totalPfS_j .section .rodata,"a",@progbits .globl _Z5totalPfS_j .p2align 3, 0x0 _Z5totalPfS_j: .quad _Z20__device_stub__totalPfS_j .size _Z5totalPfS_j, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z5totalPfS_j" .size .L__unnamed_1, 14 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z20__device_stub__totalPfS_j .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z5totalPfS_j .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include "cuda_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
code for sm_80
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include "cuda_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
.file "tmpxft_000dcbb9_00000000-6_defineGridBlock.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "grid.x: %d block.x %d \n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 movl $1024, %ecx movl $1, %edx leaq .LC0(%rip), %rbx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $512, %ecx movl $2, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $256, %ecx movl $4, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $128, %ecx movl $8, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $0, %eax popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2057: .size main, .-main .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "cuda_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
#include "hip/hip_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include "hip/hip_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include "hip/hip_runtime.h" #include "stdio.h" int main(){ // define total data elements int nElem = 1024; // define grid and block structure dim3 block(1024); dim3 grid((nElem+block.x-1)/block.x); printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 512; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 256; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); // reset block block.x = 128; grid.x = (nElem+block.x-1)/block.x; printf("grid.x: %d block.x %d \n", grid.x, block.x); }
.text .file "defineGridBlock.hip" .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rax .cfi_def_cfa_offset 16 movl $.L.str, %edi movl $1, %esi movl $1024, %edx # imm = 0x400 xorl %eax, %eax callq printf movl $.L.str, %edi movl $2, %esi movl $512, %edx # imm = 0x200 xorl %eax, %eax callq printf movl $.L.str, %edi movl $4, %esi movl $256, %edx # imm = 0x100 xorl %eax, %eax callq printf movl $.L.str, %edi movl $8, %esi movl $128, %edx xorl %eax, %eax callq printf xorl %eax, %eax popq %rcx .cfi_def_cfa_offset 8 retq .Lfunc_end0: .size main, .Lfunc_end0-main .cfi_endproc # -- End function .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "grid.x: %d block.x %d \n" .size .L.str, 26 .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80
.text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .amdgpu_metadata --- amdhsa.kernels: [] amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_000dcbb9_00000000-6_defineGridBlock.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.1,"aMS",@progbits,1 .LC0: .string "grid.x: %d block.x %d \n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 movl $1024, %ecx movl $1, %edx leaq .LC0(%rip), %rbx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $512, %ecx movl $2, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $256, %ecx movl $4, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $128, %ecx movl $8, %edx movq %rbx, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $0, %eax popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2057: .size main, .-main .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "defineGridBlock.hip" .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rax .cfi_def_cfa_offset 16 movl $.L.str, %edi movl $1, %esi movl $1024, %edx # imm = 0x400 xorl %eax, %eax callq printf movl $.L.str, %edi movl $2, %esi movl $512, %edx # imm = 0x200 xorl %eax, %eax callq printf movl $.L.str, %edi movl $4, %esi movl $256, %edx # imm = 0x100 xorl %eax, %eax callq printf movl $.L.str, %edi movl $8, %esi movl $128, %edx xorl %eax, %eax callq printf xorl %eax, %eax popq %rcx .cfi_def_cfa_offset 8 retq .Lfunc_end0: .size main, .Lfunc_end0-main .cfi_endproc # -- End function .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "grid.x: %d block.x %d \n" .size .L.str, 26 .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
# include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; cudaGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); cudaDeviceProp prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { cudaGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
code for sm_80 Function : _Z3addiiPi .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ IMAD.MOV.U32 R5, RZ, RZ, c[0x0][0x164] ; /* 0x00005900ff057624 */ /* 0x000fe200078e00ff */ /*0020*/ MOV R2, c[0x0][0x168] ; /* 0x00005a0000027a02 */ /* 0x000fe20000000f00 */ /*0030*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fe20000000a00 */ /*0040*/ MOV R3, c[0x0][0x16c] ; /* 0x00005b0000037a02 */ /* 0x000fe40000000f00 */ /*0050*/ IADD3 R5, R5, c[0x0][0x160], RZ ; /* 0x0000580005057a10 */ /* 0x000fca0007ffe0ff */ /*0060*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x000fe2000c101904 */ /*0070*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0080*/ BRA 0x80; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0100*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0110*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0120*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0130*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ .......... Function : _Z6kernelv .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0020*/ BRA 0x20; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0030*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0040*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0050*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0060*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0070*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0080*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
# include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; cudaGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); cudaDeviceProp prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { cudaGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
.file "tmpxft_00118ae7_00000000-6_cudaTest.cudafe1.cpp" .text #APP .globl _ZSt21ios_base_library_initv #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB3672: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3672: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "\nDetected %d CUDA device(s) on this computer\n" .align 8 .LC1: .string " --- General Information for Device %d ---\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "\tDevice name:\t\t %s\n" .LC3: .string "\tCompute capability:\t %d.%d\n" .LC4: .string "\tClock rate:\t\t %d\n" .section .rodata.str1.8 .align 8 .LC5: .string " --- Memory Information for device %d ---\n" .section .rodata.str1.1 .LC6: .string "\tTotal global mem:\t %f\n" .LC7: .string "\tTotal constant Mem:\t %ld\n" .section .rodata.str1.8 .align 8 .LC8: .string " --- MP Information for device %d ---\n" .section .rodata.str1.1 .LC9: .string "\tMultiprocessor count:\t %d\n" .LC10: .string "\tShared mem per mp:\t %ld\n" .LC11: .string "\tRegisters per mp:\t %d\n" .LC12: .string "\tThreads in warp:\t %d\n" .LC13: .string "\tMax threads per block:\t %d\n" .section .rodata.str1.8 .align 8 .LC14: .string "\tMax thread dimensions:\t (%d, %d, %d)\n" .align 8 .LC15: .string "\tMax grid dimensions:\t (%f, %d, %d)\n" .section .rodata.str1.1 .LC16: .string "\n" .text .globl main .type main, @function main: .LFB3669: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $1064, %rsp .cfi_def_cfa_offset 1120 movq %fs:40, %rax movq %rax, 1048(%rsp) xorl %eax, %eax leaq 12(%rsp), %rdi call cudaGetDeviceCount@PLT movl 12(%rsp), %edx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT cmpl $0, 12(%rsp) jle .L4 movl $0, %ebx leaq .LC1(%rip), %r15 leaq .LC2(%rip), %r14 leaq .LC3(%rip), %r13 leaq .LC4(%rip), %r12 .L5: leaq 16(%rsp), %rbp movl %ebx, %esi movq %rbp, %rdi call cudaGetDeviceProperties_v2@PLT movl %ebx, %edx movq %r15, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq %rbp, %rdx movq %r14, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 380(%rsp), %ecx movl 376(%rsp), %edx movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 364(%rsp), %edx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %edx leaq .LC5(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 304(%rsp), %rdx leaq .LC6(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 368(%rsp), %rdx leaq .LC7(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %edx leaq .LC8(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 404(%rsp), %edx leaq .LC9(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 312(%rsp), %rdx leaq .LC10(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 320(%rsp), %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 324(%rsp), %edx leaq .LC12(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 336(%rsp), %edx leaq .LC13(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 348(%rsp), %r8d movl 344(%rsp), %ecx movl 340(%rsp), %edx leaq .LC14(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 360(%rsp), %r8d movl 356(%rsp), %ecx movl 352(%rsp), %edx leaq .LC15(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq .LC16(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addl $1, %ebx cmpl %ebx, 12(%rsp) jg .L5 .L4: movq 1048(%rsp), %rax subq %fs:40, %rax jne .L9 movl $0, %eax addq $1064, %rsp .cfi_remember_state .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .L9: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE3669: .size main, .-main .globl _Z24__device_stub__Z6kernelvv .type _Z24__device_stub__Z6kernelvv, @function _Z24__device_stub__Z6kernelvv: .LFB3694: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L14 .L10: movq 72(%rsp), %rax subq %fs:40, %rax jne .L15 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L14: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z6kernelv(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L10 .L15: call __stack_chk_fail@PLT .cfi_endproc .LFE3694: .size _Z24__device_stub__Z6kernelvv, .-_Z24__device_stub__Z6kernelvv .globl _Z6kernelv .type _Z6kernelv, @function _Z6kernelv: .LFB3695: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z24__device_stub__Z6kernelvv addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3695: .size _Z6kernelv, .-_Z6kernelv .globl _Z24__device_stub__Z3addiiPiiiPi .type _Z24__device_stub__Z3addiiPiiiPi, @function _Z24__device_stub__Z3addiiPiiiPi: .LFB3696: .cfi_startproc endbr64 subq $120, %rsp .cfi_def_cfa_offset 128 movl %edi, 12(%rsp) movl %esi, 8(%rsp) movq %rdx, (%rsp) movq %fs:40, %rax movq %rax, 104(%rsp) xorl %eax, %eax leaq 12(%rsp), %rax movq %rax, 80(%rsp) leaq 8(%rsp), %rax movq %rax, 88(%rsp) movq %rsp, %rax movq %rax, 96(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L22 .L18: movq 104(%rsp), %rax subq %fs:40, %rax jne .L23 addq $120, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L22: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 136 pushq 24(%rsp) .cfi_def_cfa_offset 144 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z3addiiPi(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 128 jmp .L18 .L23: call __stack_chk_fail@PLT .cfi_endproc .LFE3696: .size _Z24__device_stub__Z3addiiPiiiPi, .-_Z24__device_stub__Z3addiiPiiiPi .globl _Z3addiiPi .type _Z3addiiPi, @function _Z3addiiPi: .LFB3697: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z24__device_stub__Z3addiiPiiiPi addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3697: .size _Z3addiiPi, .-_Z3addiiPi .section .rodata.str1.1 .LC17: .string "_Z3addiiPi" .LC18: .string "_Z6kernelv" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB3699: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rbx movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC17(%rip), %rdx movq %rdx, %rcx leaq _Z3addiiPi(%rip), %rsi movq %rax, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC18(%rip), %rdx movq %rdx, %rcx leaq _Z6kernelv(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3699: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
# include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; cudaGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); cudaDeviceProp prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { cudaGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
#include <hip/hip_runtime.h> # include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; hipGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); hipDeviceProp_t prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { hipGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> # include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; hipGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); hipDeviceProp_t prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { hipGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6kernelv .globl _Z6kernelv .p2align 8 .type _Z6kernelv,@function _Z6kernelv: s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6kernelv .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 0 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 0 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 1 .amdhsa_next_free_sgpr 1 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6kernelv, .Lfunc_end0-_Z6kernelv .section .AMDGPU.csdata,"",@progbits .text .protected _Z3addiiPi .globl _Z3addiiPi .p2align 8 .type _Z3addiiPi,@function _Z3addiiPi: s_load_b128 s[0:3], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_add_i32 s0, s1, s0 s_delay_alu instid0(SALU_CYCLE_1) v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s0 global_store_b32 v0, v1, s[2:3] s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z3addiiPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 16 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 2 .amdhsa_next_free_sgpr 4 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end1: .size _Z3addiiPi, .Lfunc_end1-_Z3addiiPi .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: [] .group_segment_fixed_size: 0 .kernarg_segment_align: 4 .kernarg_segment_size: 0 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6kernelv .private_segment_fixed_size: 0 .sgpr_count: 0 .sgpr_spill_count: 0 .symbol: _Z6kernelv.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 0 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 16 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z3addiiPi .private_segment_fixed_size: 0 .sgpr_count: 4 .sgpr_spill_count: 0 .symbol: _Z3addiiPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 2 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> # include <iostream> // # include "book.h" __global__ void kernel ( void ) { } __global__ void add( int a, int b, int *c ) { *c = a + b; } int main(int argc, char const *argv[]) { // kernel<<<1, 1>>>(); // printf( "Hello, World!\n"); // int c; // int *dev_c; // int N = 1; // cudaMalloc( (void**)&dev_c, N * sizeof(int) ); // add<<<1, N>>>( 2, 7 , dev_c); // cudaMemcpy ( &c, dev_c, N * sizeof(int), // cudaMemcpyDeviceToHost ); // printf("2 + 7 = %d\n", c); // cudaFree( dev_c); /* Below are the codes for detecting CUDA capable devices and * output some useful information of the deives for references * by Wu Zheshu, May 18, 2018 */ int cudaDeviceCount; hipGetDeviceCount( &cudaDeviceCount ); printf("\nDetected %d CUDA device(s) on this computer\n", cudaDeviceCount); hipDeviceProp_t prop; // cudaDeviceProp is a built-in struct // list all the detected devices for (int i=0; i < cudaDeviceCount; i++) { hipGetDeviceProperties( &prop, i ); printf(" --- General Information for Device %d ---\n", i ); printf("\tDevice name:\t\t %s\n", prop.name ); printf("\tCompute capability:\t %d.%d\n", prop.major, prop.minor ); printf("\tClock rate:\t\t %d\n", prop.clockRate ); printf(" --- Memory Information for device %d ---\n", i ); printf("\tTotal global mem:\t %f\n", prop.totalGlobalMem ); //[Caution] Here the type of totalGlobalMem is "size_t", might // output some erroneous number if the type is not set correctly printf("\tTotal constant Mem:\t %ld\n", prop.totalConstMem ); //[Caution] Here the type of totalConstMem is "size_t" printf(" --- MP Information for device %d ---\n", i ); printf("\tMultiprocessor count:\t %d\n", prop.multiProcessorCount ); printf("\tShared mem per mp:\t %ld\n", prop.sharedMemPerBlock ); printf("\tRegisters per mp:\t %d\n", prop.regsPerBlock ); printf("\tThreads in warp:\t %d\n", prop.warpSize ); printf("\tMax threads per block:\t %d\n", prop.maxThreadsPerBlock ); printf("\tMax thread dimensions:\t (%d, %d, %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] ); printf("\tMax grid dimensions:\t (%f, %d, %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2] ); printf("\n" ); } return 0; }
.text .file "cudaTest.hip" # Start of file scope inline assembly .globl _ZSt21ios_base_library_initv # End of file scope inline assembly .globl _Z21__device_stub__kernelv # -- Begin function _Z21__device_stub__kernelv .p2align 4, 0x90 .type _Z21__device_stub__kernelv,@function _Z21__device_stub__kernelv: # @_Z21__device_stub__kernelv .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z6kernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z21__device_stub__kernelv, .Lfunc_end0-_Z21__device_stub__kernelv .cfi_endproc # -- End function .globl _Z18__device_stub__addiiPi # -- Begin function _Z18__device_stub__addiiPi .p2align 4, 0x90 .type _Z18__device_stub__addiiPi,@function _Z18__device_stub__addiiPi: # @_Z18__device_stub__addiiPi .cfi_startproc # %bb.0: subq $88, %rsp .cfi_def_cfa_offset 96 movl %edi, 4(%rsp) movl %esi, (%rsp) movq %rdx, 56(%rsp) leaq 4(%rsp), %rax movq %rax, 64(%rsp) movq %rsp, %rax movq %rax, 72(%rsp) leaq 56(%rsp), %rax movq %rax, 80(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 64(%rsp), %r9 movl $_Z3addiiPi, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $104, %rsp .cfi_adjust_cfa_offset -104 retq .Lfunc_end1: .size _Z18__device_stub__addiiPi, .Lfunc_end1-_Z18__device_stub__addiiPi .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %rbx .cfi_def_cfa_offset 24 subq $1480, %rsp # imm = 0x5C8 .cfi_def_cfa_offset 1504 .cfi_offset %rbx, -24 .cfi_offset %rbp, -16 leaq 4(%rsp), %rdi callq hipGetDeviceCount movl 4(%rsp), %esi movl $.L.str, %edi xorl %eax, %eax callq printf cmpl $0, 4(%rsp) jle .LBB2_3 # %bb.1: # %.lr.ph leaq 8(%rsp), %rbx xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_2: # =>This Inner Loop Header: Depth=1 movq %rbx, %rdi movl %ebp, %esi callq hipGetDevicePropertiesR0600 movl $.L.str.1, %edi movl %ebp, %esi xorl %eax, %eax callq printf movl $.L.str.2, %edi movq %rbx, %rsi xorl %eax, %eax callq printf movl 368(%rsp), %esi movl 372(%rsp), %edx movl $.L.str.3, %edi xorl %eax, %eax callq printf movl 356(%rsp), %esi movl $.L.str.4, %edi xorl %eax, %eax callq printf movl $.L.str.5, %edi movl %ebp, %esi xorl %eax, %eax callq printf movq 296(%rsp), %rsi movl $.L.str.6, %edi xorl %eax, %eax callq printf movq 360(%rsp), %rsi movl $.L.str.7, %edi xorl %eax, %eax callq printf movl $.L.str.8, %edi movl %ebp, %esi xorl %eax, %eax callq printf movl 396(%rsp), %esi movl $.L.str.9, %edi xorl %eax, %eax callq printf movq 304(%rsp), %rsi movl $.L.str.10, %edi xorl %eax, %eax callq printf movl 312(%rsp), %esi movl $.L.str.11, %edi xorl %eax, %eax callq printf movl 316(%rsp), %esi movl $.L.str.12, %edi xorl %eax, %eax callq printf movl 328(%rsp), %esi movl $.L.str.13, %edi xorl %eax, %eax callq printf movl 332(%rsp), %esi movl 336(%rsp), %edx movl 340(%rsp), %ecx movl $.L.str.14, %edi xorl %eax, %eax callq printf movl 344(%rsp), %esi movl 348(%rsp), %edx movl 352(%rsp), %ecx movl $.L.str.15, %edi xorl %eax, %eax callq printf movl $10, %edi callq putchar@PLT incl %ebp cmpl 4(%rsp), %ebp jl .LBB2_2 .LBB2_3: # %._crit_edge xorl %eax, %eax addq $1480, %rsp # imm = 0x5C8 .cfi_def_cfa_offset 24 popq %rbx .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rbx xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6kernelv, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z3addiiPi, %esi movl $.L__unnamed_2, %edx movl $.L__unnamed_2, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z6kernelv,@object # @_Z6kernelv .section .rodata,"a",@progbits .globl _Z6kernelv .p2align 3, 0x0 _Z6kernelv: .quad _Z21__device_stub__kernelv .size _Z6kernelv, 8 .type _Z3addiiPi,@object # @_Z3addiiPi .globl _Z3addiiPi .p2align 3, 0x0 _Z3addiiPi: .quad _Z18__device_stub__addiiPi .size _Z3addiiPi, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "\nDetected %d CUDA device(s) on this computer\n" .size .L.str, 46 .type .L.str.1,@object # @.str.1 .L.str.1: .asciz " --- General Information for Device %d ---\n" .size .L.str.1, 44 .type .L.str.2,@object # @.str.2 .L.str.2: .asciz "\tDevice name:\t\t %s\n" .size .L.str.2, 20 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "\tCompute capability:\t %d.%d\n" .size .L.str.3, 29 .type .L.str.4,@object # @.str.4 .L.str.4: .asciz "\tClock rate:\t\t %d\n" .size .L.str.4, 19 .type .L.str.5,@object # @.str.5 .L.str.5: .asciz " --- Memory Information for device %d ---\n" .size .L.str.5, 43 .type .L.str.6,@object # @.str.6 .L.str.6: .asciz "\tTotal global mem:\t %f\n" .size .L.str.6, 24 .type .L.str.7,@object # @.str.7 .L.str.7: .asciz "\tTotal constant Mem:\t %ld\n" .size .L.str.7, 27 .type .L.str.8,@object # @.str.8 .L.str.8: .asciz " --- MP Information for device %d ---\n" .size .L.str.8, 39 .type .L.str.9,@object # @.str.9 .L.str.9: .asciz "\tMultiprocessor count:\t %d\n" .size .L.str.9, 28 .type .L.str.10,@object # @.str.10 .L.str.10: .asciz "\tShared mem per mp:\t %ld\n" .size .L.str.10, 26 .type .L.str.11,@object # @.str.11 .L.str.11: .asciz "\tRegisters per mp:\t %d\n" .size .L.str.11, 24 .type .L.str.12,@object # @.str.12 .L.str.12: .asciz "\tThreads in warp:\t %d\n" .size .L.str.12, 23 .type .L.str.13,@object # @.str.13 .L.str.13: .asciz "\tMax threads per block:\t %d\n" .size .L.str.13, 29 .type .L.str.14,@object # @.str.14 .L.str.14: .asciz "\tMax thread dimensions:\t (%d, %d, %d)\n" .size .L.str.14, 39 .type .L.str.15,@object # @.str.15 .L.str.15: .asciz "\tMax grid dimensions:\t (%f, %d, %d)\n" .size .L.str.15, 37 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6kernelv" .size .L__unnamed_1, 11 .type .L__unnamed_2,@object # @1 .L__unnamed_2: .asciz "_Z3addiiPi" .size .L__unnamed_2, 11 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__kernelv .addrsig_sym _Z18__device_stub__addiiPi .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6kernelv .addrsig_sym _Z3addiiPi .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z3addiiPi .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ IMAD.MOV.U32 R5, RZ, RZ, c[0x0][0x164] ; /* 0x00005900ff057624 */ /* 0x000fe200078e00ff */ /*0020*/ MOV R2, c[0x0][0x168] ; /* 0x00005a0000027a02 */ /* 0x000fe20000000f00 */ /*0030*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fe20000000a00 */ /*0040*/ MOV R3, c[0x0][0x16c] ; /* 0x00005b0000037a02 */ /* 0x000fe40000000f00 */ /*0050*/ IADD3 R5, R5, c[0x0][0x160], RZ ; /* 0x0000580005057a10 */ /* 0x000fca0007ffe0ff */ /*0060*/ STG.E [R2.64], R5 ; /* 0x0000000502007986 */ /* 0x000fe2000c101904 */ /*0070*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0080*/ BRA 0x80; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0100*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0110*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0120*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0130*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0140*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0150*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0160*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0170*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ .......... Function : _Z6kernelv .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ MOV R1, c[0x0][0x28] ; /* 0x00000a0000017a02 */ /* 0x000fe40000000f00 */ /*0010*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0020*/ BRA 0x20; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0030*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0040*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0050*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0060*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0070*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0080*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0090*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*00f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z6kernelv .globl _Z6kernelv .p2align 8 .type _Z6kernelv,@function _Z6kernelv: s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z6kernelv .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 0 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 0 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 1 .amdhsa_next_free_sgpr 1 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z6kernelv, .Lfunc_end0-_Z6kernelv .section .AMDGPU.csdata,"",@progbits .text .protected _Z3addiiPi .globl _Z3addiiPi .p2align 8 .type _Z3addiiPi,@function _Z3addiiPi: s_load_b128 s[0:3], s[0:1], 0x0 s_waitcnt lgkmcnt(0) s_add_i32 s0, s1, s0 s_delay_alu instid0(SALU_CYCLE_1) v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s0 global_store_b32 v0, v1, s[2:3] s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z3addiiPi .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 16 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 2 .amdhsa_next_free_sgpr 4 .amdhsa_reserve_vcc 0 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end1: .size _Z3addiiPi, .Lfunc_end1-_Z3addiiPi .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: [] .group_segment_fixed_size: 0 .kernarg_segment_align: 4 .kernarg_segment_size: 0 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z6kernelv .private_segment_fixed_size: 0 .sgpr_count: 0 .sgpr_spill_count: 0 .symbol: _Z6kernelv.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 0 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 - .args: - .offset: 0 .size: 4 .value_kind: by_value - .offset: 4 .size: 4 .value_kind: by_value - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 16 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z3addiiPi .private_segment_fixed_size: 0 .sgpr_count: 4 .sgpr_spill_count: 0 .symbol: _Z3addiiPi.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 2 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_00118ae7_00000000-6_cudaTest.cudafe1.cpp" .text #APP .globl _ZSt21ios_base_library_initv #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB3672: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3672: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "\nDetected %d CUDA device(s) on this computer\n" .align 8 .LC1: .string " --- General Information for Device %d ---\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "\tDevice name:\t\t %s\n" .LC3: .string "\tCompute capability:\t %d.%d\n" .LC4: .string "\tClock rate:\t\t %d\n" .section .rodata.str1.8 .align 8 .LC5: .string " --- Memory Information for device %d ---\n" .section .rodata.str1.1 .LC6: .string "\tTotal global mem:\t %f\n" .LC7: .string "\tTotal constant Mem:\t %ld\n" .section .rodata.str1.8 .align 8 .LC8: .string " --- MP Information for device %d ---\n" .section .rodata.str1.1 .LC9: .string "\tMultiprocessor count:\t %d\n" .LC10: .string "\tShared mem per mp:\t %ld\n" .LC11: .string "\tRegisters per mp:\t %d\n" .LC12: .string "\tThreads in warp:\t %d\n" .LC13: .string "\tMax threads per block:\t %d\n" .section .rodata.str1.8 .align 8 .LC14: .string "\tMax thread dimensions:\t (%d, %d, %d)\n" .align 8 .LC15: .string "\tMax grid dimensions:\t (%f, %d, %d)\n" .section .rodata.str1.1 .LC16: .string "\n" .text .globl main .type main, @function main: .LFB3669: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $1064, %rsp .cfi_def_cfa_offset 1120 movq %fs:40, %rax movq %rax, 1048(%rsp) xorl %eax, %eax leaq 12(%rsp), %rdi call cudaGetDeviceCount@PLT movl 12(%rsp), %edx leaq .LC0(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT cmpl $0, 12(%rsp) jle .L4 movl $0, %ebx leaq .LC1(%rip), %r15 leaq .LC2(%rip), %r14 leaq .LC3(%rip), %r13 leaq .LC4(%rip), %r12 .L5: leaq 16(%rsp), %rbp movl %ebx, %esi movq %rbp, %rdi call cudaGetDeviceProperties_v2@PLT movl %ebx, %edx movq %r15, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq %rbp, %rdx movq %r14, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 380(%rsp), %ecx movl 376(%rsp), %edx movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 364(%rsp), %edx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %edx leaq .LC5(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 304(%rsp), %rdx leaq .LC6(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 368(%rsp), %rdx leaq .LC7(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %edx leaq .LC8(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 404(%rsp), %edx leaq .LC9(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 312(%rsp), %rdx leaq .LC10(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 320(%rsp), %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 324(%rsp), %edx leaq .LC12(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 336(%rsp), %edx leaq .LC13(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 348(%rsp), %r8d movl 344(%rsp), %ecx movl 340(%rsp), %edx leaq .LC14(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl 360(%rsp), %r8d movl 356(%rsp), %ecx movl 352(%rsp), %edx leaq .LC15(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq .LC16(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addl $1, %ebx cmpl %ebx, 12(%rsp) jg .L5 .L4: movq 1048(%rsp), %rax subq %fs:40, %rax jne .L9 movl $0, %eax addq $1064, %rsp .cfi_remember_state .