mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-25 22:28:17 +08:00
Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.
This commit is contained in:
Rob Armstrong
@@ -42,207 +42,208 @@
|
||||
*/
|
||||
|
||||
// System includes
|
||||
#include <stdio.h>
|
||||
#include <assert.h>
|
||||
#include <stdio.h>
|
||||
|
||||
// CUDA runtime
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
#include "nvrtc_helper.h"
|
||||
|
||||
// Helper functions and utilities to work with CUDA
|
||||
#include <helper_functions.h>
|
||||
|
||||
void constantInit(float *data, int size, float val) {
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data[i] = val;
|
||||
}
|
||||
void constantInit(float *data, int size, float val)
|
||||
{
|
||||
for (int i = 0; i < size; ++i) {
|
||||
data[i] = val;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Run a simple test of matrix multiplication using CUDA
|
||||
*/
|
||||
int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
|
||||
dim3 &dimsB) {
|
||||
// Allocate host memory for matrices A and B
|
||||
unsigned int size_A = dimsA.x * dimsA.y;
|
||||
unsigned int mem_size_A = sizeof(float) * size_A;
|
||||
float *h_A = (float *)malloc(mem_size_A);
|
||||
unsigned int size_B = dimsB.x * dimsB.y;
|
||||
unsigned int mem_size_B = sizeof(float) * size_B;
|
||||
float *h_B = (float *)malloc(mem_size_B);
|
||||
int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA, dim3 &dimsB)
|
||||
{
|
||||
// Allocate host memory for matrices A and B
|
||||
unsigned int size_A = dimsA.x * dimsA.y;
|
||||
unsigned int mem_size_A = sizeof(float) * size_A;
|
||||
float *h_A = (float *)malloc(mem_size_A);
|
||||
unsigned int size_B = dimsB.x * dimsB.y;
|
||||
unsigned int mem_size_B = sizeof(float) * size_B;
|
||||
float *h_B = (float *)malloc(mem_size_B);
|
||||
|
||||
// Initialize host memory
|
||||
const float valB = 0.01f;
|
||||
constantInit(h_A, size_A, 1.0f);
|
||||
constantInit(h_B, size_B, valB);
|
||||
// Initialize host memory
|
||||
const float valB = 0.01f;
|
||||
constantInit(h_A, size_A, 1.0f);
|
||||
constantInit(h_B, size_B, valB);
|
||||
|
||||
// Allocate device memory
|
||||
CUdeviceptr d_A, d_B, d_C;
|
||||
// Allocate device memory
|
||||
CUdeviceptr d_A, d_B, d_C;
|
||||
|
||||
char *cubin, *kernel_file;
|
||||
size_t cubinSize;
|
||||
char *cubin, *kernel_file;
|
||||
size_t cubinSize;
|
||||
|
||||
kernel_file = sdkFindFilePath("matrixMul_kernel.cu", argv[0]);
|
||||
compileFileToCUBIN(kernel_file, argc, argv, &cubin, &cubinSize, 1);
|
||||
kernel_file = sdkFindFilePath("matrixMul_kernel.cu", argv[0]);
|
||||
compileFileToCUBIN(kernel_file, argc, argv, &cubin, &cubinSize, 1);
|
||||
|
||||
CUmodule module = loadCUBIN(cubin, argc, argv);
|
||||
CUmodule module = loadCUBIN(cubin, argc, argv);
|
||||
|
||||
// Allocate host matrix C
|
||||
dim3 dimsC(dimsB.x, dimsA.y, 1);
|
||||
unsigned int mem_size_C = dimsC.x * dimsC.y * sizeof(float);
|
||||
float *h_C = (float *)malloc(mem_size_C);
|
||||
// Allocate host matrix C
|
||||
dim3 dimsC(dimsB.x, dimsA.y, 1);
|
||||
unsigned int mem_size_C = dimsC.x * dimsC.y * sizeof(float);
|
||||
float *h_C = (float *)malloc(mem_size_C);
|
||||
|
||||
if (h_C == NULL) {
|
||||
fprintf(stderr, "Failed to allocate host matrix C!\n");
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
checkCudaErrors(cuMemAlloc(&d_A, mem_size_A));
|
||||
checkCudaErrors(cuMemAlloc(&d_B, mem_size_B));
|
||||
checkCudaErrors(cuMemAlloc(&d_C, mem_size_C));
|
||||
|
||||
// copy host memory to device
|
||||
checkCudaErrors(cuMemcpyHtoD(d_A, h_A, mem_size_A));
|
||||
checkCudaErrors(cuMemcpyHtoD(d_B, h_B, mem_size_B));
|
||||
|
