portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 09:15:38 +01:00
Code Releases Activity

Thread construct changes

Browse Source
This commit is contained in:
paboyle 2018-01-24 13:39:06 +00:00
parent 733f8ff0b2
commit d70709a8e8
1 changed files with 22 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
lib/lattice/Lattice_rng.h
View File

@ -247,7 +247,7 @@ public:
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l)); CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}; }
template <class distribution> inline void fill(ComplexF &l,std::vector<distribution> &dist){ template <class distribution> inline void fill(ComplexF &l,std::vector<distribution> &dist){
dist[0].reset(); dist[0].reset();
@ -346,7 +346,7 @@ public:
int osites = _grid->oSites(); // guaranteed to be <= l._grid->oSites() by a factor multiplicity int osites = _grid->oSites(); // guaranteed to be <= l._grid->oSites() by a factor multiplicity
int words = sizeof(scalar_object) / sizeof(scalar_type); int words = sizeof(scalar_object) / sizeof(scalar_type);
parallel_for(int ss=0;ss<osites;ss++){ thread_loop( (int ss=0;ss<osites;ss++), {
std::vector<scalar_object> buf(Nsimd); std::vector<scalar_object> buf(Nsimd);
for (int m = 0; m < multiplicity; m++) { // Draw from same generator multiplicity times for (int m = 0; m < multiplicity; m++) { // Draw from same generator multiplicity times
@ -363,10 +363,10 @@ public:
// merge into SIMD lanes, FIXME suboptimal implementation // merge into SIMD lanes, FIXME suboptimal implementation
merge(l._odata[sm], buf); merge(l._odata[sm], buf);
} }
} });
_time_counter += usecond()- inner_time_counter; _time_counter += usecond()- inner_time_counter;
}; }
void SeedFixedIntegers(const std::vector<int> &seeds){ void SeedFixedIntegers(const std::vector<int> &seeds){
@ -387,23 +387,23 @@ public:
//////////////////////////////////////////////// ////////////////////////////////////////////////
// Everybody loops over global volume. // Everybody loops over global volume.
parallel_for(int gidx=0;gidx<_grid->_gsites;gidx++){ thread_loop( (int gidx=0;gidx<_grid->_gsites;gidx++) , {
// Where is it?
int rank;
int o_idx;
int i_idx;
std::vector<int> gcoor;
// Where is it? _grid->GlobalIndexToGlobalCoor(gidx,gcoor);
int rank,o_idx,i_idx; _grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
std::vector<int> gcoor;
// If this is one of mine we take it
_grid->GlobalIndexToGlobalCoor(gidx,gcoor); if( rank == _grid->ThisRank() ){
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor); int l_idx=generator_idx(o_idx,i_idx);
_generators[l_idx] = master_engine;
// If this is one of mine we take it Skip(_generators[l_idx],gidx); // Skip to next RNG sequence
if( rank == _grid->ThisRank() ){ }
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 #else
//////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////
// Machine and thread decomposition dependent seeding is efficient // Machine and thread decomposition dependent seeding is efficient
@ -429,7 +429,7 @@ public:
seeders[t] = Reseed(master_engine); seeders[t] = Reseed(master_engine);
} }
parallel_for(int t=0;t<Nthread;t++) { thread_loop( (int t=0;t<Nthread;t++), {
// set up one per local site in threaded fashion // set up one per local site in threaded fashion
std::vector<uint32_t> newseeds; std::vector<uint32_t> newseeds;
std::uniform_int_distribution<uint32_t> uid; std::uniform_int_distribution<uint32_t> uid;
@ -438,7 +438,7 @@ public:
_generators[l] = Reseed(seeders[t],newseeds,uid); _generators[l] = Reseed(seeders[t],newseeds,uid);
} }
} }
} });
} }
#endif #endif
} }