| 234 | |
| 235 | |
| 236 | void DevSparseMatrix :: MultAdd (double s, const BaseVector & x, BaseVector & y) const |
| 237 | { |
| 238 | static Timer tmv("DevSparseMatrix :: MultAdd"); |
| 239 | CudaRegionTimer rt(tmv); |
| 240 | // RegionTimer reg(tmv); |
| 241 | |
| 242 | UnifiedVectorWrapper ux(x); |
| 243 | UnifiedVectorWrapper uy(y); |
| 244 | |
| 245 | ux.UpdateDevice(); |
| 246 | uy.UpdateDevice(); |
| 247 | |
| 248 | double alpha= s; |
| 249 | double beta = 1; |
| 250 | |
| 251 | cusparseDnVecDescr_t descr_x, descr_y; |
| 252 | cusparseCreateDnVec (&descr_x, ux.Size(), ux.DevData(), CUDA_R_64F); |
| 253 | cusparseCreateDnVec (&descr_y, uy.Size(), uy.DevData(), CUDA_R_64F); |
| 254 | |
| 255 | { cudaStreamCaptureStatus cap_status; |
| 256 | cudaStreamIsCapturing(ngs_cuda::ngs_cuda_stream, &cap_status); |
| 257 | // std::cerr << "[SpMV] capture status before SpMV: " << cap_status << " (1=capturing)" << std::endl; } |
| 258 | } |
| 259 | cusparseSetStream(Get_CuSparse_Handle(), ngs_cuda::ngs_cuda_stream); |
| 260 | cusparseSpMV(Get_CuSparse_Handle(), |
| 261 | CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, descr, |
| 262 | descr_x, &beta, descr_y, CUDA_R_64F, |
| 263 | CUSPARSE_SPMV_ALG_DEFAULT, spmv_buffer); |
| 264 | |
| 265 | cusparseDestroyDnVec(descr_x); |
| 266 | cusparseDestroyDnVec(descr_y); |
| 267 | uy.InvalidateHost(); |
| 268 | } |
| 269 | |
| 270 | |
| 271 | void DevSparseMatrix :: MultTransAdd (double s, const BaseVector & x, BaseVector & y) const |
nothing calls this directly
no test coverage detected