portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-03 21:44:33 +00:00
Code Releases Activity

HMC bit repro across checkpoints. Fixed parallel RNG issue with threading.

Browse Source 
Conclusion: c++11 distributions not thread safe and must us distinct dist as well as distinct engine
per site. Makes sense when you think of box muller. Also added a reset of dist on fill to ensure
repro across checkpoints.
This commit is contained in:
paboyle
2015-12-22 08:54:40 +00:00
parent 0abfbcc8eb
commit 08edbb5cbe
7 changed files with 210 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
lib/lattice/Lattice_rng.h
View File

@@ -22,13 +22,16 @@ namespace Grid {
assert(fine->_processors[d]==1);
}
// local and global volumes subdivide cleanly after SIMDization
int multiplicity=1;
for(int d=0;d<lowerdims;d++){
multiplicity=multiplicity*fine->_rdimensions[d];
}
// local and global volumes subdivide cleanly after SIMDization
for(int d=0;d<rngdims;d++){
int fd= d+lowerdims;
assert(coarse->_processors[d] == fine->_processors[fd]);
assert(coarse->_simd_layout[d] == fine->_simd_layout[fd]);
assert((fine->_rdimensions[fd] / coarse->_rdimensions[d])* coarse->_rdimensions[d]==fine->_rdimensions[fd]);
assert(((fine->_rdimensions[fd] / coarse->_rdimensions[d])* coarse->_rdimensions[d])==fine->_rdimensions[fd]);
multiplicity = multiplicity *fine->_rdimensions[fd] / coarse->_rdimensions[d];
}
@@ -76,8 +79,6 @@ namespace Grid {
public:
GridRNGbase() : _uniform{0,1}, _gaussian(0.0,1.0) {};
int _seeded;
// One generator per site.
// Uniform and Gaussian distributions from these generators.
@@ -91,13 +92,12 @@ namespace Grid {
static const int RngStateCount = std::mt19937::state_size;
#endif
std::vector<RngEngine> _generators;
std::uniform_real_distribution<double> _uniform;
std::normal_distribution<double> _gaussian;
std::vector<std::uniform_real_distribution<RealD> > _uniform;
std::vector<std::normal_distribution<RealD> > _gaussian;
void GetState(std::vector<RngStateType> & saved,int gen) {
saved.resize(RngStateCount);
std::stringstream ss;
// std::cout << _generators[gen]<<std::endl;
ss<<_generators[gen];
ss.seekg(0,ss.beg);
for(int i=0;i<RngStateCount;i++){
@@ -112,7 +112,6 @@ namespace Grid {
}
ss.seekg(0,ss.beg);
ss>>_generators[gen];
// std::cout << _generators[gen]<<std::endl;
}
};
@@ -132,12 +131,14 @@ namespace Grid {
GridSerialRNG() : GridRNGbase() {
_generators.resize(1);
_uniform.resize(1,std::uniform_real_distribution<RealD>{0,1});
_gaussian.resize(1,std::normal_distribution<RealD>(0.0,1.0) );
_seeded=0;
}
template <class sobj,class distribution> inline void fill(sobj &l,distribution &dist){
template <class sobj,class distribution> inline void fill(sobj &l,std::vector<distribution> &dist){
typedef typename sobj::scalar_type scalar_type;
@@ -145,56 +146,65 @@ namespace Grid {
scalar_type *buf = (scalar_type *) & l;
dist[0].reset();
for(int idx=0;idx<words;idx++){
fillScalar(buf[idx],dist,_generators[0]);
fillScalar(buf[idx],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
};
template <class distribution> inline void fill(ComplexF &l,distribution &dist){
fillScalar(l,dist,_generators[0]);
template <class distribution> inline void fill(ComplexF &l,std::vector<distribution> &dist){
dist[0].reset();
fillScalar(l,dist[0],_generators[0]);
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(ComplexD &l,distribution &dist){
fillScalar(l,dist,_generators[0]);
template <class distribution> inline void fill(ComplexD &l,std::vector<distribution> &dist){
dist[0].reset();
fillScalar(l,dist[0],_generators[0]);
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(RealF &l,distribution &dist){
fillScalar(l,dist,_generators[0]);
template <class distribution> inline void fill(RealF &l,std::vector<distribution> &dist){
dist[0].reset();
fillScalar(l,dist[0],_generators[0]);
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(RealD &l,distribution &dist){
fillScalar(l,dist,_generators[0]);
template <class distribution> inline void fill(RealD &l,std::vector<distribution> &dist){
dist[0].reset();
fillScalar(l,dist[0],_generators[0]);
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
// vector fill
template <class distribution> inline void fill(vComplexF &l,distribution &dist){
template <class distribution> inline void fill(vComplexF &l,std::vector<distribution> &dist){
RealF *pointer=(RealF *)&l;
dist[0].reset();
for(int i=0;i<2*vComplexF::Nsimd();i++){
fillScalar(pointer[i],dist,_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(vComplexD &l,distribution &dist){
template <class distribution> inline void fill(vComplexD &l,std::vector<distribution> &dist){
RealD *pointer=(RealD *)&l;
dist[0].reset();
for(int i=0;i<2*vComplexD::Nsimd();i++){
fillScalar(pointer[i],dist,_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(vRealF &l,distribution &dist){
template <class distribution> inline void fill(vRealF &l,std::vector<distribution> &dist){
RealF *pointer=(RealF *)&l;
dist[0].reset();
for(int i=0;i<vRealF::Nsimd();i++){
fillScalar(pointer[i],dist,_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
template <class distribution> inline void fill(vRealD &l,distribution &dist){
template <class distribution> inline void fill(vRealD &l,std::vector<distribution> &dist){
RealD *pointer=(RealD *)&l;
dist[0].reset();
for(int i=0;i<vRealD::Nsimd();i++){
fillScalar(pointer[i],dist,_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
@@ -213,10 +223,6 @@ namespace Grid {
class GridParallelRNG : public GridRNGbase {
public:
// Uniform and Gaussian distributions from these generators.
std::uniform_real_distribution<double> _uniform;
std::normal_distribution<double> _gaussian;
GridBase *_grid;
int _vol;
@@ -228,7 +234,10 @@ namespace Grid {
GridParallelRNG(GridBase *grid) : GridRNGbase() {
_grid=grid;
_vol =_grid->iSites()*_grid->oSites();
_generators.resize(_vol);
_uniform.resize(_vol,std::uniform_real_distribution<RealD>{0,1});
_gaussian.resize(_vol,std::normal_distribution<RealD>(0.0,1.0) );
_seeded=0;
}
@@ -277,7 +286,7 @@ namespace Grid {
//void SaveState(const std::string<char> &file);
//void LoadState(const std::string<char> &file);
template <class vobj,class distribution> inline void fill(Lattice<vobj> &l,distribution &dist){
template <class vobj,class distribution> inline void fill(Lattice<vobj> &l,std::vector<distribution> &dist){
typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::scalar_type scalar_type;
@@ -301,8 +310,9 @@ PARALLEL_FOR_LOOP
for(int si=0;si<Nsimd;si++){
int gdx = generator_idx(ss,si); // index of generator state
scalar_type *pointer = (scalar_type *)&buf[si];
dist[gdx].reset();
for(int idx=0;idx<words;idx++){
fillScalar(pointer[idx],dist,_generators[gdx]);
fillScalar(pointer[idx],dist[gdx],_generators[gdx]);
}
}