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .L9: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE3669: .size main, .-main .globl _Z24__device_stub__Z6kernelvv .type _Z24__device_stub__Z6kernelvv, @function _Z24__device_stub__Z6kernelvv: .LFB3694: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L14 .L10: movq 72(%rsp), %rax subq %fs:40, %rax jne .L15 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L14: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z6kernelv(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L10 .L15: call __stack_chk_fail@PLT .cfi_endproc .LFE3694: .size _Z24__device_stub__Z6kernelvv, .-_Z24__device_stub__Z6kernelvv .globl _Z6kernelv .type _Z6kernelv, @function _Z6kernelv: .LFB3695: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z24__device_stub__Z6kernelvv addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3695: .size _Z6kernelv, .-_Z6kernelv .globl _Z24__device_stub__Z3addiiPiiiPi .type _Z24__device_stub__Z3addiiPiiiPi, @function _Z24__device_stub__Z3addiiPiiiPi: .LFB3696: .cfi_startproc endbr64 subq $120, %rsp .cfi_def_cfa_offset 128 movl %edi, 12(%rsp) movl %esi, 8(%rsp) movq %rdx, (%rsp) movq %fs:40, %rax movq %rax, 104(%rsp) xorl %eax, %eax leaq 12(%rsp), %rax movq %rax, 80(%rsp) leaq 8(%rsp), %rax movq %rax, 88(%rsp) movq %rsp, %rax movq %rax, 96(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) movl $1, 40(%rsp) movl $1, 44(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) leaq 24(%rsp), %rcx leaq 16(%rsp), %rdx leaq 44(%rsp), %rsi leaq 32(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L22 .L18: movq 104(%rsp), %rax subq %fs:40, %rax jne .L23 addq $120, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L22: .cfi_restore_state pushq 24(%rsp) .cfi_def_cfa_offset 136 pushq 24(%rsp) .cfi_def_cfa_offset 144 leaq 96(%rsp), %r9 movq 60(%rsp), %rcx movl 68(%rsp), %r8d movq 48(%rsp), %rsi movl 56(%rsp), %edx leaq _Z3addiiPi(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 128 jmp .L18 .L23: call __stack_chk_fail@PLT .cfi_endproc .LFE3696: .size _Z24__device_stub__Z3addiiPiiiPi, .-_Z24__device_stub__Z3addiiPiiiPi .globl _Z3addiiPi .type _Z3addiiPi, @function _Z3addiiPi: .LFB3697: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z24__device_stub__Z3addiiPiiiPi addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3697: .size _Z3addiiPi, .-_Z3addiiPi .section .rodata.str1.1 .LC17: .string "_Z3addiiPi" .LC18: .string "_Z6kernelv" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB3699: .cfi_startproc endbr64 pushq %rbx .cfi_def_cfa_offset 16 .cfi_offset 3, -16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rbx movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC17(%rip), %rdx movq %rdx, %rcx leaq _Z3addiiPi(%rip), %rsi movq %rax, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC18(%rip), %rdx movq %rdx, %rcx leaq _Z6kernelv(%rip), %rsi movq %rbx, %rdi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT popq %rbx .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE3699: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "cudaTest.hip" # Start of file scope inline assembly .globl _ZSt21ios_base_library_initv # End of file scope inline assembly .globl _Z21__device_stub__kernelv # -- Begin function _Z21__device_stub__kernelv .p2align 4, 0x90 .type _Z21__device_stub__kernelv,@function _Z21__device_stub__kernelv: # @_Z21__device_stub__kernelv .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 32(%rsp), %rdi leaq 16(%rsp), %rsi leaq 8(%rsp), %rdx movq %rsp, %rcx callq __hipPopCallConfiguration movq 32(%rsp), %rsi movl 40(%rsp), %edx movq 16(%rsp), %rcx movl 24(%rsp), %r8d leaq 48(%rsp), %r9 movl $_Z6kernelv, %edi pushq (%rsp) .cfi_adjust_cfa_offset 8 pushq 16(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z21__device_stub__kernelv, .Lfunc_end0-_Z21__device_stub__kernelv .cfi_endproc # -- End function .globl _Z18__device_stub__addiiPi # -- Begin function _Z18__device_stub__addiiPi .p2align 4, 0x90 .type _Z18__device_stub__addiiPi,@function _Z18__device_stub__addiiPi: # @_Z18__device_stub__addiiPi .cfi_startproc # %bb.0: subq $88, %rsp .cfi_def_cfa_offset 96 movl %edi, 4(%rsp) movl %esi, (%rsp) movq %rdx, 56(%rsp) leaq 4(%rsp), %rax movq %rax, 64(%rsp) movq %rsp, %rax movq %rax, 72(%rsp) leaq 56(%rsp), %rax movq %rax, 80(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 64(%rsp), %r9 movl $_Z3addiiPi, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $104, %rsp .cfi_adjust_cfa_offset -104 retq .Lfunc_end1: .size _Z18__device_stub__addiiPi, .Lfunc_end1-_Z18__device_stub__addiiPi .cfi_endproc # -- End function .globl main # -- Begin function main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %rbx .cfi_def_cfa_offset 24 subq $1480, %rsp # imm = 0x5C8 .cfi_def_cfa_offset 1504 .cfi_offset %rbx, -24 .cfi_offset %rbp, -16 leaq 4(%rsp), %rdi callq hipGetDeviceCount movl 4(%rsp), %esi movl $.L.str, %edi xorl %eax, %eax callq printf cmpl $0, 4(%rsp) jle .LBB2_3 # %bb.1: # %.lr.ph leaq 8(%rsp), %rbx xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_2: # =>This Inner Loop Header: Depth=1 movq %rbx, %rdi movl %ebp, %esi callq hipGetDevicePropertiesR0600 movl $.L.str.1, %edi movl %ebp, %esi xorl %eax, %eax callq printf movl $.L.str.2, %edi movq %rbx, %rsi xorl %eax, %eax callq printf movl 368(%rsp), %esi movl 372(%rsp), %edx movl $.L.str.3, %edi xorl %eax, %eax callq printf movl 356(%rsp), %esi movl $.L.str.4, %edi xorl %eax, %eax callq printf movl $.L.str.5, %edi movl %ebp, %esi xorl %eax, %eax callq printf movq 296(%rsp), %rsi movl $.L.str.6, %edi xorl %eax, %eax callq printf movq 360(%rsp), %rsi movl $.L.str.7, %edi xorl %eax, %eax callq printf movl $.L.str.8, %edi movl %ebp, %esi xorl %eax, %eax callq printf movl 396(%rsp), %esi movl $.L.str.9, %edi xorl %eax, %eax callq printf movq 304(%rsp), %rsi movl $.L.str.10, %edi xorl %eax, %eax callq printf movl 312(%rsp), %esi movl $.L.str.11, %edi xorl %eax, %eax callq printf movl 316(%rsp), %esi movl $.L.str.12, %edi xorl %eax, %eax callq printf movl 328(%rsp), %esi movl $.L.str.13, %edi xorl %eax, %eax callq printf movl 332(%rsp), %esi movl 336(%rsp), %edx movl 340(%rsp), %ecx movl $.L.str.14, %edi xorl %eax, %eax callq printf movl 344(%rsp), %esi movl 348(%rsp), %edx movl 352(%rsp), %ecx movl $.L.str.15, %edi xorl %eax, %eax callq printf movl $10, %edi callq putchar@PLT incl %ebp cmpl 4(%rsp), %ebp jl .LBB2_2 .LBB2_3: # %._crit_edge xorl %eax, %eax addq $1480, %rsp # imm = 0x5C8 .cfi_def_cfa_offset 24 popq %rbx .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: pushq %rbx .cfi_def_cfa_offset 16 subq $32, %rsp .cfi_def_cfa_offset 48 .cfi_offset %rbx, -16 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rbx xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z6kernelv, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z3addiiPi, %esi movl $.L__unnamed_2, %edx movl $.L__unnamed_2, %ecx movq %rbx, %rdi movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $32, %rsp .cfi_def_cfa_offset 16 popq %rbx .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z6kernelv,@object # @_Z6kernelv .section .rodata,"a",@progbits .globl _Z6kernelv .p2align 3, 0x0 _Z6kernelv: .quad _Z21__device_stub__kernelv .size _Z6kernelv, 8 .type _Z3addiiPi,@object # @_Z3addiiPi .globl _Z3addiiPi .p2align 3, 0x0 _Z3addiiPi: .quad _Z18__device_stub__addiiPi .size _Z3addiiPi, 8 .type .L.str,@object # @.str .section .rodata.str1.1,"aMS",@progbits,1 .L.str: .asciz "\nDetected %d CUDA device(s) on this computer\n" .size .L.str, 46 .type .L.str.1,@object # @.str.1 .L.str.1: .asciz " --- General Information for Device %d ---\n" .size .L.str.1, 44 .type .L.str.2,@object # @.str.2 .L.str.2: .asciz "\tDevice name:\t\t %s\n" .size .L.str.2, 20 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "\tCompute capability:\t %d.%d\n" .size .L.str.3, 29 .type .L.str.4,@object # @.str.4 .L.str.4: .asciz "\tClock rate:\t\t %d\n" .size .L.str.4, 19 .type .L.str.5,@object # @.str.5 .L.str.5: .asciz " --- Memory Information for device %d ---\n" .size .L.str.5, 43 .type .L.str.6,@object # @.str.6 .L.str.6: .asciz "\tTotal global mem:\t %f\n" .size .L.str.6, 24 .type .L.str.7,@object # @.str.7 .L.str.7: .asciz "\tTotal constant Mem:\t %ld\n" .size .L.str.7, 27 .type .L.str.8,@object # @.str.8 .L.str.8: .asciz " --- MP Information for device %d ---\n" .size .L.str.8, 39 .type .L.str.9,@object # @.str.9 .L.str.9: .asciz "\tMultiprocessor count:\t %d\n" .size .L.str.9, 28 .type .L.str.10,@object # @.str.10 .L.str.10: .asciz "\tShared mem per mp:\t %ld\n" .size .L.str.10, 26 .type .L.str.11,@object # @.str.11 .L.str.11: .asciz "\tRegisters per mp:\t %d\n" .size .L.str.11, 24 .type .L.str.12,@object # @.str.12 .L.str.12: .asciz "\tThreads in warp:\t %d\n" .size .L.str.12, 23 .type .L.str.13,@object # @.str.13 .L.str.13: .asciz "\tMax threads per block:\t %d\n" .size .L.str.13, 29 .type .L.str.14,@object # @.str.14 .L.str.14: .asciz "\tMax thread dimensions:\t (%d, %d, %d)\n" .size .L.str.14, 39 .type .L.str.15,@object # @.str.15 .L.str.15: .asciz "\tMax grid dimensions:\t (%f, %d, %d)\n" .size .L.str.15, 37 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z6kernelv" .size .L__unnamed_1, 11 .type .L__unnamed_2,@object # @1 .L__unnamed_2: .asciz "_Z3addiiPi" .size .L__unnamed_2, 11 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z21__device_stub__kernelv .addrsig_sym _Z18__device_stub__addiiPi .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z6kernelv .addrsig_sym _Z3addiiPi .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include <cuda.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ cudaMalloc (&mat_in1_gpu, size); cudaMalloc (&mat_in2_gpu, size); cudaMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ cudaMemcpy(mat_in1_gpu, mat_in1, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_in2_gpu, mat_in2, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_out_gpu, mat_out, size, cudaMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); cudaDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ cudaMemcpy(mat_out, mat_out_gpu, size, cudaMemcpyDeviceToHost); /* free memory */ cudaFree(mat_out_gpu); cudaFree(mat_in1_gpu); cudaFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
code for sm_80 Function : _Z7kernaddPfS_S_ii .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ IABS R7, c[0x0][0x178] ; /* 0x00005e0000077a13 */ /* 0x000fe20000000000 */ /*0020*/ S2R R6, SR_CTAID.X ; /* 0x0000000000067919 */ /* 0x000e260000002500 */ /*0030*/ I2F.RP R0, R7 ; /* 0x0000000700007306 */ /* 0x000e620000209400 */ /*0040*/ S2R R5, SR_TID.X ; /* 0x0000000000057919 */ /* 0x000e2e0000002100 */ /*0050*/ MUFU.RCP R0, R0 ; /* 0x0000000000007308 */ /* 0x002e620000001000 */ /*0060*/ IMAD R6, R6, c[0x0][0x0], R5 ; /* 0x0000000006067a24 */ /* 0x001fe200078e0205 */ /*0070*/ IADD3 R2, R0, 0xffffffe, RZ ; /* 0x0ffffffe00027810 */ /* 0x002fc80007ffe0ff */ /*0080*/ IABS R0, R6 ; /* 0x0000000600007213 */ /* 0x000fe40000000000 */ /*0090*/ F2I.FTZ.U32.TRUNC.NTZ R3, R2 ; /* 0x0000000200037305 */ /* 0x000064000021f000 */ /*00a0*/ HFMA2.MMA R2, -RZ, RZ, 0, 0 ; /* 0x00000000ff027435 */ /* 0x001fe200000001ff */ /*00b0*/ IMAD.MOV R4, RZ, RZ, -R3 ; /* 0x000000ffff047224 */ /* 0x002fc800078e0a03 */ /*00c0*/ IMAD R5, R4, R7, RZ ; /* 0x0000000704057224 */ /* 0x000fca00078e02ff */ /*00d0*/ IMAD.HI.U32 R3, R3, R5, R2 ; /* 0x0000000503037227 */ /* 0x000fcc00078e0002 */ /*00e0*/ IMAD.HI.U32 R3, R3, R0, RZ ; /* 0x0000000003037227 */ /* 0x000fc800078e00ff */ /*00f0*/ IMAD.MOV R4, RZ, RZ, -R3 ; /* 0x000000ffff047224 */ /* 0x000fc800078e0a03 */ /*0100*/ IMAD R0, R7, R4, R0 ; /* 0x0000000407007224 */ /* 0x000fca00078e0200 */ /*0110*/ ISETP.GT.U32.AND P2, PT, R7, R0, PT ; /* 0x000000000700720c */ /* 0x000fda0003f44070 */ /*0120*/ @!P2 IADD3 R0, R0, -R7.reuse, RZ ; /* 0x800000070000a210 */ /* 0x080fe40007ffe0ff */ /*0130*/ @!P2 IADD3 R3, R3, 0x1, RZ ; /* 0x000000010303a810 */ /* 0x000fe40007ffe0ff */ /*0140*/ ISETP.GE.U32.AND P0, PT, R0, R7, PT ; /* 0x000000070000720c */ /* 0x000fe40003f06070 */ /*0150*/ LOP3.LUT R0, R6, c[0x0][0x178], RZ, 0x3c, !PT ; /* 0x00005e0006007a12 */ /* 0x000fe400078e3cff */ /*0160*/ ISETP.NE.AND P2, PT, RZ, c[0x0][0x178], PT ; /* 0x00005e00ff007a0c */ /* 0x000fe40003f45270 */ /*0170*/ ISETP.GE.AND P1, PT, R0, RZ, PT ; /* 0x000000ff0000720c */ /* 0x000fce0003f26270 */ /*0180*/ @P0 IADD3 R3, R3, 0x1, RZ ; /* 0x0000000103030810 */ /* 0x000fcc0007ffe0ff */ /*0190*/ @!P1 IMAD.MOV R3, RZ, RZ, -R3 ; /* 0x000000ffff039224 */ /* 0x000fe200078e0a03 */ /*01a0*/ @!P2 LOP3.LUT R3, RZ, c[0x0][0x178], RZ, 0x33, !PT ; /* 0x00005e00ff03aa12 */ /* 0x000fc800078e33ff */ /*01b0*/ IADD3 R5, -R3, RZ, RZ ; /* 0x000000ff03057210 */ /* 0x000fca0007ffe1ff */ /*01c0*/ IMAD R0, R5, c[0x0][0x178], R6 ; /* 0x00005e0005007a24 */ /* 0x000fca00078e0206 */ /*01d0*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x178], PT ; /* 0x00005e0000007a0c */ /* 0x000fc80003f06270 */ /*01e0*/ ISETP.GE.OR P0, PT, R3, c[0x0][0x17c], P0 ; /* 0x00005f0003007a0c */ /* 0x000fda0000706670 */ /*01f0*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0200*/ IMAD.MOV.U32 R7, RZ, RZ, 0x4 ; /* 0x00000004ff077424 */ /* 0x000fe200078e00ff */ /*0210*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fc60000000a00 */ /*0220*/ IMAD.WIDE R4, R6, R7, c[0x0][0x170] ; /* 0x00005c0006047625 */ /* 0x000fc800078e0207 */ /*0230*/ IMAD.WIDE R2, R6.reuse, R7.reuse, c[0x0][0x168] ; /* 0x00005a0006027625 */ /* 0x0c0fe400078e0207 */ /*0240*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ea8000c1e1900 */ /*0250*/ LDG.E R3, [R2.64] ; /* 0x0000000402037981 */ /* 0x000ea2000c1e1900 */ /*0260*/ IMAD.WIDE R6, R6, R7, c[0x0][0x160] ; /* 0x0000580006067625 */ /* 0x000fc800078e0207 */ /*0270*/ FADD R9, R4, R3 ; /* 0x0000000304097221 */ /* 0x004fca0000000000 */ /*0280*/ STG.E [R6.64], R9 ; /* 0x0000000906007986 */ /* 0x000fe2000c101904 */ /*0290*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*02a0*/ BRA 0x2a0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*02b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0300*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0310*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0320*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0330*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0340*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0350*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0360*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0370*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include <cuda.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ cudaMalloc (&mat_in1_gpu, size); cudaMalloc (&mat_in2_gpu, size); cudaMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ cudaMemcpy(mat_in1_gpu, mat_in1, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_in2_gpu, mat_in2, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_out_gpu, mat_out, size, cudaMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); cudaDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ cudaMemcpy(mat_out, mat_out_gpu, size, cudaMemcpyDeviceToHost); /* free memory */ cudaFree(mat_out_gpu); cudaFree(mat_in1_gpu); cudaFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
.file "tmpxft_0003d2e8_00000000-6_addongpu.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2061: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2061: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z4InitPfii .type _Z4InitPfii, @function _Z4InitPfii: .LFB2057: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $24, %rsp .cfi_def_cfa_offset 80 movq %rdi, 8(%rsp) testl %esi, %esi jle .L3 movl %edx, %r12d movslq %esi, %r15 leaq 0(,%r15,4), %r13 movl $0, %r14d jmp .L5 .L6: call drand48@PLT cvtsd2ss %xmm0, %xmm0 movss %xmm0, 0(%rbp) addl $1, %ebx addq %r13, %rbp cmpl %ebx, %r12d jne .L6 .L8: addq $1, %r14 cmpq %r15, %r14 je .L3 .L5: movq 8(%rsp), %rax leaq (%rax,%r14,4), %rbp movl $0, %ebx testl %r12d, %r12d jg .L6 jmp .L8 .L3: addq $24, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2057: .size _Z4InitPfii, .-_Z4InitPfii .globl _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii .type _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii, @function _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii: .LFB2083: .cfi_startproc endbr64 subq $152, %rsp .cfi_def_cfa_offset 160 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movq %rdx, 8(%rsp) movl %ecx, 4(%rsp) movl %r8d, (%rsp) movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 8(%rsp), %rax movq %rax, 112(%rsp) leaq 4(%rsp), %rax movq %rax, 120(%rsp) movq %rsp, %rax movq %rax, 128(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L16 .L12: movq 136(%rsp), %rax subq %fs:40, %rax jne .L17 addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L16: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 168 pushq 40(%rsp) .cfi_def_cfa_offset 176 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z7kernaddPfS_S_ii(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 160 jmp .L12 .L17: call __stack_chk_fail@PLT .cfi_endproc .LFE2083: .size _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii, .-_Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii .globl _Z7kernaddPfS_S_ii .type _Z7kernaddPfS_S_ii, @function _Z7kernaddPfS_S_ii: .LFB2084: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2084: .size _Z7kernaddPfS_S_ii, .-_Z7kernaddPfS_S_ii .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "FAILED\n" .LC3: .string "PASSED\n" .text .globl main .type main, @function main: .LFB2058: .cfi_startproc endbr64 pushq %r12 .cfi_def_cfa_offset 16 .cfi_offset 12, -16 pushq %rbp .cfi_def_cfa_offset 24 .cfi_offset 6, -24 pushq %rbx .cfi_def_cfa_offset 32 .cfi_offset 3, -32 subq $64, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 56(%rsp) xorl %eax, %eax movl $80000, %edi call malloc@PLT movq %rax, %r12 movl $80000, %edi call malloc@PLT movq %rax, %rbp movl $80000, %edi call malloc@PLT movq %rax, %rbx leaq 16(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT leaq 24(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT leaq 8(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT movl $100, %edx movl $200, %esi movq %r12, %rdi call _Z4InitPfii movl $100, %edx movl $200, %esi movq %rbp, %rdi call _Z4InitPfii movl $1, %ecx movl $80000, %edx movq %r12, %rsi movq 16(%rsp), %rdi call cudaMemcpy@PLT movl $1, %ecx movl $80000, %edx movq %rbp, %rsi movq 24(%rsp), %rdi call cudaMemcpy@PLT movl $1, %ecx movl $80000, %edx movq %rbx, %rsi movq 8(%rsp), %rdi call cudaMemcpy@PLT movl $256, 44(%rsp) movl $1, 48(%rsp) movl $79, 32(%rsp) movl $1, 36(%rsp) movl $0, %r9d movl $0, %r8d movq 44(%rsp), %rdx movl $1, %ecx movq 32(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L30 .L21: call cudaDeviceSynchronize@PLT movl $2, %ecx movl $80000, %edx movq 8(%rsp), %rsi movq %rbx, %rdi call cudaMemcpy@PLT movq 8(%rsp), %rdi call cudaFree@PLT movq 16(%rsp), %rdi call cudaFree@PLT movq 24(%rsp), %rdi call cudaFree@PLT movl $0, %eax movl $0, %edx movss .LC0(%rip), %xmm2 movl $1, %ecx .L23: movss (%r12,%rax), %xmm1 addss 0(%rbp,%rax), %xmm1 movss (%rbx,%rax), %xmm0 subss %xmm1, %xmm0 andps %xmm2, %xmm0 ucomiss .LC1(%rip), %xmm0 cmova %ecx, %edx addq $4, %rax cmpq $80000, %rax jne .L23 testl %edx, %edx je .L24 leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L25: movq %r12, %rdi call free@PLT movq %rbp, %rdi call free@PLT movq %rbx, %rdi call free@PLT movq 56(%rsp), %rax subq %fs:40, %rax jne .L31 movl $0, %eax addq $64, %rsp .cfi_remember_state .cfi_def_cfa_offset 32 popq %rbx .cfi_def_cfa_offset 24 popq %rbp .cfi_def_cfa_offset 16 popq %r12 .cfi_def_cfa_offset 8 ret .L30: .cfi_restore_state movl $100, %r8d movl $200, %ecx movq 24(%rsp), %rdx movq 16(%rsp), %rsi movq 8(%rsp), %rdi call _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii jmp .L21 .L24: leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L25 .L31: call __stack_chk_fail@PLT .cfi_endproc .LFE2058: .size main, .-main .section .rodata.str1.1 .LC4: .string "_Z7kernaddPfS_S_ii" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2086: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z7kernaddPfS_S_ii(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2086: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .section .rodata.cst16,"aM",@progbits,16 .align 16 .LC0: .long 2147483647 .long 0 .long 0 .long 0 .section .rodata.cst4,"aM",@progbits,4 .align 4 .LC1: .long 869711765 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include <cuda.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ cudaMalloc (&mat_in1_gpu, size); cudaMalloc (&mat_in2_gpu, size); cudaMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ cudaMemcpy(mat_in1_gpu, mat_in1, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_in2_gpu, mat_in2, size, cudaMemcpyHostToDevice); cudaMemcpy(mat_out_gpu, mat_out, size, cudaMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); cudaDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ cudaMemcpy(mat_out, mat_out_gpu, size, cudaMemcpyDeviceToHost); /* free memory */ cudaFree(mat_out_gpu); cudaFree(mat_in1_gpu); cudaFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
#include <hip/hip_runtime.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ hipMalloc (&mat_in1_gpu, size); hipMalloc (&mat_in2_gpu, size); hipMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ hipMemcpy(mat_in1_gpu, mat_in1, size, hipMemcpyHostToDevice); hipMemcpy(mat_in2_gpu, mat_in2, size, hipMemcpyHostToDevice); hipMemcpy(mat_out_gpu, mat_out, size, hipMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); hipDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ hipMemcpy(mat_out, mat_out_gpu, size, hipMemcpyDeviceToHost); /* free memory */ hipFree(mat_out_gpu); hipFree(mat_in1_gpu); hipFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ hipMalloc (&mat_in1_gpu, size); hipMalloc (&mat_in2_gpu, size); hipMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ hipMemcpy(mat_in1_gpu, mat_in1, size, hipMemcpyHostToDevice); hipMemcpy(mat_in2_gpu, mat_in2, size, hipMemcpyHostToDevice); hipMemcpy(mat_out_gpu, mat_out, size, hipMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); hipDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ hipMemcpy(mat_out, mat_out_gpu, size, hipMemcpyDeviceToHost); /* free memory */ hipFree(mat_out_gpu); hipFree(mat_in1_gpu); hipFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z7kernaddPfS_S_ii .globl _Z7kernaddPfS_S_ii .p2align 8 .type _Z7kernaddPfS_S_ii,@function _Z7kernaddPfS_S_ii: s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b32 s4, s[0:1], 0x2c s_waitcnt lgkmcnt(0) s_ashr_i32 s5, s2, 31 s_and_b32 s4, s4, 0xffff s_add_i32 s6, s2, s5 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_xor_b32 s6, s6, s5 v_cvt_f32_u32_e32 v1, s6 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1) v_rcp_iflag_f32_e32 v1, v1 s_waitcnt_depctr 0xfff v_mul_f32_e32 v1, 0x4f7ffffe, v1 v_cvt_u32_f32_e32 v3, v1 v_mad_u64_u32 v[1:2], null, s15, s4, v[0:1] s_sub_i32 s4, 0, s6 s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) v_mul_lo_u32 v0, s4, v3 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_ashrrev_i32_e32 v2, 31, v1 v_mul_hi_u32 v0, v3, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v4, v1, v2 v_xor_b32_e32 v4, v4, v2 v_xor_b32_e32 v2, s5, v2 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v0, v3, v0 v_mul_hi_u32 v0, v4, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_lo_u32 v3, v0, s6 v_sub_nc_u32_e32 v3, v4, v3 v_add_nc_u32_e32 v4, 1, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_subrev_nc_u32_e32 v5, s6, v3 v_cmp_le_u32_e32 vcc_lo, s6, v3 v_dual_cndmask_b32 v3, v3, v5 :: v_dual_cndmask_b32 v0, v0, v4 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_cmp_le_u32_e32 vcc_lo, s6, v3 v_add_nc_u32_e32 v4, 1, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_cndmask_b32_e32 v0, v0, v4, vcc_lo v_xor_b32_e32 v0, v0, v2 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_sub_nc_u32_e32 v0, v0, v2 v_mul_lo_u32 v2, v0, s2 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_sub_nc_u32_e32 v2, v1, v2 v_cmp_gt_i32_e32 vcc_lo, s2, v2 v_cmp_gt_i32_e64 s2, s3, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_and_b32 s2, s2, vcc_lo s_and_saveexec_b32 s3, s2 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b128 s[4:7], s[0:1], 0x0 s_load_b64 s[0:1], s[0:1], 0x10 v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[1:2] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s6, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s7, v1, vcc_lo v_add_co_u32 v4, vcc_lo, s0, v0 v_add_co_ci_u32_e32 v5, vcc_lo, s1, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s4, v0 global_load_b32 v2, v[2:3], off global_load_b32 v3, v[4:5], off v_add_co_ci_u32_e32 v1, vcc_lo, s5, v1, vcc_lo s_waitcnt vmcnt(0) v_add_f32_e32 v2, v2, v3 global_store_b32 v[0:1], v2, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z7kernaddPfS_S_ii .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 288 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 6 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z7kernaddPfS_S_ii, .Lfunc_end0-_Z7kernaddPfS_S_ii .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .offset: 24 .size: 4 .value_kind: by_value - .offset: 28 .size: 4 .value_kind: by_value - .offset: 32 .size: 4 .value_kind: hidden_block_count_x - .offset: 36 .size: 4 .value_kind: hidden_block_count_y - .offset: 40 .size: 4 .value_kind: hidden_block_count_z - .offset: 44 .size: 2 .value_kind: hidden_group_size_x - .offset: 46 .size: 2 .value_kind: hidden_group_size_y - .offset: 48 .size: 2 .value_kind: hidden_group_size_z - .offset: 50 .size: 2 .value_kind: hidden_remainder_x - .offset: 52 .size: 2 .value_kind: hidden_remainder_y - .offset: 54 .size: 2 .value_kind: hidden_remainder_z - .offset: 72 .size: 8 .value_kind: hidden_global_offset_x - .offset: 80 .size: 8 .value_kind: hidden_global_offset_y - .offset: 88 .size: 8 .value_kind: hidden_global_offset_z - .offset: 96 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 288 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z7kernaddPfS_S_ii .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z7kernaddPfS_S_ii.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 6 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include <stdlib.h> #include <stdio.h> #include <math.h> /* Define the matrix size */ #define NX 200 #define NY 100 /* Size of a block */ #define BLOCKSIZE 256 __global__ void kernadd (float* mout, float* min1, float *min2, int nx, int ny) { int i, j, index; index = blockDim.x*blockIdx.x+threadIdx.x; j = index/nx; i = index - j*nx; if ((i < nx) && (j < ny)) mout[index] = min1[index] + min2[index]; } /*******************************************************/ /* We initialize the vectors with random values */ /*******************************************************/ void Init(float* mat, int nx, int ny) { int i, j; for (i = 0; i < nx; i++) { /* 2D loop */ for (j = 0; j < ny; j++) { mat[i+j*nx] = drand48 (); /* position of cell (i,j) */ } } } /*******************************************************/ /* MAIN PROGRAM */ /*******************************************************/ int main () { int i=0, error=0, nx=NX, ny=NY; float diff; int size = nx * ny * sizeof(float); /* Matrix allocation */ float *mat_in1 = (float*) malloc(size); float *mat_in2 = (float*) malloc(size); float *mat_out = (float*) malloc(size); /* Matrix allocation on device */ float *mat_out_gpu, *mat_in1_gpu, *mat_in2_gpu; /* TO DO : do the allocation below, using cudaMalloc()*/ hipMalloc (&mat_in1_gpu, size); hipMalloc (&mat_in2_gpu, size); hipMalloc (&mat_out_gpu, size); /* Matrix initialization */ Init(mat_in1, nx, ny); Init(mat_in2, nx, ny); /* TO DO : write below the instructions to copy it to the device */ hipMemcpy(mat_in1_gpu, mat_in1, size, hipMemcpyHostToDevice); hipMemcpy(mat_in2_gpu, mat_in2, size, hipMemcpyHostToDevice); hipMemcpy(mat_out_gpu, mat_out, size, hipMemcpyHostToDevice); /* TO DO : complete the number of blocks below */ int numBlocks = (nx * ny + BLOCKSIZE-1) / BLOCKSIZE; /* TO DO : kernel invocation */ kernadd<<<numBlocks, BLOCKSIZE>>>(mat_out_gpu, mat_in1_gpu, mat_in2_gpu, nx, ny); hipDeviceSynchronize(); /* We now transfer back the matrix from the device to the host */ /* TO DO : write cudaMemcpy() instruction below */ hipMemcpy(mat_out, mat_out_gpu, size, hipMemcpyDeviceToHost); /* free memory */ hipFree(mat_out_gpu); hipFree(mat_in1_gpu); hipFree(mat_in2_gpu); /* We now check that the result is correct */ for (i=0; i< nx*ny; i++) { /* No need for a 2D loop, actually ! */ diff = mat_out[i] - (mat_in1[i]+mat_in2[i]); if (fabs(diff) > 0.0000001f) { error = 1; } } if (error) { printf("FAILED\n"); } else { printf("PASSED\n"); } free (mat_in1); free (mat_in2); free (mat_out); }
