portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Overlap comms compute improvements in hand op kernels, and better timing from Edison and Cori

Browse Source
This commit is contained in:
paboyle
2016-01-10 16:54:27 -08:00
parent d19321dfde
commit dafc74020c
6 changed files with 4618 additions and 2702 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
lib/Stencil.h
View File

@ -111,6 +111,7 @@ namespace Grid {
}
void Communicate(void ) {
commtime-=usecond();
for(int i=0;i<Packets.size();i++){
_grid->SendToRecvFrom(Packets[i].send_buf,
Packets[i].to_rank,
@ -118,6 +119,7 @@ namespace Grid {
Packets[i].from_rank,
Packets[i].bytes);
}
commtime+=usecond();
}
///////////////////////////////////////////
@ -181,12 +183,10 @@ PARALLEL_FOR_LOOP
/////////////////////////////////////////
#define TIMING_HACK
#ifdef TIMING_HACK
double buftime;
double jointime;
double gathertime;
double commtime;
double commstime;
double halotime;
double scattertime;
double halogtime;
double mergetime;
double gathermtime;
double splicetime;
@ -202,13 +202,11 @@ PARALLEL_FOR_LOOP
{
#ifdef TIMING_HACK
gathertime=0;
jointime=0;
commtime=0;
commstime=0;
halotime=0;
scattertime=0;
halogtime=0;
mergetime=0;
gathermtime=0;
buftime=0;
splicetime=0;
nosplicetime=0;
#endif
@ -514,7 +512,9 @@ PARALLEL_FOR_LOOP
void HaloExchangeComplete(std::thread &thr)
{
jointime-=usecond();
thr.join();
jointime+=usecond();
CommsMerge();
}
@ -522,7 +522,7 @@ PARALLEL_FOR_LOOP
{
// conformable(source._grid,_grid);
assert(source._grid==_grid);
halotime-=usecond();
halogtime-=usecond();
assert (comm_buf.size() == _unified_buffer_size );
u_comm_offset=0;
@ -582,7 +582,7 @@ PARALLEL_FOR_LOOP
}
assert(u_comm_offset==_unified_buffer_size);
halotime+=usecond();
halogtime+=usecond();
}
void Gather(const Lattice<vobj> &rhs,int dimension,int shift,int cbmask,compressor & compress)
@ -633,7 +633,6 @@ PARALLEL_FOR_LOOP
assert (recv_from_rank != _grid->ThisRank());
// FIXME Implement asynchronous send & also avoid buffer copy
commtime-=usecond();
/*
_grid->SendToRecvFrom((void *)&send_buf[0],
xmit_to_rank,
@ -647,8 +646,6 @@ PARALLEL_FOR_LOOP
recv_from_rank,
bytes);
commtime+=usecond();
u_comm_offset+=words;
}
}
@ -657,7 +654,6 @@ PARALLEL_FOR_LOOP
void GatherSimd(const Lattice<vobj> &rhs,int dimension,int shift,int cbmask,compressor &compress)
{
buftime-=usecond();
const int Nsimd = _grid->Nsimd();
int fd = _grid->_fdimensions[dimension];
@ -687,8 +683,6 @@ PARALLEL_FOR_LOOP
std::vector<scalar_object *> rpointers(Nsimd);
std::vector<scalar_object *> spointers(Nsimd);
buftime+=usecond();
///////////////////////////////////////////
// Work out what to send where
///////////////////////////////////////////
@ -745,9 +739,7 @@ PARALLEL_FOR_LOOP
_grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank);
commstime-=usecond();
AddPacket( vsp,vrp,xmit_to_rank,recv_from_rank,bytes);
commstime+=usecond();
rpointers[i] = rp;