portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Merge branch 'develop' into feature/hadrons-new-memory-model

Browse Source
This commit is contained in:
Antonin Portelli 2017-12-06 16:49:21 +01:00
commit 29e2eddea8
6 changed files with 45 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
lib/communicator/Communicator_base.h
View File

@ -276,10 +276,11 @@ class CartesianCommunicator {
assert(in.size()==out.size());
uint64_t bytes=sizeof(T);
uint64_t words=in.size()/numnode;
// std:: cout << "AllToAll buffer size "<< in.size()*sizeof(T)<<std::endl;
// std:: cout << "AllToAll datum bytes "<< bytes<<std::endl;
// std:: cout << "AllToAll datum count "<< words<<std::endl;
assert(numnode * words == in.size());
assert(words < (1ULL<<32));
assert(words < (1ULL<<31));
AllToAll(dim,(void *)&in[0],(void *)&out[0],words,bytes);
}
void AllToAll(int dim ,void *in,void *out,uint64_t words,uint64_t bytes);

42
lib/lattice/Lattice_rng.h
View File

@ -77,9 +77,6 @@ namespace Grid {
// merge of April 11 2017
//<<<<<<< HEAD
// this function is necessary for the LS vectorised field
inline int RNGfillable_general(GridBase *coarse,GridBase *fine)
{
@ -91,7 +88,6 @@ namespace Grid {
// all further divisions are local
for(int d=0;d<lowerdims;d++) assert(fine->_processors[d]==1);
for(int d=0;d<rngdims;d++) assert(coarse->_processors[d] == fine->_processors[d+lowerdims]);
// then divide the number of local sites
// check that the total number of sims agree, meanse the iSites are the same
@ -102,27 +98,6 @@ namespace Grid {
return fine->lSites() / coarse->lSites();
}
/*
// Wrap seed_seq to give common interface with random_device
class fixedSeed {
public:
typedef std::seed_seq::result_type result_type;
std::seed_seq src;
fixedSeed(const std::vector<int> &seeds) : src(seeds.begin(),seeds.end()) {};
result_type operator () (void){
std::vector<result_type> list(1);
src.generate(list.begin(),list.end());
return list[0];
}
};
=======
>>>>>>> develop
*/
// real scalars are one component
template<class scalar,class distribution,class generator>
@ -171,7 +146,7 @@ namespace Grid {
// support for parallel init
///////////////////////
#ifdef RNG_FAST_DISCARD
static void Skip(RngEngine &eng)
static void Skip(RngEngine &eng,uint64_t site)
{
/////////////////////////////////////////////////////////////////////////////////////
// Skip by 2^40 elements between successive lattice sites
@ -184,8 +159,11 @@ namespace Grid {
// and margin of safety is orders of magnitude.
// We could hack Sitmo to skip in the higher order words of state if necessary
/////////////////////////////////////////////////////////////////////////////////////
uint64_t skip = 0x1; skip = skip<<40;
// uint64_t skip = site+1; // Old init Skipped then drew. Checked compat with faster init
uint64_t skip = site;
skip = skip<<40;
eng.discard(skip);
// std::cout << " Engine " <<site << " state " <<eng<<std::endl;
}
#endif
static RngEngine Reseed(RngEngine &eng)
@ -407,15 +385,14 @@ namespace Grid {
// MT implementation does not implement fast discard even though
// in principle this is possible
////////////////////////////////////////////////
std::vector<int> gcoor;
int rank,o_idx,i_idx;
// Everybody loops over global volume.
for(int gidx=0;gidx<_grid->_gsites;gidx++){
Skip(master_engine); // Skip to next RNG sequence
parallel_for(int gidx=0;gidx<_grid->_gsites;gidx++){
// Where is it?
int rank,o_idx,i_idx;
std::vector<int> gcoor;
_grid->GlobalIndexToGlobalCoor(gidx,gcoor);
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
@ -423,6 +400,7 @@ namespace Grid {
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
}
}

12
lib/lattice/Lattice_transfer.h
View File

@ -822,6 +822,7 @@ void Grid_split(std::vector<Lattice<Vobj> > & full,Lattice<Vobj> & split)
// Loop over reordered data post A2A
parallel_for(int c=0;c<chunk;c++){
std::vector<int> coor(ndim);
for(int m=0;m<M;m++){
for(int s=0;s<sP;s++){
@ -833,7 +834,6 @@ void Grid_split(std::vector<Lattice<Vobj> > & full,Lattice<Vobj> & split)
uint64_t lex_vec = lex_fvol_vec/fvol;
// which node sets an adder to the coordinate
std::vector<int> coor(ndim);
Lexicographic::CoorFromIndex(coor, lex_fvol, ldims);
coor[d] += m*ldims[d];
Lexicographic::IndexFromCoor(coor, lex_r, rdims);
@ -940,10 +940,11 @@ void Grid_unsplit(std::vector<Lattice<Vobj> > & full,Lattice<Vobj> & split)
{
// Loop over reordered data post A2A
for(int c=0;c<chunk;c++){
parallel_for(int c=0;c<chunk;c++){
std::vector<int> coor(ndim);
for(int m=0;m<M;m++){
for(int s=0;s<sP;s++){
// addressing; use lexico
int lex_r;
uint64_t lex_c = c+chunk*m+chunk*M*s;
@ -952,7 +953,6 @@ void Grid_unsplit(std::vector<Lattice<Vobj> > & full,Lattice<Vobj> & split)
uint64_t lex_vec = lex_fvol_vec/fvol;
// which node sets an adder to the coordinate
std::vector<int> coor(ndim);
Lexicographic::CoorFromIndex(coor, lex_fvol, ldims);
coor[d] += m*ldims[d];
Lexicographic::IndexFromCoor(coor, lex_r, rdims);
@ -978,9 +978,9 @@ void Grid_unsplit(std::vector<Lattice<Vobj> > & full,Lattice<Vobj> & split)
lsites = full_grid->lSites();
for(int v=0;v<nvector;v++){
assert(v<full.size());
// assert(v<full.size());
parallel_for(int site=0;site<lsites;site++){
assert(v*lsites+site < alldata.size());
// assert(v*lsites+site < alldata.size());
scalardata[site] = alldata[v*lsites+site];
}
vectorizeFromLexOrdArray(scalardata,full[v]);