||||
// Setup execution parameters
|
||||
dim3 threads(block_size, block_size);
|
||||
dim3 grid(dimsB.x / threads.x, dimsA.y / threads.y);
|
||||
|
||||
// Create and start timer
|
||||
printf("Computing result using CUDA Kernel...\n");
|
||||
|
||||
CUfunction kernel_addr;
|
||||
if (block_size == 16) {
|
||||
checkCudaErrors(
|
||||
cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
|
||||
} else {
|
||||
checkCudaErrors(
|
||||
cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
|
||||
}
|
||||
|
||||
void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x,
|
||||
(void *)&dimsB.x};
|
||||
|
||||
// Execute the kernel
|
||||
int nIter = 300;
|
||||
|
||||
for (int j = 0; j < nIter; j++) {
|
||||
checkCudaErrors(
|
||||
cuLaunchKernel(kernel_addr, grid.x, grid.y, grid.z, /* grid dim */
|
||||
threads.x, threads.y, threads.z, /* block dim */
|
||||
0, 0, /* shared mem, stream */
|
||||
&arr[0], /* arguments */
|
||||
0));
|
||||
|
||||
checkCudaErrors(cuCtxSynchronize());
|
||||
}
|
||||
|
||||
// Copy result from device to host
|
||||
checkCudaErrors(cuMemcpyDtoH(h_C, d_C, mem_size_C));
|
||||
|
||||
printf("Checking computed result for correctness: ");
|
||||
|
||||
bool correct = true;
|
||||
|
||||
// test relative error by the formula
|
||||
// |<x, y>_cpu - <x,y>_gpu|/<|x|, |y|> < eps
|
||||
|
||||
double eps = 1.e-6; // machine zero
|
||||
|
||||
for (int i = 0; i < (int)(dimsC.x * dimsC.y); i++) {
|
||||
double abs_err = fabs(h_C[i] - (dimsA.x * valB));
|
||||
double dot_length = dimsA.x;
|
||||
double abs_val = fabs(h_C[i]);
|
||||
double rel_err = abs_err / abs_val / dot_length;
|
||||
|
||||
if (rel_err > eps) {
|
||||
printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i,
|
||||
h_C[i], dimsA.x * valB, eps);
|
||||
correct = false;
|
||||
if (h_C == NULL) {
|
||||
fprintf(stderr, "Failed to allocate host matrix C!\n");
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
}
|
||||
|
||||
printf("%s\n", correct ? "Result = PASS" : "Result = FAIL");
|
||||
checkCudaErrors(cuMemAlloc(&d_A, mem_size_A));
|
||||
checkCudaErrors(cuMemAlloc(&d_B, mem_size_B));
|
||||
checkCudaErrors(cuMemAlloc(&d_C, mem_size_C));
|
||||
|
||||
printf(
|
||||
"\nNOTE: The CUDA Samples are not meant for performance measurements. "
|
||||
"Results may vary when GPU Boost is enabled.\n");
|
||||
// copy host memory to device
|
||||
checkCudaErrors(cuMemcpyHtoD(d_A, h_A, mem_size_A));
|
||||
checkCudaErrors(cuMemcpyHtoD(d_B, h_B, mem_size_B));
|
||||
|
||||
// Clean up memory
|
||||
free(h_A);
|
||||
free(h_B);
|
||||
free(h_C);
|
||||
// Setup execution parameters
|
||||
dim3 threads(block_size, block_size);
|
||||
dim3 grid(dimsB.x / threads.x, dimsA.y / threads.y);
|
||||
|
||||
checkCudaErrors(cuMemFree(d_A));
|
||||
checkCudaErrors(cuMemFree(d_B));
|
||||
checkCudaErrors(cuMemFree(d_C));
|
||||
// Create and start timer
|
||||
printf("Computing result using CUDA Kernel...\n");
|
||||
|
||||
if (correct) {
|
||||
return EXIT_SUCCESS;
|
||||
} else {
|
||||
return EXIT_FAILURE;
|
||||
}
|
||||
CUfunction kernel_addr;
|
||||
if (block_size == 16) {
|
||||
checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
|
||||
}
|
||||
else {
|
||||
checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
|
||||
}
|
||||
|
||||
void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x, (void *)&dimsB.x};
|
||||
|
||||
// Execute the kernel
|
||||
int nIter = 300;
|
||||
|
||||
for (int j = 0; j < nIter; j++) {
|
||||
checkCudaErrors(cuLaunchKernel(kernel_addr,
|
||||
grid.x,
|
||||
grid.y,
|
||||
grid.z, /* grid dim */
|
||||
threads.x,
|
||||
threads.y,
|
||||
threads.z, /* block dim */
|
||||
