scemama/qp_plugins_scemama
1
0
Fork 0
mirror of https://gitlab.com/scemama/qp_plugins_scemama.git synced 2024-10-06 08:15:59 +02:00
Code Issues Releases Wiki Activity

Compare commits

base: scemama:4d2adddbd9d6547999ad330245dca463ab806300
Branches Tags
scemama:master
..
compare: scemama:29903111f4169e0d019f56e0cdc242e32d931ce7
Branches Tags
scemama:master

No commits in common. "4d2adddbd9d6547999ad330245dca463ab806300" and "29903111f4169e0d019f56e0cdc242e32d931ce7" have entirely different histories.
4d2adddbd9 ... 29903111f4

1 changed files with 5 additions and 6 deletions
Show Stats Expand all files Collapse all files

11
devel/ccsd_gpu/gpu_dgemm.c
View File

@ -5,7 +5,6 @@
#include <cublas_v2.h>
#include <cuda_runtime.h>
#include <assert.h>
#include "gpu.h"
#define BLOCK_SIZE 16
@ -28,13 +27,13 @@ void gpu_dgemm(char transa, char transb, int m, int n, int k, double alpha,
cublasOperation_t ta, tb;
if (transa == 'N') {
cudaStat = gpu_malloc((void**)&d_A, (size_t) lda*k*sizeof(double));
cudaStat = cudaMalloc((void**)&d_A, (size_t) lda*k*sizeof(double));
assert(cudaStat == cudaSuccess);
cudaStat = cublasSetMatrix(m, k, sizeof(double), A, lda, d_A, lda);
assert(cudaStat == cudaSuccess);
ta = CUBLAS_OP_N;
} else {
cudaStat = gpu_malloc((void**)&d_A, (size_t) lda*m*sizeof(double));
cudaStat = cudaMalloc((void**)&d_A, (size_t) lda*m*sizeof(double));
assert(cudaStat == cudaSuccess);
cudaStat = cublasSetMatrix(k, m, sizeof(double), A, lda, d_A, lda);
assert(cudaStat == cudaSuccess);
@ -42,20 +41,20 @@ void gpu_dgemm(char transa, char transb, int m, int n, int k, double alpha,
}
if (transb == 'N') {
cudaStat = gpu_malloc((void**)&d_B, (size_t) ldb*n*sizeof(double));
cudaStat = cudaMalloc((void**)&d_B, (size_t) ldb*n*sizeof(double));
assert(cudaStat == cudaSuccess);
cudaStat = cublasSetMatrix(k, n, sizeof(double), B, ldb, d_B, ldb);
assert(cudaStat == cudaSuccess);
tb = CUBLAS_OP_N;
} else {
cudaStat = gpu_malloc((void**)&d_B, (size_t) ldb*k*sizeof(double));
cudaStat = cudaMalloc((void**)&d_B, (size_t) ldb*k*sizeof(double));
assert(cudaStat == cudaSuccess);
cudaStat = cublasSetMatrix(n, k, sizeof(double), B, ldb, d_B, ldb);
assert(cudaStat == cudaSuccess);
tb = CUBLAS_OP_T;
}
cudaStat = gpu_malloc((void**)&d_C, (size_t) ldc*n*sizeof(double));
cudaStat = cudaMalloc((void**)&d_C, (size_t) ldc*n*sizeof(double));
assert(cudaStat == cudaSuccess);
if (beta != 0.) {
cudaStat = cublasSetMatrix(m, n, sizeof(double), C, ldc, d_C, ldc);