portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Merge branch 'develop' into feature/mpi3

Browse Source
This commit is contained in:
paboyle 2016-10-25 06:02:33 +01:00
commit b820076b91
5 changed files with 30 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
lib/FFT.h
View File

@ -200,18 +200,14 @@ namespace Grid {
sign,FFTW_ESTIMATE);
}
double add,mul,fma;
FFTW<scalar>::fftw_flops(p,&add,&mul,&fma);
flops_call = add+mul+2.0*fma;
GridStopWatch timer;
std::vector<int> lcoor(Nd), gcoor(Nd);
// Barrel shift and collect global pencil
for(int p=0;p<processors[dim];p++) {
for(int idx=0;idx<sgrid->lSites();idx++) {
std::vector<int> lcoor(Nd);
sgrid->LocalIndexToLocalCoor(idx,lcoor);
sobj s;
@ -228,14 +224,11 @@ namespace Grid {
// Loop over orthog coords
int NN=pencil_g.lSites();
GridStopWatch Timer;
Timer.Start();
GridStopWatch timer;
timer.Start();
PARALLEL_FOR_LOOP
for(int idx=0;idx<NN;idx++) {
std::vector<int> lcoor(Nd);
for(int idx=0;idx<NN;idx++) {
pencil_g.LocalIndexToLocalCoor(idx,lcoor);
if ( lcoor[dim] == 0 ) { // restricts loop to plane at lcoor[dim]==0
@ -245,15 +238,17 @@ PARALLEL_FOR_LOOP
}
}
Timer.Stop();
usec += Timer.useconds();
flops+= flops_call*NN;
timer.Stop();
double add,mul,fma;
FFTW<scalar>::fftw_flops(p,&add,&mul,&fma);
flops_call = add+mul+2.0*fma;
usec += timer.useconds();
flops+= flops_call*NN;
int pc = processor_coor[dim];
for(int idx=0;idx<sgrid->lSites();idx++) {
std::vector<int> lcoor(Nd);
for(int idx=0;idx<sgrid->lSites();idx++) {
sgrid->LocalIndexToLocalCoor(idx,lcoor);
std::vector<int> gcoor = lcoor;
gcoor = lcoor;
// extract the result
sobj s;
gcoor[dim] = lcoor[dim]+L*pc;

2
lib/cartesian/Cartesian_base.h
View File

@ -77,7 +77,7 @@ public:
// GridCartesian / GridRedBlackCartesian
////////////////////////////////////////////////////////////////
virtual int CheckerBoarded(int dim)=0;
virtual int CheckerBoard(std::vector<int> site)=0;
virtual int CheckerBoard(std::vector<int> &site)=0;
virtual int CheckerBoardDestination(int source_cb,int shift,int dim)=0;
virtual int CheckerBoardShift(int source_cb,int dim,int shift,int osite)=0;
virtual int CheckerBoardShiftForCB(int source_cb,int dim,int shift,int cb)=0;

2
lib/cartesian/Cartesian_full.h
View File

@ -49,7 +49,7 @@ public:
virtual int CheckerBoarded(int dim){
return 0;
}
virtual int CheckerBoard(std::vector<int> site){
virtual int CheckerBoard(std::vector<int> &site){
return 0;
}
virtual int CheckerBoardDestination(int cb,int shift,int dim){

2
lib/cartesian/Cartesian_red_black.h
View File

@ -49,7 +49,7 @@ public:
if( dim==_checker_dim) return 1;
else return 0;
}
virtual int CheckerBoard(std::vector<int> site){
virtual int CheckerBoard(std::vector<int> &site){
int linear=0;
assert(site.size()==_ndimension);
for(int d=0;d<_ndimension;d++){

27
lib/lattice/Lattice_peekpoke.h
View File

@ -154,7 +154,7 @@ PARALLEL_FOR_LOOP
template<class vobj,class sobj>
void peekLocalSite(sobj &s,const Lattice<vobj> &l,std::vector<int> &site){
GridBase *grid=l._grid;
GridBase *grid = l._grid;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
@ -164,16 +164,18 @@ PARALLEL_FOR_LOOP
assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj));
static const int words=sizeof(vobj)/sizeof(vector_type);
int odx,idx;
idx= grid->iIndex(site);
odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd);
extract(l._odata[odx],buf);
scalar_type * vp = (scalar_type *)&l._odata[odx];
scalar_type * pt = (scalar_type *)&s;
for(int w=0;w<words;w++){
pt[w] = vp[idx+w*Nsimd];
}
s = buf[idx];
return;
};
@ -190,18 +192,17 @@ PARALLEL_FOR_LOOP
assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj));
static const int words=sizeof(vobj)/sizeof(vector_type);
int odx,idx;
idx= grid->iIndex(site);
odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd);
// extract-modify-merge cycle is easiest way and this is not perf critical
extract(l._odata[odx],buf);
scalar_type * vp = (scalar_type *)&l._odata[odx];
scalar_type * pt = (scalar_type *)&s;
buf[idx] = s;
merge(l._odata[odx],buf);
for(int w=0;w<words;w++){
vp[idx+w*Nsimd] = pt[w];
}
return;
};