portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

Merge branch 'paboyle:develop' into sycl_slicesum_update

Browse Source
This commit is contained in:
dbollweg
2024-03-13 18:06:17 -04:00
committed by GitHub
parent 31f9971dbf 89c0519f83
commit fee65d7a75
8 changed files with 159 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/lattice/Lattice_reduction.h
View File

@ -281,12 +281,14 @@ inline ComplexD rankInnerProduct(const Lattice<vobj> &left,const Lattice<vobj> &
return nrm;
}
template<class vobj>
inline ComplexD innerProduct(const Lattice<vobj> &left,const Lattice<vobj> &right) {
GridBase *grid = left.Grid();
ComplexD nrm = rankInnerProduct(left,right);
// std::cerr<<"flight log " << std::hexfloat << nrm <<" "<<crc(left)<<std::endl;
// GridNormLog(real(nrm)); // Could log before and after global sum to distinguish local and MPI
grid->GlobalSum(nrm);
// GridNormLog(real(nrm));
return nrm;
}

26
Grid/lattice/Lattice_rng.h
View File

@ -411,7 +411,7 @@ public:
std::cout << GridLogMessage << "Seed SHA256: " << GridChecksum::sha256_string(seeds) << std::endl;
SeedFixedIntegers(seeds);
}
void SeedFixedIntegers(const std::vector<int> &seeds){
void SeedFixedIntegers(const std::vector<int> &seeds, int britney=0){
// Everyone generates the same seed_seq based on input seeds
CartesianCommunicator::BroadcastWorld(0,(void *)&seeds[0],sizeof(int)*seeds.size());
@ -428,7 +428,6 @@ public:
// MT implementation does not implement fast discard even though
// in principle this is possible
////////////////////////////////////////////////
#if 1
thread_for( lidx, _grid->lSites(), {
int gidx;
@ -449,29 +448,12 @@ public:
int l_idx=generator_idx(o_idx,i_idx);
_generators[l_idx] = master_engine;
Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
});
#else
// Everybody loops over global volume.
thread_for( gidx, _grid->_gsites, {
// Where is it?
int rank;
int o_idx;
int i_idx;
Coordinate gcoor;
_grid->GlobalIndexToGlobalCoor(gidx,gcoor);
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
// If this is one of mine we take it
if( rank == _grid->ThisRank() ){
int l_idx=generator_idx(o_idx,i_idx);
_generators[l_idx] = master_engine;
if ( britney ) {
Skip(_generators[l_idx],l_idx); // Skip to next RNG sequence
} else {
Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
}
});
#endif
#else
////////////////////////////////////////////////////////////////
// Machine and thread decomposition dependent seeding is efficient