0,
|
||||
0, /* shared mem, stream */
|
||||
&arr[0], /* arguments */
|
||||
0));
|
||||
|
||||
checkCudaErrors(cuCtxSynchronize());
|
||||
}
|
||||
|
||||
// Copy result from device to host
|
||||
checkCudaErrors(cuMemcpyDtoH(h_C, d_C, mem_size_C));
|
||||
|
||||
printf("Checking computed result for correctness: ");
|
||||
|
||||
bool correct = true;
|
||||
|
||||
// test relative error by the formula
|
||||
// |<x, y>_cpu - <x,y>_gpu|/<|x|, |y|> < eps
|
||||
|
||||
double eps = 1.e-6; // machine zero
|
||||
|
||||
for (int i = 0; i < (int)(dimsC.x * dimsC.y); i++) {
|
||||
double abs_err = fabs(h_C[i] - (dimsA.x * valB));
|
||||
double dot_length = dimsA.x;
|
||||
double abs_val = fabs(h_C[i]);
|
||||
double rel_err = abs_err / abs_val / dot_length;
|
||||
|
||||
if (rel_err > eps) {
|
||||
printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i, h_C[i], dimsA.x * valB, eps);
|
||||
correct = false;
|
||||
}
|
||||
}
|
||||
|
||||
printf("%s\n", correct ? "Result = PASS" : "Result = FAIL");
|
||||
|
||||
printf("\nNOTE: The CUDA Samples are not meant for performance measurements. "
|
||||
"Results may vary when GPU Boost is enabled.\n");
|
||||
|
||||
// Clean up memory
|
||||
free(h_A);
|
||||
free(h_B);
|
||||
free(h_C);
|
||||
|
||||
checkCudaErrors(cuMemFree(d_A));
|
||||
checkCudaErrors(cuMemFree(d_B));
|
||||
checkCudaErrors(cuMemFree(d_C));
|
||||
|
||||
if (correct) {
|
||||
return EXIT_SUCCESS;
|
||||
}
|
||||
else {
|
||||
return EXIT_FAILURE;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Program main
|
||||
*/
|
||||
|
||||
int main(int argc, char **argv) {
|
||||
printf("[Matrix Multiply Using CUDA] - Starting...\n");
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
printf("[Matrix Multiply Using CUDA] - Starting...\n");
|
||||
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
|
||||
checkCmdLineFlag(argc, (const char **)argv, "?")) {
|
||||
printf("Usage -device=n (n >= 0 for deviceID)\n");
|
||||
printf(" -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
|
||||
printf(" -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
|
||||
printf(
|
||||
" Note: Outer matrix dimensions of A & B matrices must be equal.\n");
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "help") || checkCmdLineFlag(argc, (const char **)argv, "?")) {
|
||||
printf("Usage -device=n (n >= 0 for deviceID)\n");
|
||||
printf(" -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
|
||||
printf(" -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
|
||||
printf(" Note: Outer matrix dimensions of A & B matrices must be equal.\n");
|
||||
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
|
||||
int block_size = 32;
|
||||
int block_size = 32;
|
||||
|
||||
// original:
|
||||
dim3 dimsA(5 * 2 * block_size, 5 * 2 * block_size, 1);
|
||||
dim3 dimsB(5 * 4 * block_size, 5 * 2 * block_size, 1);
|
||||
// original:
|
||||
dim3 dimsA(5 * 2 * block_size, 5 * 2 * block_size, 1);
|
||||
dim3 dimsB(5 * 4 * block_size, 5 * 2 * block_size, 1);
|
||||
|
||||
// reduce sizes to avoid running out of memory
|
||||
// dim3 dimsA(32,32, 1);
|
||||
// dim3 dimsB(32,32,1);
|
||||
// reduce sizes to avoid running out of memory
|
||||
// dim3 dimsA(32,32, 1);
|
||||
// dim3 dimsB(32,32,1);
|
||||
|
||||
// width of Matrix A
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "wA")) {
|
||||
dimsA.x = getCmdLineArgumentInt(argc, (const char **)argv, "wA");
|
||||
}
|
||||
// width of Matrix A
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "wA")) {
|
||||
dimsA.x = getCmdLineArgumentInt(argc, (const char **)argv, "wA");
|
||||
}
|
||||
|
||||
// height of Matrix A
