portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 01:05:38 +01:00
Code Releases Activity

Benchmark improvements from tesseract

Browse Source
This commit is contained in:
Dr Peter Boyle 2018-04-27 11:44:46 +01:00
parent d6406b13e1
commit 1dddd17e3c
2 changed files with 19 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
benchmarks/Benchmark_ITT.cc
View File

@ -158,8 +158,10 @@ public:
dbytes=0; dbytes=0;
ncomm=0; ncomm=0;
#ifdef GRID_OMP
parallel_for(int dir=0;dir<8;dir++){ #pragma omp parallel for num_threads(Grid::CartesianCommunicator::nCommThreads)
#endif
for(int dir=0;dir<8;dir++){
double tbytes; double tbytes;
int mu =dir % 4; int mu =dir % 4;
@ -175,9 +177,14 @@ public:
int comm_proc = mpi_layout[mu]-1; int comm_proc = mpi_layout[mu]-1;
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank); Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
} }
#ifdef GRID_OMP
int tid = omp_get_thread_num();
#else
int tid = dir;
#endif
tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank, tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank,
(void *)&rbuf[dir][0], recv_from_rank, (void *)&rbuf[dir][0], recv_from_rank,
bytes,dir); bytes,tid);
#ifdef GRID_OMP #ifdef GRID_OMP
#pragma omp atomic #pragma omp atomic

13
benchmarks/Benchmark_comms.cc
View File

@ -169,7 +169,11 @@ int main (int argc, char ** argv)
for(int lat=4;lat<=maxlat;lat+=4){ for(int lat=4;lat<=maxlat;lat+=4){
for(int Ls=8;Ls<=8;Ls*=2){ for(int Ls=8;Ls<=8;Ls*=2){
std::vector<int> latt_size ({lat,lat,lat,lat}); std::vector<int> latt_size ({lat*mpi_layout[0],
lat*mpi_layout[1],
lat*mpi_layout[2],
lat*mpi_layout[3]});
GridCartesian Grid(latt_size,simd_layout,mpi_layout); GridCartesian Grid(latt_size,simd_layout,mpi_layout);
RealD Nrank = Grid._Nprocessors; RealD Nrank = Grid._Nprocessors;
@ -485,7 +489,8 @@ int main (int argc, char ** argv)
dbytes=0; dbytes=0;
ncomm=0; ncomm=0;
parallel_for(int dir=0;dir<8;dir++){ #pragma omp parallel for num_threads(Grid::CartesianCommunicator::nCommThreads)
for(int dir=0;dir<8;dir++){
double tbytes; double tbytes;
int mu =dir % 4; int mu =dir % 4;
@ -502,9 +507,9 @@ int main (int argc, char ** argv)
int comm_proc = mpi_layout[mu]-1; int comm_proc = mpi_layout[mu]-1;
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank); Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
} }
int tid = omp_get_thread_num();
tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank, tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank,
(void *)&rbuf[dir][0], recv_from_rank, bytes,dir); (void *)&rbuf[dir][0], recv_from_rank, bytes,tid);
#pragma omp atomic #pragma omp atomic
dbytes+=tbytes; dbytes+=tbytes;