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

Make comms benchmark same as Benchmark_comms_host_device

Browse Source
This commit is contained in:
Peter Boyle 2021-08-09 04:06:30 -07:00
parent 80ac2a73ca
commit fe5aaf7677
1 changed files with 16 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
benchmarks/Benchmark_ITT.cc
View File

@ -133,15 +133,15 @@ public:
std::vector<HalfSpinColourVectorD *> xbuf(8); std::vector<HalfSpinColourVectorD *> xbuf(8);
std::vector<HalfSpinColourVectorD *> rbuf(8); std::vector<HalfSpinColourVectorD *> rbuf(8);
Grid.ShmBufferFreeAll(); //Grid.ShmBufferFreeAll();
uint64_t bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
for(int d=0;d<8;d++){ for(int d=0;d<8;d++){
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); xbuf[d] = (HalfSpinColourVectorD *)acceleratorAllocDevice(bytes);
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); rbuf[d] = (HalfSpinColourVectorD *)acceleratorAllocDevice(bytes);
// bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
} }
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
int ncomm; int ncomm;
double dbytes; double dbytes;
std::vector<double> times(Nloop); std::vector<double> times(Nloop);
@ -152,7 +152,7 @@ public:
dbytes=0; dbytes=0;
ncomm=0; ncomm=0;
thread_for(dir,8,{ for(int dir=0;dir<8;dir++) {
double tbytes; double tbytes;
int mu =dir % 4; int mu =dir % 4;
@ -168,15 +168,16 @@ 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);
} }
tbytes= Grid.StencilSendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank, Grid.SendToRecvFrom((void *)&xbuf[dir][0], xmit_to_rank,
(void *)&rbuf[dir][0], recv_from_rank, (void *)&rbuf[dir][0], recv_from_rank,
bytes,dir); bytes);
tbytes = bytes;
thread_critical { thread_critical {
ncomm++; ncomm++;
dbytes+=tbytes; dbytes+=tbytes;
} }
} }
}); };
Grid.Barrier(); Grid.Barrier();
double stop=usecond(); double stop=usecond();
t_time[i] = stop-start; // microseconds t_time[i] = stop-start; // microseconds
@ -196,6 +197,10 @@ public:
<< "\t\t"<< bidibytes/timestat.mean<< " " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " " << "\t\t"<< bidibytes/timestat.mean<< " " << bidibytes*timestat.err/(timestat.mean*timestat.mean) << " "
<< bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl; << bidibytes/timestat.max << " " << bidibytes/timestat.min << std::endl;
for(int d=0;d<8;d++){
acceleratorFreeDevice(xbuf[d]);
acceleratorFreeDevice(rbuf[d]);
}
} }
} }
@ -281,7 +286,6 @@ public:
uint64_t lmax=32; uint64_t lmax=32;
#define NLOOP (1000*lmax*lmax*lmax*lmax/lat/lat/lat/lat)
GridSerialRNG sRNG; sRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9})); GridSerialRNG sRNG; sRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
for(int lat=8;lat<=lmax;lat+=8){ for(int lat=8;lat<=lmax;lat+=8){