.text .file "addongpu.hip" .globl _Z22__device_stub__kernaddPfS_S_ii # -- Begin function _Z22__device_stub__kernaddPfS_S_ii .p2align 4, 0x90 .type _Z22__device_stub__kernaddPfS_S_ii,@function _Z22__device_stub__kernaddPfS_S_ii: # @_Z22__device_stub__kernaddPfS_S_ii .cfi_startproc # %bb.0: subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movq %rdx, 56(%rsp) movl %ecx, 4(%rsp) movl %r8d, (%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 56(%rsp), %rax movq %rax, 96(%rsp) leaq 4(%rsp), %rax movq %rax, 104(%rsp) movq %rsp, %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z7kernaddPfS_S_ii, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $136, %rsp .cfi_adjust_cfa_offset -136 retq .Lfunc_end0: .size _Z22__device_stub__kernaddPfS_S_ii, .Lfunc_end0-_Z22__device_stub__kernaddPfS_S_ii .cfi_endproc # -- End function .globl _Z4InitPfii # -- Begin function _Z4InitPfii .p2align 4, 0x90 .type _Z4InitPfii,@function _Z4InitPfii: # @_Z4InitPfii .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $24, %rsp .cfi_def_cfa_offset 80 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl %edx, 12(%rsp) # 4-byte Spill testl %esi, %esi jle .LBB1_6 # %bb.1: # %.preheader.lr.ph movq %rdi, %rbx movl %esi, %r15d movl 12(%rsp), %eax # 4-byte Reload movq %rax, 16(%rsp) # 8-byte Spill leaq (,%r15,4), %r13 xorl %ebp, %ebp jmp .LBB1_2 .p2align 4, 0x90 .LBB1_5: # %._crit_edge # in Loop: Header=BB1_2 Depth=1 incq %rbp addq $4, %rbx cmpq %r15, %rbp je .LBB1_6 .LBB1_2: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB1_4 Depth 2 cmpl $0, 12(%rsp) # 4-byte Folded Reload jle .LBB1_5 # %bb.3: # %.lr.ph.preheader # in Loop: Header=BB1_2 Depth=1 movq 16(%rsp), %r12 # 8-byte Reload movq %rbx, %r14 .p2align 4, 0x90 .LBB1_4: # %.lr.ph # Parent Loop BB1_2 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r14) addq %r13, %r14 decq %r12 jne .LBB1_4 jmp .LBB1_5 .LBB1_6: # %._crit_edge13 addq $24, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size _Z4InitPfii, .Lfunc_end1-_Z4InitPfii .cfi_endproc # -- End function .section .rodata.cst16,"aM",@progbits,16 .p2align 4, 0x0 # -- Begin function main .LCPI2_0: .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .section .rodata.cst4,"aM",@progbits,4 .p2align 2, 0x0 .LCPI2_1: .long 0x33d6bf95 # float 1.00000001E-7 .text .globl main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $152, %rsp .cfi_def_cfa_offset 208 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %rbx movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %r14 movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %r15 leaq 24(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc leaq 16(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc leaq 8(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc xorl %r12d, %r12d movq %rbx, %r13 .p2align 4, 0x90 .LBB2_1: # %.preheader.i # =>This Loop Header: Depth=1 # Child Loop BB2_2 Depth 2 xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_2: # %.lr.ph.i # Parent Loop BB2_1 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r13,%rbp) addq $800, %rbp # imm = 0x320 cmpq $80000, %rbp # imm = 0x13880 jne .LBB2_2 # %bb.3: # %._crit_edge.i # in Loop: Header=BB2_1 Depth=1 incq %r12 addq $4, %r13 cmpq $200, %r12 jne .LBB2_1 # %bb.4: # %.preheader.i47.preheader xorl %r12d, %r12d movq %r14, %r13 .p2align 4, 0x90 .LBB2_5: # %.preheader.i47 # =>This Loop Header: Depth=1 # Child Loop BB2_6 Depth 2 xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_6: # %.lr.ph.i49 # Parent Loop BB2_5 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r13,%rbp) addq $800, %rbp # imm = 0x320 cmpq $80000, %rbp # imm = 0x13880 jne .LBB2_6 # %bb.7: # %._crit_edge.i53 # in Loop: Header=BB2_5 Depth=1 incq %r12 addq $4, %r13 cmpq $200, %r12 jne .LBB2_5 # %bb.8: # %_Z4InitPfii.exit56 movq 24(%rsp), %rdi movl $1, %ebp movl $80000, %edx # imm = 0x13880 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy movq 16(%rsp), %rdi movl $80000, %edx # imm = 0x13880 movq %r14, %rsi movl $1, %ecx callq hipMemcpy movq 8(%rsp), %rdi movl $80000, %edx # imm = 0x13880 movq %r15, %rsi movl $1, %ecx callq hipMemcpy movabsq $4294967375, %rdi # imm = 0x10000004F leaq 177(%rdi), %rdx xorl %r12d, %r12d movl $1, %esi movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB2_10 # %bb.9: movq 8(%rsp), %rax movq 24(%rsp), %rcx movq 16(%rsp), %rdx movq %rax, 104(%rsp) movq %rcx, 96(%rsp) movq %rdx, 88(%rsp) movl $200, 36(%rsp) movl $100, 32(%rsp) leaq 104(%rsp), %rax movq %rax, 112(%rsp) leaq 96(%rsp), %rax movq %rax, 120(%rsp) leaq 88(%rsp), %rax movq %rax, 128(%rsp) leaq 36(%rsp), %rax movq %rax, 136(%rsp) leaq 32(%rsp), %rax movq %rax, 144(%rsp) leaq 72(%rsp), %rdi leaq 56(%rsp), %rsi leaq 48(%rsp), %rdx leaq 40(%rsp), %rcx callq __hipPopCallConfiguration movq 72(%rsp), %rsi movl 80(%rsp), %edx movq 56(%rsp), %rcx movl 64(%rsp), %r8d leaq 112(%rsp), %r9 movl $_Z7kernaddPfS_S_ii, %edi pushq 40(%rsp) .cfi_adjust_cfa_offset 8 pushq 56(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB2_10: callq hipDeviceSynchronize movq 8(%rsp), %rsi movl $80000, %edx # imm = 0x13880 movq %r15, %rdi movl $2, %ecx callq hipMemcpy movq 8(%rsp), %rdi callq hipFree movq 24(%rsp), %rdi callq hipFree movq 16(%rsp), %rdi callq hipFree movaps .LCPI2_0(%rip), %xmm0 # xmm0 = [NaN,NaN,NaN,NaN] movss .LCPI2_1(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero xorl %eax, %eax .p2align 4, 0x90 .LBB2_11: # =>This Inner Loop Header: Depth=1 movss (%r15,%rax,4), %xmm2 # xmm2 = mem[0],zero,zero,zero movss (%rbx,%rax,4), %xmm3 # xmm3 = mem[0],zero,zero,zero addss (%r14,%rax,4), %xmm3 subss %xmm3, %xmm2 andps %xmm0, %xmm2 ucomiss %xmm1, %xmm2 cmoval %ebp, %r12d incq %rax cmpq $20000, %rax # imm = 0x4E20 jne .LBB2_11 # %bb.12: testl %r12d, %r12d movl $.Lstr, %eax movl $.Lstr.1, %edi cmoveq %rax, %rdi callq puts@PLT movq %rbx, %rdi callq free movq %r14, %rdi callq free movq %r15, %rdi callq free xorl %eax, %eax addq $152, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z7kernaddPfS_S_ii, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z7kernaddPfS_S_ii,@object # @_Z7kernaddPfS_S_ii .section .rodata,"a",@progbits .globl _Z7kernaddPfS_S_ii .p2align 3, 0x0 _Z7kernaddPfS_S_ii: .quad _Z22__device_stub__kernaddPfS_S_ii .size _Z7kernaddPfS_S_ii, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z7kernaddPfS_S_ii" .size .L__unnamed_1, 19 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "PASSED" .size .Lstr, 7 .type .Lstr.1,@object # @str.1 .Lstr.1: .asciz "FAILED" .size .Lstr.1, 7 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z22__device_stub__kernaddPfS_S_ii .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z7kernaddPfS_S_ii .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z7kernaddPfS_S_ii .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ IABS R7, c[0x0][0x178] ; /* 0x00005e0000077a13 */ /* 0x000fe20000000000 */ /*0020*/ S2R R6, SR_CTAID.X ; /* 0x0000000000067919 */ /* 0x000e260000002500 */ /*0030*/ I2F.RP R0, R7 ; /* 0x0000000700007306 */ /* 0x000e620000209400 */ /*0040*/ S2R R5, SR_TID.X ; /* 0x0000000000057919 */ /* 0x000e2e0000002100 */ /*0050*/ MUFU.RCP R0, R0 ; /* 0x0000000000007308 */ /* 0x002e620000001000 */ /*0060*/ IMAD R6, R6, c[0x0][0x0], R5 ; /* 0x0000000006067a24 */ /* 0x001fe200078e0205 */ /*0070*/ IADD3 R2, R0, 0xffffffe, RZ ; /* 0x0ffffffe00027810 */ /* 0x002fc80007ffe0ff */ /*0080*/ IABS R0, R6 ; /* 0x0000000600007213 */ /* 0x000fe40000000000 */ /*0090*/ F2I.FTZ.U32.TRUNC.NTZ R3, R2 ; /* 0x0000000200037305 */ /* 0x000064000021f000 */ /*00a0*/ HFMA2.MMA R2, -RZ, RZ, 0, 0 ; /* 0x00000000ff027435 */ /* 0x001fe200000001ff */ /*00b0*/ IMAD.MOV R4, RZ, RZ, -R3 ; /* 0x000000ffff047224 */ /* 0x002fc800078e0a03 */ /*00c0*/ IMAD R5, R4, R7, RZ ; /* 0x0000000704057224 */ /* 0x000fca00078e02ff */ /*00d0*/ IMAD.HI.U32 R3, R3, R5, R2 ; /* 0x0000000503037227 */ /* 0x000fcc00078e0002 */ /*00e0*/ IMAD.HI.U32 R3, R3, R0, RZ ; /* 0x0000000003037227 */ /* 0x000fc800078e00ff */ /*00f0*/ IMAD.MOV R4, RZ, RZ, -R3 ; /* 0x000000ffff047224 */ /* 0x000fc800078e0a03 */ /*0100*/ IMAD R0, R7, R4, R0 ; /* 0x0000000407007224 */ /* 0x000fca00078e0200 */ /*0110*/ ISETP.GT.U32.AND P2, PT, R7, R0, PT ; /* 0x000000000700720c */ /* 0x000fda0003f44070 */ /*0120*/ @!P2 IADD3 R0, R0, -R7.reuse, RZ ; /* 0x800000070000a210 */ /* 0x080fe40007ffe0ff */ /*0130*/ @!P2 IADD3 R3, R3, 0x1, RZ ; /* 0x000000010303a810 */ /* 0x000fe40007ffe0ff */ /*0140*/ ISETP.GE.U32.AND P0, PT, R0, R7, PT ; /* 0x000000070000720c */ /* 0x000fe40003f06070 */ /*0150*/ LOP3.LUT R0, R6, c[0x0][0x178], RZ, 0x3c, !PT ; /* 0x00005e0006007a12 */ /* 0x000fe400078e3cff */ /*0160*/ ISETP.NE.AND P2, PT, RZ, c[0x0][0x178], PT ; /* 0x00005e00ff007a0c */ /* 0x000fe40003f45270 */ /*0170*/ ISETP.GE.AND P1, PT, R0, RZ, PT ; /* 0x000000ff0000720c */ /* 0x000fce0003f26270 */ /*0180*/ @P0 IADD3 R3, R3, 0x1, RZ ; /* 0x0000000103030810 */ /* 0x000fcc0007ffe0ff */ /*0190*/ @!P1 IMAD.MOV R3, RZ, RZ, -R3 ; /* 0x000000ffff039224 */ /* 0x000fe200078e0a03 */ /*01a0*/ @!P2 LOP3.LUT R3, RZ, c[0x0][0x178], RZ, 0x33, !PT ; /* 0x00005e00ff03aa12 */ /* 0x000fc800078e33ff */ /*01b0*/ IADD3 R5, -R3, RZ, RZ ; /* 0x000000ff03057210 */ /* 0x000fca0007ffe1ff */ /*01c0*/ IMAD R0, R5, c[0x0][0x178], R6 ; /* 0x00005e0005007a24 */ /* 0x000fca00078e0206 */ /*01d0*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x178], PT ; /* 0x00005e0000007a0c */ /* 0x000fc80003f06270 */ /*01e0*/ ISETP.GE.OR P0, PT, R3, c[0x0][0x17c], P0 ; /* 0x00005f0003007a0c */ /* 0x000fda0000706670 */ /*01f0*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0200*/ IMAD.MOV.U32 R7, RZ, RZ, 0x4 ; /* 0x00000004ff077424 */ /* 0x000fe200078e00ff */ /*0210*/ ULDC.64 UR4, c[0x0][0x118] ; /* 0x0000460000047ab9 */ /* 0x000fc60000000a00 */ /*0220*/ IMAD.WIDE R4, R6, R7, c[0x0][0x170] ; /* 0x00005c0006047625 */ /* 0x000fc800078e0207 */ /*0230*/ IMAD.WIDE R2, R6.reuse, R7.reuse, c[0x0][0x168] ; /* 0x00005a0006027625 */ /* 0x0c0fe400078e0207 */ /*0240*/ LDG.E R4, [R4.64] ; /* 0x0000000404047981 */ /* 0x000ea8000c1e1900 */ /*0250*/ LDG.E R3, [R2.64] ; /* 0x0000000402037981 */ /* 0x000ea2000c1e1900 */ /*0260*/ IMAD.WIDE R6, R6, R7, c[0x0][0x160] ; /* 0x0000580006067625 */ /* 0x000fc800078e0207 */ /*0270*/ FADD R9, R4, R3 ; /* 0x0000000304097221 */ /* 0x004fca0000000000 */ /*0280*/ STG.E [R6.64], R9 ; /* 0x0000000906007986 */ /* 0x000fe2000c101904 */ /*0290*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*02a0*/ BRA 0x2a0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*02b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*02f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0300*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0310*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0320*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0330*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0340*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0350*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0360*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0370*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z7kernaddPfS_S_ii .globl _Z7kernaddPfS_S_ii .p2align 8 .type _Z7kernaddPfS_S_ii,@function _Z7kernaddPfS_S_ii: s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b32 s4, s[0:1], 0x2c s_waitcnt lgkmcnt(0) s_ashr_i32 s5, s2, 31 s_and_b32 s4, s4, 0xffff s_add_i32 s6, s2, s5 s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_xor_b32 s6, s6, s5 v_cvt_f32_u32_e32 v1, s6 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1) v_rcp_iflag_f32_e32 v1, v1 s_waitcnt_depctr 0xfff v_mul_f32_e32 v1, 0x4f7ffffe, v1 v_cvt_u32_f32_e32 v3, v1 v_mad_u64_u32 v[1:2], null, s15, s4, v[0:1] s_sub_i32 s4, 0, s6 s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) v_mul_lo_u32 v0, s4, v3 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_ashrrev_i32_e32 v2, 31, v1 v_mul_hi_u32 v0, v3, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v4, v1, v2 v_xor_b32_e32 v4, v4, v2 v_xor_b32_e32 v2, s5, v2 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v0, v3, v0 v_mul_hi_u32 v0, v4, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_lo_u32 v3, v0, s6 v_sub_nc_u32_e32 v3, v4, v3 v_add_nc_u32_e32 v4, 1, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_subrev_nc_u32_e32 v5, s6, v3 v_cmp_le_u32_e32 vcc_lo, s6, v3 v_dual_cndmask_b32 v3, v3, v5 :: v_dual_cndmask_b32 v0, v0, v4 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_cmp_le_u32_e32 vcc_lo, s6, v3 v_add_nc_u32_e32 v4, 1, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_cndmask_b32_e32 v0, v0, v4, vcc_lo v_xor_b32_e32 v0, v0, v2 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_sub_nc_u32_e32 v0, v0, v2 v_mul_lo_u32 v2, v0, s2 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_sub_nc_u32_e32 v2, v1, v2 v_cmp_gt_i32_e32 vcc_lo, s2, v2 v_cmp_gt_i32_e64 s2, s3, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) s_and_b32 s2, s2, vcc_lo s_and_saveexec_b32 s3, s2 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b128 s[4:7], s[0:1], 0x0 s_load_b64 s[0:1], s[0:1], 0x10 v_ashrrev_i32_e32 v2, 31, v1 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[1:2] s_waitcnt lgkmcnt(0) v_add_co_u32 v2, vcc_lo, s6, v0 s_delay_alu instid0(VALU_DEP_2) v_add_co_ci_u32_e32 v3, vcc_lo, s7, v1, vcc_lo v_add_co_u32 v4, vcc_lo, s0, v0 v_add_co_ci_u32_e32 v5, vcc_lo, s1, v1, vcc_lo v_add_co_u32 v0, vcc_lo, s4, v0 global_load_b32 v2, v[2:3], off global_load_b32 v3, v[4:5], off v_add_co_ci_u32_e32 v1, vcc_lo, s5, v1, vcc_lo s_waitcnt vmcnt(0) v_add_f32_e32 v2, v2, v3 global_store_b32 v[0:1], v2, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z7kernaddPfS_S_ii .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 288 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 6 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z7kernaddPfS_S_ii, .Lfunc_end0-_Z7kernaddPfS_S_ii .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .offset: 24 .size: 4 .value_kind: by_value - .offset: 28 .size: 4 .value_kind: by_value - .offset: 32 .size: 4 .value_kind: hidden_block_count_x - .offset: 36 .size: 4 .value_kind: hidden_block_count_y - .offset: 40 .size: 4 .value_kind: hidden_block_count_z - .offset: 44 .size: 2 .value_kind: hidden_group_size_x - .offset: 46 .size: 2 .value_kind: hidden_group_size_y - .offset: 48 .size: 2 .value_kind: hidden_group_size_z - .offset: 50 .size: 2 .value_kind: hidden_remainder_x - .offset: 52 .size: 2 .value_kind: hidden_remainder_y - .offset: 54 .size: 2 .value_kind: hidden_remainder_z - .offset: 72 .size: 8 .value_kind: hidden_global_offset_x - .offset: 80 .size: 8 .value_kind: hidden_global_offset_y - .offset: 88 .size: 8 .value_kind: hidden_global_offset_z - .offset: 96 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 288 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z7kernaddPfS_S_ii .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z7kernaddPfS_S_ii.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 6 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_0003d2e8_00000000-6_addongpu.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2061: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2061: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z4InitPfii .type _Z4InitPfii, @function _Z4InitPfii: .LFB2057: .cfi_startproc endbr64 pushq %r15 .cfi_def_cfa_offset 16 .cfi_offset 15, -16 pushq %r14 .cfi_def_cfa_offset 24 .cfi_offset 14, -24 pushq %r13 .cfi_def_cfa_offset 32 .cfi_offset 13, -32 pushq %r12 .cfi_def_cfa_offset 40 .cfi_offset 12, -40 pushq %rbp .cfi_def_cfa_offset 48 .cfi_offset 6, -48 pushq %rbx .cfi_def_cfa_offset 56 .cfi_offset 3, -56 subq $24, %rsp .cfi_def_cfa_offset 80 movq %rdi, 8(%rsp) testl %esi, %esi jle .L3 movl %edx, %r12d movslq %esi, %r15 leaq 0(,%r15,4), %r13 movl $0, %r14d jmp .L5 .L6: call drand48@PLT cvtsd2ss %xmm0, %xmm0 movss %xmm0, 0(%rbp) addl $1, %ebx addq %r13, %rbp cmpl %ebx, %r12d jne .L6 .L8: addq $1, %r14 cmpq %r15, %r14 je .L3 .L5: movq 8(%rsp), %rax leaq (%rax,%r14,4), %rbp movl $0, %ebx testl %r12d, %r12d jg .L6 jmp .L8 .L3: addq $24, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %rbp .cfi_def_cfa_offset 40 popq %r12 .cfi_def_cfa_offset 32 popq %r13 .cfi_def_cfa_offset 24 popq %r14 .cfi_def_cfa_offset 16 popq %r15 .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2057: .size _Z4InitPfii, .-_Z4InitPfii .globl _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii .type _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii, @function _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii: .LFB2083: .cfi_startproc endbr64 subq $152, %rsp .cfi_def_cfa_offset 160 movq %rdi, 24(%rsp) movq %rsi, 16(%rsp) movq %rdx, 8(%rsp) movl %ecx, 4(%rsp) movl %r8d, (%rsp) movq %fs:40, %rax movq %rax, 136(%rsp) xorl %eax, %eax leaq 24(%rsp), %rax movq %rax, 96(%rsp) leaq 16(%rsp), %rax movq %rax, 104(%rsp) leaq 8(%rsp), %rax movq %rax, 112(%rsp) leaq 4(%rsp), %rax movq %rax, 120(%rsp) movq %rsp, %rax movq %rax, 128(%rsp) movl $1, 48(%rsp) movl $1, 52(%rsp) movl $1, 56(%rsp) movl $1, 60(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) leaq 40(%rsp), %rcx leaq 32(%rsp), %rdx leaq 60(%rsp), %rsi leaq 48(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L16 .L12: movq 136(%rsp), %rax subq %fs:40, %rax jne .L17 addq $152, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L16: .cfi_restore_state pushq 40(%rsp) .cfi_def_cfa_offset 168 pushq 40(%rsp) .cfi_def_cfa_offset 176 leaq 112(%rsp), %r9 movq 76(%rsp), %rcx movl 84(%rsp), %r8d movq 64(%rsp), %rsi movl 72(%rsp), %edx leaq _Z7kernaddPfS_S_ii(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 160 jmp .L12 .L17: call __stack_chk_fail@PLT .cfi_endproc .LFE2083: .size _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii, .-_Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii .globl _Z7kernaddPfS_S_ii .type _Z7kernaddPfS_S_ii, @function _Z7kernaddPfS_S_ii: .LFB2084: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 call _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2084: .size _Z7kernaddPfS_S_ii, .-_Z7kernaddPfS_S_ii .section .rodata.str1.1,"aMS",@progbits,1 .LC2: .string "FAILED\n" .LC3: .string "PASSED\n" .text .globl main .type main, @function main: .LFB2058: .cfi_startproc endbr64 pushq %r12 .cfi_def_cfa_offset 16 .cfi_offset 12, -16 pushq %rbp .cfi_def_cfa_offset 24 .cfi_offset 6, -24 pushq %rbx .cfi_def_cfa_offset 32 .cfi_offset 3, -32 subq $64, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 56(%rsp) xorl %eax, %eax movl $80000, %edi call malloc@PLT movq %rax, %r12 movl $80000, %edi call malloc@PLT movq %rax, %rbp movl $80000, %edi call malloc@PLT movq %rax, %rbx leaq 16(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT leaq 24(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT leaq 8(%rsp), %rdi movl $80000, %esi call cudaMalloc@PLT movl $100, %edx movl $200, %esi movq %r12, %rdi call _Z4InitPfii movl $100, %edx movl $200, %esi movq %rbp, %rdi call _Z4InitPfii movl $1, %ecx movl $80000, %edx movq %r12, %rsi movq 16(%rsp), %rdi call cudaMemcpy@PLT movl $1, %ecx movl $80000, %edx movq %rbp, %rsi movq 24(%rsp), %rdi call cudaMemcpy@PLT movl $1, %ecx movl $80000, %edx movq %rbx, %rsi movq 8(%rsp), %rdi call cudaMemcpy@PLT movl $256, 44(%rsp) movl $1, 48(%rsp) movl $79, 32(%rsp) movl $1, 36(%rsp) movl $0, %r9d movl $0, %r8d movq 44(%rsp), %rdx movl $1, %ecx movq 32(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L30 .L21: call cudaDeviceSynchronize@PLT movl $2, %ecx movl $80000, %edx movq 8(%rsp), %rsi movq %rbx, %rdi call cudaMemcpy@PLT movq 8(%rsp), %rdi call cudaFree@PLT movq 16(%rsp), %rdi call cudaFree@PLT movq 24(%rsp), %rdi call cudaFree@PLT movl $0, %eax movl $0, %edx movss .LC0(%rip), %xmm2 movl $1, %ecx .L23: movss (%r12,%rax), %xmm1 addss 0(%rbp,%rax), %xmm1 movss (%rbx,%rax), %xmm0 subss %xmm1, %xmm0 andps %xmm2, %xmm0 ucomiss .LC1(%rip), %xmm0 cmova %ecx, %edx addq $4, %rax cmpq $80000, %rax jne .L23 testl %edx, %edx je .L24 leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L25: movq %r12, %rdi call free@PLT movq %rbp, %rdi call free@PLT movq %rbx, %rdi call free@PLT movq 56(%rsp), %rax subq %fs:40, %rax jne .L31 movl $0, %eax addq $64, %rsp .cfi_remember_state .cfi_def_cfa_offset 32 popq %rbx .cfi_def_cfa_offset 24 popq %rbp .cfi_def_cfa_offset 16 popq %r12 .cfi_def_cfa_offset 8 ret .L30: .cfi_restore_state movl $100, %r8d movl $200, %ecx movq 24(%rsp), %rdx movq 16(%rsp), %rsi movq 8(%rsp), %rdi call _Z32__device_stub__Z7kernaddPfS_S_iiPfS_S_ii jmp .L21 .L24: leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L25 .L31: call __stack_chk_fail@PLT .cfi_endproc .LFE2058: .size main, .-main .section .rodata.str1.1 .LC4: .string "_Z7kernaddPfS_S_ii" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2086: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC4(%rip), %rdx movq %rdx, %rcx leaq _Z7kernaddPfS_S_ii(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2086: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .section .rodata.cst16,"aM",@progbits,16 .align 16 .LC0: .long 2147483647 .long 0 .long 0 .long 0 .section .rodata.cst4,"aM",@progbits,4 .align 4 .LC1: .long 869711765 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "addongpu.hip" .globl _Z22__device_stub__kernaddPfS_S_ii # -- Begin function _Z22__device_stub__kernaddPfS_S_ii .p2align 4, 0x90 .type _Z22__device_stub__kernaddPfS_S_ii,@function _Z22__device_stub__kernaddPfS_S_ii: # @_Z22__device_stub__kernaddPfS_S_ii .cfi_startproc # %bb.0: subq $120, %rsp .cfi_def_cfa_offset 128 movq %rdi, 72(%rsp) movq %rsi, 64(%rsp) movq %rdx, 56(%rsp) movl %ecx, 4(%rsp) movl %r8d, (%rsp) leaq 72(%rsp), %rax movq %rax, 80(%rsp) leaq 64(%rsp), %rax movq %rax, 88(%rsp) leaq 56(%rsp), %rax movq %rax, 96(%rsp) leaq 4(%rsp), %rax movq %rax, 104(%rsp) movq %rsp, %rax movq %rax, 112(%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d leaq 80(%rsp), %r9 movl $_Z7kernaddPfS_S_ii, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $136, %rsp .cfi_adjust_cfa_offset -136 retq .Lfunc_end0: .size _Z22__device_stub__kernaddPfS_S_ii, .Lfunc_end0-_Z22__device_stub__kernaddPfS_S_ii .cfi_endproc # -- End function .globl _Z4InitPfii # -- Begin function _Z4InitPfii .p2align 4, 0x90 .type _Z4InitPfii,@function _Z4InitPfii: # @_Z4InitPfii .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $24, %rsp .cfi_def_cfa_offset 80 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl %edx, 12(%rsp) # 4-byte Spill testl %esi, %esi jle .LBB1_6 # %bb.1: # %.preheader.lr.ph movq %rdi, %rbx movl %esi, %r15d movl 12(%rsp), %eax # 4-byte Reload movq %rax, 16(%rsp) # 8-byte Spill leaq (,%r15,4), %r13 xorl %ebp, %ebp jmp .LBB1_2 .p2align 4, 0x90 .LBB1_5: # %._crit_edge # in Loop: Header=BB1_2 Depth=1 incq %rbp addq $4, %rbx cmpq %r15, %rbp je .LBB1_6 .LBB1_2: # %.preheader # =>This Loop Header: Depth=1 # Child Loop BB1_4 Depth 2 cmpl $0, 12(%rsp) # 4-byte Folded Reload jle .LBB1_5 # %bb.3: # %.lr.ph.preheader # in Loop: Header=BB1_2 Depth=1 movq 16(%rsp), %r12 # 8-byte Reload movq %rbx, %r14 .p2align 4, 0x90 .LBB1_4: # %.lr.ph # Parent Loop BB1_2 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r14) addq %r13, %r14 decq %r12 jne .LBB1_4 jmp .LBB1_5 .LBB1_6: # %._crit_edge13 addq $24, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size _Z4InitPfii, .Lfunc_end1-_Z4InitPfii .cfi_endproc # -- End function .section .rodata.cst16,"aM",@progbits,16 .p2align 4, 0x0 # -- Begin function main .LCPI2_0: .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .long 0x7fffffff # float NaN .section .rodata.cst4,"aM",@progbits,4 .p2align 2, 0x0 .LCPI2_1: .long 0x33d6bf95 # float 1.00000001E-7 .text .globl main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $152, %rsp .cfi_def_cfa_offset 208 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %rbx movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %r14 movl $80000, %edi # imm = 0x13880 callq malloc movq %rax, %r15 leaq 24(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc leaq 16(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc leaq 8(%rsp), %rdi movl $80000, %esi # imm = 0x13880 callq hipMalloc xorl %r12d, %r12d movq %rbx, %r13 .p2align 4, 0x90 .LBB2_1: # %.preheader.i # =>This Loop Header: Depth=1 # Child Loop BB2_2 Depth 2 xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_2: # %.lr.ph.i # Parent Loop BB2_1 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r13,%rbp) addq $800, %rbp # imm = 0x320 cmpq $80000, %rbp # imm = 0x13880 jne .LBB2_2 # %bb.3: # %._crit_edge.i # in Loop: Header=BB2_1 Depth=1 incq %r12 addq $4, %r13 cmpq $200, %r12 jne .LBB2_1 # %bb.4: # %.preheader.i47.preheader xorl %r12d, %r12d movq %r14, %r13 .p2align 4, 0x90 .LBB2_5: # %.preheader.i47 # =>This Loop Header: Depth=1 # Child Loop BB2_6 Depth 2 xorl %ebp, %ebp .p2align 4, 0x90 .LBB2_6: # %.lr.ph.i49 # Parent Loop BB2_5 Depth=1 # => This Inner Loop Header: Depth=2 callq drand48 cvtsd2ss %xmm0, %xmm0 movss %xmm0, (%r13,%rbp) addq $800, %rbp # imm = 0x320 cmpq $80000, %rbp # imm = 0x13880 jne .LBB2_6 # %bb.7: # %._crit_edge.i53 # in Loop: Header=BB2_5 Depth=1 incq %r12 addq $4, %r13 cmpq $200, %r12 jne .LBB2_5 # %bb.8: # %_Z4InitPfii.exit56 movq 24(%rsp), %rdi movl $1, %ebp movl $80000, %edx # imm = 0x13880 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy movq 16(%rsp), %rdi movl $80000, %edx # imm = 0x13880 movq %r14, %rsi movl $1, %ecx callq hipMemcpy movq 8(%rsp), %rdi movl $80000, %edx # imm = 0x13880 movq %r15, %rsi movl $1, %ecx callq hipMemcpy movabsq $4294967375, %rdi # imm = 0x10000004F leaq 177(%rdi), %rdx xorl %r12d, %r12d movl $1, %esi movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB2_10 # %bb.9: movq 8(%rsp), %rax movq 24(%rsp), %rcx movq 16(%rsp), %rdx movq %rax, 104(%rsp) movq %rcx, 96(%rsp) movq %rdx, 88(%rsp) movl $200, 36(%rsp) movl $100, 32(%rsp) leaq 104(%rsp), %rax movq %rax, 112(%rsp) leaq 96(%rsp), %rax movq %rax, 120(%rsp) leaq 88(%rsp), %rax movq %rax, 128(%rsp) leaq 36(%rsp), %rax movq %rax, 136(%rsp) leaq 32(%rsp), %rax movq %rax, 144(%rsp) leaq 72(%rsp), %rdi leaq 56(%rsp), %rsi leaq 48(%rsp), %rdx leaq 40(%rsp), %rcx callq __hipPopCallConfiguration movq 72(%rsp), %rsi movl 80(%rsp), %edx movq 56(%rsp), %rcx movl 64(%rsp), %r8d leaq 112(%rsp), %r9 movl $_Z7kernaddPfS_S_ii, %edi pushq 40(%rsp) .cfi_adjust_cfa_offset 8 pushq 56(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB2_10: callq hipDeviceSynchronize movq 8(%rsp), %rsi movl $80000, %edx # imm = 0x13880 movq %r15, %rdi movl $2, %ecx callq hipMemcpy movq 8(%rsp), %rdi callq hipFree movq 24(%rsp), %rdi callq hipFree movq 16(%rsp), %rdi callq hipFree movaps .LCPI2_0(%rip), %xmm0 # xmm0 = [NaN,NaN,NaN,NaN] movss .LCPI2_1(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero xorl %eax, %eax .p2align 4, 0x90 .LBB2_11: # =>This Inner Loop Header: Depth=1 movss (%r15,%rax,4), %xmm2 # xmm2 = mem[0],zero,zero,zero movss (%rbx,%rax,4), %xmm3 # xmm3 = mem[0],zero,zero,zero addss (%r14,%rax,4), %xmm3 subss %xmm3, %xmm2 andps %xmm0, %xmm2 ucomiss %xmm1, %xmm2 cmoval %ebp, %r12d incq %rax cmpq $20000, %rax # imm = 0x4E20 jne .LBB2_11 # %bb.12: testl %r12d, %r12d movl $.Lstr, %eax movl $.Lstr.1, %edi cmoveq %rax, %rdi callq puts@PLT movq %rbx, %rdi callq free movq %r14, %rdi callq free movq %r15, %rdi callq free xorl %eax, %eax addq $152, %rsp .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end2: .size main, .Lfunc_end2-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB3_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB3_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z7kernaddPfS_S_ii, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end3: .size __hip_module_ctor, .Lfunc_end3-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB4_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB4_2: retq .Lfunc_end4: .size __hip_module_dtor, .Lfunc_end4-__hip_module_dtor .cfi_endproc # -- End function .type _Z7kernaddPfS_S_ii,@object # @_Z7kernaddPfS_S_ii .section .rodata,"a",@progbits .globl _Z7kernaddPfS_S_ii .p2align 3, 0x0 _Z7kernaddPfS_S_ii: .quad _Z22__device_stub__kernaddPfS_S_ii .size _Z7kernaddPfS_S_ii, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z7kernaddPfS_S_ii" .size .L__unnamed_1, 19 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "PASSED" .size .Lstr, 7 .type .Lstr.1,@object # @str.1 .Lstr.1: .asciz "FAILED" .size .Lstr.1, 7 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z22__device_stub__kernaddPfS_S_ii .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z7kernaddPfS_S_ii .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
