portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-12 20:27:06 +01:00
Code Releases Activity

Global changes to parallel_for structure.

Browse Source 
Move the comms flags to more sensible names
This commit is contained in:
paboyle
2017-02-21 05:24:27 -05:00
parent 3906cd2149
commit 3ae92fa2e6
43 changed files with 271 additions and 513 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
lib/lattice/Lattice_arith.h
View File

@ -39,8 +39,7 @@ namespace Grid {
ret.checkerboard = lhs.checkerboard;
conformable(ret,rhs);
conformable(lhs,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
mult(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@ -56,8 +55,7 @@ PARALLEL_FOR_LOOP
ret.checkerboard = lhs.checkerboard;
conformable(ret,rhs);
conformable(lhs,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
mac(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@ -73,8 +71,7 @@ PARALLEL_FOR_LOOP
ret.checkerboard = lhs.checkerboard;
conformable(ret,rhs);
conformable(lhs,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
sub(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@ -89,8 +86,7 @@ PARALLEL_FOR_LOOP
ret.checkerboard = lhs.checkerboard;
conformable(ret,rhs);
conformable(lhs,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
add(&tmp,&lhs._odata[ss],&rhs._odata[ss]);
@ -108,8 +104,7 @@ PARALLEL_FOR_LOOP
void mult(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.checkerboard = lhs.checkerboard;
conformable(lhs,ret);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
obj1 tmp;
mult(&tmp,&lhs._odata[ss],&rhs);
vstream(ret._odata[ss],tmp);
@ -120,8 +115,7 @@ PARALLEL_FOR_LOOP
void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.checkerboard = lhs.checkerboard;
conformable(ret,lhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
obj1 tmp;
mac(&tmp,&lhs._odata[ss],&rhs);
vstream(ret._odata[ss],tmp);
@ -132,8 +126,7 @@ PARALLEL_FOR_LOOP
void sub(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.checkerboard = lhs.checkerboard;
conformable(ret,lhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
sub(&tmp,&lhs._odata[ss],&rhs);
@ -147,8 +140,7 @@ PARALLEL_FOR_LOOP
void add(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
ret.checkerboard = lhs.checkerboard;
conformable(lhs,ret);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
add(&tmp,&lhs._odata[ss],&rhs);
@ -166,8 +158,7 @@ PARALLEL_FOR_LOOP
void mult(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
mult(&tmp,&lhs,&rhs._odata[ss]);
@ -182,8 +173,7 @@ PARALLEL_FOR_LOOP
void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
mac(&tmp,&lhs,&rhs._odata[ss]);
@ -198,8 +188,7 @@ PARALLEL_FOR_LOOP
void sub(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
sub(&tmp,&lhs,&rhs._odata[ss]);
@ -213,8 +202,7 @@ PARALLEL_FOR_LOOP
void add(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
#ifdef STREAMING_STORES
obj1 tmp;
add(&tmp,&lhs,&rhs._odata[ss]);
@ -230,8 +218,7 @@ PARALLEL_FOR_LOOP
ret.checkerboard = x.checkerboard;
conformable(ret,x);
conformable(x,y);
PARALLEL_FOR_LOOP
for(int ss=0;ss<x._grid->oSites();ss++){
parallel_for(int ss=0;ss<x._grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = a*x._odata[ss]+y._odata[ss];
vstream(ret._odata[ss],tmp);
@ -245,8 +232,7 @@ PARALLEL_FOR_LOOP
ret.checkerboard = x.checkerboard;
conformable(ret,x);
conformable(x,y);
PARALLEL_FOR_LOOP
for(int ss=0;ss<x._grid->oSites();ss++){
parallel_for(int ss=0;ss<x._grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = a*x._odata[ss]+b*y._odata[ss];
vstream(ret._odata[ss],tmp);

