Skip to content

Commit

Permalink
Cusolver stream update
Browse files Browse the repository at this point in the history
  • Loading branch information
TeachRaccooon committed Aug 21, 2024
1 parent 2a06106 commit e5d07e3
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions RandLAPACK/drivers/rl_cqrrp_gpu.hh
Original file line number Diff line number Diff line change
Expand Up @@ -220,6 +220,7 @@ int CQRRP_blocked_GPU<T, RNG>::call(
// Create cusolver handle - used for the ORMQR call
cusolverDnHandle_t cusolverH = nullptr;
cusolverDnCreate(&cusolverH);
cusolverDnSetStream(cusolverH, strm);

/******************************WORKSPACE PARAMETERS*********************************/
char* d_work_getrf, * d_work_geqrf;
Expand Down Expand Up @@ -504,9 +505,8 @@ int CQRRP_blocked_GPU<T, RNG>::call(
cudaMalloc(reinterpret_cast<void **>(&d_work_ormqr), sizeof(double) * lwork_ormqr);
}
cusolverDnDormqr(cusolverH, CUBLAS_SIDE_LEFT, CUBLAS_OP_T, rows, cols - b_sz, b_sz, A_work, lda, &tau[iter * b_sz], Work1, lda, d_work_ormqr, lwork_ormqr, d_info_cusolver);
cusolverDnGetStream(cusolverH, &stream_cusolver);
cudaStreamSynchronize(stream_cusolver);
if(this -> timing) {
cudaStreamSynchronize(strm);
nvtxRangePop();
updating_A_t_stop = high_resolution_clock::now();
updating_A_t_dur += duration_cast<microseconds>(updating_A_t_stop - updating_A_t_start).count();
Expand Down

0 comments on commit e5d07e3

Please sign in to comment.