#include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
code for sm_80 Function : _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R0, SR_TID.X ; /* 0x0000000000007919 */ /* 0x000e280000002100 */ /*0020*/ S2R R3, SR_CTAID.X ; /* 0x0000000000037919 */ /* 0x000e620000002500 */ /*0030*/ LOP3.LUT R0, R0, 0x1f, RZ, 0xc0, !PT ; /* 0x0000001f00007812 */ /* 0x001fca00078ec0ff */ /*0040*/ IMAD R0, R3, c[0x0][0x0], R0 ; /* 0x0000000003007a24 */ /* 0x002fca00078e0200 */ /*0050*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x180], PT ; /* 0x0000600000007a0c */ /* 0x000fda0003f06270 */ /*0060*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0070*/ IMAD.MOV.U32 R5, RZ, RZ, 0x4 ; /* 0x00000004ff057424 */ /* 0x000fe200078e00ff */ /*0080*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fc60000000a00 */ /*0090*/ IMAD.WIDE R2, R0, R5, c[0x0][0x178] ; /* 0x00005e0000027625 */ /* 0x000fca00078e0205 */ /*00a0*/ LDG.E R0, [R2.64] ; /* 0x0000000602007981 */ /* 0x000ea2000c1e1900 */ /*00b0*/ ULDC UR4, c[0x0][0x194] ; /* 0x0000650000047ab9 */ /* 0x000fe40000000800 */ /*00c0*/ ULDC UR5, c[0x0][0x184] ; /* 0x0000610000057ab9 */ /* 0x000fe40000000800 */ /*00d0*/ UIMAD UR4, UR4, UR5, URZ ; /* 0x00000005040472a4 */ /* 0x000fd2000f8e023f */ /*00e0*/ I2F.U32 R7, UR4 ; /* 0x0000000400077d06 */ /* 0x000e220008201000 */ /*00f0*/ IMAD.WIDE R4, R0, R5, c[0x0][0x160] ; /* 0x0000580000047625 */ /* 0x004fcc00078e0205 */ /*0100*/ LDG.E R4, [R4.64] ; /* 0x0000000604047981 */ /* 0x000ea2000c1e1900 */ /*0110*/ MUFU.RCP R6, R7 ; /* 0x0000000700067308 */ /* 0x001e220000001000 */ /*0120*/ BSSY B0, 0x1e0 ; /* 0x000000b000007945 */ /* 0x000fe20003800000 */ /*0130*/ FFMA R9, -R7, R6, 1 ; /* 0x3f80000007097423 */ /* 0x001fc80000000106 */ /*0140*/ FFMA R9, R6, R9, R6 ; /* 0x0000000906097223 */ /* 0x000fe40000000006 */ /*0150*/ FCHK P0, -R4, R7 ; /* 0x0000000704007302 */ /* 0x004e240000000100 */ /*0160*/ FFMA R2, -R4, R9, RZ ; /* 0x0000000904027223 */ /* 0x000fc800000001ff */ /*0170*/ FFMA R3, -R7, R2, -R4 ; /* 0x0000000207037223 */ /* 0x000fc80000000904 */ /*0180*/ FFMA R9, R9, R3, R2 ; /* 0x0000000309097223 */ /* 0x000fe20000000002 */ /*0190*/ @!P0 BRA 0x1d0 ; /* 0x0000003000008947 */ /* 0x001fea0003800000 */ /*01a0*/ FADD R5, -R4, -RZ ; /* 0x800000ff04057221 */ /* 0x000fe20000000100 */ /*01b0*/ MOV R2, 0x1d0 ; /* 0x000001d000027802 */ /* 0x000fe40000000f00 */ /*01c0*/ CALL.REL.NOINC 0x220 ; /* 0x0000005000007944 */ /* 0x000fea0003c00000 */ /*01d0*/ BSYNC B0 ; /* 0x0000000000007941 */ /* 0x000fea0003800000 */ /*01e0*/ IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff037424 */ /* 0x000fc800078e00ff */ /*01f0*/ IMAD.WIDE R2, R0, R3, c[0x0][0x168] ; /* 0x00005a0000027625 */ /* 0x000fca00078e0203 */ /*0200*/ STG.E [R2.64], R9 ; /* 0x0000000902007986 */ /* 0x000fe2000c101906 */ /*0210*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0220*/ SHF.R.U32.HI R4, RZ, 0x17, R7 ; /* 0x00000017ff047819 */ /* 0x000fe20000011607 */ /*0230*/ BSSY B1, 0x870 ; /* 0x0000063000017945 */ /* 0x000fe20003800000 */ /*0240*/ SHF.R.U32.HI R3, RZ, 0x17, R5 ; /* 0x00000017ff037819 */ /* 0x000fe20000011605 */ /*0250*/ IMAD.MOV.U32 R6, RZ, RZ, R7 ; /* 0x000000ffff067224 */ /* 0x000fe200078e0007 */ /*0260*/ LOP3.LUT R4, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04047812 */ /* 0x000fe400078ec0ff */ /*0270*/ LOP3.LUT R3, R3, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff03037812 */ /* 0x000fe400078ec0ff */ /*0280*/ IADD3 R10, R4, -0x1, RZ ; /* 0xffffffff040a7810 */ /* 0x000fe40007ffe0ff */ /*0290*/ IADD3 R9, R3, -0x1, RZ ; /* 0xffffffff03097810 */ /* 0x000fc40007ffe0ff */ /*02a0*/ ISETP.GT.U32.AND P0, PT, R10, 0xfd, PT ; /* 0x000000fd0a00780c */ /* 0x000fc80003f04070 */ /*02b0*/ ISETP.GT.U32.OR P0, PT, R9, 0xfd, P0 ; /* 0x000000fd0900780c */ /* 0x000fda0000704470 */ /*02c0*/ @!P0 IMAD.MOV.U32 R8, RZ, RZ, RZ ; /* 0x000000ffff088224 */ /* 0x000fe200078e00ff */ /*02d0*/ @!P0 BRA 0x450 ; /* 0x0000017000008947 */ /* 0x000fea0003800000 */ /*02e0*/ FSETP.GTU.FTZ.AND P0, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fe40003f1c200 */ /*02f0*/ FSETP.GTU.FTZ.AND P1, PT, |R7|, +INF , PT ; /* 0x7f8000000700780b */ /* 0x000fc80003f3c200 */ /*0300*/ PLOP3.LUT P0, PT, P0, P1, PT, 0xa8, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000703570 */ /*0310*/ @P0 BRA 0x850 ; /* 0x0000053000000947 */ /* 0x000fea0003800000 */ /*0320*/ LOP3.LUT P0, RZ, R6, 0x7fffffff, R5, 0xc8, !PT ; /* 0x7fffffff06ff7812 */ /* 0x000fda000780c805 */ /*0330*/ @!P0 BRA 0x830 ; /* 0x000004f000008947 */ /* 0x000fea0003800000 */ /*0340*/ FSETP.NEU.FTZ.AND P2, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fe40003f5d200 */ /*0350*/ FSETP.NEU.FTZ.AND P1, PT, |R7|, +INF , PT ; /* 0x7f8000000700780b */ /* 0x000fe40003f3d200 */ /*0360*/ FSETP.NEU.FTZ.AND P0, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fd60003f1d200 */ /*0370*/ @!P1 BRA !P2, 0x830 ; /* 0x000004b000009947 */ /* 0x000fea0005000000 */ /*0380*/ LOP3.LUT P2, RZ, R5, 0x7fffffff, RZ, 0xc0, !PT ; /* 0x7fffffff05ff7812 */ /* 0x000fc8000784c0ff */ /*0390*/ PLOP3.LUT P1, PT, P1, P2, PT, 0x2a, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000f24572 */ /*03a0*/ @P1 BRA 0x810 ; /* 0x0000046000001947 */ /* 0x000fea0003800000 */ /*03b0*/ LOP3.LUT P1, RZ, R6, 0x7fffffff, RZ, 0xc0, !PT ; /* 0x7fffffff06ff7812 */ /* 0x000fc8000782c0ff */ /*03c0*/ PLOP3.LUT P0, PT, P0, P1, PT, 0x2a, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000702572 */ /*03d0*/ @P0 BRA 0x7e0 ; /* 0x0000040000000947 */ /* 0x000fea0003800000 */ /*03e0*/ ISETP.GE.AND P0, PT, R9, RZ, PT ; /* 0x000000ff0900720c */ /* 0x000fe40003f06270 */ /*03f0*/ ISETP.GE.AND P1, PT, R10, RZ, PT ; /* 0x000000ff0a00720c */ /* 0x000fd60003f26270 */ /*0400*/ @P0 IMAD.MOV.U32 R8, RZ, RZ, RZ ; /* 0x000000ffff080224 */ /* 0x000fe400078e00ff */ /*0410*/ @!P0 IMAD.MOV.U32 R8, RZ, RZ, -0x40 ; /* 0xffffffc0ff088424 */ /* 0x000fe400078e00ff */ /*0420*/ @!P0 FFMA R5, R5, 1.84467440737095516160e+19, RZ ; /* 0x5f80000005058823 */ /* 0x000fe400000000ff */ /*0430*/ @!P1 FFMA R6, R7, 1.84467440737095516160e+19, RZ ; /* 0x5f80000007069823 */ /* 0x000fe200000000ff */ /*0440*/ @!P1 IADD3 R8, R8, 0x40, RZ ; /* 0x0000004008089810 */ /* 0x000fe40007ffe0ff */ /*0450*/ LEA R7, R4, 0xc0800000, 0x17 ; /* 0xc080000004077811 */ /* 0x000fe200078eb8ff */ /*0460*/ BSSY B2, 0x7d0 ; /* 0x0000036000027945 */ /* 0x000fe80003800000 */ /*0470*/ IMAD.IADD R7, R6, 0x1, -R7 ; /* 0x0000000106077824 */ /* 0x000fe200078e0a07 */ /*0480*/ IADD3 R6, R3, -0x7f, RZ ; /* 0xffffff8103067810 */ /* 0x000fc60007ffe0ff */ /*0490*/ MUFU.RCP R9, R7 ; /* 0x0000000700097308 */ /* 0x000e220000001000 */ /*04a0*/ FADD.FTZ R10, -R7, -RZ ; /* 0x800000ff070a7221 */ /* 0x000fe40000010100 */ /*04b0*/ IMAD R5, R6, -0x800000, R5 ; /* 0xff80000006057824 */ /* 0x000fe400078e0205 */ /*04c0*/ FFMA R12, R9, R10, 1 ; /* 0x3f800000090c7423 */ /* 0x001fc8000000000a */ /*04d0*/ FFMA R14, R9, R12, R9 ; /* 0x0000000c090e7223 */ /* 0x000fc80000000009 */ /*04e0*/ FFMA R3, R5, R14, RZ ; /* 0x0000000e05037223 */ /* 0x000fc800000000ff */ /*04f0*/ FFMA R12, R10, R3, R5 ; /* 0x000000030a0c7223 */ /* 0x000fc80000000005 */ /*0500*/ FFMA R9, R14, R12, R3 ; /* 0x0000000c0e097223 */ /* 0x000fc80000000003 */ /*0510*/ FFMA R10, R10, R9, R5 ; /* 0x000000090a0a7223 */ /* 0x000fe20000000005 */ /*0520*/ IADD3 R5, R6, 0x7f, -R4 ; /* 0x0000007f06057810 */ /* 0x000fc60007ffe804 */ /*0530*/ FFMA R3, R14, R10, R9 ; /* 0x0000000a0e037223 */ /* 0x000fe40000000009 */ /*0540*/ IMAD.IADD R5, R5, 0x1, R8 ; /* 0x0000000105057824 */ /* 0x000fc600078e0208 */ /*0550*/ SHF.R.U32.HI R4, RZ, 0x17, R3 ; /* 0x00000017ff047819 */ /* 0x000fc80000011603 */ /*0560*/ LOP3.LUT R4, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04047812 */ /* 0x000fca00078ec0ff */ /*0570*/ IMAD.IADD R8, R4, 0x1, R5 ; /* 0x0000000104087824 */ /* 0x000fca00078e0205 */ /*0580*/ IADD3 R4, R8, -0x1, RZ ; /* 0xffffffff08047810 */ /* 0x000fc80007ffe0ff */ /*0590*/ ISETP.GE.U32.AND P0, PT, R4, 0xfe, PT ; /* 0x000000fe0400780c */ /* 0x000fda0003f06070 */ /*05a0*/ @!P0 BRA 0x7b0 ; /* 0x0000020000008947 */ /* 0x000fea0003800000 */ /*05b0*/ ISETP.GT.AND P0, PT, R8, 0xfe, PT ; /* 0x000000fe0800780c */ /* 0x000fda0003f04270 */ /*05c0*/ @P0 BRA 0x780 ; /* 0x000001b000000947 */ /* 0x000fea0003800000 */ /*05d0*/ ISETP.GE.AND P0, PT, R8, 0x1, PT ; /* 0x000000010800780c */ /* 0x000fda0003f06270 */ /*05e0*/ @P0 BRA 0x7c0 ; /* 0x000001d000000947 */ /* 0x000fea0003800000 */ /*05f0*/ ISETP.GE.AND P0, PT, R8, -0x18, PT ; /* 0xffffffe80800780c */ /* 0x000fe40003f06270 */ /*0600*/ LOP3.LUT R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fd600078ec0ff */ /*0610*/ @!P0 BRA 0x7c0 ; /* 0x000001a000008947 */ /* 0x000fea0003800000 */ /*0620*/ FFMA.RZ R4, R14, R10.reuse, R9.reuse ; /* 0x0000000a0e047223 */ /* 0x180fe2000000c009 */ /*0630*/ IADD3 R7, R8, 0x20, RZ ; /* 0x0000002008077810 */ /* 0x000fe20007ffe0ff */ /*0640*/ FFMA.RM R5, R14, R10.reuse, R9.reuse ; /* 0x0000000a0e057223 */ /* 0x180fe20000004009 */ /*0650*/ ISETP.NE.AND P2, PT, R8, RZ, PT ; /* 0x000000ff0800720c */ /* 0x000fe40003f45270 */ /*0660*/ LOP3.LUT R6, R4, 0x7fffff, RZ, 0xc0, !PT ; /* 0x007fffff04067812 */ /* 0x000fe200078ec0ff */ /*0670*/ FFMA.RP R4, R14, R10, R9 ; /* 0x0000000a0e047223 */ /* 0x000fe20000008009 */ /*0680*/ ISETP.NE.AND P1, PT, R8, RZ, PT ; /* 0x000000ff0800720c */ /* 0x000fe20003f25270 */ /*0690*/ IMAD.MOV R8, RZ, RZ, -R8 ; /* 0x000000ffff087224 */ /* 0x000fe200078e0a08 */ /*06a0*/ LOP3.LUT R6, R6, 0x800000, RZ, 0xfc, !PT ; /* 0x0080000006067812 */ /* 0x000fe400078efcff */ /*06b0*/ FSETP.NEU.FTZ.AND P0, PT, R4, R5, PT ; /* 0x000000050400720b */ /* 0x000fc40003f1d000 */ /*06c0*/ SHF.L.U32 R7, R6, R7, RZ ; /* 0x0000000706077219 */ /* 0x000fe400000006ff */ /*06d0*/ SEL R5, R8, RZ, P2 ; /* 0x000000ff08057207 */ /* 0x000fe40001000000 */ /*06e0*/ ISETP.NE.AND P1, PT, R7, RZ, P1 ; /* 0x000000ff0700720c */ /* 0x000fe40000f25270 */ /*06f0*/ SHF.R.U32.HI R5, RZ, R5, R6 ; /* 0x00000005ff057219 */ /* 0x000fe40000011606 */ /*0700*/ PLOP3.LUT P0, PT, P0, P1, PT, 0xa8, 0x0 ; /* 0x000000000000781c */ /* 0x000fe40000703570 */ /*0710*/ SHF.R.U32.HI R7, RZ, 0x1, R5 ; /* 0x00000001ff077819 */ /* 0x000fc40000011605 */ /*0720*/ SEL R4, RZ, 0x1, !P0 ; /* 0x00000001ff047807 */ /* 0x000fc80004000000 */ /*0730*/ LOP3.LUT R4, R4, 0x1, R7, 0xf8, !PT ; /* 0x0000000104047812 */ /* 0x000fc800078ef807 */ /*0740*/ LOP3.LUT R4, R4, R5, RZ, 0xc0, !PT ; /* 0x0000000504047212 */ /* 0x000fca00078ec0ff */ /*0750*/ IMAD.IADD R4, R7, 0x1, R4 ; /* 0x0000000107047824 */ /* 0x000fca00078e0204 */ /*0760*/ LOP3.LUT R3, R4, R3, RZ, 0xfc, !PT ; /* 0x0000000304037212 */ /* 0x000fe200078efcff */ /*0770*/ BRA 0x7c0 ; /* 0x0000004000007947 */ /* 0x000fea0003800000 */ /*0780*/ LOP3.LUT R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fc800078ec0ff */ /*0790*/ LOP3.LUT R3, R3, 0x7f800000, RZ, 0xfc, !PT ; /* 0x7f80000003037812 */ /* 0x000fe200078efcff */ /*07a0*/ BRA 0x7c0 ; /* 0x0000001000007947 */ /* 0x000fea0003800000 */ /*07b0*/ IMAD R3, R5, 0x800000, R3 ; /* 0x0080000005037824 */ /* 0x000fe400078e0203 */ /*07c0*/ BSYNC B2 ; /* 0x0000000000027941 */ /* 0x000fea0003800000 */ /*07d0*/ BRA 0x860 ; /* 0x0000008000007947 */ /* 0x000fea0003800000 */ /*07e0*/ LOP3.LUT R3, R6, 0x80000000, R5, 0x48, !PT ; /* 0x8000000006037812 */ /* 0x000fc800078e4805 */ /*07f0*/ LOP3.LUT R3, R3, 0x7f800000, RZ, 0xfc, !PT ; /* 0x7f80000003037812 */ /* 0x000fe200078efcff */ /*0800*/ BRA 0x860 ; /* 0x0000005000007947 */ /* 0x000fea0003800000 */ /*0810*/ LOP3.LUT R3, R6, 0x80000000, R5, 0x48, !PT ; /* 0x8000000006037812 */ /* 0x000fe200078e4805 */ /*0820*/ BRA 0x860 ; /* 0x0000003000007947 */ /* 0x000fea0003800000 */ /*0830*/ MUFU.RSQ R3, -QNAN ; /* 0xffc0000000037908 */ /* 0x000e220000001400 */ /*0840*/ BRA 0x860 ; /* 0x0000001000007947 */ /* 0x000fea0003800000 */ /*0850*/ FADD.FTZ R3, R5, R7 ; /* 0x0000000705037221 */ /* 0x000fe40000010000 */ /*0860*/ BSYNC B1 ; /* 0x0000000000017941 */ /* 0x000fea0003800000 */ /*0870*/ IMAD.MOV.U32 R9, RZ, RZ, R3 ; /* 0x000000ffff097224 */ /* 0x001fe400078e0003 */ /*0880*/ IMAD.MOV.U32 R3, RZ, RZ, 0x0 ; /* 0x00000000ff037424 */ /* 0x000fc800078e00ff */ /*0890*/ RET.REL.NODEC R2 0x0 ; /* 0xfffff76002007950 */ /* 0x000fea0003c3ffff */ /*08a0*/ BRA 0x8a0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*08b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0900*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0910*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0920*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0930*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0940*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0950*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0960*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0970*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
#include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
.file "tmpxft_0019613a_00000000-6_cudaSAnchorBackPropagateSSD_NegSamples_kernel.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj .type _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj, @function _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj: .LFB2051: .cfi_startproc endbr64 subq $216, %rsp .cfi_def_cfa_offset 224 movq %rdi, 40(%rsp) movq %rsi, 32(%rsp) movq %rdx, 24(%rsp) movq %rcx, 16(%rsp) movl %r8d, 12(%rsp) movl %r9d, 8(%rsp) movq %fs:40, %rax movq %rax, 200(%rsp) xorl %eax, %eax leaq 40(%rsp), %rax movq %rax, 112(%rsp) leaq 32(%rsp), %rax movq %rax, 120(%rsp) leaq 24(%rsp), %rax movq %rax, 128(%rsp) leaq 16(%rsp), %rax movq %rax, 136(%rsp) leaq 12(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) leaq 224(%rsp), %rax movq %rax, 160(%rsp) leaq 232(%rsp), %rax movq %rax, 168(%rsp) leaq 240(%rsp), %rax movq %rax, 176(%rsp) leaq 248(%rsp), %rax movq %rax, 184(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 200(%rsp), %rax subq %fs:40, %rax jne .L8 addq $216, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 232 pushq 56(%rsp) .cfi_def_cfa_offset 240 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 224 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj, .-_Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, @function _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 24 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 32 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 40 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 48 call _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj addq $40, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .-_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
#include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
#include <hip/hip_runtime.h> #include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
#include <hip/hip_runtime.h> #include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 8 .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@function _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: s_clause 0x1 s_load_b32 s2, s[0:1], 0x44 s_load_b32 s3, s[0:1], 0x20 v_and_b32_e32 v2, 31, v0 s_waitcnt lgkmcnt(0) s_and_b32 s2, s2, 0xffff s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[0:1], null, s15, s2, v[2:3] s_mov_b32 s2, exec_lo s_delay_alu instid0(VALU_DEP_1) v_cmpx_gt_i32_e64 s3, v0 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b128 s[4:7], s[0:1], 0x0 v_ashrrev_i32_e32 v1, 31, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[0:1] s_waitcnt lgkmcnt(0) v_add_co_u32 v0, vcc_lo, s2, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1) v_add_co_ci_u32_e32 v1, vcc_lo, s3, v1, vcc_lo global_load_b32 v0, v[0:1], off s_waitcnt vmcnt(0) v_ashrrev_i32_e32 v1, 31, v0 v_lshlrev_b64 v[0:1], 2, v[0:1] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v2, vcc_lo, s4, v0 v_add_co_ci_u32_e32 v3, vcc_lo, s5, v1, vcc_lo global_load_b32 v2, v[2:3], off s_clause 0x1 s_load_b32 s2, s[0:1], 0x24 s_load_b32 s0, s[0:1], 0x34 s_waitcnt lgkmcnt(0) s_mul_i32 s0, s0, s2 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_cvt_f32_u32_e32 v3, s0 s_waitcnt vmcnt(0) v_div_scale_f32 v4, null, v3, v3, -v2 v_div_scale_f32 v7, vcc_lo, -v2, v3, -v2 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) v_rcp_f32_e32 v5, v4 s_waitcnt_depctr 0xfff v_fma_f32 v6, -v4, v5, 1.0 v_fmac_f32_e32 v5, v6, v5 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_f32_e32 v6, v7, v5 v_fma_f32 v8, -v4, v6, v7 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmac_f32_e32 v6, v8, v5 v_fma_f32 v4, -v4, v6, v7 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) v_div_fmas_f32 v4, v4, v5, v6 v_add_co_u32 v0, vcc_lo, s6, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s7, v1, vcc_lo v_div_fixup_f32 v2, v4, v3, -v2 global_store_b32 v[0:1], v2, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 312 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 9 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .Lfunc_end0-_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .offset: 32 .size: 4 .value_kind: by_value - .offset: 36 .size: 4 .value_kind: by_value - .offset: 40 .size: 4 .value_kind: by_value - .offset: 44 .size: 4 .value_kind: by_value - .offset: 48 .size: 4 .value_kind: by_value - .offset: 52 .size: 4 .value_kind: by_value - .offset: 56 .size: 4 .value_kind: hidden_block_count_x - .offset: 60 .size: 4 .value_kind: hidden_block_count_y - .offset: 64 .size: 4 .value_kind: hidden_block_count_z - .offset: 68 .size: 2 .value_kind: hidden_group_size_x - .offset: 70 .size: 2 .value_kind: hidden_group_size_y - .offset: 72 .size: 2 .value_kind: hidden_group_size_z - .offset: 74 .size: 2 .value_kind: hidden_remainder_x - .offset: 76 .size: 2 .value_kind: hidden_remainder_y - .offset: 78 .size: 2 .value_kind: hidden_remainder_z - .offset: 96 .size: 8 .value_kind: hidden_global_offset_x - .offset: 104 .size: 8 .value_kind: hidden_global_offset_y - .offset: 112 .size: 8 .value_kind: hidden_global_offset_z - .offset: 120 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 312 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 9 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
#include <hip/hip_runtime.h> #include "includes.h" __global__ void cudaSAnchorBackPropagateSSD_NegSamples_kernel(const float* inputCls, float* diffOutputsCls, const float* confSamples, const int* keySamples, const int nbSamples, const int nbPositive, const unsigned int nbAnchors, const unsigned int outputsHeight, const unsigned int outputsWidth, const unsigned int batchSize) { const int index = (threadIdx.x & 0x1f) + blockIdx.x*blockDim.x; if(index < nbSamples) { const int indexSamples = keySamples[index]; const float error = inputCls[indexSamples]; //diffOutputsCls[indexSamples] = -inputCls[index] / (nbPositive * batchSize); //printf("error[%d]: %f\n", indexSamples, error); diffOutputsCls[indexSamples] = -error / (nbPositive * batchSize); } }
.text .file "cudaSAnchorBackPropagateSSD_NegSamples_kernel.hip" .globl _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj # -- Begin function _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 4, 0x90 .type _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@function _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: # @_Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .cfi_startproc # %bb.0: subq $184, %rsp .cfi_def_cfa_offset 192 movq %rdi, 88(%rsp) movq %rsi, 80(%rsp) movq %rdx, 72(%rsp) movq %rcx, 64(%rsp) movl %r8d, 12(%rsp) movl %r9d, 8(%rsp) leaq 88(%rsp), %rax movq %rax, 96(%rsp) leaq 80(%rsp), %rax movq %rax, 104(%rsp) leaq 72(%rsp), %rax movq %rax, 112(%rsp) leaq 64(%rsp), %rax movq %rax, 120(%rsp) leaq 12(%rsp), %rax movq %rax, 128(%rsp) leaq 8(%rsp), %rax movq %rax, 136(%rsp) leaq 192(%rsp), %rax movq %rax, 144(%rsp) leaq 200(%rsp), %rax movq %rax, 152(%rsp) leaq 208(%rsp), %rax movq %rax, 160(%rsp) leaq 216(%rsp), %rax movq %rax, 168(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $200, %rsp .cfi_adjust_cfa_offset -200 retq .Lfunc_end0: .size _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .Lfunc_end0-_Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@object # @_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .rodata,"a",@progbits .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 3, 0x0 _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: .quad _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj" .size .L__unnamed_1, 67 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R0, SR_TID.X ; /* 0x0000000000007919 */ /* 0x000e280000002100 */ /*0020*/ S2R R3, SR_CTAID.X ; /* 0x0000000000037919 */ /* 0x000e620000002500 */ /*0030*/ LOP3.LUT R0, R0, 0x1f, RZ, 0xc0, !PT ; /* 0x0000001f00007812 */ /* 0x001fca00078ec0ff */ /*0040*/ IMAD R0, R3, c[0x0][0x0], R0 ; /* 0x0000000003007a24 */ /* 0x002fca00078e0200 */ /*0050*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x180], PT ; /* 0x0000600000007a0c */ /* 0x000fda0003f06270 */ /*0060*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0070*/ IMAD.MOV.U32 R5, RZ, RZ, 0x4 ; /* 0x00000004ff057424 */ /* 0x000fe200078e00ff */ /*0080*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fc60000000a00 */ /*0090*/ IMAD.WIDE R2, R0, R5, c[0x0][0x178] ; /* 0x00005e0000027625 */ /* 0x000fca00078e0205 */ /*00a0*/ LDG.E R0, [R2.64] ; /* 0x0000000602007981 */ /* 0x000ea2000c1e1900 */ /*00b0*/ ULDC UR4, c[0x0][0x194] ; /* 0x0000650000047ab9 */ /* 0x000fe40000000800 */ /*00c0*/ ULDC UR5, c[0x0][0x184] ; /* 0x0000610000057ab9 */ /* 0x000fe40000000800 */ /*00d0*/ UIMAD UR4, UR4, UR5, URZ ; /* 0x00000005040472a4 */ /* 0x000fd2000f8e023f */ /*00e0*/ I2F.U32 R7, UR4 ; /* 0x0000000400077d06 */ /* 0x000e220008201000 */ /*00f0*/ IMAD.WIDE R4, R0, R5, c[0x0][0x160] ; /* 0x0000580000047625 */ /* 0x004fcc00078e0205 */ /*0100*/ LDG.E R4, [R4.64] ; /* 0x0000000604047981 */ /* 0x000ea2000c1e1900 */ /*0110*/ MUFU.RCP R6, R7 ; /* 0x0000000700067308 */ /* 0x001e220000001000 */ /*0120*/ BSSY B0, 0x1e0 ; /* 0x000000b000007945 */ /* 0x000fe20003800000 */ /*0130*/ FFMA R9, -R7, R6, 1 ; /* 0x3f80000007097423 */ /* 0x001fc80000000106 */ /*0140*/ FFMA R9, R6, R9, R6 ; /* 0x0000000906097223 */ /* 0x000fe40000000006 */ /*0150*/ FCHK P0, -R4, R7 ; /* 0x0000000704007302 */ /* 0x004e240000000100 */ /*0160*/ FFMA R2, -R4, R9, RZ ; /* 0x0000000904027223 */ /* 0x000fc800000001ff */ /*0170*/ FFMA R3, -R7, R2, -R4 ; /* 0x0000000207037223 */ /* 0x000fc80000000904 */ /*0180*/ FFMA R9, R9, R3, R2 ; /* 0x0000000309097223 */ /* 0x000fe20000000002 */ /*0190*/ @!P0 BRA 0x1d0 ; /* 0x0000003000008947 */ /* 0x001fea0003800000 */ /*01a0*/ FADD R5, -R4, -RZ ; /* 0x800000ff04057221 */ /* 0x000fe20000000100 */ /*01b0*/ MOV R2, 0x1d0 ; /* 0x000001d000027802 */ /* 0x000fe40000000f00 */ /*01c0*/ CALL.REL.NOINC 0x220 ; /* 0x0000005000007944 */ /* 0x000fea0003c00000 */ /*01d0*/ BSYNC B0 ; /* 0x0000000000007941 */ /* 0x000fea0003800000 */ /*01e0*/ IMAD.MOV.U32 R3, RZ, RZ, 0x4 ; /* 0x00000004ff037424 */ /* 0x000fc800078e00ff */ /*01f0*/ IMAD.WIDE R2, R0, R3, c[0x0][0x168] ; /* 0x00005a0000027625 */ /* 0x000fca00078e0203 */ /*0200*/ STG.E [R2.64], R9 ; /* 0x0000000902007986 */ /* 0x000fe2000c101906 */ /*0210*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0220*/ SHF.R.U32.HI R4, RZ, 0x17, R7 ; /* 0x00000017ff047819 */ /* 0x000fe20000011607 */ /*0230*/ BSSY B1, 0x870 ; /* 0x0000063000017945 */ /* 0x000fe20003800000 */ /*0240*/ SHF.R.U32.HI R3, RZ, 0x17, R5 ; /* 0x00000017ff037819 */ /* 0x000fe20000011605 */ /*0250*/ IMAD.MOV.U32 R6, RZ, RZ, R7 ; /* 0x000000ffff067224 */ /* 0x000fe200078e0007 */ /*0260*/ LOP3.LUT R4, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04047812 */ /* 0x000fe400078ec0ff */ /*0270*/ LOP3.LUT R3, R3, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff03037812 */ /* 0x000fe400078ec0ff */ /*0280*/ IADD3 R10, R4, -0x1, RZ ; /* 0xffffffff040a7810 */ /* 0x000fe40007ffe0ff */ /*0290*/ IADD3 R9, R3, -0x1, RZ ; /* 0xffffffff03097810 */ /* 0x000fc40007ffe0ff */ /*02a0*/ ISETP.GT.U32.AND P0, PT, R10, 0xfd, PT ; /* 0x000000fd0a00780c */ /* 0x000fc80003f04070 */ /*02b0*/ ISETP.GT.U32.OR P0, PT, R9, 0xfd, P0 ; /* 0x000000fd0900780c */ /* 0x000fda0000704470 */ /*02c0*/ @!P0 IMAD.MOV.U32 R8, RZ, RZ, RZ ; /* 0x000000ffff088224 */ /* 0x000fe200078e00ff */ /*02d0*/ @!P0 BRA 0x450 ; /* 0x0000017000008947 */ /* 0x000fea0003800000 */ /*02e0*/ FSETP.GTU.FTZ.AND P0, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fe40003f1c200 */ /*02f0*/ FSETP.GTU.FTZ.AND P1, PT, |R7|, +INF , PT ; /* 0x7f8000000700780b */ /* 0x000fc80003f3c200 */ /*0300*/ PLOP3.LUT P0, PT, P0, P1, PT, 0xa8, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000703570 */ /*0310*/ @P0 BRA 0x850 ; /* 0x0000053000000947 */ /* 0x000fea0003800000 */ /*0320*/ LOP3.LUT P0, RZ, R6, 0x7fffffff, R5, 0xc8, !PT ; /* 0x7fffffff06ff7812 */ /* 0x000fda000780c805 */ /*0330*/ @!P0 BRA 0x830 ; /* 0x000004f000008947 */ /* 0x000fea0003800000 */ /*0340*/ FSETP.NEU.FTZ.AND P2, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fe40003f5d200 */ /*0350*/ FSETP.NEU.FTZ.AND P1, PT, |R7|, +INF , PT ; /* 0x7f8000000700780b */ /* 0x000fe40003f3d200 */ /*0360*/ FSETP.NEU.FTZ.AND P0, PT, |R5|, +INF , PT ; /* 0x7f8000000500780b */ /* 0x000fd60003f1d200 */ /*0370*/ @!P1 BRA !P2, 0x830 ; /* 0x000004b000009947 */ /* 0x000fea0005000000 */ /*0380*/ LOP3.LUT P2, RZ, R5, 0x7fffffff, RZ, 0xc0, !PT ; /* 0x7fffffff05ff7812 */ /* 0x000fc8000784c0ff */ /*0390*/ PLOP3.LUT P1, PT, P1, P2, PT, 0x2a, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000f24572 */ /*03a0*/ @P1 BRA 0x810 ; /* 0x0000046000001947 */ /* 0x000fea0003800000 */ /*03b0*/ LOP3.LUT P1, RZ, R6, 0x7fffffff, RZ, 0xc0, !PT ; /* 0x7fffffff06ff7812 */ /* 0x000fc8000782c0ff */ /*03c0*/ PLOP3.LUT P0, PT, P0, P1, PT, 0x2a, 0x0 ; /* 0x000000000000781c */ /* 0x000fda0000702572 */ /*03d0*/ @P0 BRA 0x7e0 ; /* 0x0000040000000947 */ /* 0x000fea0003800000 */ /*03e0*/ ISETP.GE.AND P0, PT, R9, RZ, PT ; /* 0x000000ff0900720c */ /* 0x000fe40003f06270 */ /*03f0*/ ISETP.GE.AND P1, PT, R10, RZ, PT ; /* 0x000000ff0a00720c */ /* 0x000fd60003f26270 */ /*0400*/ @P0 IMAD.MOV.U32 R8, RZ, RZ, RZ ; /* 0x000000ffff080224 */ /* 0x000fe400078e00ff */ /*0410*/ @!P0 IMAD.MOV.U32 R8, RZ, RZ, -0x40 ; /* 0xffffffc0ff088424 */ /* 0x000fe400078e00ff */ /*0420*/ @!P0 FFMA R5, R5, 1.84467440737095516160e+19, RZ ; /* 0x5f80000005058823 */ /* 0x000fe400000000ff */ /*0430*/ @!P1 FFMA R6, R7, 1.84467440737095516160e+19, RZ ; /* 0x5f80000007069823 */ /* 0x000fe200000000ff */ /*0440*/ @!P1 IADD3 R8, R8, 0x40, RZ ; /* 0x0000004008089810 */ /* 0x000fe40007ffe0ff */ /*0450*/ LEA R7, R4, 0xc0800000, 0x17 ; /* 0xc080000004077811 */ /* 0x000fe200078eb8ff */ /*0460*/ BSSY B2, 0x7d0 ; /* 0x0000036000027945 */ /* 0x000fe80003800000 */ /*0470*/ IMAD.IADD R7, R6, 0x1, -R7 ; /* 0x0000000106077824 */ /* 0x000fe200078e0a07 */ /*0480*/ IADD3 R6, R3, -0x7f, RZ ; /* 0xffffff8103067810 */ /* 0x000fc60007ffe0ff */ /*0490*/ MUFU.RCP R9, R7 ; /* 0x0000000700097308 */ /* 0x000e220000001000 */ /*04a0*/ FADD.FTZ R10, -R7, -RZ ; /* 0x800000ff070a7221 */ /* 0x000fe40000010100 */ /*04b0*/ IMAD R5, R6, -0x800000, R5 ; /* 0xff80000006057824 */ /* 0x000fe400078e0205 */ /*04c0*/ FFMA R12, R9, R10, 1 ; /* 0x3f800000090c7423 */ /* 0x001fc8000000000a */ /*04d0*/ FFMA R14, R9, R12, R9 ; /* 0x0000000c090e7223 */ /* 0x000fc80000000009 */ /*04e0*/ FFMA R3, R5, R14, RZ ; /* 0x0000000e05037223 */ /* 0x000fc800000000ff */ /*04f0*/ FFMA R12, R10, R3, R5 ; /* 0x000000030a0c7223 */ /* 0x000fc80000000005 */ /*0500*/ FFMA R9, R14, R12, R3 ; /* 0x0000000c0e097223 */ /* 0x000fc80000000003 */ /*0510*/ FFMA R10, R10, R9, R5 ; /* 0x000000090a0a7223 */ /* 0x000fe20000000005 */ /*0520*/ IADD3 R5, R6, 0x7f, -R4 ; /* 0x0000007f06057810 */ /* 0x000fc60007ffe804 */ /*0530*/ FFMA R3, R14, R10, R9 ; /* 0x0000000a0e037223 */ /* 0x000fe40000000009 */ /*0540*/ IMAD.IADD R5, R5, 0x1, R8 ; /* 0x0000000105057824 */ /* 0x000fc600078e0208 */ /*0550*/ SHF.R.U32.HI R4, RZ, 0x17, R3 ; /* 0x00000017ff047819 */ /* 0x000fc80000011603 */ /*0560*/ LOP3.LUT R4, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04047812 */ /* 0x000fca00078ec0ff */ /*0570*/ IMAD.IADD R8, R4, 0x1, R5 ; /* 0x0000000104087824 */ /* 0x000fca00078e0205 */ /*0580*/ IADD3 R4, R8, -0x1, RZ ; /* 0xffffffff08047810 */ /* 0x000fc80007ffe0ff */ /*0590*/ ISETP.GE.U32.AND P0, PT, R4, 0xfe, PT ; /* 0x000000fe0400780c */ /* 0x000fda0003f06070 */ /*05a0*/ @!P0 BRA 0x7b0 ; /* 0x0000020000008947 */ /* 0x000fea0003800000 */ /*05b0*/ ISETP.GT.AND P0, PT, R8, 0xfe, PT ; /* 0x000000fe0800780c */ /* 0x000fda0003f04270 */ /*05c0*/ @P0 BRA 0x780 ; /* 0x000001b000000947 */ /* 0x000fea0003800000 */ /*05d0*/ ISETP.GE.AND P0, PT, R8, 0x1, PT ; /* 0x000000010800780c */ /* 0x000fda0003f06270 */ /*05e0*/ @P0 BRA 0x7c0 ; /* 0x000001d000000947 */ /* 0x000fea0003800000 */ /*05f0*/ ISETP.GE.AND P0, PT, R8, -0x18, PT ; /* 0xffffffe80800780c */ /* 0x000fe40003f06270 */ /*0600*/ LOP3.LUT R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fd600078ec0ff */ /*0610*/ @!P0 BRA 0x7c0 ; /* 0x000001a000008947 */ /* 0x000fea0003800000 */ /*0620*/ FFMA.RZ R4, R14, R10.reuse, R9.reuse ; /* 0x0000000a0e047223 */ /* 0x180fe2000000c009 */ /*0630*/ IADD3 R7, R8, 0x20, RZ ; /* 0x0000002008077810 */ /* 0x000fe20007ffe0ff */ /*0640*/ FFMA.RM R5, R14, R10.reuse, R9.reuse ; /* 0x0000000a0e057223 */ /* 0x180fe20000004009 */ /*0650*/ ISETP.NE.AND P2, PT, R8, RZ, PT ; /* 0x000000ff0800720c */ /* 0x000fe40003f45270 */ /*0660*/ LOP3.LUT R6, R4, 0x7fffff, RZ, 0xc0, !PT ; /* 0x007fffff04067812 */ /* 0x000fe200078ec0ff */ /*0670*/ FFMA.RP R4, R14, R10, R9 ; /* 0x0000000a0e047223 */ /* 0x000fe20000008009 */ /*0680*/ ISETP.NE.AND P1, PT, R8, RZ, PT ; /* 0x000000ff0800720c */ /* 0x000fe20003f25270 */ /*0690*/ IMAD.MOV R8, RZ, RZ, -R8 ; /* 0x000000ffff087224 */ /* 0x000fe200078e0a08 */ /*06a0*/ LOP3.LUT R6, R6, 0x800000, RZ, 0xfc, !PT ; /* 0x0080000006067812 */ /* 0x000fe400078efcff */ /*06b0*/ FSETP.NEU.FTZ.AND P0, PT, R4, R5, PT ; /* 0x000000050400720b */ /* 0x000fc40003f1d000 */ /*06c0*/ SHF.L.U32 R7, R6, R7, RZ ; /* 0x0000000706077219 */ /* 0x000fe400000006ff */ /*06d0*/ SEL R5, R8, RZ, P2 ; /* 0x000000ff08057207 */ /* 0x000fe40001000000 */ /*06e0*/ ISETP.NE.AND P1, PT, R7, RZ, P1 ; /* 0x000000ff0700720c */ /* 0x000fe40000f25270 */ /*06f0*/ SHF.R.U32.HI R5, RZ, R5, R6 ; /* 0x00000005ff057219 */ /* 0x000fe40000011606 */ /*0700*/ PLOP3.LUT P0, PT, P0, P1, PT, 0xa8, 0x0 ; /* 0x000000000000781c */ /* 0x000fe40000703570 */ /*0710*/ SHF.R.U32.HI R7, RZ, 0x1, R5 ; /* 0x00000001ff077819 */ /* 0x000fc40000011605 */ /*0720*/ SEL R4, RZ, 0x1, !P0 ; /* 0x00000001ff047807 */ /* 0x000fc80004000000 */ /*0730*/ LOP3.LUT R4, R4, 0x1, R7, 0xf8, !PT ; /* 0x0000000104047812 */ /* 0x000fc800078ef807 */ /*0740*/ LOP3.LUT R4, R4, R5, RZ, 0xc0, !PT ; /* 0x0000000504047212 */ /* 0x000fca00078ec0ff */ /*0750*/ IMAD.IADD R4, R7, 0x1, R4 ; /* 0x0000000107047824 */ /* 0x000fca00078e0204 */ /*0760*/ LOP3.LUT R3, R4, R3, RZ, 0xfc, !PT ; /* 0x0000000304037212 */ /* 0x000fe200078efcff */ /*0770*/ BRA 0x7c0 ; /* 0x0000004000007947 */ /* 0x000fea0003800000 */ /*0780*/ LOP3.LUT R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fc800078ec0ff */ /*0790*/ LOP3.LUT R3, R3, 0x7f800000, RZ, 0xfc, !PT ; /* 0x7f80000003037812 */ /* 0x000fe200078efcff */ /*07a0*/ BRA 0x7c0 ; /* 0x0000001000007947 */ /* 0x000fea0003800000 */ /*07b0*/ IMAD R3, R5, 0x800000, R3 ; /* 0x0080000005037824 */ /* 0x000fe400078e0203 */ /*07c0*/ BSYNC B2 ; /* 0x0000000000027941 */ /* 0x000fea0003800000 */ /*07d0*/ BRA 0x860 ; /* 0x0000008000007947 */ /* 0x000fea0003800000 */ /*07e0*/ LOP3.LUT R3, R6, 0x80000000, R5, 0x48, !PT ; /* 0x8000000006037812 */ /* 0x000fc800078e4805 */ /*07f0*/ LOP3.LUT R3, R3, 0x7f800000, RZ, 0xfc, !PT ; /* 0x7f80000003037812 */ /* 0x000fe200078efcff */ /*0800*/ BRA 0x860 ; /* 0x0000005000007947 */ /* 0x000fea0003800000 */ /*0810*/ LOP3.LUT R3, R6, 0x80000000, R5, 0x48, !PT ; /* 0x8000000006037812 */ /* 0x000fe200078e4805 */ /*0820*/ BRA 0x860 ; /* 0x0000003000007947 */ /* 0x000fea0003800000 */ /*0830*/ MUFU.RSQ R3, -QNAN ; /* 0xffc0000000037908 */ /* 0x000e220000001400 */ /*0840*/ BRA 0x860 ; /* 0x0000001000007947 */ /* 0x000fea0003800000 */ /*0850*/ FADD.FTZ R3, R5, R7 ; /* 0x0000000705037221 */ /* 0x000fe40000010000 */ /*0860*/ BSYNC B1 ; /* 0x0000000000017941 */ /* 0x000fea0003800000 */ /*0870*/ IMAD.MOV.U32 R9, RZ, RZ, R3 ; /* 0x000000ffff097224 */ /* 0x001fe400078e0003 */ /*0880*/ IMAD.MOV.U32 R3, RZ, RZ, 0x0 ; /* 0x00000000ff037424 */ /* 0x000fc800078e00ff */ /*0890*/ RET.REL.NODEC R2 0x0 ; /* 0xfffff76002007950 */ /* 0x000fea0003c3ffff */ /*08a0*/ BRA 0x8a0; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*08b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*08f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0900*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0910*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0920*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0930*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0940*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0950*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0960*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0970*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 8 .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@function _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: s_clause 0x1 s_load_b32 s2, s[0:1], 0x44 s_load_b32 s3, s[0:1], 0x20 v_and_b32_e32 v2, 31, v0 s_waitcnt lgkmcnt(0) s_and_b32 s2, s2, 0xffff s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) v_mad_u64_u32 v[0:1], null, s15, s2, v[2:3] s_mov_b32 s2, exec_lo s_delay_alu instid0(VALU_DEP_1) v_cmpx_gt_i32_e64 s3, v0 s_cbranch_execz .LBB0_2 s_clause 0x1 s_load_b64 s[2:3], s[0:1], 0x18 s_load_b128 s[4:7], s[0:1], 0x0 v_ashrrev_i32_e32 v1, 31, v0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_lshlrev_b64 v[0:1], 2, v[0:1] s_waitcnt lgkmcnt(0) v_add_co_u32 v0, vcc_lo, s2, v0 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_1) v_add_co_ci_u32_e32 v1, vcc_lo, s3, v1, vcc_lo global_load_b32 v0, v[0:1], off s_waitcnt vmcnt(0) v_ashrrev_i32_e32 v1, 31, v0 v_lshlrev_b64 v[0:1], 2, v[0:1] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_co_u32 v2, vcc_lo, s4, v0 v_add_co_ci_u32_e32 v3, vcc_lo, s5, v1, vcc_lo global_load_b32 v2, v[2:3], off s_clause 0x1 s_load_b32 s2, s[0:1], 0x24 s_load_b32 s0, s[0:1], 0x34 s_waitcnt lgkmcnt(0) s_mul_i32 s0, s0, s2 s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) v_cvt_f32_u32_e32 v3, s0 s_waitcnt vmcnt(0) v_div_scale_f32 v4, null, v3, v3, -v2 v_div_scale_f32 v7, vcc_lo, -v2, v3, -v2 s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) v_rcp_f32_e32 v5, v4 s_waitcnt_depctr 0xfff v_fma_f32 v6, -v4, v5, 1.0 v_fmac_f32_e32 v5, v6, v5 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mul_f32_e32 v6, v7, v5 v_fma_f32 v8, -v4, v6, v7 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmac_f32_e32 v6, v8, v5 v_fma_f32 v4, -v4, v6, v7 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) v_div_fmas_f32 v4, v4, v5, v6 v_add_co_u32 v0, vcc_lo, s6, v0 v_add_co_ci_u32_e32 v1, vcc_lo, s7, v1, vcc_lo v_div_fixup_f32 v2, v4, v3, -v2 global_store_b32 v[0:1], v2, off .LBB0_2: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 312 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 9 .amdhsa_next_free_sgpr 16 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .Lfunc_end0-_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .address_space: global .offset: 0 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 8 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 16 .size: 8 .value_kind: global_buffer - .address_space: global .offset: 24 .size: 8 .value_kind: global_buffer - .offset: 32 .size: 4 .value_kind: by_value - .offset: 36 .size: 4 .value_kind: by_value - .offset: 40 .size: 4 .value_kind: by_value - .offset: 44 .size: 4 .value_kind: by_value - .offset: 48 .size: 4 .value_kind: by_value - .offset: 52 .size: 4 .value_kind: by_value - .offset: 56 .size: 4 .value_kind: hidden_block_count_x - .offset: 60 .size: 4 .value_kind: hidden_block_count_y - .offset: 64 .size: 4 .value_kind: hidden_block_count_z - .offset: 68 .size: 2 .value_kind: hidden_group_size_x - .offset: 70 .size: 2 .value_kind: hidden_group_size_y - .offset: 72 .size: 2 .value_kind: hidden_group_size_z - .offset: 74 .size: 2 .value_kind: hidden_remainder_x - .offset: 76 .size: 2 .value_kind: hidden_remainder_y - .offset: 78 .size: 2 .value_kind: hidden_remainder_z - .offset: 96 .size: 8 .value_kind: hidden_global_offset_x - .offset: 104 .size: 8 .value_kind: hidden_global_offset_y - .offset: 112 .size: 8 .value_kind: hidden_global_offset_z - .offset: 120 .size: 2 .value_kind: hidden_grid_dims .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 312 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .private_segment_fixed_size: 0 .sgpr_count: 18 .sgpr_spill_count: 0 .symbol: _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 9 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_0019613a_00000000-6_cudaSAnchorBackPropagateSSD_NegSamples_kernel.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2029: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2029: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj .type _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj, @function _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj: .LFB2051: .cfi_startproc endbr64 subq $216, %rsp .cfi_def_cfa_offset 224 movq %rdi, 40(%rsp) movq %rsi, 32(%rsp) movq %rdx, 24(%rsp) movq %rcx, 16(%rsp) movl %r8d, 12(%rsp) movl %r9d, 8(%rsp) movq %fs:40, %rax movq %rax, 200(%rsp) xorl %eax, %eax leaq 40(%rsp), %rax movq %rax, 112(%rsp) leaq 32(%rsp), %rax movq %rax, 120(%rsp) leaq 24(%rsp), %rax movq %rax, 128(%rsp) leaq 16(%rsp), %rax movq %rax, 136(%rsp) leaq 12(%rsp), %rax movq %rax, 144(%rsp) leaq 8(%rsp), %rax movq %rax, 152(%rsp) leaq 224(%rsp), %rax movq %rax, 160(%rsp) leaq 232(%rsp), %rax movq %rax, 168(%rsp) leaq 240(%rsp), %rax movq %rax, 176(%rsp) leaq 248(%rsp), %rax movq %rax, 184(%rsp) movl $1, 64(%rsp) movl $1, 68(%rsp) movl $1, 72(%rsp) movl $1, 76(%rsp) movl $1, 80(%rsp) movl $1, 84(%rsp) leaq 56(%rsp), %rcx leaq 48(%rsp), %rdx leaq 76(%rsp), %rsi leaq 64(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 200(%rsp), %rax subq %fs:40, %rax jne .L8 addq $216, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 56(%rsp) .cfi_def_cfa_offset 232 pushq 56(%rsp) .cfi_def_cfa_offset 240 leaq 128(%rsp), %r9 movq 92(%rsp), %rcx movl 100(%rsp), %r8d movq 80(%rsp), %rsi movl 88(%rsp), %edx leaq _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 224 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2051: .size _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj, .-_Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, @function _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: .LFB2052: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 24 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 32 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 40 movl 40(%rsp), %eax pushq %rax .cfi_def_cfa_offset 48 call _Z80__device_stub__Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjjPKfPfS0_PKiiijjjj addq $40, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2052: .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .-_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2054: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC0(%rip), %rdx movq %rdx, %rcx leaq _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2054: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "cudaSAnchorBackPropagateSSD_NegSamples_kernel.hip" .globl _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj # -- Begin function _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 4, 0x90 .type _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@function _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: # @_Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .cfi_startproc # %bb.0: subq $184, %rsp .cfi_def_cfa_offset 192 movq %rdi, 88(%rsp) movq %rsi, 80(%rsp) movq %rdx, 72(%rsp) movq %rcx, 64(%rsp) movl %r8d, 12(%rsp) movl %r9d, 8(%rsp) leaq 88(%rsp), %rax movq %rax, 96(%rsp) leaq 80(%rsp), %rax movq %rax, 104(%rsp) leaq 72(%rsp), %rax movq %rax, 112(%rsp) leaq 64(%rsp), %rax movq %rax, 120(%rsp) leaq 12(%rsp), %rax movq %rax, 128(%rsp) leaq 8(%rsp), %rax movq %rax, 136(%rsp) leaq 192(%rsp), %rax movq %rax, 144(%rsp) leaq 200(%rsp), %rax movq %rax, 152(%rsp) leaq 208(%rsp), %rax movq %rax, 160(%rsp) leaq 216(%rsp), %rax movq %rax, 168(%rsp) leaq 48(%rsp), %rdi leaq 32(%rsp), %rsi leaq 24(%rsp), %rdx leaq 16(%rsp), %rcx callq __hipPopCallConfiguration movq 48(%rsp), %rsi movl 56(%rsp), %edx movq 32(%rsp), %rcx movl 40(%rsp), %r8d leaq 96(%rsp), %r9 movl $_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, %edi pushq 16(%rsp) .cfi_adjust_cfa_offset 8 pushq 32(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $200, %rsp .cfi_adjust_cfa_offset -200 retq .Lfunc_end0: .size _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, .Lfunc_end0-_Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB1_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB1_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end1: .size __hip_module_ctor, .Lfunc_end1-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB2_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB2_2: retq .Lfunc_end2: .size __hip_module_dtor, .Lfunc_end2-__hip_module_dtor .cfi_endproc # -- End function .type _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj,@object # @_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .section .rodata,"a",@progbits .globl _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .p2align 3, 0x0 _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj: .quad _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .size _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj, 8 .type .L__unnamed_1,@object # @0 .section .rodata.str1.1,"aMS",@progbits,1 .L__unnamed_1: .asciz "_Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj" .size .L__unnamed_1, 67 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z60__device_stub__cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z45cudaSAnchorBackPropagateSSD_NegSamples_kernelPKfPfS0_PKiiijjjj .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting CUDA source code to CUDA device assembly.
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<cuda.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; cudaGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); cudaDeviceProp devProp; cudaGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: cudaError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMemcpy(gpu_arg, arg, N*sizeof(arg[0]), cudaMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = cudaMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), cudaMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); cudaFree(gpu_arg); cudaFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
code for sm_80 Function : _Z8cos_main9cosParams .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R0, SR_TID.X ; /* 0x0000000000007919 */ /* 0x000e240000002100 */ /*0020*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x001fda0003f06270 */ /*0030*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0040*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fe40000000a00 */ /*0050*/ IMAD.MOV.U32 R7, RZ, RZ, 0x4 ; /* 0x00000004ff077424 */ /* 0x000fc800078e00ff */ /*0060*/ IMAD.WIDE R6, R0, R7, c[0x0][0x160] ; /* 0x0000580000067625 */ /* 0x000fca00078e0207 */ /*0070*/ LDG.E R3, [R6.64] ; /* 0x0000000606037981 */ /* 0x001ea2000c1e1900 */ /*0080*/ BSSY B0, 0x7f0 ; /* 0x0000076000007945 */ /* 0x000fe20003800000 */ /*0090*/ FMUL R4, R3.reuse, 0.63661974668502807617 ; /* 0x3f22f98303047820 */ /* 0x044fe20000400000 */ /*00a0*/ FSETP.GE.AND P0, PT, |R3|, 105615, PT ; /* 0x47ce47800300780b */ /* 0x000fca0003f06200 */ /*00b0*/ F2I.NTZ R4, R4 ; /* 0x0000000400047305 */ /* 0x000e300000203100 */ /*00c0*/ I2F R8, R4 ; /* 0x0000000400087306 */ /* 0x001e240000201400 */ /*00d0*/ FFMA R5, R8, -1.5707962512969970703, R3 ; /* 0xbfc90fda08057823 */ /* 0x001fc80000000003 */ /*00e0*/ FFMA R9, R8, -7.5497894158615963534e-08, R5 ; /* 0xb3a2216808097823 */ /* 0x000fe20000000005 */ /*00f0*/ SHF.R.S32.HI R5, RZ, 0x1f, R0 ; /* 0x0000001fff057819 */ /* 0x000fc60000011400 */ /*0100*/ FFMA R17, R8, -5.3903029534742383927e-15, R9 ; /* 0xa7c234c508117823 */ /* 0x000fe20000000009 */ /*0110*/ @!P0 BRA 0x7e0 ; /* 0x000006c000008947 */ /* 0x000fea0003800000 */ /*0120*/ FSETP.NEU.AND P0, PT, |R3|, +INF , PT ; /* 0x7f8000000300780b */ /* 0x000fda0003f0d200 */ /*0130*/ @!P0 BRA 0x7c0 ; /* 0x0000068000008947 */ /* 0x000fea0003800000 */ /*0140*/ SHF.R.U32.HI R4, RZ, 0x17, R3 ; /* 0x00000017ff047819 */ /* 0x000fe20000011603 */ /*0150*/ IMAD.SHL.U32 R7, R3, 0x100, RZ ; /* 0x0000010003077824 */ /* 0x000fe200078e00ff */ /*0160*/ UMOV UR4, URZ ; /* 0x0000003f00047c82 */ /* 0x000fe20008000000 */ /*0170*/ IMAD.MOV.U32 R10, RZ, RZ, RZ ; /* 0x000000ffff0a7224 */ /* 0x000fe200078e00ff */ /*0180*/ LOP3.LUT R6, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04067812 */ /* 0x000fe200078ec0ff */ /*0190*/ IMAD.MOV.U32 R4, RZ, RZ, RZ ; /* 0x000000ffff047224 */ /* 0x000fe200078e00ff */ /*01a0*/ LOP3.LUT R17, R7, 0x80000000, RZ, 0xfc, !PT ; /* 0x8000000007117812 */ /* 0x000fe200078efcff */ /*01b0*/ IMAD.MOV.U32 R19, RZ, RZ, RZ ; /* 0x000000ffff137224 */ /* 0x000fe200078e00ff */ /*01c0*/ IADD3 R11, R6, -0x80, RZ ; /* 0xffffff80060b7810 */ /* 0x000fe20007ffe0ff */ /*01d0*/ ULDC.64 UR8, c[0x4][0x0] ; /* 0x0100000000087ab9 */ /* 0x000fc60000000a00 */ /*01e0*/ SHF.R.U32.HI R18, RZ, 0x5, R11 ; /* 0x00000005ff127819 */ /* 0x000fe4000001160b */ /*01f0*/ IMAD.U32 R8, RZ, RZ, UR8 ; /* 0x00000008ff087e24 */ /* 0x000fe4000f8e00ff */ /*0200*/ IMAD.U32 R9, RZ, RZ, UR9 ; /* 0x00000009ff097e24 */ /* 0x000fca000f8e00ff */ /*0210*/ LDG.E.CONSTANT R6, [R8.64] ; /* 0x0000000608067981 */ /* 0x000ea2000c1e9900 */ /*0220*/ IADD3 R10, R10, 0x1, RZ ; /* 0x000000010a0a7810 */ /* 0x000fe20007ffe0ff */ /*0230*/ UIADD3 UR8, UP0, UR8, 0x4, URZ ; /* 0x0000000408087890 */ /* 0x000fe2000ff1e03f */ /*0240*/ ISETP.EQ.AND P0, PT, R19.reuse, RZ, PT ; /* 0x000000ff1300720c */ /* 0x040fe40003f02270 */ /*0250*/ ISETP.EQ.AND P5, PT, R19.reuse, 0x4, PT ; /* 0x000000041300780c */ /* 0x040fe20003fa2270 */ /*0260*/ UIADD3.X UR9, URZ, UR9, URZ, UP0, !UPT ; /* 0x000000093f097290 */ /* 0x000fe200087fe43f */ /*0270*/ ISETP.EQ.AND P4, PT, R19.reuse, 0x8, PT ; /* 0x000000081300780c */ /* 0x040fe40003f82270 */ /*0280*/ ISETP.EQ.AND P3, PT, R19.reuse, 0xc, PT ; /* 0x0000000c1300780c */ /* 0x040fe40003f62270 */ /*0290*/ ISETP.EQ.AND P2, PT, R19, 0x10, PT ; /* 0x000000101300780c */ /* 0x000fc40003f42270 */ /*02a0*/ ISETP.EQ.AND P1, PT, R19.reuse, 0x14, PT ; /* 0x000000141300780c */ /* 0x040fe40003f22270 */ /*02b0*/ IADD3 R19, R19, 0x4, RZ ; /* 0x0000000413137810 */ /* 0x000fe20007ffe0ff */ /*02c0*/ IMAD.WIDE.U32 R6, R6, R17, RZ ; /* 0x0000001106067225 */ /* 0x004fca00078e00ff */ /*02d0*/ IADD3 R6, P6, R6, R4, RZ ; /* 0x0000000406067210 */ /* 0x000fc80007fde0ff */ /*02e0*/ IADD3.X R4, R7, UR4, RZ, P6, !PT ; /* 0x0000000407047c10 */ /* 0x000fe2000b7fe4ff */ /*02f0*/ @P0 IMAD.MOV.U32 R2, RZ, RZ, R6.reuse ; /* 0x000000ffff020224 */ /* 0x100fe200078e0006 */ /*0300*/ ISETP.NE.AND P6, PT, R10, 0x6, PT ; /* 0x000000060a00780c */ /* 0x000fe20003fc5270 */ /*0310*/ @P5 IMAD.MOV.U32 R12, RZ, RZ, R6.reuse ; /* 0x000000ffff0c5224 */ /* 0x100fe400078e0006 */ /*0320*/ @P4 IMAD.MOV.U32 R13, RZ, RZ, R6.reuse ; /* 0x000000ffff0d4224 */ /* 0x100fe400078e0006 */ /*0330*/ @P3 IMAD.MOV.U32 R14, RZ, RZ, R6.reuse ; /* 0x000000ffff0e3224 */ /* 0x100fe400078e0006 */ /*0340*/ @P2 IMAD.MOV.U32 R15, RZ, RZ, R6.reuse ; /* 0x000000ffff0f2224 */ /* 0x100fe400078e0006 */ /*0350*/ @P1 IMAD.MOV.U32 R16, RZ, RZ, R6 ; /* 0x000000ffff101224 */ /* 0x000fc800078e0006 */ /*0360*/ @P6 BRA 0x1f0 ; /* 0xfffffe8000006947 */ /* 0x000fea000383ffff */ /*0370*/ IADD3 R6, -R18, 0x6, RZ ; /* 0x0000000612067810 */ /* 0x000fe20007ffe1ff */ /*0380*/ BSSY B1, 0x6b0 ; /* 0x0000032000017945 */ /* 0x000fe80003800000 */ /*0390*/ IMAD.SHL.U32 R6, R6, 0x4, RZ ; /* 0x0000000406067824 */ /* 0x000fca00078e00ff */ /*03a0*/ ISETP.EQ.AND P0, PT, R6.reuse, RZ, PT ; /* 0x000000ff0600720c */ /* 0x040fe40003f02270 */ /*03b0*/ ISETP.EQ.AND P3, PT, R6.reuse, 0x4, PT ; /* 0x000000040600780c */ /* 0x040fe40003f62270 */ /*03c0*/ ISETP.EQ.AND P4, PT, R6.reuse, 0x8, PT ; /* 0x000000080600780c */ /* 0x040fe40003f82270 */ /*03d0*/ ISETP.EQ.AND P2, PT, R6.reuse, 0xc, PT ; /* 0x0000000c0600780c */ /* 0x040fe40003f42270 */ /*03e0*/ ISETP.EQ.AND P1, PT, R6, 0x10, PT ; /* 0x000000100600780c */ /* 0x000fca0003f22270 */ /*03f0*/ @P0 IMAD.MOV.U32 R7, RZ, RZ, R2.reuse ; /* 0x000000ffff070224 */ /* 0x100fe200078e0002 */ /*0400*/ ISETP.EQ.AND P0, PT, R6.reuse, 0x14, PT ; /* 0x000000140600780c */ /* 0x040fe20003f02270 */ /*0410*/ @P3 IMAD.MOV.U32 R8, RZ, RZ, R2 ; /* 0x000000ffff083224 */ /* 0x000fe400078e0002 */ /*0420*/ @P3 IMAD.MOV.U32 R7, RZ, RZ, R12.reuse ; /* 0x000000ffff073224 */ /* 0x100fe200078e000c */ /*0430*/ ISETP.EQ.AND P3, PT, R6.reuse, 0x18, PT ; /* 0x000000180600780c */ /* 0x040fe20003f62270 */ /*0440*/ @P4 IMAD.MOV.U32 R8, RZ, RZ, R12 ; /* 0x000000ffff084224 */ /* 0x000fe400078e000c */ /*0450*/ @P4 IMAD.MOV.U32 R7, RZ, RZ, R13.reuse ; /* 0x000000ffff074224 */ /* 0x100fe200078e000d */ /*0460*/ ISETP.EQ.AND P4, PT, R6, 0x1c, PT ; /* 0x0000001c0600780c */ /* 0x000fe20003f82270 */ /*0470*/ @P2 IMAD.MOV.U32 R8, RZ, RZ, R13 ; /* 0x000000ffff082224 */ /* 0x000fc400078e000d */ /*0480*/ @P2 IMAD.MOV.U32 R7, RZ, RZ, R14.reuse ; /* 0x000000ffff072224 */ /* 0x100fe400078e000e */ /*0490*/ @P1 IMAD.MOV.U32 R8, RZ, RZ, R14 ; /* 0x000000ffff081224 */ /* 0x000fe400078e000e */ /*04a0*/ @P1 IMAD.MOV.U32 R7, RZ, RZ, R15.reuse ; /* 0x000000ffff071224 */ /* 0x100fe200078e000f */ /*04b0*/ LOP3.LUT P1, R11, R11, 0x1f, RZ, 0xc0, !PT ; /* 0x0000001f0b0b7812 */ /* 0x000fe2000782c0ff */ /*04c0*/ @P0 IMAD.MOV.U32 R7, RZ, RZ, R16 ; /* 0x000000ffff070224 */ /* 0x000fe400078e0010 */ /*04d0*/ @P3 IMAD.MOV.U32 R7, RZ, RZ, R4 ; /* 0x000000ffff073224 */ /* 0x000fe400078e0004 */ /*04e0*/ @P0 IMAD.MOV.U32 R8, RZ, RZ, R15 ; /* 0x000000ffff080224 */ /* 0x000fc400078e000f */ /*04f0*/ @P3 IMAD.MOV.U32 R8, RZ, RZ, R16 ; /* 0x000000ffff083224 */ /* 0x000fe400078e0010 */ /*0500*/ @P4 IMAD.MOV.U32 R8, RZ, RZ, R4 ; /* 0x000000ffff084224 */ /* 0x000fe400078e0004 */ /*0510*/ IMAD.MOV.U32 R6, RZ, RZ, R7 ; /* 0x000000ffff067224 */ /* 0x000fe400078e0007 */ /*0520*/ @!P1 BRA 0x6a0 ; /* 0x0000017000009947 */ /* 0x000fea0003800000 */ /*0530*/ IADD3 R7, -R18, 0x4, RZ ; /* 0x0000000412077810 */ /* 0x000fe40007ffe1ff */ /*0540*/ SHF.L.U32 R9, R6, R11, RZ ; /* 0x0000000b06097219 */ /* 0x000fc600000006ff */ /*0550*/ IMAD.SHL.U32 R7, R7, 0x4, RZ ; /* 0x0000000407077824 */ /* 0x000fca00078e00ff */ /*0560*/ ISETP.EQ.AND P0, PT, R7.reuse, RZ, PT ; /* 0x000000ff0700720c */ /* 0x040fe40003f02270 */ /*0570*/ ISETP.EQ.AND P1, PT, R7.reuse, 0x4, PT ; /* 0x000000040700780c */ /* 0x040fe40003f22270 */ /*0580*/ ISETP.EQ.AND P2, PT, R7.reuse, 0x8, PT ; /* 0x000000080700780c */ /* 0x040fe40003f42270 */ /*0590*/ ISETP.EQ.AND P3, PT, R7, 0xc, PT ; /* 0x0000000c0700780c */ /* 0x000fce0003f62270 */ /*05a0*/ @P0 IMAD.MOV.U32 R10, RZ, RZ, R2 ; /* 0x000000ffff0a0224 */ /* 0x000fe200078e0002 */ /*05b0*/ ISETP.EQ.AND P0, PT, R7.reuse, 0x10, PT ; /* 0x000000100700780c */ /* 0x040fe20003f02270 */ /*05c0*/ @P1 IMAD.MOV.U32 R10, RZ, RZ, R12 ; /* 0x000000ffff0a1224 */ /* 0x000fe200078e000c */ /*05d0*/ ISETP.EQ.AND P1, PT, R7.reuse, 0x14, PT ; /* 0x000000140700780c */ /* 0x040fe20003f22270 */ /*05e0*/ @P2 IMAD.MOV.U32 R10, RZ, RZ, R13 ; /* 0x000000ffff0a2224 */ /* 0x000fe200078e000d */ /*05f0*/ ISETP.EQ.AND P2, PT, R7, 0x18, PT ; /* 0x000000180700780c */ /* 0x000fe20003f42270 */ /*0600*/ @P3 IMAD.MOV.U32 R10, RZ, RZ, R14 ; /* 0x000000ffff0a3224 */ /* 0x000fe200078e000e */ /*0610*/ IADD3 R7, -R11, 0x20, RZ ; /* 0x000000200b077810 */ /* 0x000fc80007ffe1ff */ /*0620*/ SHF.R.U32.HI R6, RZ, R7, R8 ; /* 0x00000007ff067219 */ /* 0x000fc60000011608 */ /*0630*/ @P0 IMAD.MOV.U32 R10, RZ, RZ, R15 ; /* 0x000000ffff0a0224 */ /* 0x000fe400078e000f */ /*0640*/ @P1 IMAD.MOV.U32 R10, RZ, RZ, R16 ; /* 0x000000ffff0a1224 */ /* 0x000fe400078e0010 */ /*0650*/ @P2 IMAD.MOV.U32 R10, RZ, RZ, R4 ; /* 0x000000ffff0a2224 */ /* 0x000fe200078e0004 */ /*0660*/ SHF.L.U32 R4, R8, R11, RZ ; /* 0x0000000b08047219 */ /* 0x000fe200000006ff */ /*0670*/ IMAD.IADD R6, R6, 0x1, R9 ; /* 0x0000000106067824 */ /* 0x000fc600078e0209 */ /*0680*/ SHF.R.U32.HI R7, RZ, R7, R10 ; /* 0x00000007ff077219 */ /* 0x000fca000001160a */ /*0690*/ IMAD.IADD R8, R7, 0x1, R4 ; /* 0x0000000107087824 */ /* 0x000fe400078e0204 */ /*06a0*/ BSYNC B1 ; /* 0x0000000000017941 */ /* 0x000fea0003800000 */ /*06b0*/ SHF.L.U32.HI R11, R8.reuse, 0x2, R6 ; /* 0x00000002080b7819 */ /* 0x040fe20000010606 */ /*06c0*/ IMAD.SHL.U32 R10, R8, 0x4, RZ ; /* 0x00000004080a7824 */ /* 0x000fe200078e00ff */ /*06d0*/ LOP3.LUT P1, R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fe4000782c0ff */ /*06e0*/ SHF.R.U32.HI R7, RZ, 0x1f, R11 ; /* 0x0000001fff077819 */ /* 0x000fc8000001160b */ /*06f0*/ ISETP.NE.AND P0, PT, R7, RZ, PT ; /* 0x000000ff0700720c */ /* 0x000fe40003f05270 */ /*0700*/ LEA.HI R4, R6, R7, RZ, 0x2 ; /* 0x0000000706047211 */ /* 0x000fd600078f10ff */ /*0710*/ @P0 LOP3.LUT R11, RZ, R11, RZ, 0x33, !PT ; /* 0x0000000bff0b0212 */ /* 0x000fe400078e33ff */ /*0720*/ @P0 LOP3.LUT R10, RZ, R10, RZ, 0x33, !PT ; /* 0x0000000aff0a0212 */ /* 0x000fe400078e33ff */ /*0730*/ @P0 LOP3.LUT R3, R3, 0x80000000, RZ, 0x3c, !PT ; /* 0x8000000003030812 */ /* 0x000fe400078e3cff */ /*0740*/ I2F.F64.S64 R8, R10 ; /* 0x0000000a00087312 */ /* 0x000e240000301c00 */ /*0750*/ ISETP.NE.AND P0, PT, R3, RZ, PT ; /* 0x000000ff0300720c */ /* 0x000fe20003f05270 */ /*0760*/ IMAD.MOV R3, RZ, RZ, -R4 ; /* 0x000000ffff037224 */ /* 0x000fc800078e0a04 */ /*0770*/ @P1 IMAD.MOV.U32 R4, RZ, RZ, R3 ; /* 0x000000ffff041224 */ /* 0x000fe200078e0003 */ /*0780*/ DMUL R8, R8, c[0x2][0x0] ; /* 0x0080000008087a28 */ /* 0x001e140000000000 */ /*0790*/ F2F.F32.F64 R8, R8 ; /* 0x0000000800087310 */ /* 0x001e240000301000 */ /*07a0*/ FSEL R17, R8, -R8, !P0 ; /* 0x8000000808117208 */ /* 0x001fe20004000000 */ /*07b0*/ BRA 0x7e0 ; /* 0x0000002000007947 */ /* 0x000fea0003800000 */ /*07c0*/ FMUL R17, RZ, R3 ; /* 0x00000003ff117220 */ /* 0x000fe40000400000 */ /*07d0*/ IMAD.MOV.U32 R4, RZ, RZ, RZ ; /* 0x000000ffff047224 */ /* 0x000fe400078e00ff */ /*07e0*/ BSYNC B0 ; /* 0x0000000000007941 */ /* 0x000fea0003800000 */ /*07f0*/ IADD3 R8, R4, 0x1, RZ ; /* 0x0000000104087810 */ /* 0x000fe20007ffe0ff */ /*0800*/ IMAD.MOV.U32 R4, RZ, RZ, 0x3c0885e4 ; /* 0x3c0885e4ff047424 */ /* 0x000fe400078e00ff */ /*0810*/ FMUL R9, R17, R17 ; /* 0x0000001111097220 */ /* 0x000fe20000400000 */ /*0820*/ LOP3.LUT P1, RZ, R8.reuse, 0x1, RZ, 0xc0, !PT ; /* 0x0000000108ff7812 */ /* 0x040fe2000782c0ff */ /*0830*/ IMAD.MOV.U32 R7, RZ, RZ, 0x3e2aaaa8 ; /* 0x3e2aaaa8ff077424 */ /* 0x000fe200078e00ff */ /*0840*/ LOP3.LUT P0, RZ, R8, 0x2, RZ, 0xc0, !PT ; /* 0x0000000208ff7812 */ /* 0x000fe2000780c0ff */ /*0850*/ IMAD.MOV.U32 R3, RZ, RZ, -0x46b2bead ; /* 0xb94d4153ff037424 */ /* 0x000fe200078e00ff */ /*0860*/ FSEL R4, R4, 0.041666727513074874878, !P1 ; /* 0x3d2aaabb04047808 */ /* 0x000fc40004800000 */ /*0870*/ FSEL R17, R17, 1, !P1 ; /* 0x3f80000011117808 */ /* 0x000fca0004800000 */ /*0880*/ FFMA R8, R17, R9, RZ ; /* 0x0000000911087223 */ /* 0x000fe400000000ff */ /*0890*/ @P1 IMAD.MOV.U32 R6, RZ, RZ, 0x37cbac00 ; /* 0x37cbac00ff061424 */ /* 0x000fc800078e00ff */ /*08a0*/ @P1 FFMA R3, R9, R6, -0.0013887860113754868507 ; /* 0xbab607ed09031423 */ /* 0x000fe20000000006 */ /*08b0*/ FSEL R6, -R7, -0.4999999701976776123, !P1 ; /* 0xbeffffff07067808 */ /* 0x000fc60004800100 */ /*08c0*/ FFMA R3, R9, R3, R4 ; /* 0x0000000309037223 */ /* 0x000fe20000000004 */ /*08d0*/ LEA R4, P1, R0, c[0x0][0x168], 0x2 ; /* 0x00005a0000047a11 */ /* 0x000fc600078210ff */ /*08e0*/ FFMA R3, R9, R3, R6 ; /* 0x0000000309037223 */ /* 0x000fe20000000006 */ /*08f0*/ LEA.HI.X R5, R0.reuse, c[0x0][0x16c], R5, 0x2, P1 ; /* 0x00005b0000057a11 */ /* 0x040fe400008f1405 */ /*0900*/ IADD3 R0, R0, 0x200, RZ ; /* 0x0000020000007810 */ /* 0x000fe20007ffe0ff */ /*0910*/ FFMA R3, R3, R8, R17 ; /* 0x0000000803037223 */ /* 0x000fc80000000011 */ /*0920*/ @P0 FFMA R3, R3, -1, RZ ; /* 0xbf80000003030823 */ /* 0x000fe200000000ff */ /*0930*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x000fc80003f06270 */ /*0940*/ STG.E [R4.64], R3 ; /* 0x0000000304007986 */ /* 0x0001f2000c101906 */ /*0950*/ @!P0 BRA 0x50 ; /* 0xfffff6f000008947 */ /* 0x000fea000383ffff */ /*0960*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0970*/ BRA 0x970; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0980*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0990*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
You are a helpful coding assistant specialized in converting CUDA source code to CUDA host assembly.