27
lib/lattice/Lattice_base.h
View File

@ -121,8 +121,7 @@ public:
assert( (cb==Odd) || (cb==Even));
checkerboard=cb;
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = eval(ss,expr);
vstream(_odata[ss] ,tmp);
@ -144,8 +143,7 @@ PARALLEL_FOR_LOOP
assert( (cb==Odd) || (cb==Even));
checkerboard=cb;
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = eval(ss,expr);
vstream(_odata[ss] ,tmp);
@ -167,8 +165,7 @@ PARALLEL_FOR_LOOP
assert( (cb==Odd) || (cb==Even));
checkerboard=cb;
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
#ifdef STREAMING_STORES
//vobj tmp = eval(ss,expr);
vstream(_odata[ss] ,eval(ss,expr));
@ -191,8 +188,7 @@ PARALLEL_FOR_LOOP
checkerboard=cb;
_odata.resize(_grid->oSites());
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = eval(ss,expr);
vstream(_odata[ss] ,tmp);
@ -213,8 +209,7 @@ PARALLEL_FOR_LOOP
checkerboard=cb;
_odata.resize(_grid->oSites());
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
#ifdef STREAMING_STORES
vobj tmp = eval(ss,expr);
vstream(_odata[ss] ,tmp);
@ -235,8 +230,7 @@ PARALLEL_FOR_LOOP
checkerboard=cb;
_odata.resize(_grid->oSites());
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
vstream(_odata[ss] ,eval(ss,expr));
}
};
@ -258,8 +252,7 @@ PARALLEL_FOR_LOOP
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
_odata[ss]=r._odata[ss];
}
}
@ -269,8 +262,7 @@ PARALLEL_FOR_LOOP
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++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r;
}
return *this;
@ -279,8 +271,7 @@ PARALLEL_FOR_LOOP
this->checkerboard = r.checkerboard;
conformable(*this,r);
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
return *this;

154
lib/lattice/Lattice_comparison.h
View File

@ -45,90 +45,87 @@ namespace Grid {
//////////////////////////////////////////////////////////////////////////
template<class vfunctor,class lobj,class robj>
inline Lattice<vInteger> LLComparison(vfunctor op,const Lattice<lobj> &lhs,const Lattice<robj> &rhs)
{
Lattice<vInteger> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs._odata[ss]);
}
return ret;
{
Lattice<vInteger> ret(rhs._grid);
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs._odata[ss]);
}
return ret;
}
//////////////////////////////////////////////////////////////////////////
// compare lattice to scalar
//////////////////////////////////////////////////////////////////////////
template<class vfunctor,class lobj,class robj>
template<class vfunctor,class lobj,class robj>
inline Lattice<vInteger> LSComparison(vfunctor op,const Lattice<lobj> &lhs,const robj &rhs)
{
Lattice<vInteger> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs);
}
return ret;
{
Lattice<vInteger> ret(lhs._grid);
parallel_for(int ss=0;ss<lhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs);
}
return ret;
}
//////////////////////////////////////////////////////////////////////////
// compare scalar to lattice
//////////////////////////////////////////////////////////////////////////
template<class vfunctor,class lobj,class robj>
template<class vfunctor,class lobj,class robj>
inline Lattice<vInteger> SLComparison(vfunctor op,const lobj &lhs,const Lattice<robj> &rhs)
{
Lattice<vInteger> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs);
}
return ret;
{
Lattice<vInteger> ret(rhs._grid);
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=op(lhs._odata[ss],rhs);
}
return ret;
}
//////////////////////////////////////////////////////////////////////////
// Map to functors
//////////////////////////////////////////////////////////////////////////
// Less than
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vlt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vlt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vlt<lobj,robj>(),lhs,rhs);
}
// Less than equal
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vle<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vle<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vle<lobj,robj>(),lhs,rhs);
}
// Greater than
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vgt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vgt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const lobj & lhs, const Lattice<robj> & rhs) {
// Less than
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vlt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vlt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator < (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vlt<lobj,robj>(),lhs,rhs);
}
// Less than equal
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vle<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vle<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator <= (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vle<lobj,robj>(),lhs,rhs);
}
// Greater than
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vgt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vgt<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator > (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vgt<lobj,robj>(),lhs,rhs);
}
// Greater than equal
}
// Greater than equal
template<class lobj,class robj>
inline Lattice<vInteger> operator >= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator >= (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vge<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
@ -136,38 +133,37 @@ PARALLEL_FOR_LOOP
return LSComparison(vge<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator >= (const lobj & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator >= (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vge<lobj,robj>(),lhs,rhs);
}
// equal
template<class lobj,class robj>
inline Lattice<vInteger> operator == (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator == (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(veq<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator == (const Lattice<lobj> & lhs, const robj & rhs) {
inline Lattice<vInteger> operator == (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(veq<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator == (const lobj & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator == (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(veq<lobj,robj>(),lhs,rhs);
}
// not equal
template<class lobj,class robj>
inline Lattice<vInteger> operator != (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator != (const Lattice<lobj> & lhs, const Lattice<robj> & rhs) {
return LLComparison(vne<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator != (const Lattice<lobj> & lhs, const robj & rhs) {
inline Lattice<vInteger> operator != (const Lattice<lobj> & lhs, const robj & rhs) {
return LSComparison(vne<lobj,robj>(),lhs,rhs);
}
template<class lobj,class robj>
inline Lattice<vInteger> operator != (const lobj & lhs, const Lattice<robj> & rhs) {
inline Lattice<vInteger> operator != (const lobj & lhs, const Lattice<robj> & rhs) {
return SLComparison(vne<lobj,robj>(),lhs,rhs);
}
}
#endif

