Created
February 3, 2014 23:25
-
-
Save y-tag/8794544 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #include <cstdio> | |
| #include <cstdlib> | |
| #include <sys/time.h> | |
| #include "blas.h" | |
| #include "utility.h" | |
| // $ g++ -O3 -o bench_sgemm bench_sgemm.cc -lblas -lpthread -Wall -Werror | |
| int main(int argc, char **argv) { | |
| int loop_num = 10; | |
| int m = 1 << 11; | |
| int n = 1 << 11; | |
| int k = 1 << 11; | |
| float *A = new float[m * k]; | |
| float *B = new float[k * n]; | |
| float *C = new float[m * n]; | |
| if (A == NULL || B == NULL || C == NULL) { | |
| fprintf(stderr, "malloc error!\n"); | |
| return 1; | |
| } | |
| float alpha = 1.0f; | |
| float beta = 1.0f; | |
| double elapsed_time = 0.0; | |
| double start = 0.0; | |
| for (int l = 0; l < loop_num; ++l) { | |
| randn(A, m * k); | |
| randn(B, k * n); | |
| randn(C, m * n); | |
| start = gettimeofday_sec(); | |
| sgemm_("N", "N", &m, &n, &k, &alpha, A, &m, B, &k, &beta, C, &m); | |
| elapsed_time += gettimeofday_sec() - start; | |
| } | |
| long ops_num = 2L * m * n * k * loop_num; | |
| double gflops = ops_num / (1000 * 1000 * 1000 * elapsed_time); | |
| fprintf(stdout, "elapsed_time: %f\n", elapsed_time); | |
| fprintf(stdout, "GFLOPS: %f\n", gflops); | |
| delete[] A; A = NULL; | |
| delete[] B; B = NULL; | |
| delete[] C; C = NULL; | |
| return 0; | |
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #include <cstdio> | |
| #include <cstdlib> | |
| #include <sys/time.h> | |
| #include <cuda_runtime.h> | |
| #include <cublas_v2.h> | |
| #include "utility.h" | |
| // $ nvcc -o bench_sgemm_cublas bench_sgemm_cublas.cc -I/usr/local/cuda-5.5/include -lcublas | |
| int main(int argc, char **argv) { | |
| int loop_num = 10; | |
| int m = 1 << 11; | |
| int n = 1 << 11; | |
| int k = 1 << 11; | |
| float *A = new float[m * k]; | |
| float *B = new float[k * n]; | |
| float *C = new float[m * n]; | |
| if (A == NULL || B == NULL || C == NULL) { | |
| fprintf(stderr, "malloc error!\n"); | |
| return 1; | |
| } | |
| float alpha = 1.0f; | |
| float beta = 1.0f; | |
| double elapsed_time = 0.0; | |
| double start = 0.0; | |
| cudaError_t cuda_stat; | |
| cublasStatus_t stat; | |
| cublasHandle_t handle; | |
| float *devA; | |
| float *devB; | |
| float *devC; | |
| cuda_stat = cudaMalloc((void**)&devA, m * k * sizeof(*A)); | |
| if (cuda_stat != cudaSuccess) { fprintf(stderr, "cudaMalloc error! (devA)\n"); return 1; } | |
| cuda_stat = cudaMalloc((void**)&devB, k * n * sizeof(*B)); | |
| if (cuda_stat != cudaSuccess) { fprintf(stderr, "cudaMalloc error! (devB)\n"); return 1; } | |
| cuda_stat = cudaMalloc((void**)&devC, m * n * sizeof(*C)); | |
| if (cuda_stat != cudaSuccess) { fprintf(stderr, "cudaMalloc error! (devC)\n"); return 1; } | |
| stat = cublasCreate(&handle); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { | |
| fprintf(stderr, "cublasCreate error!\n"); | |
| cudaFree(devA); cudaFree(devB); cudaFree(devC); | |
| cublasDestroy(handle); | |
| return 1; | |
| } | |
| int l = 0; | |
| for (l = 0; l < loop_num; ++l) { | |
| randn(A, m * k); | |
| randn(B, k * n); | |
| randn(C, m * n); | |
| start = gettimeofday_sec(); | |
| stat = cublasSetMatrix(m, k, sizeof(*A), A, m, devA, m); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasSetMatrix error!(devA)\n"); break; } | |
| stat = cublasSetMatrix(k, n, sizeof(*B), B, k, devB, k); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasSetMatrix error!(devB)\n"); break; } | |
| stat = cublasSetMatrix(m, n, sizeof(*C), C, m, devC, m); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasSetMatrix error!(devC)\n"); break; } | |
| stat = cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, &alpha, devA, m, devB, k, &beta, devC, m); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { | |
| fprintf(stderr, "cublasSgemm error!\n"); | |
| break; | |
| } | |
| stat = cublasGetMatrix(m, k, sizeof(*A), devA, m, A, m); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasGetMatrix error!(devA)\n"); break; } | |
| stat = cublasGetMatrix(k, n, sizeof(*B), devB, k, B, k); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasGetMatrix error!(devB)\n"); break; } | |
| stat = cublasGetMatrix(m, n, sizeof(*C), devC, m, C, m); | |
| if (stat != CUBLAS_STATUS_SUCCESS) { fprintf(stderr, "cublasGetMatrix error!(devC)\n"); break; } | |
| elapsed_time += gettimeofday_sec() - start; | |
| } | |
| if (l == loop_num) { | |
| long ops_num = 2L * m * n * k * loop_num; | |
| double gflops = ops_num / (1000 * 1000 * 1000 * elapsed_time); | |
| fprintf(stdout, "elapsed_time: %f\n", elapsed_time); | |
| fprintf(stdout, "GFLOPS: %f\n", gflops); | |
| } | |
| cudaFree(devA); devA = NULL; | |
| cudaFree(devB); devB = NULL; | |
| cudaFree(devC); devC = NULL; | |
| cublasDestroy(handle); | |
| delete[] A; A = NULL; | |
| delete[] B; B = NULL; | |
| delete[] C; C = NULL; | |
| return 0; | |
| } |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment