portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-07-25 17:07:07 +01:00
Code Releases Activity

Merge remote-tracking branch 'origin/develop' into temporary-smearing

Browse Source
This commit is contained in:
Guido Cossu
2016-07-04 17:28:40 +01:00
parent 6ce174cd60 d6737e4bd8
commit 9cb90f714e
107 changed files with 7839 additions and 4572 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
lib/lattice/Lattice_base.h
View File

@@ -55,7 +55,13 @@ extern int GridCshiftPermuteMap[4][16];
// Basic expressions used in Expression Template
////////////////////////////////////////////////
class LatticeBase {};
class LatticeBase
{
public:
virtual ~LatticeBase(void) = default;
GridBase *_grid;
};
class LatticeExpressionBase {};
template<class T> using Vector = std::vector<T,alignedAllocator<T> >; // Aligned allocator??
@@ -88,8 +94,6 @@ template<class vobj>
class Lattice : public LatticeBase
{
public:
GridBase *_grid;
int checkerboard;
Vector<vobj> _odata;
@@ -177,8 +181,8 @@ PARALLEL_FOR_LOOP
}
//GridFromExpression is tricky to do
template<class Op,class T1>
Lattice(const LatticeUnaryExpression<Op,T1> & expr): _grid(nullptr){
Lattice(const LatticeUnaryExpression<Op,T1> & expr) {
_grid = nullptr;
GridFromExpression(_grid,expr);
assert(_grid!=nullptr);
@@ -199,7 +203,8 @@ PARALLEL_FOR_LOOP
}
};
template<class Op,class T1, class T2>
Lattice(const LatticeBinaryExpression<Op,T1,T2> & expr): _grid(nullptr){
Lattice(const LatticeBinaryExpression<Op,T1,T2> & expr) {
_grid = nullptr;
GridFromExpression(_grid,expr);
assert(_grid!=nullptr);
@@ -220,7 +225,8 @@ PARALLEL_FOR_LOOP
}
};
template<class Op,class T1, class T2, class T3>
Lattice(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr): _grid(nullptr){
Lattice(const LatticeTrinaryExpression<Op,T1,T2,T3> & expr) {
_grid = nullptr;
GridFromExpression(_grid,expr);
assert(_grid!=nullptr);
@@ -240,7 +246,8 @@ PARALLEL_FOR_LOOP
// Constructor requires "grid" passed.
// what about a default grid?
//////////////////////////////////////////////////////////////////
Lattice(GridBase *grid) : _grid(grid), _odata(_grid->oSites()) {
Lattice(GridBase *grid) : _odata(grid->oSites()) {
_grid = grid;
// _odata.reserve(_grid->oSites());
// _odata.resize(_grid->oSites());
// std::cout << "Constructing lattice object with Grid pointer "<<_grid<<std::endl;
@@ -248,6 +255,8 @@ PARALLEL_FOR_LOOP
checkerboard=0;
}
virtual ~Lattice(void) = default;
template<class sobj> strong_inline Lattice<vobj> & operator = (const sobj & r){
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){

2
lib/lattice/Lattice_peekpoke.h
View File

@@ -152,7 +152,7 @@ PARALLEL_FOR_LOOP
// Peek a scalar object from the SIMD array
//////////////////////////////////////////////////////////
template<class vobj,class sobj>
void peekLocalSite(sobj &s,Lattice<vobj> &l,std::vector<int> &site){
void peekLocalSite(sobj &s,const Lattice<vobj> &l,std::vector<int> &site){
GridBase *grid=l._grid;