55
lib/lattice/Lattice_local.h
View File

@ -34,47 +34,42 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
namespace Grid {
/////////////////////////////////////////////////////
// Non site, reduced locally reduced routines
/////////////////////////////////////////////////////
// localNorm2,
template<class vobj>
/////////////////////////////////////////////////////
// Non site, reduced locally reduced routines
/////////////////////////////////////////////////////
// localNorm2,
template<class vobj>
inline auto localNorm2 (const Lattice<vobj> &rhs)-> Lattice<typename vobj::tensor_reduced>
{
Lattice<typename vobj::tensor_reduced> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=innerProduct(rhs._odata[ss],rhs._odata[ss]);
}
return ret;
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=innerProduct(rhs._odata[ss],rhs._odata[ss]);
}
return ret;
}
// localInnerProduct
template<class vobj>
// localInnerProduct
template<class vobj>
inline auto localInnerProduct (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs) -> Lattice<typename vobj::tensor_reduced>
{
Lattice<typename vobj::tensor_reduced> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=innerProduct(lhs._odata[ss],rhs._odata[ss]);
}
return ret;
}
// outerProduct Scalar x Scalar -> Scalar
// Vector x Vector -> Matrix
template<class ll,class rr>
// outerProduct Scalar x Scalar -> Scalar
// Vector x Vector -> Matrix
template<class ll,class rr>
inline auto outerProduct (const Lattice<ll> &lhs,const Lattice<rr> &rhs) -> Lattice<decltype(outerProduct(lhs._odata[0],rhs._odata[0]))>
{
Lattice<decltype(outerProduct(lhs._odata[0],rhs._odata[0]))> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=outerProduct(lhs._odata[ss],rhs._odata[ss]);
}
return ret;
}
{
Lattice<decltype(outerProduct(lhs._odata[0],rhs._odata[0]))> ret(rhs._grid);
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
ret._odata[ss]=outerProduct(lhs._odata[ss],rhs._odata[ss]);
}
return ret;
}
}
#endif

