scemama/qp_plugins_scemama
1
0
Fork 0
mirror of https://gitlab.com/scemama/qp_plugins_scemama.git synced 2024-07-25 20:27:35 +02:00
Code Issues Releases Wiki Activity

GPU acceleration

Browse Source
This commit is contained in:
Anthony Scemama 2023-07-16 13:50:40 +02:00
parent 7847ccc674
commit 0c6e6a1ca0
1 changed files with 15 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
devel/ccsd_gpu/gpu.c
View File

@ -105,28 +105,27 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
C=d_tmpB1 ; ldc=nV*BLOCK_SIZE;
cublasDgemm(handle, CUBLAS_OP_T, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, lda, &beta, C, ldc);
for (size_t bet=iblock ; bet<(nV < iblock+BLOCK_SIZE ? nV : iblock+BLOCK_SIZE) ; ++bet) {
for (size_t bet=iblock ; bet<(nV < iblock+BLOCK_SIZE ? nV : iblock+BLOCK_SIZE) ; ++bet)
{
alpha = 1.0;
beta = 0.0;
A = &(d_tmpB1[nV*(bet-iblock)]); lda = nV*BLOCK_SIZE;
B = d_tmpB1; ldb = nV;
C = &(d_B1[nV*nV*(bet-iblock)]) ; ldc = nV;
cublasDgeam(handle, CUBLAS_OP_N, CUBLAS_OP_N, nV, nV, &alpha, A, lda, &beta, B, ldb, C, ldc);
alpha = 1.0;
beta = 0.0;
A = &(d_tmpB1[nV*(bet-iblock)]); lda = nV*BLOCK_SIZE;
B = d_tmpB1; ldb = nV;
C = &(d_B1[nV*nV*(bet-iblock)]) ; ldc = nV;
cublasDgeam(handle, CUBLAS_OP_N, CUBLAS_OP_N, nV, nV, &alpha, A, lda, &beta, B, ldb, C, ldc);
}
alpha=1.0; beta=1.0;
m=nO*nO; n=mbs; k=nV*nV;
alpha=1.0; beta=1.0;
m=nO*nO; n=mbs; k=nV*nV;
A=d_tau; lda=nO*nO;
B=d_B1 ; ldb=nV*nV;
C=&(d_r2[nO*nO*(iblock + nV*gam)]); ldc=nO*nO;
cublasDgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, ldb, &beta, C, ldc);
A=d_tau; lda=nO*nO;
B=d_B1 ; ldb=nV*nV;
C=&(d_r2[nO*nO*(iblock + nV*gam)]); ldc=nO*nO;
cublasDgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, ldb, &beta, C, ldc);
}
}
free(B1);
free(tmpB1);
}
lda=nO*nO;
cublasGetMatrix(nO*nO, nV*nV, sizeof(double), d_r2, lda, r2, lda);