4
lib/lattice/Lattice_reduction.h
View File

@@ -152,7 +152,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
assert(grid!=NULL);
// FIXME
std::cout<<GridLogMessage<<"WARNING ! SliceSum is unthreaded "<<grid->SumArraySize()<<" threads "<<std::endl;
// std::cout<<GridLogMessage<<"WARNING ! SliceSum is unthreaded "<<grid->SumArraySize()<<" threads "<<std::endl;
const int Nd = grid->_ndimension;
const int Nsimd = grid->Nsimd();
@@ -178,7 +178,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
// sum over reduced dimension planes, breaking out orthog dir
for(int ss=0;ss<grid->oSites();ss++){
GridBase::CoorFromIndex(coor,ss,grid->_rdimensions);
Lexicographic::CoorFromIndex(coor,ss,grid->_rdimensions);
int r = coor[orthogdim];
lvSum[r]=lvSum[r]+Data._odata[ss];
}

25
lib/lattice/Lattice_rng.h
View File

@@ -75,7 +75,7 @@ namespace Grid {
std::seed_seq src;
fixedSeed(std::vector<int> &seeds) : src(seeds.begin(),seeds.end()) {};
fixedSeed(const std::vector<int> &seeds) : src(seeds.begin(),seeds.end()) {};
result_type operator () (void){
@@ -119,9 +119,10 @@ namespace Grid {
typedef uint32_t RngStateType;
static const int RngStateCount = std::mt19937::state_size;
#endif
std::vector<RngEngine> _generators;
std::vector<std::uniform_real_distribution<RealD> > _uniform;
std::vector<std::normal_distribution<RealD> > _gaussian;
std::vector<RngEngine> _generators;
std::vector<std::uniform_real_distribution<RealD>> _uniform;
std::vector<std::normal_distribution<RealD>> _gaussian;
std::vector<std::discrete_distribution<int32_t>> _bernoulli;
void GetState(std::vector<RngStateType> & saved,int gen) {
saved.resize(RngStateCount);
@@ -161,6 +162,7 @@ namespace Grid {
_generators.resize(1);
_uniform.resize(1,std::uniform_real_distribution<RealD>{0,1});
_gaussian.resize(1,std::normal_distribution<RealD>(0.0,1.0) );
_bernoulli.resize(1,std::discrete_distribution<int32_t>{1,1});
_seeded=0;
}
@@ -242,7 +244,7 @@ namespace Grid {
std::random_device rd;
Seed(rd);
}
void SeedFixedIntegers(std::vector<int> &seeds){
void SeedFixedIntegers(const std::vector<int> &seeds){
fixedSeed src(seeds);
Seed(src);
}
@@ -266,6 +268,7 @@ namespace Grid {
_generators.resize(_vol);
_uniform.resize(_vol,std::uniform_real_distribution<RealD>{0,1});
_gaussian.resize(_vol,std::normal_distribution<RealD>(0.0,1.0) );
_bernoulli.resize(_vol,std::discrete_distribution<int32_t>{1,1});
_seeded=0;
}
@@ -354,7 +357,7 @@ PARALLEL_FOR_LOOP
std::random_device rd;
Seed(rd);
}
void SeedFixedIntegers(std::vector<int> &seeds){
void SeedFixedIntegers(const std::vector<int> &seeds){
fixedSeed src(seeds);
Seed(src);
}
@@ -368,14 +371,22 @@ PARALLEL_FOR_LOOP
template <class vobj> inline void gaussian(GridParallelRNG &rng,Lattice<vobj> &l){
rng.fill(l,rng._gaussian);
}
template <class vobj> inline void bernoulli(GridParallelRNG &rng,Lattice<vobj> &l){
rng.fill(l,rng._bernoulli);
}
template <class sobj> inline void random(GridSerialRNG &rng,sobj &l){
rng.fill(l,rng._uniform);
}
template <class sobj> inline void gaussian(GridSerialRNG &rng,sobj &l){
rng.fill(l,rng._gaussian);
}
template <class sobj> inline void bernoulli(GridSerialRNG &rng,sobj &l){
rng.fill(l,rng._bernoulli);
}
}
#endif