24
lib/lattice/Lattice_overload.h
View File

@ -37,8 +37,7 @@ namespace Grid {
inline Lattice<vobj> operator -(const Lattice<vobj> &r)
{
Lattice<vobj> ret(r._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<r._grid->oSites();ss++){
parallel_for(int ss=0;ss<r._grid->oSites();ss++){
vstream(ret._odata[ss], -r._odata[ss]);
}
return ret;
@ -74,8 +73,7 @@ PARALLEL_FOR_LOOP
inline auto operator * (const left &lhs,const Lattice<right> &rhs) -> Lattice<decltype(lhs*rhs._odata[0])>
{
Lattice<decltype(lhs*rhs._odata[0])> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
decltype(lhs*rhs._odata[0]) tmp=lhs*rhs._odata[ss];
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs*rhs._odata[ss];
@ -86,8 +84,7 @@ PARALLEL_FOR_LOOP
inline auto operator + (const left &lhs,const Lattice<right> &rhs) -> Lattice<decltype(lhs+rhs._odata[0])>
{
Lattice<decltype(lhs+rhs._odata[0])> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
decltype(lhs+rhs._odata[0]) tmp =lhs-rhs._odata[ss];
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs+rhs._odata[ss];
@ -98,11 +95,9 @@ PARALLEL_FOR_LOOP
inline auto operator - (const left &lhs,const Lattice<right> &rhs) -> Lattice<decltype(lhs-rhs._odata[0])>
{
Lattice<decltype(lhs-rhs._odata[0])> ret(rhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
decltype(lhs-rhs._odata[0]) tmp=lhs-rhs._odata[ss];
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs-rhs._odata[ss];
}
return ret;
}
@ -110,8 +105,7 @@ PARALLEL_FOR_LOOP
inline auto operator * (const Lattice<left> &lhs,const right &rhs) -> Lattice<decltype(lhs._odata[0]*rhs)>
{
Lattice<decltype(lhs._odata[0]*rhs)> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites(); ss++){
decltype(lhs._odata[0]*rhs) tmp =lhs._odata[ss]*rhs;
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs._odata[ss]*rhs;
@ -122,8 +116,7 @@ PARALLEL_FOR_LOOP
inline auto operator + (const Lattice<left> &lhs,const right &rhs) -> Lattice<decltype(lhs._odata[0]+rhs)>
{
Lattice<decltype(lhs._odata[0]+rhs)> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
decltype(lhs._odata[0]+rhs) tmp=lhs._odata[ss]+rhs;
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs._odata[ss]+rhs;
@ -134,15 +127,12 @@ PARALLEL_FOR_LOOP
inline auto operator - (const Lattice<left> &lhs,const right &rhs) -> Lattice<decltype(lhs._odata[0]-rhs)>
{
Lattice<decltype(lhs._odata[0]-rhs)> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites(); ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites(); ss++){
decltype(lhs._odata[0]-rhs) tmp=lhs._odata[ss]-rhs;
vstream(ret._odata[ss],tmp);
// ret._odata[ss]=lhs._odata[ss]-rhs;
}
return ret;
}
}
#endif

37
lib/lattice/Lattice_peekpoke.h
View File

@ -44,22 +44,20 @@ namespace Grid {
{
Lattice<decltype(peekIndex<Index>(lhs._odata[0],i))> ret(lhs._grid);
ret.checkerboard=lhs.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = peekIndex<Index>(lhs._odata[ss],i);
}
return ret;
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = peekIndex<Index>(lhs._odata[ss],i);
}
return ret;
};
template<int Index,class vobj>
auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekIndex<Index>(lhs._odata[0],i,j))>
auto PeekIndex(const Lattice<vobj> &lhs,int i,int j) -> Lattice<decltype(peekIndex<Index>(lhs._odata[0],i,j))>
{
Lattice<decltype(peekIndex<Index>(lhs._odata[0],i,j))> ret(lhs._grid);
ret.checkerboard=lhs.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = peekIndex<Index>(lhs._odata[ss],i,j);
}
return ret;
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = peekIndex<Index>(lhs._odata[ss],i,j);
}
return ret;
};
////////////////////////////////////////////////////////////////////////////////////////////////////
@ -68,18 +66,16 @@ PARALLEL_FOR_LOOP
template<int Index,class vobj>
void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(lhs._odata[0],0))> & rhs,int i)
{
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
pokeIndex<Index>(lhs._odata[ss],rhs._odata[ss],i);
}
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
pokeIndex<Index>(lhs._odata[ss],rhs._odata[ss],i);
}
}
template<int Index,class vobj>
void PokeIndex(Lattice<vobj> &lhs,const Lattice<decltype(peekIndex<Index>(lhs._odata[0],0,0))> & rhs,int i,int j)
{
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
pokeIndex<Index>(lhs._odata[ss],rhs._odata[ss],i,j);
}
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
pokeIndex<Index>(lhs._odata[ss],rhs._odata[ss],i,j);
}
}
//////////////////////////////////////////////////////
@ -131,9 +127,6 @@ PARALLEL_FOR_LOOP
assert( l.checkerboard == l._grid->CheckerBoard(site));
// FIXME
// assert( sizeof(sobj)*Nsimd == sizeof(vobj));
int rank,odx,idx;
grid->GlobalCoorToRankIndex(rank,odx,idx,site);

10
lib/lattice/Lattice_reality.h
View File

@ -40,8 +40,7 @@ namespace Grid {
template<class vobj> inline Lattice<vobj> adj(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = adj(lhs._odata[ss]);
}
return ret;
@ -49,13 +48,10 @@ PARALLEL_FOR_LOOP
template<class vobj> inline Lattice<vobj> conjugate(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = conjugate(lhs._odata[ss]);
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = conjugate(lhs._odata[ss]);
}
return ret;
};
}
#endif

