portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-13 01:05:36 +00:00
Code Releases Activity

Overlap cudamemcpy - didn't set up stream right

Browse Source
This commit is contained in:
Peter Boyle 2021-10-11 13:31:26 -07:00
parent cda915a345
commit 16c2a99965
3 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Grid/communicator/Communicator_mpi3.cc
View File

@ -389,7 +389,6 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
void *shm = (void *) this->ShmBufferTranslate(dest,recv); void *shm = (void *) this->ShmBufferTranslate(dest,recv);
assert(shm!=NULL); assert(shm!=NULL);
acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes); acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
acceleratorCopySynchronise(); // MPI prob slower
} }
if ( CommunicatorPolicy == CommunicatorPolicySequential ) { if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
@ -405,6 +404,7 @@ void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsReque
if (nreq==0) return; if (nreq==0) return;
std::vector<MPI_Status> status(nreq); std::vector<MPI_Status> status(nreq);
acceleratorCopySynchronise();
int ierr = MPI_Waitall(nreq,&list[0],&status[0]); int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
assert(ierr==0); assert(ierr==0);
list.resize(0); list.resize(0);

2
Grid/threads/Accelerator.cc
View File

@ -95,7 +95,7 @@ void acceleratorInit(void)
#endif #endif
cudaSetDevice(device); cudaSetDevice(device);
cudaStreamCreate(&copyStream);
const int len=64; const int len=64;
char busid[len]; char busid[len];
if( rank == world_rank ) { if( rank == world_rank ) {

1
Grid/threads/Accelerator.h
View File

@ -95,6 +95,7 @@ void acceleratorInit(void);
////////////////////////////////////////////// //////////////////////////////////////////////
#ifdef GRID_CUDA #ifdef GRID_CUDA
#include <cuda.h> #include <cuda.h>
#ifdef __CUDA_ARCH__ #ifdef __CUDA_ARCH__