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<cuda.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; cudaGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); cudaDeviceProp devProp; cudaGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: cudaError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMemcpy(gpu_arg, arg, N*sizeof(arg[0]), cudaMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = cudaMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), cudaMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); cudaFree(gpu_arg); cudaFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
.file "tmpxft_000d8b16_00000000-6_CUDA_example_01.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z35__device_stub__Z8cos_main9cosParamsR9cosParams .type _Z35__device_stub__Z8cos_main9cosParamsR9cosParams, @function _Z35__device_stub__Z8cos_main9cosParamsR9cosParams: .LFB2082: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movq %rdi, 64(%rsp) movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 72(%rsp), %rax subq %fs:40, %rax jne .L8 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z8cos_main9cosParams(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z35__device_stub__Z8cos_main9cosParamsR9cosParams, .-_Z35__device_stub__Z8cos_main9cosParamsR9cosParams .globl _Z8cos_main9cosParams .type _Z8cos_main9cosParams, @function _Z8cos_main9cosParams: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq 16(%rsp), %rdi call _Z35__device_stub__Z8cos_main9cosParamsR9cosParams addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z8cos_main9cosParams, .-_Z8cos_main9cosParams .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "\nBeginning CUDA example code on the CPU...\n" .align 8 .LC1: .string "Getting some info about CUDA devices on your system...\n" .align 8 .LC2: .string "There is %d CUDA device on your system.\n" .align 8 .LC3: .string "There are %d CUDA devices on your system.\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC4: .string "\nCUDA Device #%d\n" .LC5: .string "Name: %s\n" .LC6: .string "\n" .section .rodata.str1.8 .align 8 .LC7: .string "Allocating memory on the CPU... (n=%d)\n" .align 8 .LC10: .string "Allocating memory on the GPU... (n=%d)\n" .align 8 .LC11: .string " value = %d : Memory Allocation on GPU Device failed\n" .align 8 .LC12: .string "done. Allocating more memory on the GPU...\n" .align 8 .LC13: .string "done again. Copying stuff from host (CPU) to device (GPU)...\n" .align 8 .LC14: .string " Memory Copy from Host to Device failed.\n" .section .rodata.str1.1 .LC15: .string "successful.\n" .LC16: .string "Launching kernel on GPU...\n" .section .rodata.str1.8 .align 8 .LC17: .string "GPU computations finished. Copying result back to CPU...\n" .align 8 .LC18: .string " Memory Copy from Device to Host failed.\n" .align 8 .LC19: .string "Copy back successful. Printing result...\n" .section .rodata.str1.1 .LC20: .string "cos(%f) = %f\n" .LC21: .string "Freeing memory...\n" .LC22: .string "\n\nFinished. :)\n\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %r13 .cfi_def_cfa_offset 16 .cfi_offset 13, -16 pushq %r12 .cfi_def_cfa_offset 24 .cfi_offset 12, -24 pushq %rbp .cfi_def_cfa_offset 32 .cfi_offset 6, -32 pushq %rbx .cfi_def_cfa_offset 40 .cfi_offset 3, -40 subq $1128, %rsp .cfi_def_cfa_offset 1168 movq %fs:40, %rax movq %rax, 1112(%rsp) xorl %eax, %eax leaq .LC0(%rip), %rsi movl $2, %edi call __printf_chk@PLT leaq .LC1(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq 4(%rsp), %rdi call cudaGetDeviceCount@PLT movl 4(%rsp), %edx cmpl $1, %edx je .L35 leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L15: movl $0, %ebx leaq .LC4(%rip), %r13 leaq 80(%rsp), %rbp leaq .LC5(%rip), %r12 cmpl $0, 4(%rsp) jle .L14 .L13: movl %ebx, %edx movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %esi movq %rbp, %rdi call cudaGetDeviceProperties_v2@PLT movq %rbp, %rdx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addl $1, %ebx cmpl %ebx, 4(%rsp) jg .L13 .L14: leaq .LC6(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq $0, 8(%rsp) movq $0, 16(%rsp) movl $1000000, %edx leaq .LC7(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $4000000, %edi call malloc@PLT movq %rax, %rbp movl $4000000, %edi call malloc@PLT movq %rax, %r12 movl $0, %eax movss .LC8(%rip), %xmm2 movss .LC9(%rip), %xmm1 .L16: pxor %xmm0, %xmm0 cvtsi2ssl %eax, %xmm0 mulss %xmm2, %xmm0 divss %xmm1, %xmm0 movss %xmm0, 0(%rbp,%rax,4) addq $1, %rax cmpq $1000000, %rax jne .L16 movl $1000000, %edx leaq .LC10(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq 16(%rsp), %rdi movl $4000000, %esi call cudaMalloc@PLT testl %eax, %eax je .L17 movl %eax, %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L18: leaq 8(%rsp), %rdi movl $4000000, %esi call cudaMalloc@PLT testl %eax, %eax je .L19 movl %eax, %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L20: movl $1, %ecx movl $4000000, %edx movq %rbp, %rsi movq 16(%rsp), %rdi call cudaMemcpy@PLT testl %eax, %eax je .L21 leaq .LC14(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L22: movq 8(%rsp), %rbx movq 16(%rsp), %r13 leaq .LC16(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $512, 36(%rsp) movl $1, 40(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $0, %r9d movl $0, %r8d movq 36(%rsp), %rdx movl $1, %ecx movq 24(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L36 .L23: leaq .LC17(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $2, %ecx movl $4000000, %edx movq 8(%rsp), %rsi movq %r12, %rdi call cudaMemcpy@PLT testl %eax, %eax je .L24 leaq .LC18(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L26: movl $0, %ebx leaq .LC20(%rip), %r13 jmp .L25 .L35: leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L15 .L17: leaq .LC12(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L18 .L19: leaq .LC13(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L20 .L21: leaq .LC15(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L22 .L36: movq %r13, 48(%rsp) movq %rbx, 56(%rsp) movl $1000000, 64(%rsp) leaq 48(%rsp), %rdi call _Z35__device_stub__Z8cos_main9cosParamsR9cosParams jmp .L23 .L24: leaq .LC19(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L26 .L27: addq $1, %rbx cmpq $1000000, %rbx je .L37 .L25: movslq %ebx, %rax imulq $351843721, %rax, %rax sarq $45, %rax movl %ebx, %edx sarl $31, %edx subl %edx, %eax imull $100000, %eax, %eax cmpl %ebx, %eax jne .L27 pxor %xmm0, %xmm0 cvtss2sd 0(%rbp,%rbx,4), %xmm0 pxor %xmm1, %xmm1 cvtss2sd (%r12,%rbx,4), %xmm1 movq %r13, %rsi movl $2, %edi movl $2, %eax call __printf_chk@PLT jmp .L27 .L37: leaq .LC21(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq %rbp, %rdi call free@PLT movq %r12, %rdi call free@PLT movq 16(%rsp), %rdi call cudaFree@PLT movq 8(%rsp), %rdi call cudaFree@PLT leaq .LC22(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 1112(%rsp), %rax subq %fs:40, %rax jne .L38 movl $0, %eax addq $1128, %rsp .cfi_remember_state .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %rbp .cfi_def_cfa_offset 24 popq %r12 .cfi_def_cfa_offset 16 popq %r13 .cfi_def_cfa_offset 8 ret .L38: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC23: .string "_Z8cos_main9cosParams" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC23(%rip), %rdx movq %rdx, %rcx leaq _Z8cos_main9cosParams(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .section .rodata.cst4,"aM",@progbits,4 .align 4 .LC8: .long 1086918618 .align 4 .LC9: .long 1232348160 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
You are a helpful coding assistant specialized in converting CUDA source code to HIP/AMD source code.
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<cuda.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; cudaGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); cudaDeviceProp devProp; cudaGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: cudaError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = cudaMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = cudaMemcpy(gpu_arg, arg, N*sizeof(arg[0]), cudaMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = cudaMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), cudaMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); cudaFree(gpu_arg); cudaFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<hip/hip_runtime.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; hipGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); hipDeviceProp_t devProp; hipGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: hipError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = hipMemcpy(gpu_arg, arg, N*sizeof(arg[0]), hipMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = hipMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), hipMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); hipFree(gpu_arg); hipFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
You are a helpful coding assistant specialized in converting HIP source code to AMD device assembly.
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<hip/hip_runtime.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; hipGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); hipDeviceProp_t devProp; hipGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: hipError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = hipMemcpy(gpu_arg, arg, N*sizeof(arg[0]), hipMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = hipMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), hipMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); hipFree(gpu_arg); hipFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z8cos_main9cosParams .globl _Z8cos_main9cosParams .p2align 8 .type _Z8cos_main9cosParams,@function _Z8cos_main9cosParams: s_load_b32 s2, s[0:1], 0x10 s_mov_b32 s3, exec_lo s_waitcnt lgkmcnt(0) v_cmpx_gt_i32_e64 s2, v0 s_cbranch_execz .LBB0_7 s_load_b128 s[4:7], s[0:1], 0x0 v_lshlrev_b32_e32 v3, 2, v0 s_mov_b32 s3, 0 s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_co_u32 v1, s0, s4, v3 v_add_co_ci_u32_e64 v2, null, s5, 0, s0 v_add_co_u32 v3, s0, s6, v3 s_delay_alu instid0(VALU_DEP_1) v_add_co_ci_u32_e64 v4, null, s7, 0, s0 s_mov_b32 s4, 0x7fffff s_mov_b32 s5, 0xb94c1982 s_mov_b32 s6, 0x37d75334 s_branch .LBB0_3 .LBB0_2: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) v_dual_mul_f32 v8, v6, v6 :: v_dual_and_b32 v11, 1, v7 v_add_nc_u32_e32 v0, 0x200, v0 v_lshlrev_b32_e32 v7, 30, v7 v_fmaak_f32 v9, s5, v8, 0x3c0881c4 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3) v_cmp_eq_u32_e32 vcc_lo, 0, v11 v_and_b32_e32 v7, 0x80000000, v7 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmaak_f32 v9, v8, v9, 0xbe2aaa9d v_dual_fmaak_f32 v10, s6, v8, 0xbab64f3b :: v_dual_mul_f32 v9, v8, v9 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_fmaak_f32 v10, v8, v10, 0x3d2aabf7 v_fmac_f32_e32 v6, v6, v9 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmaak_f32 v10, v8, v10, 0xbf000004 v_fma_f32 v8, v8, v10, 1.0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_cndmask_b32_e64 v6, -v6, v8, vcc_lo v_cmp_class_f32_e64 vcc_lo, v5, 0x1f8 v_xor_b32_e32 v6, v7, v6 s_delay_alu instid0(VALU_DEP_1) v_cndmask_b32_e32 v5, 0x7fc00000, v6, vcc_lo v_add_co_u32 v1, vcc_lo, v1, 0x800 v_add_co_ci_u32_e32 v2, vcc_lo, 0, v2, vcc_lo v_cmp_le_i32_e32 vcc_lo, s2, v0 global_store_b32 v[3:4], v5, off v_add_co_u32 v3, s0, v3, 0x800 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) v_add_co_ci_u32_e64 v4, s0, 0, v4, s0 s_or_b32 s3, vcc_lo, s3 s_and_not1_b32 exec_lo, exec_lo, s3 s_cbranch_execz .LBB0_7 .LBB0_3: global_load_b32 v5, v[1:2], off s_mov_b32 s1, exec_lo s_waitcnt vmcnt(0) v_cmpx_ngt_f32_e64 0x48000000, |v5| s_xor_b32 s7, exec_lo, s1 s_cbranch_execz .LBB0_5 v_dual_mov_b32 v8, 0 :: v_dual_and_b32 v13, 0x7fffffff, v5 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_and_or_b32 v16, v13, s4, 0x800000 v_lshrrev_b32_e32 v13, 23, v13 v_mad_u64_u32 v[6:7], null, v16, 0xfe5163ab, 0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v14, 0xffffff88, v13 v_cmp_lt_u32_e32 vcc_lo, 63, v14 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_mad_u64_u32 v[9:10], null, v16, 0x3c439041, v[7:8] v_cndmask_b32_e64 v15, 0, 0xffffffc0, vcc_lo v_mov_b32_e32 v7, v10 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_nc_u32_e32 v15, v15, v14 v_mad_u64_u32 v[10:11], null, v16, 0xdb629599, v[7:8] s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_cmp_lt_u32_e64 s0, 31, v15 v_cndmask_b32_e64 v17, 0, 0xffffffe0, s0 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) v_dual_mov_b32 v7, v11 :: v_dual_cndmask_b32 v6, v10, v6 v_add_nc_u32_e32 v17, v17, v15 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_mad_u64_u32 v[11:12], null, v16, 0xf534ddc0, v[7:8] v_cmp_lt_u32_e64 s1, 31, v17 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) v_mov_b32_e32 v7, v12 v_cndmask_b32_e32 v9, v11, v9, vcc_lo s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_mad_u64_u32 v[12:13], null, v16, 0xfc2757d1, v[7:8] v_cndmask_b32_e64 v6, v9, v6, s0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_mov_b32_e32 v7, v13 v_mad_u64_u32 v[13:14], null, v16, 0x4e441529, v[7:8] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mov_b32_e32 v7, v14 v_mad_u64_u32 v[14:15], null, v16, 0xa2f9836e, v[7:8] v_cndmask_b32_e64 v7, 0, 0xffffffe0, s1 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_dual_cndmask_b32 v8, v13, v11 :: v_dual_add_nc_u32 v7, v7, v17 v_dual_cndmask_b32 v14, v14, v12 :: v_dual_cndmask_b32 v13, v15, v13 v_cndmask_b32_e32 v12, v12, v10, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_cmp_eq_u32_e32 vcc_lo, 0, v7 v_cndmask_b32_e64 v11, v14, v8, s0 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_cndmask_b32_e64 v13, v13, v14, s0 v_cndmask_b32_e64 v8, v8, v12, s0 v_sub_nc_u32_e32 v14, 32, v7 v_cndmask_b32_e64 v12, v12, v9, s0 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_cndmask_b32_e64 v13, v13, v11, s1 v_cndmask_b32_e64 v11, v11, v8, s1 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_cndmask_b32_e64 v8, v8, v12, s1 v_cndmask_b32_e64 v6, v12, v6, s1 v_alignbit_b32 v15, v13, v11, v14 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) v_alignbit_b32 v10, v11, v8, v14 v_cndmask_b32_e32 v7, v15, v13, vcc_lo s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v13, v8, v6, v14 v_cndmask_b32_e32 v9, v10, v11, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_bfe_u32 v10, v7, 29, 1 v_cndmask_b32_e32 v8, v13, v8, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v11, v7, v9, 30 v_sub_nc_u32_e32 v12, 0, v10 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_alignbit_b32 v9, v9, v8, 30 v_alignbit_b32 v6, v8, v6, 30 v_xor_b32_e32 v11, v11, v12 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_xor_b32_e32 v8, v9, v12 v_xor_b32_e32 v6, v6, v12 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) v_clz_i32_u32_e32 v13, v11 v_min_u32_e32 v13, 32, v13 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v9, 31, v13 v_lshlrev_b32_e32 v15, 23, v13 v_alignbit_b32 v11, v11, v8, v9 v_alignbit_b32 v6, v8, v6, v9 v_lshrrev_b32_e32 v9, 29, v7 v_lshrrev_b32_e32 v7, 30, v7 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v8, v11, v6, 9 v_lshlrev_b32_e32 v9, 31, v9 v_lshrrev_b32_e32 v11, 9, v11 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_add_nc_u32_e32 v7, v10, v7 v_clz_i32_u32_e32 v12, v8 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2) v_or_b32_e32 v14, 0.5, v9 v_min_u32_e32 v12, 32, v12 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v14, v14, v15 v_sub_nc_u32_e32 v16, 31, v12 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v6, v8, v6, v16 v_or_b32_e32 v8, v11, v14 v_add_lshl_u32 v11, v12, v13, 23 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_lshrrev_b32_e32 v6, 9, v6 v_mul_f32_e32 v12, 0x3fc90fda, v8 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v6, v6, v11 v_fma_f32 v11, v8, 0x3fc90fda, -v12 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_nc_u32_e32 v6, 0x33000000, v6 v_fmac_f32_e32 v11, 0x33a22168, v8 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_or_b32_e32 v6, v6, v9 v_fmac_f32_e32 v11, 0x3fc90fda, v6 s_delay_alu instid0(VALU_DEP_1) v_add_f32_e32 v6, v12, v11 .LBB0_5: s_and_not1_saveexec_b32 s0, s7 s_cbranch_execz .LBB0_2 v_mul_f32_e64 v6, 0x3f22f983, |v5| s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_rndne_f32_e32 v7, v6 v_fma_f32 v6, v7, 0xbfc90fda, |v5| s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmac_f32_e32 v6, 0xb3a22168, v7 v_fmac_f32_e32 v6, 0xa7c234c4, v7 v_cvt_i32_f32_e32 v7, v7 s_branch .LBB0_2 .LBB0_7: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z8cos_main9cosParams .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 24 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 18 .amdhsa_next_free_sgpr 8 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z8cos_main9cosParams, .Lfunc_end0-_Z8cos_main9cosParams .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .offset: 0 .size: 24 .value_kind: by_value .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 24 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z8cos_main9cosParams .private_segment_fixed_size: 0 .sgpr_count: 10 .sgpr_spill_count: 0 .symbol: _Z8cos_main9cosParams.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 18 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting HIP source code to AMD host assembly.
/* Cuda Program for finding cos(0), cos(1*2*pi/N), ... , cos((N-1)*2*pi/N) */ // nvcc CUDA_example_01.cu -use_fast_math -o CUDA_example_01.out /* --------------------------- header secton ----------------------------*/ #include<stdio.h> #include<hip/hip_runtime.h> #define PRINT_RESULT 1 #define COS_THREAD_CNT 512 #define N 1000000 #define TWO_PI 6.283185 /* --------------------------- target code ------------------------------*/ struct cosParams { float *arg; float *res; int n; }; // CUDA kernel: __global__ void cos_main(struct cosParams params) { // Computes: // res[threadIdx.x + 0*COS_THREAD_CNT] = cos(threadIdx.x + 0*COS_THREAD_CNT) // res[threadIdx.x + 1*COS_THREAD_CNT] = cos(threadIdx.x + 1*COS_THREAD_CNT) // res[threadIdx.x + 2*COS_THREAD_CNT] = cos(threadIdx.x + 2*COS_THREAD_CNT) // ...etc... // // if COS_THREAD_CNT = 1, this computes all of the cosines in one go, // otherwise, it spreads it out across threads... // // Choice of either: // // __cosf: // Calculate the fast approximate cosine of the input argument x, measured in radians. // // cosf: // Calculate the cosine of the input argument x (measured in radians). int i; for (i=threadIdx.x; i<params.n; i+=COS_THREAD_CNT){ // params.res[i] = __cosf(params.arg[i]); params.res[i] = cosf(params.arg[i]); } return; } /* --------------------------- host code ------------------------------*/ int main (int argc, char *argv[]) { printf("\nBeginning CUDA example code on the CPU...\n"); printf("Getting some info about CUDA devices on your system...\n"); int i, devCount; hipGetDeviceCount(&devCount); if (devCount == 1){ printf("There is %d CUDA device on your system.\n", devCount); } else { printf("There are %d CUDA devices on your system.\n", devCount); } // Iterate through devices for (i=0; i<devCount; i++) { // Get device properties printf("\nCUDA Device #%d\n", i); hipDeviceProp_t devProp; hipGetDeviceProperties(&devProp, i); printf("Name: %s\n", devProp.name); } printf("\n"); // Begin cosine code: hipError_t cudaStat; float* gpu_res = 0; float* gpu_arg = 0; printf("Allocating memory on the CPU... (n=%d)\n",(int)N); // Allocate the vector 1,...,N: float *arg = (float *) malloc(N*sizeof(float)); // Allocate vector of length N to store the result: float *res = (float *) malloc(N*sizeof(float)); int print_every = (int)((double)N/10.0); struct cosParams funcParams; // Populate arg with 1,...,N for(i=0; i<N; i++){ arg[i] = (float)i*(float)TWO_PI/(float)N; } printf("Allocating memory on the GPU... (n=%d)\n",(int)N); // Key function: // cudaError_t cudaMalloc(void ** devPtr, size_t size); // Allocate N floats on the GPU for the argument 0,1,...,N-1, and make gpu_arg a pointer to that memory: // NOTE: sizeof(gpu_arg[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_arg, N*sizeof(gpu_arg[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", cudaStat); } else { printf("done. Allocating more memory on the GPU...\n"); } // Allocate N floats on the GPU to store the result, and make gpu_res a pointer to that memory: // NOTE: sizeof(gpu_res[0]) is equivalent to sizeof(float) here cudaStat = hipMalloc((void **)&gpu_res, N*sizeof(gpu_res[0])); if (cudaStat){ printf(" value = %d : Memory Allocation on GPU Device failed\n", (int)cudaStat); } else { printf("done again. Copying stuff from host (CPU) to device (GPU)...\n"); } // Key function: // cudaError_t cudaMemcpy(void * dst, const void * src, size_t count, enum cudaMemcpyKind kind); // Copy the vector 0,1,...,N-1 from arg (on the host) to gpu_arg (on the device) // NOTE: sizeof(arg[0]) is equivalent to sizeof(float) here cudaStat = hipMemcpy(gpu_arg, arg, N*sizeof(arg[0]), hipMemcpyHostToDevice); if (cudaStat){ printf(" Memory Copy from Host to Device failed.\n"); } else { printf("successful.\n"); } // Set up the parameters for the GPU kernel: funcParams.res = gpu_res; funcParams.arg = gpu_arg; funcParams.n = N; printf("Launching kernel on GPU...\n"); // Launch the GPU kernel... // Key code: // KernelFunction<<<dimGrid, dimBlock>>>(args); cos_main<<<1,COS_THREAD_CNT>>>(funcParams); printf("GPU computations finished. Copying result back to CPU...\n"); // Copy the vector cos(0),cos(1),...,cos(N-1) from gpu_res (on the device) to res (on the host) cudaStat = hipMemcpy(res, gpu_res, N*sizeof(gpu_res[0]), hipMemcpyDeviceToHost); if (cudaStat){ printf(" Memory Copy from Device to Host failed.\n"); } else { printf("Copy back successful. Printing result...\n"); } if (PRINT_RESULT){ // Print the result? for(i=0; i < N; i++ ){ if (i%print_every == 0 ){ printf("cos(%f) = %f\n", arg[i], res[i] ); } } } printf("Freeing memory...\n"); free(arg); free(res); hipFree(gpu_arg); hipFree(gpu_res); printf("\n\nFinished. :)\n\n"); return 0; }
.text .file "CUDA_example_01.hip" .globl _Z23__device_stub__cos_main9cosParams # -- Begin function _Z23__device_stub__cos_main9cosParams .p2align 4, 0x90 .type _Z23__device_stub__cos_main9cosParams,@function _Z23__device_stub__cos_main9cosParams: # @_Z23__device_stub__cos_main9cosParams .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 64(%rsp), %rax movq %rax, (%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d movq %rsp, %r9 movl $_Z8cos_main9cosParams, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z23__device_stub__cos_main9cosParams, .Lfunc_end0-_Z23__device_stub__cos_main9cosParams .cfi_endproc # -- End function .section .rodata.cst4,"aM",@progbits,4 .p2align 2, 0x0 # -- Begin function main .LCPI1_0: .long 0x40c90fda # float 6.283185 .LCPI1_1: .long 0x49742400 # float 1.0E+6 .text .globl main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $1560, %rsp # imm = 0x618 .cfi_def_cfa_offset 1616 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl $.Lstr, %edi callq puts@PLT movl $.Lstr.1, %edi callq puts@PLT leaq 12(%rsp), %rdi callq hipGetDeviceCount movl 12(%rsp), %esi cmpl $1, %esi jne .LBB1_2 # %bb.1: movl $.L.str.2, %edi movl $1, %esi jmp .LBB1_3 .LBB1_2: movl $.L.str.3, %edi .LBB1_3: xorl %eax, %eax callq printf cmpl $0, 12(%rsp) jle .LBB1_6 # %bb.4: # %.lr.ph.preheader leaq 88(%rsp), %rbx xorl %ebp, %ebp .p2align 4, 0x90 .LBB1_5: # %.lr.ph # =>This Inner Loop Header: Depth=1 movl $.L.str.4, %edi movl %ebp, %esi xorl %eax, %eax callq printf movq %rbx, %rdi movl %ebp, %esi callq hipGetDevicePropertiesR0600 movl $.L.str.5, %edi movq %rbx, %rsi xorl %eax, %eax callq printf incl %ebp cmpl 12(%rsp), %ebp jl .LBB1_5 .LBB1_6: # %._crit_edge movl $10, %edi callq putchar@PLT movq $0, 24(%rsp) movq $0, 16(%rsp) xorl %r15d, %r15d movl $.L.str.7, %edi movl $1000000, %esi # imm = 0xF4240 xorl %eax, %eax callq printf movl $4000000, %edi # imm = 0x3D0900 callq malloc movq %rax, %rbx movl $4000000, %edi # imm = 0x3D0900 callq malloc movq %rax, %r14 movss .LCPI1_0(%rip), %xmm0 # xmm0 = mem[0],zero,zero,zero movss .LCPI1_1(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero .p2align 4, 0x90 .LBB1_7: # =>This Inner Loop Header: Depth=1 xorps %xmm2, %xmm2 cvtsi2ss %r15d, %xmm2 mulss %xmm0, %xmm2 divss %xmm1, %xmm2 movss %xmm2, (%rbx,%r15,4) incq %r15 cmpq $1000000, %r15 # imm = 0xF4240 jne .LBB1_7 # %bb.8: movl $.L.str.8, %edi movl $1000000, %esi # imm = 0xF4240 xorl %eax, %eax callq printf leaq 16(%rsp), %rdi movl $4000000, %esi # imm = 0x3D0900 callq hipMalloc testl %eax, %eax je .LBB1_10 # %bb.9: movl $.L.str.9, %edi movl %eax, %esi xorl %eax, %eax callq printf jmp .LBB1_11 .LBB1_10: movl $.Lstr.2, %edi callq puts@PLT .LBB1_11: leaq 24(%rsp), %rdi movl $4000000, %esi # imm = 0x3D0900 callq hipMalloc testl %eax, %eax je .LBB1_13 # %bb.12: movl $.L.str.9, %edi movl %eax, %esi xorl %eax, %eax callq printf jmp .LBB1_14 .LBB1_13: movl $.Lstr.3, %edi callq puts@PLT .LBB1_14: movq 16(%rsp), %rdi movl $4000000, %edx # imm = 0x3D0900 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy testl %eax, %eax movl $.Lstr.4, %eax movl $.Lstr.5, %edi cmoveq %rax, %rdi callq puts@PLT movq 24(%rsp), %r12 movq 16(%rsp), %r13 movl $.Lstr.6, %edi callq puts@PLT movabsq $4294967297, %rdi # imm = 0x100000001 leaq 511(%rdi), %rdx xorl %r15d, %r15d movl $1, %esi movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_16 # %bb.15: movq %r13, 88(%rsp) movq %r12, 96(%rsp) movl $1000000, 104(%rsp) # imm = 0xF4240 leaq 88(%rsp), %rax movq %rax, 32(%rsp) leaq 72(%rsp), %rdi leaq 56(%rsp), %rsi leaq 48(%rsp), %rdx leaq 40(%rsp), %rcx callq __hipPopCallConfiguration movq 72(%rsp), %rsi movl 80(%rsp), %edx movq 56(%rsp), %rcx movl 64(%rsp), %r8d leaq 32(%rsp), %r9 movl $_Z8cos_main9cosParams, %edi pushq 40(%rsp) .cfi_adjust_cfa_offset 8 pushq 56(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_16: movl $.Lstr.7, %edi callq puts@PLT movq 24(%rsp), %rsi movl $4000000, %edx # imm = 0x3D0900 movq %r14, %rdi movl $2, %ecx callq hipMemcpy testl %eax, %eax movl $.Lstr.8, %eax movl $.Lstr.9, %edi cmoveq %rax, %rdi callq puts@PLT xorl %r12d, %r12d jmp .LBB1_17 .p2align 4, 0x90 .LBB1_19: # in Loop: Header=BB1_17 Depth=1 incq %r12 decl %r15d cmpq $1000000, %r12 # imm = 0xF4240 je .LBB1_20 .LBB1_17: # =>This Inner Loop Header: Depth=1 movl %r12d, %eax shrl $5, %eax imulq $175921861, %rax, %rax # imm = 0xA7C5AC5 shrq $39, %rax imull $100000, %eax, %eax # imm = 0x186A0 addl %r15d, %eax jne .LBB1_19 # %bb.18: # in Loop: Header=BB1_17 Depth=1 movss (%rbx,%r12,4), %xmm0 # xmm0 = mem[0],zero,zero,zero cvtss2sd %xmm0, %xmm0 movss (%r14,%r12,4), %xmm1 # xmm1 = mem[0],zero,zero,zero cvtss2sd %xmm1, %xmm1 movl $.L.str.18, %edi movb $2, %al callq printf jmp .LBB1_19 .LBB1_20: movl $.Lstr.10, %edi callq puts@PLT movq %rbx, %rdi callq free movq %r14, %rdi callq free movq 16(%rsp), %rdi callq hipFree movq 24(%rsp), %rdi callq hipFree movl $.Lstr.11, %edi callq puts@PLT xorl %eax, %eax addq $1560, %rsp # imm = 0x618 .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z8cos_main9cosParams, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z8cos_main9cosParams,@object # @_Z8cos_main9cosParams .section .rodata,"a",@progbits .globl _Z8cos_main9cosParams .p2align 3, 0x0 _Z8cos_main9cosParams: .quad _Z23__device_stub__cos_main9cosParams .size _Z8cos_main9cosParams, 8 .type .L.str.2,@object # @.str.2 .section .rodata.str1.1,"aMS",@progbits,1 .L.str.2: .asciz "There is %d CUDA device on your system.\n" .size .L.str.2, 41 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "There are %d CUDA devices on your system.\n" .size .L.str.3, 43 .type .L.str.4,@object # @.str.4 .L.str.4: .asciz "\nCUDA Device #%d\n" .size .L.str.4, 18 .type .L.str.5,@object # @.str.5 .L.str.5: .asciz "Name: %s\n" .size .L.str.5, 11 .type .L.str.7,@object # @.str.7 .L.str.7: .asciz "Allocating memory on the CPU... (n=%d)\n" .size .L.str.7, 40 .type .L.str.8,@object # @.str.8 .L.str.8: .asciz "Allocating memory on the GPU... (n=%d)\n" .size .L.str.8, 40 .type .L.str.9,@object # @.str.9 .L.str.9: .asciz " value = %d : Memory Allocation on GPU Device failed\n" .size .L.str.9, 54 .type .L.str.18,@object # @.str.18 .L.str.18: .asciz "cos(%f) = %f\n" .size .L.str.18, 14 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z8cos_main9cosParams" .size .L__unnamed_1, 22 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "\nBeginning CUDA example code on the CPU..." .size .Lstr, 43 .type .Lstr.1,@object # @str.1 .Lstr.1: .asciz "Getting some info about CUDA devices on your system..." .size .Lstr.1, 55 .type .Lstr.2,@object # @str.2 .Lstr.2: .asciz "done. Allocating more memory on the GPU..." .size .Lstr.2, 43 .type .Lstr.3,@object # @str.3 .Lstr.3: .asciz "done again. Copying stuff from host (CPU) to device (GPU)..." .size .Lstr.3, 61 .type .Lstr.4,@object # @str.4 .Lstr.4: .asciz "successful." .size .Lstr.4, 12 .type .Lstr.5,@object # @str.5 .Lstr.5: .asciz " Memory Copy from Host to Device failed." .size .Lstr.5, 41 .type .Lstr.6,@object # @str.6 .Lstr.6: .asciz "Launching kernel on GPU..." .size .Lstr.6, 27 .type .Lstr.7,@object # @str.7 .Lstr.7: .asciz "GPU computations finished. Copying result back to CPU..." .size .Lstr.7, 57 .type .Lstr.8,@object # @str.8 .Lstr.8: .asciz "Copy back successful. Printing result..." .size .Lstr.8, 41 .type .Lstr.9,@object # @str.9 .Lstr.9: .asciz " Memory Copy from Device to Host failed." .size .Lstr.9, 41 .type .Lstr.10,@object # @str.10 .Lstr.10: .asciz "Freeing memory..." .size .Lstr.10, 18 .type .Lstr.11,@object # @str.11 .Lstr.11: .asciz "\n\nFinished. :)\n" .size .Lstr.11, 16 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z23__device_stub__cos_main9cosParams .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z8cos_main9cosParams .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_
You are a helpful coding assistant specialized in converting from CUDA device assembly to AMD device assembly.