12
lib/lattice/Lattice_reduction.h
View File

@ -57,8 +57,7 @@ namespace Grid {
sumarray[i]=zero;
}
PARALLEL_FOR_LOOP
for(int thr=0;thr<grid->SumArraySize();thr++){
parallel_for(int thr=0;thr<grid->SumArraySize();thr++){
int nwork, mywork, myoff;
GridThread::GetWork(left._grid->oSites(),thr,mywork,myoff);
@ -68,7 +67,7 @@ PARALLEL_FOR_LOOP
}
sumarray[thr]=TensorRemove(vnrm) ;
}
vector_type vvnrm; vvnrm=zero; // sum across threads
for(int i=0;i<grid->SumArraySize();i++){
vvnrm = vvnrm+sumarray[i];
@ -114,18 +113,17 @@ PARALLEL_FOR_LOOP
sumarray[i]=zero;
}
PARALLEL_FOR_LOOP
for(int thr=0;thr<grid->SumArraySize();thr++){
parallel_for(int thr=0;thr<grid->SumArraySize();thr++){
int nwork, mywork, myoff;
GridThread::GetWork(grid->oSites(),thr,mywork,myoff);
vobj vvsum=zero;
for(int ss=myoff;ss<mywork+myoff; ss++){
vvsum = vvsum + arg._odata[ss];
}
sumarray[thr]=vvsum;
}
vobj vsum=zero; // sum across threads
for(int i=0;i<grid->SumArraySize();i++){
vsum = vsum+sumarray[i];

3
lib/lattice/Lattice_rng.h
View File

@ -302,8 +302,7 @@ namespace Grid {
int words=sizeof(scalar_object)/sizeof(scalar_type);
PARALLEL_FOR_LOOP
for(int ss=0;ss<osites;ss++){
parallel_for(int ss=0;ss<osites;ss++){
std::vector<scalar_object> buf(Nsimd);
for(int m=0;m<multiplicity;m++) {// Draw from same generator multiplicity times

6
lib/lattice/Lattice_trace.h
View File

@ -42,8 +42,7 @@ namespace Grid {
-> Lattice<decltype(trace(lhs._odata[0]))>
{
Lattice<decltype(trace(lhs._odata[0]))> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = trace(lhs._odata[ss]);
}
return ret;
@ -56,8 +55,7 @@ PARALLEL_FOR_LOOP
inline auto TraceIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(traceIndex<Index>(lhs._odata[0]))>
{
Lattice<decltype(traceIndex<Index>(lhs._odata[0]))> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = traceIndex<Index>(lhs._odata[ss]);
}
return ret;

