portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-15 14:27:06 +01:00
Code Releases Activity

Merge remote-tracking branch 'upstream/develop' into feature/wilsonmg

Browse Source
This commit is contained in:
Daniel Richtmann
2018-05-15 14:57:18 +02:00
parent 73ced656eb a61e0df54b
commit 61812ab7f1
57 changed files with 4559 additions and 2279 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
lib/lattice/Lattice_base.h
View File

@ -256,9 +256,42 @@ public:
_odata[ss]=r._odata[ss];
}
}
Lattice(Lattice&& r){ // move constructor
_grid = r._grid;
checkerboard = r.checkerboard;
_odata=std::move(r._odata);
}
inline Lattice<vobj> & operator = (Lattice<vobj> && r)
{
_grid = r._grid;
checkerboard = r.checkerboard;
_odata =std::move(r._odata);
return *this;
}
inline Lattice<vobj> & operator = (const Lattice<vobj> & r){
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
parallel_for(int ss=0;ss<_grid->oSites();ss++){
_odata[ss]=r._odata[ss];
}
return *this;
}
template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
this->checkerboard = r.checkerboard;
conformable(*this,r);
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
return *this;
}
virtual ~Lattice(void) = default;
void reset(GridBase* grid) {
@ -277,15 +310,6 @@ public:
return *this;
}
template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
this->checkerboard = r.checkerboard;
conformable(*this,r);
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
return *this;
}
// *=,+=,-= operators inherit behvour from correspond */+/- operation
template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) {

18
lib/lattice/Lattice_comparison_utils.h
View File

@ -179,7 +179,7 @@ namespace Grid {
return ret;
}
#define DECLARE_RELATIONAL(op,functor) \
#define DECLARE_RELATIONAL_EQ(op,functor) \
template<class vsimd,IfSimd<vsimd> = 0>\
inline vInteger operator op (const vsimd & lhs, const vsimd & rhs)\
{\
@ -198,11 +198,6 @@ namespace Grid {
typedef typename vsimd::scalar_type scalar;\
return Comparison(functor<scalar,scalar>(),lhs,rhs);\
}\
template<class vsimd,IfSimd<vsimd> = 0>\
inline vInteger operator op(const iScalar<vsimd> &lhs,const iScalar<vsimd> &rhs)\
{ \
return lhs._internal op rhs._internal; \
} \
template<class vsimd>\
inline vInteger operator op(const iScalar<vsimd> &lhs,const typename vsimd::scalar_type &rhs) \
{ \
@ -212,14 +207,21 @@ namespace Grid {
inline vInteger operator op(const typename vsimd::scalar_type &lhs,const iScalar<vsimd> &rhs) \
{ \
return lhs op rhs._internal; \
}
} \
#define DECLARE_RELATIONAL(op,functor) \
DECLARE_RELATIONAL_EQ(op,functor) \
template<class vsimd>\
inline vInteger operator op(const iScalar<vsimd> &lhs,const iScalar<vsimd> &rhs)\
{ \
return lhs._internal op rhs._internal; \
}
DECLARE_RELATIONAL(<,slt);
DECLARE_RELATIONAL(<=,sle);
DECLARE_RELATIONAL(>,sgt);
DECLARE_RELATIONAL(>=,sge);
DECLARE_RELATIONAL(==,seq);
DECLARE_RELATIONAL_EQ(==,seq);
DECLARE_RELATIONAL(!=,sne);
#undef DECLARE_RELATIONAL

93
lib/lattice/Lattice_transfer.h
View File

@ -606,6 +606,51 @@ unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
extract1(in_vobj, out_ptrs, 0);
}
}
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type
unvectorizeToRevLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
{
typedef typename vobj::vector_type vtype;
GridBase* in_grid = in._grid;
out.resize(in_grid->lSites());
int ndim = in_grid->Nd();
int in_nsimd = vtype::Nsimd();
std::vector<std::vector<int> > in_icoor(in_nsimd);
for(int lane=0; lane < in_nsimd; lane++){
in_icoor[lane].resize(ndim);
in_grid->iCoorFromIindex(in_icoor[lane], lane);
}
parallel_for(int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++){ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> out_ptrs(in_nsimd);
std::vector<int> in_ocoor(ndim);
in_grid->oCoorFromOindex(in_ocoor, in_oidx);
std::vector<int> lcoor(in_grid->Nd());
for(int lane=0; lane < in_nsimd; lane++){
for(int mu=0;mu<ndim;mu++)
lcoor[mu] = in_ocoor[mu] + in_grid->_rdimensions[mu]*in_icoor[lane][mu];
int lex;
Lexicographic::IndexFromCoorReversed(lcoor, lex, in_grid->_ldimensions);
out_ptrs[lane] = &out[lex];
}
//Unpack into those ptrs
const vobj & in_vobj = in._odata[in_oidx];
extract1(in_vobj, out_ptrs, 0);
}
}
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value
@ -655,6 +700,54 @@ vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
}
}
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value
&& !isSIMDvectorized<sobj>::value, void>::type
vectorizeFromRevLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
{
typedef typename vobj::vector_type vtype;
GridBase* grid = out._grid;
assert(in.size()==grid->lSites());
int ndim = grid->Nd();
int nsimd = vtype::Nsimd();
std::vector<std::vector<int> > icoor(nsimd);
for(int lane=0; lane < nsimd; lane++){
icoor[lane].resize(ndim);
grid->iCoorFromIindex(icoor[lane],lane);
}
parallel_for(uint64_t oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> ptrs(nsimd);
std::vector<int> ocoor(ndim);
grid->oCoorFromOindex(ocoor, oidx);
std::vector<int> lcoor(grid->Nd());
for(int lane=0; lane < nsimd; lane++){
for(int mu=0;mu<ndim;mu++){
lcoor[mu] = ocoor[mu] + grid->_rdimensions[mu]*icoor[lane][mu];
}
int lex;
Lexicographic::IndexFromCoorReversed(lcoor, lex, grid->_ldimensions);
ptrs[lane] = &in[lex];
}
//pack from those ptrs
vobj vecobj;
merge1(vecobj, ptrs, 0);
out._odata[oidx] = vecobj;
}
}
//Convert a Lattice from one precision to another
template<class VobjOut, class VobjIn>
void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){