code for sm_80 Function : _Z8cos_main9cosParams .headerflags @"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM80 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM80)" /*0000*/ IMAD.MOV.U32 R1, RZ, RZ, c[0x0][0x28] ; /* 0x00000a00ff017624 */ /* 0x000fe400078e00ff */ /*0010*/ S2R R0, SR_TID.X ; /* 0x0000000000007919 */ /* 0x000e240000002100 */ /*0020*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x001fda0003f06270 */ /*0030*/ @P0 EXIT ; /* 0x000000000000094d */ /* 0x000fea0003800000 */ /*0040*/ ULDC.64 UR6, c[0x0][0x118] ; /* 0x0000460000067ab9 */ /* 0x000fe40000000a00 */ /*0050*/ IMAD.MOV.U32 R7, RZ, RZ, 0x4 ; /* 0x00000004ff077424 */ /* 0x000fc800078e00ff */ /*0060*/ IMAD.WIDE R6, R0, R7, c[0x0][0x160] ; /* 0x0000580000067625 */ /* 0x000fca00078e0207 */ /*0070*/ LDG.E R3, [R6.64] ; /* 0x0000000606037981 */ /* 0x001ea2000c1e1900 */ /*0080*/ BSSY B0, 0x7f0 ; /* 0x0000076000007945 */ /* 0x000fe20003800000 */ /*0090*/ FMUL R4, R3.reuse, 0.63661974668502807617 ; /* 0x3f22f98303047820 */ /* 0x044fe20000400000 */ /*00a0*/ FSETP.GE.AND P0, PT, |R3|, 105615, PT ; /* 0x47ce47800300780b */ /* 0x000fca0003f06200 */ /*00b0*/ F2I.NTZ R4, R4 ; /* 0x0000000400047305 */ /* 0x000e300000203100 */ /*00c0*/ I2F R8, R4 ; /* 0x0000000400087306 */ /* 0x001e240000201400 */ /*00d0*/ FFMA R5, R8, -1.5707962512969970703, R3 ; /* 0xbfc90fda08057823 */ /* 0x001fc80000000003 */ /*00e0*/ FFMA R9, R8, -7.5497894158615963534e-08, R5 ; /* 0xb3a2216808097823 */ /* 0x000fe20000000005 */ /*00f0*/ SHF.R.S32.HI R5, RZ, 0x1f, R0 ; /* 0x0000001fff057819 */ /* 0x000fc60000011400 */ /*0100*/ FFMA R17, R8, -5.3903029534742383927e-15, R9 ; /* 0xa7c234c508117823 */ /* 0x000fe20000000009 */ /*0110*/ @!P0 BRA 0x7e0 ; /* 0x000006c000008947 */ /* 0x000fea0003800000 */ /*0120*/ FSETP.NEU.AND P0, PT, |R3|, +INF , PT ; /* 0x7f8000000300780b */ /* 0x000fda0003f0d200 */ /*0130*/ @!P0 BRA 0x7c0 ; /* 0x0000068000008947 */ /* 0x000fea0003800000 */ /*0140*/ SHF.R.U32.HI R4, RZ, 0x17, R3 ; /* 0x00000017ff047819 */ /* 0x000fe20000011603 */ /*0150*/ IMAD.SHL.U32 R7, R3, 0x100, RZ ; /* 0x0000010003077824 */ /* 0x000fe200078e00ff */ /*0160*/ UMOV UR4, URZ ; /* 0x0000003f00047c82 */ /* 0x000fe20008000000 */ /*0170*/ IMAD.MOV.U32 R10, RZ, RZ, RZ ; /* 0x000000ffff0a7224 */ /* 0x000fe200078e00ff */ /*0180*/ LOP3.LUT R6, R4, 0xff, RZ, 0xc0, !PT ; /* 0x000000ff04067812 */ /* 0x000fe200078ec0ff */ /*0190*/ IMAD.MOV.U32 R4, RZ, RZ, RZ ; /* 0x000000ffff047224 */ /* 0x000fe200078e00ff */ /*01a0*/ LOP3.LUT R17, R7, 0x80000000, RZ, 0xfc, !PT ; /* 0x8000000007117812 */ /* 0x000fe200078efcff */ /*01b0*/ IMAD.MOV.U32 R19, RZ, RZ, RZ ; /* 0x000000ffff137224 */ /* 0x000fe200078e00ff */ /*01c0*/ IADD3 R11, R6, -0x80, RZ ; /* 0xffffff80060b7810 */ /* 0x000fe20007ffe0ff */ /*01d0*/ ULDC.64 UR8, c[0x4][0x0] ; /* 0x0100000000087ab9 */ /* 0x000fc60000000a00 */ /*01e0*/ SHF.R.U32.HI R18, RZ, 0x5, R11 ; /* 0x00000005ff127819 */ /* 0x000fe4000001160b */ /*01f0*/ IMAD.U32 R8, RZ, RZ, UR8 ; /* 0x00000008ff087e24 */ /* 0x000fe4000f8e00ff */ /*0200*/ IMAD.U32 R9, RZ, RZ, UR9 ; /* 0x00000009ff097e24 */ /* 0x000fca000f8e00ff */ /*0210*/ LDG.E.CONSTANT R6, [R8.64] ; /* 0x0000000608067981 */ /* 0x000ea2000c1e9900 */ /*0220*/ IADD3 R10, R10, 0x1, RZ ; /* 0x000000010a0a7810 */ /* 0x000fe20007ffe0ff */ /*0230*/ UIADD3 UR8, UP0, UR8, 0x4, URZ ; /* 0x0000000408087890 */ /* 0x000fe2000ff1e03f */ /*0240*/ ISETP.EQ.AND P0, PT, R19.reuse, RZ, PT ; /* 0x000000ff1300720c */ /* 0x040fe40003f02270 */ /*0250*/ ISETP.EQ.AND P5, PT, R19.reuse, 0x4, PT ; /* 0x000000041300780c */ /* 0x040fe20003fa2270 */ /*0260*/ UIADD3.X UR9, URZ, UR9, URZ, UP0, !UPT ; /* 0x000000093f097290 */ /* 0x000fe200087fe43f */ /*0270*/ ISETP.EQ.AND P4, PT, R19.reuse, 0x8, PT ; /* 0x000000081300780c */ /* 0x040fe40003f82270 */ /*0280*/ ISETP.EQ.AND P3, PT, R19.reuse, 0xc, PT ; /* 0x0000000c1300780c */ /* 0x040fe40003f62270 */ /*0290*/ ISETP.EQ.AND P2, PT, R19, 0x10, PT ; /* 0x000000101300780c */ /* 0x000fc40003f42270 */ /*02a0*/ ISETP.EQ.AND P1, PT, R19.reuse, 0x14, PT ; /* 0x000000141300780c */ /* 0x040fe40003f22270 */ /*02b0*/ IADD3 R19, R19, 0x4, RZ ; /* 0x0000000413137810 */ /* 0x000fe20007ffe0ff */ /*02c0*/ IMAD.WIDE.U32 R6, R6, R17, RZ ; /* 0x0000001106067225 */ /* 0x004fca00078e00ff */ /*02d0*/ IADD3 R6, P6, R6, R4, RZ ; /* 0x0000000406067210 */ /* 0x000fc80007fde0ff */ /*02e0*/ IADD3.X R4, R7, UR4, RZ, P6, !PT ; /* 0x0000000407047c10 */ /* 0x000fe2000b7fe4ff */ /*02f0*/ @P0 IMAD.MOV.U32 R2, RZ, RZ, R6.reuse ; /* 0x000000ffff020224 */ /* 0x100fe200078e0006 */ /*0300*/ ISETP.NE.AND P6, PT, R10, 0x6, PT ; /* 0x000000060a00780c */ /* 0x000fe20003fc5270 */ /*0310*/ @P5 IMAD.MOV.U32 R12, RZ, RZ, R6.reuse ; /* 0x000000ffff0c5224 */ /* 0x100fe400078e0006 */ /*0320*/ @P4 IMAD.MOV.U32 R13, RZ, RZ, R6.reuse ; /* 0x000000ffff0d4224 */ /* 0x100fe400078e0006 */ /*0330*/ @P3 IMAD.MOV.U32 R14, RZ, RZ, R6.reuse ; /* 0x000000ffff0e3224 */ /* 0x100fe400078e0006 */ /*0340*/ @P2 IMAD.MOV.U32 R15, RZ, RZ, R6.reuse ; /* 0x000000ffff0f2224 */ /* 0x100fe400078e0006 */ /*0350*/ @P1 IMAD.MOV.U32 R16, RZ, RZ, R6 ; /* 0x000000ffff101224 */ /* 0x000fc800078e0006 */ /*0360*/ @P6 BRA 0x1f0 ; /* 0xfffffe8000006947 */ /* 0x000fea000383ffff */ /*0370*/ IADD3 R6, -R18, 0x6, RZ ; /* 0x0000000612067810 */ /* 0x000fe20007ffe1ff */ /*0380*/ BSSY B1, 0x6b0 ; /* 0x0000032000017945 */ /* 0x000fe80003800000 */ /*0390*/ IMAD.SHL.U32 R6, R6, 0x4, RZ ; /* 0x0000000406067824 */ /* 0x000fca00078e00ff */ /*03a0*/ ISETP.EQ.AND P0, PT, R6.reuse, RZ, PT ; /* 0x000000ff0600720c */ /* 0x040fe40003f02270 */ /*03b0*/ ISETP.EQ.AND P3, PT, R6.reuse, 0x4, PT ; /* 0x000000040600780c */ /* 0x040fe40003f62270 */ /*03c0*/ ISETP.EQ.AND P4, PT, R6.reuse, 0x8, PT ; /* 0x000000080600780c */ /* 0x040fe40003f82270 */ /*03d0*/ ISETP.EQ.AND P2, PT, R6.reuse, 0xc, PT ; /* 0x0000000c0600780c */ /* 0x040fe40003f42270 */ /*03e0*/ ISETP.EQ.AND P1, PT, R6, 0x10, PT ; /* 0x000000100600780c */ /* 0x000fca0003f22270 */ /*03f0*/ @P0 IMAD.MOV.U32 R7, RZ, RZ, R2.reuse ; /* 0x000000ffff070224 */ /* 0x100fe200078e0002 */ /*0400*/ ISETP.EQ.AND P0, PT, R6.reuse, 0x14, PT ; /* 0x000000140600780c */ /* 0x040fe20003f02270 */ /*0410*/ @P3 IMAD.MOV.U32 R8, RZ, RZ, R2 ; /* 0x000000ffff083224 */ /* 0x000fe400078e0002 */ /*0420*/ @P3 IMAD.MOV.U32 R7, RZ, RZ, R12.reuse ; /* 0x000000ffff073224 */ /* 0x100fe200078e000c */ /*0430*/ ISETP.EQ.AND P3, PT, R6.reuse, 0x18, PT ; /* 0x000000180600780c */ /* 0x040fe20003f62270 */ /*0440*/ @P4 IMAD.MOV.U32 R8, RZ, RZ, R12 ; /* 0x000000ffff084224 */ /* 0x000fe400078e000c */ /*0450*/ @P4 IMAD.MOV.U32 R7, RZ, RZ, R13.reuse ; /* 0x000000ffff074224 */ /* 0x100fe200078e000d */ /*0460*/ ISETP.EQ.AND P4, PT, R6, 0x1c, PT ; /* 0x0000001c0600780c */ /* 0x000fe20003f82270 */ /*0470*/ @P2 IMAD.MOV.U32 R8, RZ, RZ, R13 ; /* 0x000000ffff082224 */ /* 0x000fc400078e000d */ /*0480*/ @P2 IMAD.MOV.U32 R7, RZ, RZ, R14.reuse ; /* 0x000000ffff072224 */ /* 0x100fe400078e000e */ /*0490*/ @P1 IMAD.MOV.U32 R8, RZ, RZ, R14 ; /* 0x000000ffff081224 */ /* 0x000fe400078e000e */ /*04a0*/ @P1 IMAD.MOV.U32 R7, RZ, RZ, R15.reuse ; /* 0x000000ffff071224 */ /* 0x100fe200078e000f */ /*04b0*/ LOP3.LUT P1, R11, R11, 0x1f, RZ, 0xc0, !PT ; /* 0x0000001f0b0b7812 */ /* 0x000fe2000782c0ff */ /*04c0*/ @P0 IMAD.MOV.U32 R7, RZ, RZ, R16 ; /* 0x000000ffff070224 */ /* 0x000fe400078e0010 */ /*04d0*/ @P3 IMAD.MOV.U32 R7, RZ, RZ, R4 ; /* 0x000000ffff073224 */ /* 0x000fe400078e0004 */ /*04e0*/ @P0 IMAD.MOV.U32 R8, RZ, RZ, R15 ; /* 0x000000ffff080224 */ /* 0x000fc400078e000f */ /*04f0*/ @P3 IMAD.MOV.U32 R8, RZ, RZ, R16 ; /* 0x000000ffff083224 */ /* 0x000fe400078e0010 */ /*0500*/ @P4 IMAD.MOV.U32 R8, RZ, RZ, R4 ; /* 0x000000ffff084224 */ /* 0x000fe400078e0004 */ /*0510*/ IMAD.MOV.U32 R6, RZ, RZ, R7 ; /* 0x000000ffff067224 */ /* 0x000fe400078e0007 */ /*0520*/ @!P1 BRA 0x6a0 ; /* 0x0000017000009947 */ /* 0x000fea0003800000 */ /*0530*/ IADD3 R7, -R18, 0x4, RZ ; /* 0x0000000412077810 */ /* 0x000fe40007ffe1ff */ /*0540*/ SHF.L.U32 R9, R6, R11, RZ ; /* 0x0000000b06097219 */ /* 0x000fc600000006ff */ /*0550*/ IMAD.SHL.U32 R7, R7, 0x4, RZ ; /* 0x0000000407077824 */ /* 0x000fca00078e00ff */ /*0560*/ ISETP.EQ.AND P0, PT, R7.reuse, RZ, PT ; /* 0x000000ff0700720c */ /* 0x040fe40003f02270 */ /*0570*/ ISETP.EQ.AND P1, PT, R7.reuse, 0x4, PT ; /* 0x000000040700780c */ /* 0x040fe40003f22270 */ /*0580*/ ISETP.EQ.AND P2, PT, R7.reuse, 0x8, PT ; /* 0x000000080700780c */ /* 0x040fe40003f42270 */ /*0590*/ ISETP.EQ.AND P3, PT, R7, 0xc, PT ; /* 0x0000000c0700780c */ /* 0x000fce0003f62270 */ /*05a0*/ @P0 IMAD.MOV.U32 R10, RZ, RZ, R2 ; /* 0x000000ffff0a0224 */ /* 0x000fe200078e0002 */ /*05b0*/ ISETP.EQ.AND P0, PT, R7.reuse, 0x10, PT ; /* 0x000000100700780c */ /* 0x040fe20003f02270 */ /*05c0*/ @P1 IMAD.MOV.U32 R10, RZ, RZ, R12 ; /* 0x000000ffff0a1224 */ /* 0x000fe200078e000c */ /*05d0*/ ISETP.EQ.AND P1, PT, R7.reuse, 0x14, PT ; /* 0x000000140700780c */ /* 0x040fe20003f22270 */ /*05e0*/ @P2 IMAD.MOV.U32 R10, RZ, RZ, R13 ; /* 0x000000ffff0a2224 */ /* 0x000fe200078e000d */ /*05f0*/ ISETP.EQ.AND P2, PT, R7, 0x18, PT ; /* 0x000000180700780c */ /* 0x000fe20003f42270 */ /*0600*/ @P3 IMAD.MOV.U32 R10, RZ, RZ, R14 ; /* 0x000000ffff0a3224 */ /* 0x000fe200078e000e */ /*0610*/ IADD3 R7, -R11, 0x20, RZ ; /* 0x000000200b077810 */ /* 0x000fc80007ffe1ff */ /*0620*/ SHF.R.U32.HI R6, RZ, R7, R8 ; /* 0x00000007ff067219 */ /* 0x000fc60000011608 */ /*0630*/ @P0 IMAD.MOV.U32 R10, RZ, RZ, R15 ; /* 0x000000ffff0a0224 */ /* 0x000fe400078e000f */ /*0640*/ @P1 IMAD.MOV.U32 R10, RZ, RZ, R16 ; /* 0x000000ffff0a1224 */ /* 0x000fe400078e0010 */ /*0650*/ @P2 IMAD.MOV.U32 R10, RZ, RZ, R4 ; /* 0x000000ffff0a2224 */ /* 0x000fe200078e0004 */ /*0660*/ SHF.L.U32 R4, R8, R11, RZ ; /* 0x0000000b08047219 */ /* 0x000fe200000006ff */ /*0670*/ IMAD.IADD R6, R6, 0x1, R9 ; /* 0x0000000106067824 */ /* 0x000fc600078e0209 */ /*0680*/ SHF.R.U32.HI R7, RZ, R7, R10 ; /* 0x00000007ff077219 */ /* 0x000fca000001160a */ /*0690*/ IMAD.IADD R8, R7, 0x1, R4 ; /* 0x0000000107087824 */ /* 0x000fe400078e0204 */ /*06a0*/ BSYNC B1 ; /* 0x0000000000017941 */ /* 0x000fea0003800000 */ /*06b0*/ SHF.L.U32.HI R11, R8.reuse, 0x2, R6 ; /* 0x00000002080b7819 */ /* 0x040fe20000010606 */ /*06c0*/ IMAD.SHL.U32 R10, R8, 0x4, RZ ; /* 0x00000004080a7824 */ /* 0x000fe200078e00ff */ /*06d0*/ LOP3.LUT P1, R3, R3, 0x80000000, RZ, 0xc0, !PT ; /* 0x8000000003037812 */ /* 0x000fe4000782c0ff */ /*06e0*/ SHF.R.U32.HI R7, RZ, 0x1f, R11 ; /* 0x0000001fff077819 */ /* 0x000fc8000001160b */ /*06f0*/ ISETP.NE.AND P0, PT, R7, RZ, PT ; /* 0x000000ff0700720c */ /* 0x000fe40003f05270 */ /*0700*/ LEA.HI R4, R6, R7, RZ, 0x2 ; /* 0x0000000706047211 */ /* 0x000fd600078f10ff */ /*0710*/ @P0 LOP3.LUT R11, RZ, R11, RZ, 0x33, !PT ; /* 0x0000000bff0b0212 */ /* 0x000fe400078e33ff */ /*0720*/ @P0 LOP3.LUT R10, RZ, R10, RZ, 0x33, !PT ; /* 0x0000000aff0a0212 */ /* 0x000fe400078e33ff */ /*0730*/ @P0 LOP3.LUT R3, R3, 0x80000000, RZ, 0x3c, !PT ; /* 0x8000000003030812 */ /* 0x000fe400078e3cff */ /*0740*/ I2F.F64.S64 R8, R10 ; /* 0x0000000a00087312 */ /* 0x000e240000301c00 */ /*0750*/ ISETP.NE.AND P0, PT, R3, RZ, PT ; /* 0x000000ff0300720c */ /* 0x000fe20003f05270 */ /*0760*/ IMAD.MOV R3, RZ, RZ, -R4 ; /* 0x000000ffff037224 */ /* 0x000fc800078e0a04 */ /*0770*/ @P1 IMAD.MOV.U32 R4, RZ, RZ, R3 ; /* 0x000000ffff041224 */ /* 0x000fe200078e0003 */ /*0780*/ DMUL R8, R8, c[0x2][0x0] ; /* 0x0080000008087a28 */ /* 0x001e140000000000 */ /*0790*/ F2F.F32.F64 R8, R8 ; /* 0x0000000800087310 */ /* 0x001e240000301000 */ /*07a0*/ FSEL R17, R8, -R8, !P0 ; /* 0x8000000808117208 */ /* 0x001fe20004000000 */ /*07b0*/ BRA 0x7e0 ; /* 0x0000002000007947 */ /* 0x000fea0003800000 */ /*07c0*/ FMUL R17, RZ, R3 ; /* 0x00000003ff117220 */ /* 0x000fe40000400000 */ /*07d0*/ IMAD.MOV.U32 R4, RZ, RZ, RZ ; /* 0x000000ffff047224 */ /* 0x000fe400078e00ff */ /*07e0*/ BSYNC B0 ; /* 0x0000000000007941 */ /* 0x000fea0003800000 */ /*07f0*/ IADD3 R8, R4, 0x1, RZ ; /* 0x0000000104087810 */ /* 0x000fe20007ffe0ff */ /*0800*/ IMAD.MOV.U32 R4, RZ, RZ, 0x3c0885e4 ; /* 0x3c0885e4ff047424 */ /* 0x000fe400078e00ff */ /*0810*/ FMUL R9, R17, R17 ; /* 0x0000001111097220 */ /* 0x000fe20000400000 */ /*0820*/ LOP3.LUT P1, RZ, R8.reuse, 0x1, RZ, 0xc0, !PT ; /* 0x0000000108ff7812 */ /* 0x040fe2000782c0ff */ /*0830*/ IMAD.MOV.U32 R7, RZ, RZ, 0x3e2aaaa8 ; /* 0x3e2aaaa8ff077424 */ /* 0x000fe200078e00ff */ /*0840*/ LOP3.LUT P0, RZ, R8, 0x2, RZ, 0xc0, !PT ; /* 0x0000000208ff7812 */ /* 0x000fe2000780c0ff */ /*0850*/ IMAD.MOV.U32 R3, RZ, RZ, -0x46b2bead ; /* 0xb94d4153ff037424 */ /* 0x000fe200078e00ff */ /*0860*/ FSEL R4, R4, 0.041666727513074874878, !P1 ; /* 0x3d2aaabb04047808 */ /* 0x000fc40004800000 */ /*0870*/ FSEL R17, R17, 1, !P1 ; /* 0x3f80000011117808 */ /* 0x000fca0004800000 */ /*0880*/ FFMA R8, R17, R9, RZ ; /* 0x0000000911087223 */ /* 0x000fe400000000ff */ /*0890*/ @P1 IMAD.MOV.U32 R6, RZ, RZ, 0x37cbac00 ; /* 0x37cbac00ff061424 */ /* 0x000fc800078e00ff */ /*08a0*/ @P1 FFMA R3, R9, R6, -0.0013887860113754868507 ; /* 0xbab607ed09031423 */ /* 0x000fe20000000006 */ /*08b0*/ FSEL R6, -R7, -0.4999999701976776123, !P1 ; /* 0xbeffffff07067808 */ /* 0x000fc60004800100 */ /*08c0*/ FFMA R3, R9, R3, R4 ; /* 0x0000000309037223 */ /* 0x000fe20000000004 */ /*08d0*/ LEA R4, P1, R0, c[0x0][0x168], 0x2 ; /* 0x00005a0000047a11 */ /* 0x000fc600078210ff */ /*08e0*/ FFMA R3, R9, R3, R6 ; /* 0x0000000309037223 */ /* 0x000fe20000000006 */ /*08f0*/ LEA.HI.X R5, R0.reuse, c[0x0][0x16c], R5, 0x2, P1 ; /* 0x00005b0000057a11 */ /* 0x040fe400008f1405 */ /*0900*/ IADD3 R0, R0, 0x200, RZ ; /* 0x0000020000007810 */ /* 0x000fe20007ffe0ff */ /*0910*/ FFMA R3, R3, R8, R17 ; /* 0x0000000803037223 */ /* 0x000fc80000000011 */ /*0920*/ @P0 FFMA R3, R3, -1, RZ ; /* 0xbf80000003030823 */ /* 0x000fe200000000ff */ /*0930*/ ISETP.GE.AND P0, PT, R0, c[0x0][0x170], PT ; /* 0x00005c0000007a0c */ /* 0x000fc80003f06270 */ /*0940*/ STG.E [R4.64], R3 ; /* 0x0000000304007986 */ /* 0x0001f2000c101906 */ /*0950*/ @!P0 BRA 0x50 ; /* 0xfffff6f000008947 */ /* 0x000fea000383ffff */ /*0960*/ EXIT ; /* 0x000000000000794d */ /* 0x000fea0003800000 */ /*0970*/ BRA 0x970; /* 0xfffffff000007947 */ /* 0x000fc0000383ffff */ /*0980*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*0990*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09a0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09b0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09c0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09d0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09e0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ /*09f0*/ NOP; /* 0x0000000000007918 */ /* 0x000fc00000000000 */ ..........
