portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-25 21:25:56 +01:00
Code Releases Activity

Bounce MPI through host

Browse Source
This commit is contained in:
Peter Boyle 2025-02-12 19:34:59 +00:00
parent 4c3dd82d84
commit ba9bbe0221
1 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
Grid/lattice/PaddedCell.h
View File

@ -466,6 +466,12 @@ public:
static deviceVector<vobj> recv_buf;
send_buf.resize(buffer_size*2*depth);
recv_buf.resize(buffer_size*2*depth);
#ifndef ACCELERATOR_AWARE_MPI
static hostVector<vobj> hsend_buf;
static hostVector<vobj> hrecv_buf;
hsend_buf.resize(buffer_size*2*depth);
hrecv_buf.resize(buffer_size*2*depth);
#endif
std::vector<MpiCommsRequest_t> fwd_req;
std::vector<MpiCommsRequest_t> bwd_req;
@ -495,9 +501,17 @@ public:
t_gather+=usecond()-t;
t=usecond();
#ifdef ACCELERATOR_AWARE_MPI
grid->SendToRecvFromBegin(fwd_req,
(void *)&send_buf[d*buffer_size], xmit_to_rank,
(void *)&recv_buf[d*buffer_size], recv_from_rank, bytes, tag);
#else
acceleratorCopyFromDevice(&send_buf[d*buffer_size],&hsend_buf[d*buffer_size],bytes);
grid->SendToRecvFromBegin(fwd_req,
(void *)&hsend_buf[d*buffer_size], xmit_to_rank,
(void *)&hrecv_buf[d*buffer_size], recv_from_rank, bytes, tag);
acceleratorCopyToDevice(&hrecv_buf[d*buffer_size],&recv_buf[d*buffer_size],bytes);
#endif
t_comms+=usecond()-t;
}
for ( int d=0;d < depth ; d ++ ) {
@ -508,9 +522,17 @@ public:
t_gather+= usecond() - t;
t=usecond();
#ifdef ACCELERATOR_AWARE_MPI
grid->SendToRecvFromBegin(bwd_req,
(void *)&send_buf[(d+depth)*buffer_size], recv_from_rank,
(void *)&recv_buf[(d+depth)*buffer_size], xmit_to_rank, bytes,tag);
#else
acceleratorCopyFromDevice(&send_buf[(d+depth)*buffer_size],&hsend_buf[(d+depth)*buffer_size],bytes);
grid->SendToRecvFromBegin(bwd_req,
(void *)&hsend_buf[(d+depth)*buffer_size], recv_from_rank,
(void *)&hrecv_buf[(d+depth)*buffer_size], xmit_to_rank, bytes,tag);
acceleratorCopyToDevice(&hrecv_buf[(d+depth)*buffer_size],&recv_buf[(d+depth)*buffer_size],bytes);
#endif
t_comms+=usecond()-t;
}