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "hA")) {
|
||||
dimsA.y = getCmdLineArgumentInt(argc, (const char **)argv, "hA");
|
||||
}
|
||||
// height of Matrix A
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "hA")) {
|
||||
dimsA.y = getCmdLineArgumentInt(argc, (const char **)argv, "hA");
|
||||
}
|
||||
|
||||
// width of Matrix B
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "wB")) {
|
||||
dimsB.x = getCmdLineArgumentInt(argc, (const char **)argv, "wB");
|
||||
}
|
||||
// width of Matrix B
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "wB")) {
|
||||
dimsB.x = getCmdLineArgumentInt(argc, (const char **)argv, "wB");
|
||||
}
|
||||
|
||||
// height of Matrix B
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "hB")) {
|
||||
dimsB.y = getCmdLineArgumentInt(argc, (const char **)argv, "hB");
|
||||
}
|
||||
// height of Matrix B
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "hB")) {
|
||||
dimsB.y = getCmdLineArgumentInt(argc, (const char **)argv, "hB");
|
||||
}
|
||||
|
||||
if (dimsA.x != dimsB.y) {
|
||||
printf("Error: outer matrix dimensions must be equal. (%d != %d)\n",
|
||||
dimsA.x, dimsB.y);
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
if (dimsA.x != dimsB.y) {
|
||||
printf("Error: outer matrix dimensions must be equal. (%d != %d)\n", dimsA.x, dimsB.y);
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x,
|
||||
dimsB.y);
|
||||
printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x, dimsB.y);
|
||||
|
||||
int matrix_result = matrixMultiply(argc, argv, block_size, dimsA, dimsB);
|
||||
int matrix_result = matrixMultiply(argc, argv, block_size, dimsA, dimsB);
|
||||
|
||||
exit(matrix_result);
|
||||
exit(matrix_result);
|
||||
}
|
||||
|
||||
@@ -48,84 +48,83 @@
|
||||
|
||||
#include <cooperative_groups.h>
|
||||
|
||||
template <int BLOCK_SIZE>
|
||||
__device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB) {
|
||||
// Handle to thread block group
|
||||
cooperative_groups::thread_block cta =
|
||||
cooperative_groups::this_thread_block();
|
||||
// Block index
|
||||
int bx = blockIdx.x;
|
||||
int by = blockIdx.y;
|
||||
template <int BLOCK_SIZE> __device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB)
|
||||
{
|
||||
// Handle to thread block group
|
||||
cooperative_groups::thread_block cta = cooperative_groups::this_thread_block();
|
||||
// Block index
|
||||
int bx = blockIdx.x;
|
||||
int by = blockIdx.y;
|
||||
|
||||
// Thread index
|
||||
int tx = threadIdx.x;
|
||||
int ty = threadIdx.y;
|
||||
// Thread index
|
||||
int tx = threadIdx.x;
|
||||
int ty = threadIdx.y;
|
||||
|
||||
// Index of the first sub-matrix of A processed by the block
|
||||
int aBegin = wA * BLOCK_SIZE * by;
|
||||
// Index of the first sub-matrix of A processed by the block
|
||||
int aBegin = wA * BLOCK_SIZE * by;
|
||||
|
||||
// Index of the last sub-matrix of A processed by the block
|
||||
int aEnd = aBegin + wA - 1;
|
||||
// Index of the last sub-matrix of A processed by the block
|
||||
int aEnd = aBegin + wA - 1;
|
||||
|
||||
// Step size used to iterate through the sub-matrices of A
|
||||
int aStep = BLOCK_SIZE;
|
||||
// Step size used to iterate through the sub-matrices of A
|
||||
int aStep = BLOCK_SIZE;
|
||||
|
||||
// Index of the first sub-matrix of B processed by the block
|
||||
int bBegin = BLOCK_SIZE * bx;
|
||||
// Index of the first sub-matrix of B processed by the block
|
||||
int bBegin = BLOCK_SIZE * bx;
|
||||
|
||||
// Step size used to iterate through the sub-matrices of B
|
||||
int bStep = BLOCK_SIZE * wB;
|
||||
// Step size used to iterate through the sub-matrices of B
|
||||
int bStep = BLOCK_SIZE * wB;
|