.text .amdgcn_target "amdgcn-amd-amdhsa--gfx1100" .protected _Z8cos_main9cosParams .globl _Z8cos_main9cosParams .p2align 8 .type _Z8cos_main9cosParams,@function _Z8cos_main9cosParams: s_load_b32 s2, s[0:1], 0x10 s_mov_b32 s3, exec_lo s_waitcnt lgkmcnt(0) v_cmpx_gt_i32_e64 s2, v0 s_cbranch_execz .LBB0_7 s_load_b128 s[4:7], s[0:1], 0x0 v_lshlrev_b32_e32 v3, 2, v0 s_mov_b32 s3, 0 s_waitcnt lgkmcnt(0) s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_co_u32 v1, s0, s4, v3 v_add_co_ci_u32_e64 v2, null, s5, 0, s0 v_add_co_u32 v3, s0, s6, v3 s_delay_alu instid0(VALU_DEP_1) v_add_co_ci_u32_e64 v4, null, s7, 0, s0 s_mov_b32 s4, 0x7fffff s_mov_b32 s5, 0xb94c1982 s_mov_b32 s6, 0x37d75334 s_branch .LBB0_3 .LBB0_2: s_or_b32 exec_lo, exec_lo, s0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) v_dual_mul_f32 v8, v6, v6 :: v_dual_and_b32 v11, 1, v7 v_add_nc_u32_e32 v0, 0x200, v0 v_lshlrev_b32_e32 v7, 30, v7 v_fmaak_f32 v9, s5, v8, 0x3c0881c4 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3) v_cmp_eq_u32_e32 vcc_lo, 0, v11 v_and_b32_e32 v7, 0x80000000, v7 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmaak_f32 v9, v8, v9, 0xbe2aaa9d v_dual_fmaak_f32 v10, s6, v8, 0xbab64f3b :: v_dual_mul_f32 v9, v8, v9 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) v_fmaak_f32 v10, v8, v10, 0x3d2aabf7 v_fmac_f32_e32 v6, v6, v9 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmaak_f32 v10, v8, v10, 0xbf000004 v_fma_f32 v8, v8, v10, 1.0 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_cndmask_b32_e64 v6, -v6, v8, vcc_lo v_cmp_class_f32_e64 vcc_lo, v5, 0x1f8 v_xor_b32_e32 v6, v7, v6 s_delay_alu instid0(VALU_DEP_1) v_cndmask_b32_e32 v5, 0x7fc00000, v6, vcc_lo v_add_co_u32 v1, vcc_lo, v1, 0x800 v_add_co_ci_u32_e32 v2, vcc_lo, 0, v2, vcc_lo v_cmp_le_i32_e32 vcc_lo, s2, v0 global_store_b32 v[3:4], v5, off v_add_co_u32 v3, s0, v3, 0x800 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) v_add_co_ci_u32_e64 v4, s0, 0, v4, s0 s_or_b32 s3, vcc_lo, s3 s_and_not1_b32 exec_lo, exec_lo, s3 s_cbranch_execz .LBB0_7 .LBB0_3: global_load_b32 v5, v[1:2], off s_mov_b32 s1, exec_lo s_waitcnt vmcnt(0) v_cmpx_ngt_f32_e64 0x48000000, |v5| s_xor_b32 s7, exec_lo, s1 s_cbranch_execz .LBB0_5 v_dual_mov_b32 v8, 0 :: v_dual_and_b32 v13, 0x7fffffff, v5 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_and_or_b32 v16, v13, s4, 0x800000 v_lshrrev_b32_e32 v13, 23, v13 v_mad_u64_u32 v[6:7], null, v16, 0xfe5163ab, 0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_add_nc_u32_e32 v14, 0xffffff88, v13 v_cmp_lt_u32_e32 vcc_lo, 63, v14 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_mad_u64_u32 v[9:10], null, v16, 0x3c439041, v[7:8] v_cndmask_b32_e64 v15, 0, 0xffffffc0, vcc_lo v_mov_b32_e32 v7, v10 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_nc_u32_e32 v15, v15, v14 v_mad_u64_u32 v[10:11], null, v16, 0xdb629599, v[7:8] s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_cmp_lt_u32_e64 s0, 31, v15 v_cndmask_b32_e64 v17, 0, 0xffffffe0, s0 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) v_dual_mov_b32 v7, v11 :: v_dual_cndmask_b32 v6, v10, v6 v_add_nc_u32_e32 v17, v17, v15 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_mad_u64_u32 v[11:12], null, v16, 0xf534ddc0, v[7:8] v_cmp_lt_u32_e64 s1, 31, v17 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) v_mov_b32_e32 v7, v12 v_cndmask_b32_e32 v9, v11, v9, vcc_lo s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_mad_u64_u32 v[12:13], null, v16, 0xfc2757d1, v[7:8] v_cndmask_b32_e64 v6, v9, v6, s0 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_mov_b32_e32 v7, v13 v_mad_u64_u32 v[13:14], null, v16, 0x4e441529, v[7:8] s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_mov_b32_e32 v7, v14 v_mad_u64_u32 v[14:15], null, v16, 0xa2f9836e, v[7:8] v_cndmask_b32_e64 v7, 0, 0xffffffe0, s1 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_dual_cndmask_b32 v8, v13, v11 :: v_dual_add_nc_u32 v7, v7, v17 v_dual_cndmask_b32 v14, v14, v12 :: v_dual_cndmask_b32 v13, v15, v13 v_cndmask_b32_e32 v12, v12, v10, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_cmp_eq_u32_e32 vcc_lo, 0, v7 v_cndmask_b32_e64 v11, v14, v8, s0 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_cndmask_b32_e64 v13, v13, v14, s0 v_cndmask_b32_e64 v8, v8, v12, s0 v_sub_nc_u32_e32 v14, 32, v7 v_cndmask_b32_e64 v12, v12, v9, s0 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_cndmask_b32_e64 v13, v13, v11, s1 v_cndmask_b32_e64 v11, v11, v8, s1 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_cndmask_b32_e64 v8, v8, v12, s1 v_cndmask_b32_e64 v6, v12, v6, s1 v_alignbit_b32 v15, v13, v11, v14 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) v_alignbit_b32 v10, v11, v8, v14 v_cndmask_b32_e32 v7, v15, v13, vcc_lo s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v13, v8, v6, v14 v_cndmask_b32_e32 v9, v10, v11, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_bfe_u32 v10, v7, 29, 1 v_cndmask_b32_e32 v8, v13, v8, vcc_lo s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v11, v7, v9, 30 v_sub_nc_u32_e32 v12, 0, v10 s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) v_alignbit_b32 v9, v9, v8, 30 v_alignbit_b32 v6, v8, v6, 30 v_xor_b32_e32 v11, v11, v12 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_xor_b32_e32 v8, v9, v12 v_xor_b32_e32 v6, v6, v12 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) v_clz_i32_u32_e32 v13, v11 v_min_u32_e32 v13, 32, v13 s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v9, 31, v13 v_lshlrev_b32_e32 v15, 23, v13 v_alignbit_b32 v11, v11, v8, v9 v_alignbit_b32 v6, v8, v6, v9 v_lshrrev_b32_e32 v9, 29, v7 v_lshrrev_b32_e32 v7, 30, v7 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v8, v11, v6, 9 v_lshlrev_b32_e32 v9, 31, v9 v_lshrrev_b32_e32 v11, 9, v11 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) v_add_nc_u32_e32 v7, v10, v7 v_clz_i32_u32_e32 v12, v8 s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2) v_or_b32_e32 v14, 0.5, v9 v_min_u32_e32 v12, 32, v12 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v14, v14, v15 v_sub_nc_u32_e32 v16, 31, v12 s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3) v_alignbit_b32 v6, v8, v6, v16 v_or_b32_e32 v8, v11, v14 v_add_lshl_u32 v11, v12, v13, 23 s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) v_lshrrev_b32_e32 v6, 9, v6 v_mul_f32_e32 v12, 0x3fc90fda, v8 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_sub_nc_u32_e32 v6, v6, v11 v_fma_f32 v11, v8, 0x3fc90fda, -v12 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) v_add_nc_u32_e32 v6, 0x33000000, v6 v_fmac_f32_e32 v11, 0x33a22168, v8 s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) v_or_b32_e32 v6, v6, v9 v_fmac_f32_e32 v11, 0x3fc90fda, v6 s_delay_alu instid0(VALU_DEP_1) v_add_f32_e32 v6, v12, v11 .LBB0_5: s_and_not1_saveexec_b32 s0, s7 s_cbranch_execz .LBB0_2 v_mul_f32_e64 v6, 0x3f22f983, |v5| s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_rndne_f32_e32 v7, v6 v_fma_f32 v6, v7, 0xbfc90fda, |v5| s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) v_fmac_f32_e32 v6, 0xb3a22168, v7 v_fmac_f32_e32 v6, 0xa7c234c4, v7 v_cvt_i32_f32_e32 v7, v7 s_branch .LBB0_2 .LBB0_7: s_nop 0 s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) s_endpgm .section .rodata,"a",@progbits .p2align 6, 0x0 .amdhsa_kernel _Z8cos_main9cosParams .amdhsa_group_segment_fixed_size 0 .amdhsa_private_segment_fixed_size 0 .amdhsa_kernarg_size 24 .amdhsa_user_sgpr_count 15 .amdhsa_user_sgpr_dispatch_ptr 0 .amdhsa_user_sgpr_queue_ptr 0 .amdhsa_user_sgpr_kernarg_segment_ptr 1 .amdhsa_user_sgpr_dispatch_id 0 .amdhsa_user_sgpr_private_segment_size 0 .amdhsa_wavefront_size32 1 .amdhsa_uses_dynamic_stack 0 .amdhsa_enable_private_segment 0 .amdhsa_system_sgpr_workgroup_id_x 1 .amdhsa_system_sgpr_workgroup_id_y 0 .amdhsa_system_sgpr_workgroup_id_z 0 .amdhsa_system_sgpr_workgroup_info 0 .amdhsa_system_vgpr_workitem_id 0 .amdhsa_next_free_vgpr 18 .amdhsa_next_free_sgpr 8 .amdhsa_float_round_mode_32 0 .amdhsa_float_round_mode_16_64 0 .amdhsa_float_denorm_mode_32 3 .amdhsa_float_denorm_mode_16_64 3 .amdhsa_dx10_clamp 1 .amdhsa_ieee_mode 1 .amdhsa_fp16_overflow 0 .amdhsa_workgroup_processor_mode 1 .amdhsa_memory_ordered 1 .amdhsa_forward_progress 0 .amdhsa_shared_vgpr_count 0 .amdhsa_exception_fp_ieee_invalid_op 0 .amdhsa_exception_fp_denorm_src 0 .amdhsa_exception_fp_ieee_div_zero 0 .amdhsa_exception_fp_ieee_overflow 0 .amdhsa_exception_fp_ieee_underflow 0 .amdhsa_exception_fp_ieee_inexact 0 .amdhsa_exception_int_div_zero 0 .end_amdhsa_kernel .text .Lfunc_end0: .size _Z8cos_main9cosParams, .Lfunc_end0-_Z8cos_main9cosParams .section .AMDGPU.csdata,"",@progbits .text .p2alignl 7, 3214868480 .fill 96, 4, 3214868480 .type __hip_cuid_,@object .section .bss,"aw",@nobits .globl __hip_cuid_ __hip_cuid_: .byte 0 .size __hip_cuid_, 1 .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym __hip_cuid_ .amdgpu_metadata --- amdhsa.kernels: - .args: - .offset: 0 .size: 24 .value_kind: by_value .group_segment_fixed_size: 0 .kernarg_segment_align: 8 .kernarg_segment_size: 24 .language: OpenCL C .language_version: - 2 - 0 .max_flat_workgroup_size: 1024 .name: _Z8cos_main9cosParams .private_segment_fixed_size: 0 .sgpr_count: 10 .sgpr_spill_count: 0 .symbol: _Z8cos_main9cosParams.kd .uniform_work_group_size: 1 .uses_dynamic_stack: false .vgpr_count: 18 .vgpr_spill_count: 0 .wavefront_size: 32 .workgroup_processor_mode: 1 amdhsa.target: amdgcn-amd-amdhsa--gfx1100 amdhsa.version: - 1 - 2 ... .end_amdgpu_metadata
You are a helpful coding assistant specialized in converting from CUDA host assembly to AMD host assembly.
.file "tmpxft_000d8b16_00000000-6_CUDA_example_01.cudafe1.cpp" .text #APP #NO_APP .type _ZL26__cudaUnregisterBinaryUtilv, @function _ZL26__cudaUnregisterBinaryUtilv: .LFB2060: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaUnregisterFatBinary@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2060: .size _ZL26__cudaUnregisterBinaryUtilv, .-_ZL26__cudaUnregisterBinaryUtilv .globl _Z35__device_stub__Z8cos_main9cosParamsR9cosParams .type _Z35__device_stub__Z8cos_main9cosParamsR9cosParams, @function _Z35__device_stub__Z8cos_main9cosParamsR9cosParams: .LFB2082: .cfi_startproc endbr64 subq $88, %rsp .cfi_def_cfa_offset 96 movq %fs:40, %rax movq %rax, 72(%rsp) xorl %eax, %eax movq %rdi, 64(%rsp) movl $1, 16(%rsp) movl $1, 20(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $1, 32(%rsp) movl $1, 36(%rsp) leaq 8(%rsp), %rcx movq %rsp, %rdx leaq 28(%rsp), %rsi leaq 16(%rsp), %rdi call __cudaPopCallConfiguration@PLT testl %eax, %eax je .L7 .L3: movq 72(%rsp), %rax subq %fs:40, %rax jne .L8 addq $88, %rsp .cfi_remember_state .cfi_def_cfa_offset 8 ret .L7: .cfi_restore_state pushq 8(%rsp) .cfi_def_cfa_offset 104 pushq 8(%rsp) .cfi_def_cfa_offset 112 leaq 80(%rsp), %r9 movq 44(%rsp), %rcx movl 52(%rsp), %r8d movq 32(%rsp), %rsi movl 40(%rsp), %edx leaq _Z8cos_main9cosParams(%rip), %rdi call cudaLaunchKernel@PLT addq $16, %rsp .cfi_def_cfa_offset 96 jmp .L3 .L8: call __stack_chk_fail@PLT .cfi_endproc .LFE2082: .size _Z35__device_stub__Z8cos_main9cosParamsR9cosParams, .-_Z35__device_stub__Z8cos_main9cosParamsR9cosParams .globl _Z8cos_main9cosParams .type _Z8cos_main9cosParams, @function _Z8cos_main9cosParams: .LFB2083: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq 16(%rsp), %rdi call _Z35__device_stub__Z8cos_main9cosParamsR9cosParams addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2083: .size _Z8cos_main9cosParams, .-_Z8cos_main9cosParams .section .rodata.str1.8,"aMS",@progbits,1 .align 8 .LC0: .string "\nBeginning CUDA example code on the CPU...\n" .align 8 .LC1: .string "Getting some info about CUDA devices on your system...\n" .align 8 .LC2: .string "There is %d CUDA device on your system.\n" .align 8 .LC3: .string "There are %d CUDA devices on your system.\n" .section .rodata.str1.1,"aMS",@progbits,1 .LC4: .string "\nCUDA Device #%d\n" .LC5: .string "Name: %s\n" .LC6: .string "\n" .section .rodata.str1.8 .align 8 .LC7: .string "Allocating memory on the CPU... (n=%d)\n" .align 8 .LC10: .string "Allocating memory on the GPU... (n=%d)\n" .align 8 .LC11: .string " value = %d : Memory Allocation on GPU Device failed\n" .align 8 .LC12: .string "done. Allocating more memory on the GPU...\n" .align 8 .LC13: .string "done again. Copying stuff from host (CPU) to device (GPU)...\n" .align 8 .LC14: .string " Memory Copy from Host to Device failed.\n" .section .rodata.str1.1 .LC15: .string "successful.\n" .LC16: .string "Launching kernel on GPU...\n" .section .rodata.str1.8 .align 8 .LC17: .string "GPU computations finished. Copying result back to CPU...\n" .align 8 .LC18: .string " Memory Copy from Device to Host failed.\n" .align 8 .LC19: .string "Copy back successful. Printing result...\n" .section .rodata.str1.1 .LC20: .string "cos(%f) = %f\n" .LC21: .string "Freeing memory...\n" .LC22: .string "\n\nFinished. :)\n\n" .text .globl main .type main, @function main: .LFB2057: .cfi_startproc endbr64 pushq %r13 .cfi_def_cfa_offset 16 .cfi_offset 13, -16 pushq %r12 .cfi_def_cfa_offset 24 .cfi_offset 12, -24 pushq %rbp .cfi_def_cfa_offset 32 .cfi_offset 6, -32 pushq %rbx .cfi_def_cfa_offset 40 .cfi_offset 3, -40 subq $1128, %rsp .cfi_def_cfa_offset 1168 movq %fs:40, %rax movq %rax, 1112(%rsp) xorl %eax, %eax leaq .LC0(%rip), %rsi movl $2, %edi call __printf_chk@PLT leaq .LC1(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq 4(%rsp), %rdi call cudaGetDeviceCount@PLT movl 4(%rsp), %edx cmpl $1, %edx je .L35 leaq .LC3(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L15: movl $0, %ebx leaq .LC4(%rip), %r13 leaq 80(%rsp), %rbp leaq .LC5(%rip), %r12 cmpl $0, 4(%rsp) jle .L14 .L13: movl %ebx, %edx movq %r13, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl %ebx, %esi movq %rbp, %rdi call cudaGetDeviceProperties_v2@PLT movq %rbp, %rdx movq %r12, %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT addl $1, %ebx cmpl %ebx, 4(%rsp) jg .L13 .L14: leaq .LC6(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq $0, 8(%rsp) movq $0, 16(%rsp) movl $1000000, %edx leaq .LC7(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $4000000, %edi call malloc@PLT movq %rax, %rbp movl $4000000, %edi call malloc@PLT movq %rax, %r12 movl $0, %eax movss .LC8(%rip), %xmm2 movss .LC9(%rip), %xmm1 .L16: pxor %xmm0, %xmm0 cvtsi2ssl %eax, %xmm0 mulss %xmm2, %xmm0 divss %xmm1, %xmm0 movss %xmm0, 0(%rbp,%rax,4) addq $1, %rax cmpq $1000000, %rax jne .L16 movl $1000000, %edx leaq .LC10(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT leaq 16(%rsp), %rdi movl $4000000, %esi call cudaMalloc@PLT testl %eax, %eax je .L17 movl %eax, %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L18: leaq 8(%rsp), %rdi movl $4000000, %esi call cudaMalloc@PLT testl %eax, %eax je .L19 movl %eax, %edx leaq .LC11(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L20: movl $1, %ecx movl $4000000, %edx movq %rbp, %rsi movq 16(%rsp), %rdi call cudaMemcpy@PLT testl %eax, %eax je .L21 leaq .LC14(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L22: movq 8(%rsp), %rbx movq 16(%rsp), %r13 leaq .LC16(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $512, 36(%rsp) movl $1, 40(%rsp) movl $1, 24(%rsp) movl $1, 28(%rsp) movl $0, %r9d movl $0, %r8d movq 36(%rsp), %rdx movl $1, %ecx movq 24(%rsp), %rdi movl $1, %esi call __cudaPushCallConfiguration@PLT testl %eax, %eax je .L36 .L23: leaq .LC17(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movl $2, %ecx movl $4000000, %edx movq 8(%rsp), %rsi movq %r12, %rdi call cudaMemcpy@PLT testl %eax, %eax je .L24 leaq .LC18(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT .L26: movl $0, %ebx leaq .LC20(%rip), %r13 jmp .L25 .L35: leaq .LC2(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L15 .L17: leaq .LC12(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L18 .L19: leaq .LC13(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L20 .L21: leaq .LC15(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L22 .L36: movq %r13, 48(%rsp) movq %rbx, 56(%rsp) movl $1000000, 64(%rsp) leaq 48(%rsp), %rdi call _Z35__device_stub__Z8cos_main9cosParamsR9cosParams jmp .L23 .L24: leaq .LC19(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT jmp .L26 .L27: addq $1, %rbx cmpq $1000000, %rbx je .L37 .L25: movslq %ebx, %rax imulq $351843721, %rax, %rax sarq $45, %rax movl %ebx, %edx sarl $31, %edx subl %edx, %eax imull $100000, %eax, %eax cmpl %ebx, %eax jne .L27 pxor %xmm0, %xmm0 cvtss2sd 0(%rbp,%rbx,4), %xmm0 pxor %xmm1, %xmm1 cvtss2sd (%r12,%rbx,4), %xmm1 movq %r13, %rsi movl $2, %edi movl $2, %eax call __printf_chk@PLT jmp .L27 .L37: leaq .LC21(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq %rbp, %rdi call free@PLT movq %r12, %rdi call free@PLT movq 16(%rsp), %rdi call cudaFree@PLT movq 8(%rsp), %rdi call cudaFree@PLT leaq .LC22(%rip), %rsi movl $2, %edi movl $0, %eax call __printf_chk@PLT movq 1112(%rsp), %rax subq %fs:40, %rax jne .L38 movl $0, %eax addq $1128, %rsp .cfi_remember_state .cfi_def_cfa_offset 40 popq %rbx .cfi_def_cfa_offset 32 popq %rbp .cfi_def_cfa_offset 24 popq %r12 .cfi_def_cfa_offset 16 popq %r13 .cfi_def_cfa_offset 8 ret .L38: .cfi_restore_state call __stack_chk_fail@PLT .cfi_endproc .LFE2057: .size main, .-main .section .rodata.str1.1 .LC23: .string "_Z8cos_main9cosParams" .text .type _ZL24__sti____cudaRegisterAllv, @function _ZL24__sti____cudaRegisterAllv: .LFB2085: .cfi_startproc endbr64 subq $8, %rsp .cfi_def_cfa_offset 16 leaq _ZL15__fatDeviceText(%rip), %rdi call __cudaRegisterFatBinary@PLT movq %rax, %rdi movq %rax, _ZL20__cudaFatCubinHandle(%rip) pushq $0 .cfi_def_cfa_offset 24 pushq $0 .cfi_def_cfa_offset 32 pushq $0 .cfi_def_cfa_offset 40 pushq $0 .cfi_def_cfa_offset 48 movl $0, %r9d movl $-1, %r8d leaq .LC23(%rip), %rdx movq %rdx, %rcx leaq _Z8cos_main9cosParams(%rip), %rsi call __cudaRegisterFunction@PLT addq $32, %rsp .cfi_def_cfa_offset 16 movq _ZL20__cudaFatCubinHandle(%rip), %rdi call __cudaRegisterFatBinaryEnd@PLT leaq _ZL26__cudaUnregisterBinaryUtilv(%rip), %rdi call atexit@PLT addq $8, %rsp .cfi_def_cfa_offset 8 ret .cfi_endproc .LFE2085: .size _ZL24__sti____cudaRegisterAllv, .-_ZL24__sti____cudaRegisterAllv .section .init_array,"aw" .align 8 .quad _ZL24__sti____cudaRegisterAllv .section .nvFatBinSegment,"aw" .align 8 .type _ZL15__fatDeviceText, @object .size _ZL15__fatDeviceText, 24 _ZL15__fatDeviceText: .long 1180844977 .long 1 .quad fatbinData .quad 0 .local _ZL20__cudaFatCubinHandle .comm _ZL20__cudaFatCubinHandle,8,8 .section .rodata.cst4,"aM",@progbits,4 .align 4 .LC8: .long 1086918618 .align 4 .LC9: .long 1232348160 .ident "GCC: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0" .section .note.GNU-stack,"",@progbits .section .note.gnu.property,"a" .align 8 .long 1f - 0f .long 4f - 1f .long 5 0: .string "GNU" 1: .align 8 .long 0xc0000002 .long 3f - 2f 2: .long 0x3 3: .align 8 4:
.text .file "CUDA_example_01.hip" .globl _Z23__device_stub__cos_main9cosParams # -- Begin function _Z23__device_stub__cos_main9cosParams .p2align 4, 0x90 .type _Z23__device_stub__cos_main9cosParams,@function _Z23__device_stub__cos_main9cosParams: # @_Z23__device_stub__cos_main9cosParams .cfi_startproc # %bb.0: subq $56, %rsp .cfi_def_cfa_offset 64 leaq 64(%rsp), %rax movq %rax, (%rsp) leaq 40(%rsp), %rdi leaq 24(%rsp), %rsi leaq 16(%rsp), %rdx leaq 8(%rsp), %rcx callq __hipPopCallConfiguration movq 40(%rsp), %rsi movl 48(%rsp), %edx movq 24(%rsp), %rcx movl 32(%rsp), %r8d movq %rsp, %r9 movl $_Z8cos_main9cosParams, %edi pushq 8(%rsp) .cfi_adjust_cfa_offset 8 pushq 24(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $72, %rsp .cfi_adjust_cfa_offset -72 retq .Lfunc_end0: .size _Z23__device_stub__cos_main9cosParams, .Lfunc_end0-_Z23__device_stub__cos_main9cosParams .cfi_endproc # -- End function .section .rodata.cst4,"aM",@progbits,4 .p2align 2, 0x0 # -- Begin function main .LCPI1_0: .long 0x40c90fda # float 6.283185 .LCPI1_1: .long 0x49742400 # float 1.0E+6 .text .globl main .p2align 4, 0x90 .type main,@function main: # @main .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 pushq %r15 .cfi_def_cfa_offset 24 pushq %r14 .cfi_def_cfa_offset 32 pushq %r13 .cfi_def_cfa_offset 40 pushq %r12 .cfi_def_cfa_offset 48 pushq %rbx .cfi_def_cfa_offset 56 subq $1560, %rsp # imm = 0x618 .cfi_def_cfa_offset 1616 .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 .cfi_offset %rbp, -16 movl $.Lstr, %edi callq puts@PLT movl $.Lstr.1, %edi callq puts@PLT leaq 12(%rsp), %rdi callq hipGetDeviceCount movl 12(%rsp), %esi cmpl $1, %esi jne .LBB1_2 # %bb.1: movl $.L.str.2, %edi movl $1, %esi jmp .LBB1_3 .LBB1_2: movl $.L.str.3, %edi .LBB1_3: xorl %eax, %eax callq printf cmpl $0, 12(%rsp) jle .LBB1_6 # %bb.4: # %.lr.ph.preheader leaq 88(%rsp), %rbx xorl %ebp, %ebp .p2align 4, 0x90 .LBB1_5: # %.lr.ph # =>This Inner Loop Header: Depth=1 movl $.L.str.4, %edi movl %ebp, %esi xorl %eax, %eax callq printf movq %rbx, %rdi movl %ebp, %esi callq hipGetDevicePropertiesR0600 movl $.L.str.5, %edi movq %rbx, %rsi xorl %eax, %eax callq printf incl %ebp cmpl 12(%rsp), %ebp jl .LBB1_5 .LBB1_6: # %._crit_edge movl $10, %edi callq putchar@PLT movq $0, 24(%rsp) movq $0, 16(%rsp) xorl %r15d, %r15d movl $.L.str.7, %edi movl $1000000, %esi # imm = 0xF4240 xorl %eax, %eax callq printf movl $4000000, %edi # imm = 0x3D0900 callq malloc movq %rax, %rbx movl $4000000, %edi # imm = 0x3D0900 callq malloc movq %rax, %r14 movss .LCPI1_0(%rip), %xmm0 # xmm0 = mem[0],zero,zero,zero movss .LCPI1_1(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero .p2align 4, 0x90 .LBB1_7: # =>This Inner Loop Header: Depth=1 xorps %xmm2, %xmm2 cvtsi2ss %r15d, %xmm2 mulss %xmm0, %xmm2 divss %xmm1, %xmm2 movss %xmm2, (%rbx,%r15,4) incq %r15 cmpq $1000000, %r15 # imm = 0xF4240 jne .LBB1_7 # %bb.8: movl $.L.str.8, %edi movl $1000000, %esi # imm = 0xF4240 xorl %eax, %eax callq printf leaq 16(%rsp), %rdi movl $4000000, %esi # imm = 0x3D0900 callq hipMalloc testl %eax, %eax je .LBB1_10 # %bb.9: movl $.L.str.9, %edi movl %eax, %esi xorl %eax, %eax callq printf jmp .LBB1_11 .LBB1_10: movl $.Lstr.2, %edi callq puts@PLT .LBB1_11: leaq 24(%rsp), %rdi movl $4000000, %esi # imm = 0x3D0900 callq hipMalloc testl %eax, %eax je .LBB1_13 # %bb.12: movl $.L.str.9, %edi movl %eax, %esi xorl %eax, %eax callq printf jmp .LBB1_14 .LBB1_13: movl $.Lstr.3, %edi callq puts@PLT .LBB1_14: movq 16(%rsp), %rdi movl $4000000, %edx # imm = 0x3D0900 movq %rbx, %rsi movl $1, %ecx callq hipMemcpy testl %eax, %eax movl $.Lstr.4, %eax movl $.Lstr.5, %edi cmoveq %rax, %rdi callq puts@PLT movq 24(%rsp), %r12 movq 16(%rsp), %r13 movl $.Lstr.6, %edi callq puts@PLT movabsq $4294967297, %rdi # imm = 0x100000001 leaq 511(%rdi), %rdx xorl %r15d, %r15d movl $1, %esi movl $1, %ecx xorl %r8d, %r8d xorl %r9d, %r9d callq __hipPushCallConfiguration testl %eax, %eax jne .LBB1_16 # %bb.15: movq %r13, 88(%rsp) movq %r12, 96(%rsp) movl $1000000, 104(%rsp) # imm = 0xF4240 leaq 88(%rsp), %rax movq %rax, 32(%rsp) leaq 72(%rsp), %rdi leaq 56(%rsp), %rsi leaq 48(%rsp), %rdx leaq 40(%rsp), %rcx callq __hipPopCallConfiguration movq 72(%rsp), %rsi movl 80(%rsp), %edx movq 56(%rsp), %rcx movl 64(%rsp), %r8d leaq 32(%rsp), %r9 movl $_Z8cos_main9cosParams, %edi pushq 40(%rsp) .cfi_adjust_cfa_offset 8 pushq 56(%rsp) .cfi_adjust_cfa_offset 8 callq hipLaunchKernel addq $16, %rsp .cfi_adjust_cfa_offset -16 .LBB1_16: movl $.Lstr.7, %edi callq puts@PLT movq 24(%rsp), %rsi movl $4000000, %edx # imm = 0x3D0900 movq %r14, %rdi movl $2, %ecx callq hipMemcpy testl %eax, %eax movl $.Lstr.8, %eax movl $.Lstr.9, %edi cmoveq %rax, %rdi callq puts@PLT xorl %r12d, %r12d jmp .LBB1_17 .p2align 4, 0x90 .LBB1_19: # in Loop: Header=BB1_17 Depth=1 incq %r12 decl %r15d cmpq $1000000, %r12 # imm = 0xF4240 je .LBB1_20 .LBB1_17: # =>This Inner Loop Header: Depth=1 movl %r12d, %eax shrl $5, %eax imulq $175921861, %rax, %rax # imm = 0xA7C5AC5 shrq $39, %rax imull $100000, %eax, %eax # imm = 0x186A0 addl %r15d, %eax jne .LBB1_19 # %bb.18: # in Loop: Header=BB1_17 Depth=1 movss (%rbx,%r12,4), %xmm0 # xmm0 = mem[0],zero,zero,zero cvtss2sd %xmm0, %xmm0 movss (%r14,%r12,4), %xmm1 # xmm1 = mem[0],zero,zero,zero cvtss2sd %xmm1, %xmm1 movl $.L.str.18, %edi movb $2, %al callq printf jmp .LBB1_19 .LBB1_20: movl $.Lstr.10, %edi callq puts@PLT movq %rbx, %rdi callq free movq %r14, %rdi callq free movq 16(%rsp), %rdi callq hipFree movq 24(%rsp), %rdi callq hipFree movl $.Lstr.11, %edi callq puts@PLT xorl %eax, %eax addq $1560, %rsp # imm = 0x618 .cfi_def_cfa_offset 56 popq %rbx .cfi_def_cfa_offset 48 popq %r12 .cfi_def_cfa_offset 40 popq %r13 .cfi_def_cfa_offset 32 popq %r14 .cfi_def_cfa_offset 24 popq %r15 .cfi_def_cfa_offset 16 popq %rbp .cfi_def_cfa_offset 8 retq .Lfunc_end1: .size main, .Lfunc_end1-main .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_ctor .type __hip_module_ctor,@function __hip_module_ctor: # @__hip_module_ctor .cfi_startproc # %bb.0: subq $40, %rsp .cfi_def_cfa_offset 48 cmpq $0, __hip_gpubin_handle(%rip) jne .LBB2_2 # %bb.1: movl $__hip_fatbin_wrapper, %edi callq __hipRegisterFatBinary movq %rax, __hip_gpubin_handle(%rip) .LBB2_2: movq __hip_gpubin_handle(%rip), %rdi xorps %xmm0, %xmm0 movups %xmm0, 16(%rsp) movups %xmm0, (%rsp) movl $_Z8cos_main9cosParams, %esi movl $.L__unnamed_1, %edx movl $.L__unnamed_1, %ecx movl $-1, %r8d xorl %r9d, %r9d callq __hipRegisterFunction movl $__hip_module_dtor, %edi addq $40, %rsp .cfi_def_cfa_offset 8 jmp atexit # TAILCALL .Lfunc_end2: .size __hip_module_ctor, .Lfunc_end2-__hip_module_ctor .cfi_endproc # -- End function .p2align 4, 0x90 # -- Begin function __hip_module_dtor .type __hip_module_dtor,@function __hip_module_dtor: # @__hip_module_dtor .cfi_startproc # %bb.0: movq __hip_gpubin_handle(%rip), %rdi testq %rdi, %rdi je .LBB3_2 # %bb.1: pushq %rax .cfi_def_cfa_offset 16 callq __hipUnregisterFatBinary movq $0, __hip_gpubin_handle(%rip) addq $8, %rsp .cfi_def_cfa_offset 8 .LBB3_2: retq .Lfunc_end3: .size __hip_module_dtor, .Lfunc_end3-__hip_module_dtor .cfi_endproc # -- End function .type _Z8cos_main9cosParams,@object # @_Z8cos_main9cosParams .section .rodata,"a",@progbits .globl _Z8cos_main9cosParams .p2align 3, 0x0 _Z8cos_main9cosParams: .quad _Z23__device_stub__cos_main9cosParams .size _Z8cos_main9cosParams, 8 .type .L.str.2,@object # @.str.2 .section .rodata.str1.1,"aMS",@progbits,1 .L.str.2: .asciz "There is %d CUDA device on your system.\n" .size .L.str.2, 41 .type .L.str.3,@object # @.str.3 .L.str.3: .asciz "There are %d CUDA devices on your system.\n" .size .L.str.3, 43 .type .L.str.4,@object # @.str.4 .L.str.4: .asciz "\nCUDA Device #%d\n" .size .L.str.4, 18 .type .L.str.5,@object # @.str.5 .L.str.5: .asciz "Name: %s\n" .size .L.str.5, 11 .type .L.str.7,@object # @.str.7 .L.str.7: .asciz "Allocating memory on the CPU... (n=%d)\n" .size .L.str.7, 40 .type .L.str.8,@object # @.str.8 .L.str.8: .asciz "Allocating memory on the GPU... (n=%d)\n" .size .L.str.8, 40 .type .L.str.9,@object # @.str.9 .L.str.9: .asciz " value = %d : Memory Allocation on GPU Device failed\n" .size .L.str.9, 54 .type .L.str.18,@object # @.str.18 .L.str.18: .asciz "cos(%f) = %f\n" .size .L.str.18, 14 .type .L__unnamed_1,@object # @0 .L__unnamed_1: .asciz "_Z8cos_main9cosParams" .size .L__unnamed_1, 22 .type __hip_fatbin_wrapper,@object # @__hip_fatbin_wrapper .section .hipFatBinSegment,"a",@progbits .p2align 3, 0x0 __hip_fatbin_wrapper: .long 1212764230 # 0x48495046 .long 1 # 0x1 .quad __hip_fatbin .quad 0 .size __hip_fatbin_wrapper, 24 .type __hip_gpubin_handle,@object # @__hip_gpubin_handle .local __hip_gpubin_handle .comm __hip_gpubin_handle,8,8 .section .init_array,"aw",@init_array .p2align 3, 0x0 .quad __hip_module_ctor .type __hip_cuid_,@object # @__hip_cuid_ .bss .globl __hip_cuid_ __hip_cuid_: .byte 0 # 0x0 .size __hip_cuid_, 1 .type .Lstr,@object # @str .section .rodata.str1.1,"aMS",@progbits,1 .Lstr: .asciz "\nBeginning CUDA example code on the CPU..." .size .Lstr, 43 .type .Lstr.1,@object # @str.1 .Lstr.1: .asciz "Getting some info about CUDA devices on your system..." .size .Lstr.1, 55 .type .Lstr.2,@object # @str.2 .Lstr.2: .asciz "done. Allocating more memory on the GPU..." .size .Lstr.2, 43 .type .Lstr.3,@object # @str.3 .Lstr.3: .asciz "done again. Copying stuff from host (CPU) to device (GPU)..." .size .Lstr.3, 61 .type .Lstr.4,@object # @str.4 .Lstr.4: .asciz "successful." .size .Lstr.4, 12 .type .Lstr.5,@object # @str.5 .Lstr.5: .asciz " Memory Copy from Host to Device failed." .size .Lstr.5, 41 .type .Lstr.6,@object # @str.6 .Lstr.6: .asciz "Launching kernel on GPU..." .size .Lstr.6, 27 .type .Lstr.7,@object # @str.7 .Lstr.7: .asciz "GPU computations finished. Copying result back to CPU..." .size .Lstr.7, 57 .type .Lstr.8,@object # @str.8 .Lstr.8: .asciz "Copy back successful. Printing result..." .size .Lstr.8, 41 .type .Lstr.9,@object # @str.9 .Lstr.9: .asciz " Memory Copy from Device to Host failed." .size .Lstr.9, 41 .type .Lstr.10,@object # @str.10 .Lstr.10: .asciz "Freeing memory..." .size .Lstr.10, 18 .type .Lstr.11,@object # @str.11 .Lstr.11: .asciz "\n\nFinished. :)\n" .size .Lstr.11, 16 .section ".linker-options","e",@llvm_linker_options .ident "AMD clang version 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.2 25012 e5bf7e55c91490b07c49d8960fa7983d864936c4)" .section ".note.GNU-stack","",@progbits .addrsig .addrsig_sym _Z23__device_stub__cos_main9cosParams .addrsig_sym __hip_module_ctor .addrsig_sym __hip_module_dtor .addrsig_sym _Z8cos_main9cosParams .addrsig_sym __hip_fatbin .addrsig_sym __hip_fatbin_wrapper .addrsig_sym __hip_cuid_