portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-08-05 05:57:11 +01:00
Code Releases Activity

Update to reduce memory footpring in benchmark test

Browse Source
This commit is contained in:
Peter Boyle
2021-12-07 09:02:02 -08:00
parent 5398b7e7e3
commit 2bf3b4d576
3 changed files with 46 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/communicator/Communicator_mpi3.cc
View File

@@ -388,6 +388,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
// TODO : make a OMP loop on CPU, call threaded bcopy
void *shm = (void *) this->ShmBufferTranslate(dest,recv);
assert(shm!=NULL);
std::cout <<"acceleratorCopyDeviceToDeviceAsynch"<< std::endl;
acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
}
@@ -399,12 +400,13 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
}
void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
{
acceleratorCopySynchronise(); std::cout << "Copy Synchronised\n"<<std::endl;
int nreq=list.size();
if (nreq==0) return;
std::vector<MPI_Status> status(nreq);
acceleratorCopySynchronise();
int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
assert(ierr==0);
list.resize(0);

2
Grid/threads/Accelerator.h
View File

@@ -306,7 +306,7 @@ inline void acceleratorFreeDevice(void *ptr){free(ptr,*theGridAccelerator);};
inline void acceleratorCopyDeviceToDeviceAsynch(void *from,void *to,size_t bytes) {
theGridAccelerator->memcpy(to,from,bytes);
}
inline void acceleratorCopySynchronise(void) { theGridAccelerator->wait(); }
inline void acceleratorCopySynchronise(void) { theGridAccelerator->wait(); std::cout<<"acceleratorCopySynchronise() wait "<<std::endl; }
inline void acceleratorCopyToDevice(void *from,void *to,size_t bytes) { theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
inline void acceleratorCopyFromDevice(void *from,void *to,size_t bytes){ theGridAccelerator->memcpy(to,from,bytes); theGridAccelerator->wait();}
inline void acceleratorMemSet(void *base,int value,size_t bytes) { theGridAccelerator->memset(base,value,bytes); theGridAccelerator->wait();}