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

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
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
benchmarks/Benchmark_memory_asynch.cc
View File

@ -77,8 +77,7 @@ int main (int argc, char ** argv)
}
double start=usecond();
PARALLEL_FOR_LOOP
for(int t=0;t<threads;t++){
parallel_for(int t=0;t<threads;t++){
sum[t] = x[t]._odata[0];
for(int i=0;i<Nloop;i++){

8
lib/Init.cc
View File

@ -342,11 +342,11 @@ void Grid_init(int *argc,char ***argv)
} else {
QCD::WilsonKernelsStatic::Comms = QCD::WilsonKernelsStatic::CommsThenCompute;
}
if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-isend") ){
CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicyIsend);
if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-concurrent") ){
CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicyConcurrent);
}
if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-sendrecv") ){
CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicySendrecv);
if( GridCmdOptionExists(*argv,*argv+*argc,"--comms-sequential") ){
CartesianCommunicator::SetCommunicatorPolicy(CartesianCommunicator::CommunicatorPolicySequential);
}
if( GridCmdOptionExists(*argv,*argv+*argc,"--lebesgue") ){
LebesgueOrder::UseLebesgueOrder=1;

83
lib/Stencil.h
View File

@ -81,77 +81,14 @@ template<class vobj,class cobj,class compressor>
void Gather_plane_simple_table (std::vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so)
{
int num=table.size();
PARALLEL_FOR_LOOP
for(int i=0;i<num;i++){
parallel_for(int i=0;i<num;i++){
vstream(buffer[off+table[i].first],compress(rhs._odata[so+table[i].second]));
// buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
}
}
///////////////////////////////////////////////////////////////////
// Gather for when there *is* need to SIMD split with compression
///////////////////////////////////////////////////////////////////
/*
template<class cobj,class vobj,class compressor> double
Gather_plane_exchange(const Lattice<vobj> &rhs,
std::vector<cobj *> pointers,int dimension,int plane,int cbmask,compressor &compress,int type)
{
int rd = rhs._grid->_rdimensions[dimension];
double t1,t2;
if ( !rhs._grid->CheckerBoarded(dimension) ) {
cbmask = 0x3;
}
int so = plane*rhs._grid->_ostride[dimension]; // base offset for start of plane
int e1 =rhs._grid->_slice_nblock[dimension];
int e2 =rhs._grid->_slice_block [dimension];
int n1 =rhs._grid->_slice_stride[dimension];
// Need to switch to a table loop
std::vector<std::pair<int,int> > table;
if ( cbmask ==0x3){
for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o = n*n1;
int offset = b+n*e2;
table.push_back(std::pair<int,int> (offset,o+b));
}
}
} else {
// Case of SIMD split AND checker dim cannot currently be hit, except in
// Test_cshift_red_black code.
for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o=n*n1;
int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);
int offset = b+n*e2;
if ( ocb & cbmask ) {
table.push_back(std::pair<int,int> (offset,o+b));
}
}
}
}
assert( (table.size()&0x1)==0);
t1=usecond();
PARALLEL_FOR_LOOP
for(int j=0;j<table.size()/2;j++){
// buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
cobj temp1 =compress(rhs._odata[so+table[2*j].second]);
cobj temp2 =compress(rhs._odata[so+table[2*j+1].second]);
cobj temp3;
cobj temp4;
exchange(temp3,temp4,temp1,temp2,type);
vstream(pointers[0][j],temp3);
vstream(pointers[1][j],temp4);
}
t2=usecond();
return t2-t1;
}
*/
template<class cobj,class vobj,class compressor>
void Gather_plane_exchange_table(const Lattice<vobj> &rhs,
std::vector<cobj *> pointers,int dimension,int plane,int cbmask,compressor &compress,int type) __attribute__((noinline));
@ -164,8 +101,7 @@ void Gather_plane_exchange_table(std::vector<std::pair<int,int> >& table,const L
assert( (table.size()&0x1)==0);
int num=table.size()/2;
int so = plane*rhs._grid->_ostride[dimension]; // base offset for start of plane
PARALLEL_FOR_LOOP
for(int j=0;j<num;j++){
parallel_for(int j=0;j<num;j++){
// buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
cobj temp1 =compress(rhs._odata[so+table[2*j].second]);
cobj temp2 =compress(rhs._odata[so+table[2*j+1].second]);
@ -235,19 +171,14 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
Packets[i].recv_buf,
Packets[i].from_rank,
Packets[i].bytes);
if( _grid->CommunicatorPolicy == CartesianCommunicator::CommunicatorPolicySendrecv ) {
_grid->StencilSendToRecvFromComplete(reqs[i]);
}
}
commtime+=usecond();
}
void CommunicateComplete(std::vector<std::vector<CommsRequest_t> > &reqs)
{
commtime-=usecond();
if( _grid->CommunicatorPolicy == CartesianCommunicator::CommunicatorPolicyIsend ) {
for(int i=0;i<Packets.size();i++){
_grid->StencilSendToRecvFromComplete(reqs[i]);
}
for(int i=0;i<Packets.size();i++){
_grid->StencilSendToRecvFromComplete(reqs[i]);
}
_grid->StencilBarrier();// Synch shared memory on a single nodes
commtime+=usecond();
@ -327,14 +258,12 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
// std::ofstream fout(fname);
if ( Mergers[i].exchange == 0 ) {
PARALLEL_FOR_LOOP
for(int o=0;o<Mergers[i].buffer_size;o++){
parallel_for(int o=0;o<Mergers[i].buffer_size;o++){
merge1(Mergers[i].mpointer[o],Mergers[i].rpointers,o);
// fout<<o<<" "<<Mergers[i].mpointer[o]<<std::endl;
}
} else {
PARALLEL_FOR_LOOP
for(int o=0;o<Mergers[i].buffer_size/2;o++){
parallel_for(int o=0;o<Mergers[i].buffer_size/2;o++){
exchange(Mergers[i].mpointer[2*o],Mergers[i].mpointer[2*o+1],
Mergers[i].vpointers[0][o],Mergers[i].vpointers[1][o],Mergers[i].type);
// cobj temp1,temp2;

