portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-08-16 19:21:53 +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

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