||||
|
||||
// Csub is used to store the element of the block sub-matrix
|
||||
// that is computed by the thread
|
||||
float Csub = 0;
|
||||
// Csub is used to store the element of the block sub-matrix
|
||||
// that is computed by the thread
|
||||
float Csub = 0;
|
||||
|
||||
// Loop over all the sub-matrices of A and B
|
||||
// required to compute the block sub-matrix
|
||||
for (int a = aBegin, b = bBegin; a <= aEnd; a += aStep, b += bStep) {
|
||||
// Declaration of the shared memory array As used to
|
||||
// store the sub-matrix of A
|
||||
__shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
|
||||
// Loop over all the sub-matrices of A and B
|
||||
// required to compute the block sub-matrix
|
||||
for (int a = aBegin, b = bBegin; a <= aEnd; a += aStep, b += bStep) {
|
||||
// Declaration of the shared memory array As used to
|
||||
// store the sub-matrix of A
|
||||
__shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
|
||||
|
||||
// Declaration of the shared memory array Bs used to
|
||||
// store the sub-matrix of B
|
||||
__shared__ float Bs[BLOCK_SIZE][BLOCK_SIZE];
|
||||
// Declaration of the shared memory array Bs used to
|
||||
// store the sub-matrix of B
|
||||
__shared__ float Bs[BLOCK_SIZE][BLOCK_SIZE];
|
||||
|
||||
// Load the matrices from device memory
|
||||
// to shared memory; each thread loads
|
||||
// one element of each matrix
|
||||
As[ty][tx] = A[a + wA * ty + tx];
|
||||
Bs[ty][tx] = B[b + wB * ty + tx];
|
||||
// Load the matrices from device memory
|
||||
// to shared memory; each thread loads
|
||||
// one element of each matrix
|
||||
As[ty][tx] = A[a + wA * ty + tx];
|
||||
Bs[ty][tx] = B[b + wB * ty + tx];
|
||||
|
||||
// Synchronize to make sure the matrices are loaded
|
||||
cooperative_groups::sync(cta);
|
||||
// Synchronize to make sure the matrices are loaded
|
||||
cooperative_groups::sync(cta);
|
||||
|
||||
// Multiply the two matrices together;
|
||||
// each thread computes one element
|
||||
// of the block sub-matrix
|
||||
#pragma unroll
|
||||
for (int k = 0; k < BLOCK_SIZE; ++k) {
|
||||
Csub += As[ty][k] * Bs[k][tx];
|
||||
for (int k = 0; k < BLOCK_SIZE; ++k) {
|
||||
Csub += As[ty][k] * Bs[k][tx];
|
||||
}
|
||||
|
||||
// Synchronize to make sure that the preceding
|
||||
// computation is done before loading two new
|
||||
// sub-matrices of A and B in the next iteration
|
||||
cooperative_groups::sync(cta);
|
||||
}
|
||||
|
||||
// Synchronize to make sure that the preceding
|
||||
// computation is done before loading two new
|
||||
// sub-matrices of A and B in the next iteration
|
||||
cooperative_groups::sync(cta);
|
||||
}
|
||||
|
||||
// Write the block sub-matrix to device memory;
|
||||
// each thread writes one element
|
||||
int c = wB * BLOCK_SIZE * by + BLOCK_SIZE * bx;
|
||||
C[c + wB * ty + tx] = Csub;
|
||||
// Write the block sub-matrix to device memory;
|
||||
// each thread writes one element
|
||||
int c = wB * BLOCK_SIZE * by + BLOCK_SIZE * bx;
|
||||
C[c + wB * ty + tx] = Csub;
|
||||
}
|
||||
|
||||
extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B,
|
||||
int wA, int wB) {
|
||||
matrixMulCUDA<16>(C, A, B, wA, wB);
|
||||
extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B, int wA, int wB)
|
||||
{
|
||||
matrixMulCUDA<16>(C, A, B, wA, wB);
|
||||
}
|
||||
|
||||
extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B,
|
||||
int wA, int wB) {
|
||||
matrixMulCUDA<32>(C, A, B, wA, wB);
|
||||
extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B, int wA, int wB)
|
||||
{
|
||||
matrixMulCUDA<32>(C, A, B, wA, wB);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user