3
lib/Threads.h
View File

@ -51,6 +51,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define PARALLEL_CRITICAL
#endif
#define parallel_for PARALLEL_FOR_LOOP for
#define parallel_for_nest2 PARALLEL_NESTED_LOOP2 for
namespace Grid {
// Introduce a class to gain deterministic bit reproducible reduction.

6
lib/algorithms/CoarsenedMatrix.h
View File

@ -267,8 +267,7 @@ namespace Grid {
SimpleCompressor<siteVector> compressor;
Stencil.HaloExchange(in,compressor);
PARALLEL_FOR_LOOP
for(int ss=0;ss<Grid()->oSites();ss++){
parallel_for(int ss=0;ss<Grid()->oSites();ss++){
siteVector res = zero;
siteVector nbr;
int ptype;
@ -380,8 +379,7 @@ PARALLEL_FOR_LOOP
Subspace.ProjectToSubspace(oProj,oblock);
// blockProject(iProj,iblock,Subspace.subspace);
// blockProject(oProj,oblock,Subspace.subspace);
PARALLEL_FOR_LOOP
for(int ss=0;ss<Grid()->oSites();ss++){
parallel_for(int ss=0;ss<Grid()->oSites();ss++){
for(int j=0;j<nbasis;j++){
if( disp!= 0 ) {
A[p]._odata[ss](j,i) = oProj._odata[ss](j);

2
lib/communicator/Communicator_base.cc
View File

@ -33,7 +33,7 @@ namespace Grid {
///////////////////////////////////////////////////////////////
void * CartesianCommunicator::ShmCommBuf;
uint64_t CartesianCommunicator::MAX_MPI_SHM_BYTES = 128*1024*1024;
CartesianCommunicator::CommunicatorPolicy_t CartesianCommunicator::CommunicatorPolicy= CartesianCommunicator::CommunicatorPolicySendrecv;
CartesianCommunicator::CommunicatorPolicy_t CartesianCommunicator::CommunicatorPolicy= CartesianCommunicator::CommunicatorPolicyConcurrent;
/////////////////////////////////
// Alloc, free shmem region