1
lib/qcd/action/fermion/FermionOperator.h
View File

@ -47,6 +47,7 @@ namespace Grid {
INHERIT_IMPL_TYPES(Impl);
FermionOperator(const ImplParams &p= ImplParams()) : Impl(p) {};
virtual ~FermionOperator(void) = default;
virtual FermionField &tmp(void) = 0;

6
lib/qcd/utils/SUn.h
View File

@ -746,7 +746,7 @@ template<typename GaugeField,typename GaugeMat>
}
}
template<typename GaugeField>
static void ColdConfiguration(GridParallelRNG &pRNG,GaugeField &out){
static void ColdConfiguration(GaugeField &out){
typedef typename GaugeField::vector_type vector_type;
typedef iSUnMatrix<vector_type> vMatrixType;
typedef Lattice<vMatrixType> LatticeMatrixType;
@ -757,6 +757,10 @@ template<typename GaugeField,typename GaugeMat>
PokeIndex<LorentzIndex>(out,Umu,mu);
}
}
template<typename GaugeField>
static void ColdConfiguration(GridParallelRNG &pRNG,GaugeField &out){
ColdConfiguration(out);
}
template<typename LatticeMatrixType>
static void taProj( const LatticeMatrixType &in, LatticeMatrixType &out){

34
tests/solver/Test_dwf_mrhs_cg_mpi.cc
View File

@ -81,21 +81,20 @@ int main (int argc, char ** argv)
GridCartesian * SFGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,SGrid);
GridRedBlackCartesian * SrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(SGrid);
GridRedBlackCartesian * SFrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,SGrid);
std::cout << GridLogMessage << "Made the grids"<<std::endl;
///////////////////////////////////////////////
// Set up the problem as a 4d spreadout job
///////////////////////////////////////////////
std::vector<int> seeds({1,2,3,4});
GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds);
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
std::vector<FermionField> src(nrhs,FGrid);
std::vector<FermionField> src_chk(nrhs,FGrid);
std::vector<FermionField> result(nrhs,FGrid);
FermionField tmp(FGrid);
std::cout << GridLogMessage << "Made the Fermion Fields"<<std::endl;
for(int s=0;s<nrhs;s++) result[s]=zero;
#define LEXICO_TEST
#undef LEXICO_TEST
#ifdef LEXICO_TEST
{
LatticeFermion lex(FGrid); lex = zero;
@ -117,6 +116,7 @@ int main (int argc, char ** argv)
}
}
#else
GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds);
for(int s=0;s<nrhs;s++) {
random(pRNG5,src[s]);
tmp = 100.0*s;
@ -124,13 +124,21 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << " src ["<<s<<"] "<<norm2(src[s])<<std::endl;
}
#endif
std::cout << GridLogMessage << "Intialised the Fermion Fields"<<std::endl;
for(int n =0 ; n< nrhs ; n++) {
// std::cout << " src"<<n<<"\n"<< src[n] <<std::endl;
LatticeGaugeField Umu(UGrid);
if(1) {
GridParallelRNG pRNG(UGrid );
std::cout << GridLogMessage << "Intialising 4D RNG "<<std::endl;
pRNG.SeedFixedIntegers(seeds);
std::cout << GridLogMessage << "Intialised 4D RNG "<<std::endl;
SU3::HotConfiguration(pRNG,Umu);
std::cout << "Intialised the HOT Gauge Field"<<std::endl;
// std::cout << " Site zero "<< Umu._odata[0] <<std::endl;
} else {
SU3::ColdConfiguration(Umu);
std::cout << GridLogMessage << "Intialised the COLD Gauge Field"<<std::endl;
}
LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu);
/////////////////
// MPI only sends
/////////////////
@ -139,13 +147,13 @@ int main (int argc, char ** argv)
FermionField s_tmp(SFGrid);
FermionField s_res(SFGrid);
std::cout << GridLogMessage << "Made the split grid fields"<<std::endl;
///////////////////////////////////////////////////////////////
// split the source out using MPI instead of I/O
///////////////////////////////////////////////////////////////
Grid_split (Umu,s_Umu);
Grid_split (src,s_src);
std::cout << GridLogMessage << " split rank " <<me << " s_src "<<norm2(s_src)<<std::endl;
// std::cout << " s_src\n "<< s_src <<std::endl;
#ifdef LEXICO_TEST
FermionField s_src_tmp(SFGrid);
@ -169,16 +177,12 @@ int main (int argc, char ** argv)
}
s_src_diff = s_src_tmp - s_src;
std::cout << GridLogMessage <<" LEXICO test: s_src_diff " << norm2(s_src_diff)<<std::endl;
// std::cout << " s_src \n" << s_src << std::endl;
// std::cout << " s_src_tmp \n" << s_src_tmp << std::endl;
// std::cout << " s_src_diff \n" << s_src_diff << std::endl;
// exit(0);
#endif
///////////////////////////////////////////////////////////////
// Set up N-solvers as trivially parallel
///////////////////////////////////////////////////////////////
std::cout << GridLogMessage << " Building the solvers"<<std::endl;
RealD mass=0.01;
RealD M5=1.8;
DomainWallFermionR Dchk(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5);