35
lib/lattice/Lattice_transfer.h
View File

@ -51,7 +51,7 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full){
half.checkerboard = cb;
int ssh=0;
//PARALLEL_FOR_LOOP
//parallel_for
for(int ss=0;ss<full._grid->oSites();ss++){
std::vector<int> coor;
int cbos;
@ -68,7 +68,7 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half){
int cb = half.checkerboard;
int ssh=0;
//PARALLEL_FOR_LOOP
//parallel_for
for(int ss=0;ss<full._grid->oSites();ss++){
std::vector<int> coor;
int cbos;
@ -153,8 +153,7 @@ inline void blockZAXPY(Lattice<vobj> &fineZ,
assert(block_r[d]*coarse->_rdimensions[d]==fine->_rdimensions[d]);
}
PARALLEL_FOR_LOOP
for(int sf=0;sf<fine->oSites();sf++){
parallel_for(int sf=0;sf<fine->oSites();sf++){
int sc;
std::vector<int> coor_c(_ndimension);
@ -186,8 +185,7 @@ template<class vobj,class CComplex>
fine_inner = localInnerProduct(fineX,fineY);
blockSum(coarse_inner,fine_inner);
PARALLEL_FOR_LOOP
for(int ss=0;ss<coarse->oSites();ss++){
parallel_for(int ss=0;ss<coarse->oSites();ss++){
CoarseInner._odata[ss] = coarse_inner._odata[ss];
}
}
@ -347,8 +345,7 @@ void localConvert(const Lattice<vobj> &in,Lattice<vvobj> &out)
assert(ig->lSites() == og->lSites());
}
PARALLEL_FOR_LOOP
for(int idx=0;idx<ig->lSites();idx++){
parallel_for(int idx=0;idx<ig->lSites();idx++){
sobj s;
ssobj ss;
@ -386,8 +383,7 @@ void InsertSlice(Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
parallel_for(int idx=0;idx<lg->lSites();idx++){
sobj s;
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -428,8 +424,7 @@ void ExtractSlice(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice, in
}
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
parallel_for(int idx=0;idx<lg->lSites();idx++){
sobj s;
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -468,8 +463,7 @@ void InsertSliceLocal(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
parallel_for(int idx=0;idx<lg->lSites();idx++){
sobj s;
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -504,8 +498,7 @@ void ExtractSliceLocal(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slic
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
parallel_for(int idx=0;idx<lg->lSites();idx++){
sobj s;
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -574,8 +567,7 @@ typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>
in_grid->iCoorFromIindex(in_icoor[lane], lane);
}
PARALLEL_FOR_LOOP
for(int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++){ //loop over outer index
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);
@ -623,8 +615,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
std::vector<SobjOut> in_slex_conv(in_grid->lSites());
unvectorizeToLexOrdArray(in_slex_conv, in);
PARALLEL_FOR_LOOP
for(int out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
parallel_for(int out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
std::vector<int> out_ocoor(ndim);
out_grid->oCoorFromOindex(out_ocoor, out_oidx);
@ -642,10 +633,6 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
merge(out._odata[out_oidx], ptrs, 0);
}
}
}
#endif

37
lib/lattice/Lattice_transpose.h
View File

@ -40,27 +40,24 @@ namespace Grid {
////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vobj>
inline Lattice<vobj> transpose(const Lattice<vobj> &lhs){
Lattice<vobj> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = transpose(lhs._odata[ss]);
}
return ret;
};
Lattice<vobj> ret(lhs._grid);
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = transpose(lhs._odata[ss]);
}
return ret;
};
////////////////////////////////////////////////////////////////////////////////////////////////////
// Index level dependent transpose
////////////////////////////////////////////////////////////////////////////////////////////////////
template<int Index,class vobj>
////////////////////////////////////////////////////////////////////////////////////////////////////
// Index level dependent transpose
////////////////////////////////////////////////////////////////////////////////////////////////////
template<int Index,class vobj>
inline auto TransposeIndex(const Lattice<vobj> &lhs) -> Lattice<decltype(transposeIndex<Index>(lhs._odata[0]))>
{
Lattice<decltype(transposeIndex<Index>(lhs._odata[0]))> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = transposeIndex<Index>(lhs._odata[ss]);
}
return ret;
};
{
Lattice<decltype(transposeIndex<Index>(lhs._odata[0]))> ret(lhs._grid);
parallel_for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = transposeIndex<Index>(lhs._odata[ss]);
}
return ret;
};
}
#endif

12
lib/lattice/Lattice_unary.h
View File

@ -37,8 +37,7 @@ namespace Grid {
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=pow(rhs._odata[ss],y);
}
return ret;
@ -47,8 +46,7 @@ PARALLEL_FOR_LOOP
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=mod(rhs._odata[ss],y);
}
return ret;
@ -58,8 +56,7 @@ PARALLEL_FOR_LOOP
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=div(rhs._odata[ss],y);
}
return ret;
@ -69,8 +66,7 @@ PARALLEL_FOR_LOOP
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=Exponentiate(rhs._odata[ss],alpha, Nexp);
}
return ret;

3
lib/lattice/Lattice_where.h
View File

@ -56,8 +56,7 @@ inline void whereWolf(Lattice<vobj> &ret,const Lattice<iobj> &predicate,Lattice<
std::vector<scalar_object> truevals (Nsimd);
std::vector<scalar_object> falsevals(Nsimd);
PARALLEL_FOR_LOOP
for(int ss=0;ss<iftrue._grid->oSites(); ss++){
parallel_for(int ss=0;ss<iftrue._grid->oSites(); ss++){
extract(iftrue._odata[ss] ,truevals);
extract(iffalse._odata[ss] ,falsevals);