138
lib/lattice/Lattice_transfer.h
View File

@@ -44,7 +44,7 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
}
}
////////////////////////////////////////////////////////////////////////////////////////////
// remove and insert a half checkerboard
////////////////////////////////////////////////////////////////////////////////////////////
@@ -115,9 +115,9 @@ inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
int sc;
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
for(int i=0;i<nbasis;i++) {
@@ -160,9 +160,9 @@ PARALLEL_FOR_LOOP
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
// z = A x + y
fineZ._odata[sf]=coarseA._odata[sc]*fineX._odata[sf]+fineY._odata[sf];
@@ -225,9 +225,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
coarseData._odata[sc]=coarseData._odata[sc]+fineData._odata[sf];
@@ -311,9 +311,9 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
for(int i=0;i<nbasis;i++) {
if(i==0) fineData._odata[sf]=coarseData._odata[sc](i) * Basis[i]._odata[sf];
@@ -325,6 +325,126 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
}
// Useful for precision conversion, or indeed anything where an operator= does a conversion on scalars.
// Simd layouts need not match since we use peek/poke Local
template<class vobj,class vvobj>
void localConvert(const Lattice<vobj> &in,Lattice<vvobj> &out)
{
typedef typename vobj::scalar_object sobj;
typedef typename vvobj::scalar_object ssobj;
sobj s;
ssobj ss;
GridBase *ig = in._grid;
GridBase *og = out._grid;
int ni = ig->_ndimension;
int no = og->_ndimension;
assert(ni == no);
for(int d=0;d<no;d++){
assert(ig->_processors[d] == og->_processors[d]);
assert(ig->_ldimensions[d] == og->_ldimensions[d]);
}
PARALLEL_FOR_LOOP
for(int idx=0;idx<ig->lSites();idx++){
std::vector<int> lcoor(ni);
ig->LocalIndexToLocalCoor(idx,lcoor);
peekLocalSite(s,in,lcoor);
ss=s;
pokeLocalSite(ss,out,lcoor);
}
}
template<class vobj>
void InsertSlice(Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int orthog)
{
typedef typename vobj::scalar_object sobj;
sobj s;
GridBase *lg = lowDim._grid;
GridBase *hg = higherDim._grid;
int nl = lg->_ndimension;
int nh = hg->_ndimension;
assert(nl+1 == nh);
assert(orthog<nh);
assert(orthog>=0);
assert(hg->_processors[orthog]==1);
int dl; dl = 0;
for(int d=0;d<nh;d++){
if ( d != orthog) {
assert(lg->_processors[dl] == hg->_processors[d]);
assert(lg->_ldimensions[dl] == hg->_ldimensions[d]);
dl++;
}
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
lg->LocalIndexToLocalCoor(idx,lcoor);
dl=0;
hcoor[orthog] = slice;
for(int d=0;d<nh;d++){
if ( d!=orthog ) {
hcoor[d]=lcoor[dl++];
}
}
peekLocalSite(s,lowDim,lcoor);
pokeLocalSite(s,higherDim,hcoor);
}
}
template<class vobj>
void ExtractSlice(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice, int orthog)
{
typedef typename vobj::scalar_object sobj;
sobj s;
GridBase *lg = lowDim._grid;
GridBase *hg = higherDim._grid;
int nl = lg->_ndimension;
int nh = hg->_ndimension;
assert(nl+1 == nh);
assert(orthog<nh);
assert(orthog>=0);
assert(hg->_processors[orthog]==1);
int dl; dl = 0;
for(int d=0;d<nh;d++){
if ( d != orthog) {
assert(lg->_processors[dl] == hg->_processors[d]);
assert(lg->_ldimensions[dl] == hg->_ldimensions[d]);
dl++;
}
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
lg->LocalIndexToLocalCoor(idx,lcoor);
dl=0;
hcoor[orthog] = slice;
for(int d=0;d<nh;d++){
if ( d!=orthog ) {
hcoor[d]=lcoor[dl++];
}
}
peekLocalSite(s,higherDim,hcoor);
pokeLocalSite(s,lowDim,lcoor);
}
}
template<class vobj>
void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)