2
lib/communicator/Communicator_base.h
View File

@ -118,7 +118,7 @@ class CartesianCommunicator {
static void * ShmCommBuf;
// Isend/Irecv/Wait, or Sendrecv blocking
enum CommunicatorPolicy_t { CommunicatorPolicyIsend , CommunicatorPolicySendrecv };
enum CommunicatorPolicy_t { CommunicatorPolicyConcurrent, CommunicatorPolicySequential };
static CommunicatorPolicy_t CommunicatorPolicy;
static void SetCommunicatorPolicy(CommunicatorPolicy_t policy ) { CommunicatorPolicy = policy; }

4
lib/communicator/Communicator_mpi.cc
View File

@ -158,7 +158,7 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
{
int myrank = _processor;
int ierr;
if ( CommunicatorPolicy == CommunicatorPolicyIsend ) {
if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) {
MPI_Request xrq;
MPI_Request rrq;
@ -178,7 +178,7 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
}
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{
if ( CommunicatorPolicy == CommunicatorPolicyIsend ) {
if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) {
int nreq=list.size();
std::vector<MPI_Status> status(nreq);
int ierr = MPI_Waitall(nreq,&list[0],&status[0]);

9
lib/communicator/Communicator_mpi3.cc
View File

@ -511,7 +511,7 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
int myrank = _processor;
int ierr;
if ( CommunicatorPolicy == CommunicatorPolicyIsend ) {
if ( CommunicatorPolicy == CommunicatorPolicyConcurrent ) {
MPI_Request xrq;
MPI_Request rrq;
@ -567,6 +567,11 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
list.push_back(xrq);
off_node_bytes+=bytes;
}
if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
this->StencilSendToRecvFromComplete(list);
}
return off_node_bytes;
}
void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall)
@ -585,8 +590,8 @@ void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &
std::vector<MPI_Status> status(nreq);
int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
list.resize(0);
assert(ierr==0);
list.resize(0);
}
void CartesianCommunicator::Barrier(void)
{

31
lib/cshift/Cshift_common.h
View File

@ -58,8 +58,7 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimen
int stride=rhs._grid->_slice_stride[dimension];
if ( cbmask == 0x3 ) {
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o = n*stride;
int bo = n*e2;
@ -78,8 +77,7 @@ PARALLEL_NESTED_LOOP2
}
}
}
PARALLEL_FOR_LOOP
for(int i=0;i<table.size();i++){
parallel_for(int i=0;i<table.size();i++){
buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
}
}
@ -105,8 +103,7 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
int n1=rhs._grid->_slice_stride[dimension];
if ( cbmask ==0x3){
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o = n*n1;
@ -122,8 +119,7 @@ PARALLEL_NESTED_LOOP2
// Case of SIMD split AND checker dim cannot currently be hit, except in
// Test_cshift_red_black code.
std::cout << " Dense packed buffer WARNING " <<std::endl;
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o=n*n1;
@ -175,8 +171,7 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
int stride=rhs._grid->_slice_stride[dimension];
if ( cbmask ==0x3 ) {
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o =n*rhs._grid->_slice_stride[dimension];
int bo =n*rhs._grid->_slice_block[dimension];
@ -195,8 +190,7 @@ PARALLEL_NESTED_LOOP2
}
}
}
PARALLEL_FOR_LOOP
for(int i=0;i<table.size();i++){
parallel_for(int i=0;i<table.size();i++){
// std::cout << "Rcv"<< table[i].first << " " << table[i].second << " " <<buffer[table[i].second]<<std::endl;
rhs._odata[table[i].first]=buffer[table[i].second];
}
@ -220,8 +214,7 @@ PARALLEL_FOR_LOOP
int e2=rhs._grid->_slice_block[dimension];
if(cbmask ==0x3 ) {
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o = n*rhs._grid->_slice_stride[dimension];
int offset = b+n*rhs._grid->_slice_block[dimension];
@ -265,8 +258,7 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
int e2=rhs._grid->_slice_block[dimension];
int stride = rhs._grid->_slice_stride[dimension];
if(cbmask == 0x3 ){
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o =n*stride+b;
@ -275,8 +267,7 @@ PARALLEL_NESTED_LOOP2
}
}
} else {
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o =n*stride+b;
@ -306,8 +297,8 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
int e1=rhs._grid->_slice_nblock[dimension];
int e2=rhs._grid->_slice_block [dimension];
int stride = rhs._grid->_slice_stride[dimension];
PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){
parallel_for_nest2(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o =n*stride;

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);

