64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Trim constraint-only GPU downloads

Browse Source
This commit is contained in:
CGH0S7
2026-04-09 19:36:19 +08:00
parent 7cd2414faa
commit 46e94d1248
3 changed files with 22 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
AMSS_NCKU_source/bssn_gpu.cu
View File

@@ -4034,9 +4034,9 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
// Main GPU evolution path now consumes RHS directly from device caches.
// Avoiding this bulk D2H copy removes one full round-trip per RK stage.
}
else if(calledby == CALLED_BY_CONSTRAINT)
{
cudaMemcpy(Gamxxx, Mh_ Gamxxx, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
else if(calledby == CALLED_BY_CONSTRAINT)
{
cudaMemcpy(Gamxxx, Mh_ Gamxxx, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gamxxy, Mh_ Gamxxy, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gamxxz, Mh_ Gamxxz, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gamxyy, Mh_ Gamxyy, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
@@ -4065,9 +4065,19 @@ int gpu_rhs(int calledby, int mpi_rank, int *ex, double &T,double *X, double *Y,
cudaMemcpy(movy_Res, Mh_ movy_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(movz_Res, Mh_ movz_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmx_Res, Mh_ Gmx_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmy_Res, Mh_ Gmy_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmz_Res, Mh_ Gmz_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
}
cudaMemcpy(Gmy_Res, Mh_ Gmy_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmz_Res, Mh_ Gmz_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
}
else if(calledby == CALLED_BY_CONSTRAINT_CONS_ONLY)
{
cudaMemcpy(ham_Res, Mh_ ham_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(movx_Res, Mh_ movx_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(movy_Res, Mh_ movy_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(movz_Res, Mh_ movz_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmx_Res, Mh_ Gmx_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmy_Res, Mh_ Gmy_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
cudaMemcpy(Gmz_Res, Mh_ Gmz_Res, matrix_size * sizeof(double), cudaMemcpyDeviceToHost);
}
//-----------------------------------------------------
//-------------------FOR GPU TEST----------------------