14
lib/qcd/action/fermion/CayleyFermion5Dcache.cc
View File

@ -54,8 +54,8 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
// Flops = 6.0*(Nc*Ns) *Ls*vol
M5Dcalls++;
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
for(int s=0;s<Ls;s++){
auto tmp = psi._odata[0];
if ( s==0 ) {
@ -98,8 +98,8 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
// Flops = 6.0*(Nc*Ns) *Ls*vol
M5Dcalls++;
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
for(int s=0;s<Ls;s++){
if ( s==0 ) {
@ -137,8 +137,7 @@ void CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi, FermionField &
MooeeInvCalls++;
MooeeInvTime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
// flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls = 12*Ls * (9) = 108*Ls flops
@ -184,8 +183,7 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
MooeeInvCalls++;
MooeeInvTime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];

12
lib/qcd/action/fermion/CayleyFermion5Dvec.cc
View File

@ -91,8 +91,7 @@ void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
assert(Nc==3);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
parallel_for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
#if 0
alignas(64) SiteHalfSpinor hp;
alignas(64) SiteHalfSpinor hm;
@ -232,8 +231,7 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
M5Dcalls++;
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
parallel_for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
#if 0
alignas(64) SiteHalfSpinor hp;
alignas(64) SiteHalfSpinor hm;
@ -792,13 +790,11 @@ void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField
MooeeInvTime-=usecond();
if ( switcheroo<Coeff_t>::iscomplex() ) {
PARALLEL_FOR_LOOP
for(auto site=0;site<vol;site++){
parallel_for(auto site=0;site<vol;site++){
MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
}
} else {
PARALLEL_FOR_LOOP
for(auto site=0;site<vol;site++){
parallel_for(auto site=0;site<vol;site++){
MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
}
}

18
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -194,8 +194,7 @@ namespace QCD {
GaugeLinkField tmp(mat._grid);
tmp = zero;
PARALLEL_FOR_LOOP
for(int sss=0;sss<tmp._grid->oSites();sss++){
parallel_for(int sss=0;sss<tmp._grid->oSites();sss++){
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
@ -445,8 +444,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
Uconj = where(coor==neglink,-Uconj,Uconj);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu) = U[ss]();
Uds[ss](1)(mu) = Uconj[ss]();
}
@ -459,8 +457,7 @@ PARALLEL_FOR_LOOP
Utmp = where(coor==0,Uconj,Utmp);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu+4) = Utmp[ss]();
}
@ -469,8 +466,7 @@ PARALLEL_FOR_LOOP
Utmp = where(coor==0,U,Utmp);
}
PARALLEL_FOR_LOOP
for(auto ss=U.begin();ss<U.end();ss++){
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](1)(mu+4) = Utmp[ss]();
}
@ -484,8 +480,7 @@ PARALLEL_FOR_LOOP
GaugeLinkField link(mat._grid);
// use lorentz for flavour as hack.
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
PARALLEL_FOR_LOOP
for (auto ss = tmp.begin(); ss < tmp.end(); ss++) {
parallel_for(auto ss = tmp.begin(); ss < tmp.end(); ss++) {
link[ss]() = tmp[ss](0, 0) - conjugate(tmp[ss](1, 1));
}
PokeIndex<LorentzIndex>(mat, link, mu);
@ -498,8 +493,7 @@ PARALLEL_FOR_LOOP
GaugeLinkField tmp(mat._grid);
tmp = zero;
PARALLEL_FOR_LOOP
for (int ss = 0; ss < tmp._grid->oSites(); ss++) {
parallel_for(int ss = 0; ss < tmp._grid->oSites(); ss++) {
for (int s = 0; s < Ls; s++) {
int sF = s + Ls * ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF]));

12
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -222,8 +222,7 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
////////////////////////
// Call the single hop
////////////////////////
PARALLEL_FOR_LOOP
for (int sss = 0; sss < B._grid->oSites(); sss++) {
parallel_for (int sss = 0; sss < B._grid->oSites(); sss++) {
Kernels::DhopDir(st, U, st.CommBuf(), sss, sss, B, Btilde, mu,
gamma);
}
@ -333,8 +332,7 @@ void WilsonFermion<Impl>::DhopDirDisp(const FermionField &in, FermionField &out,
Stencil.HaloExchange(in, compressor);
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DhopDir(Stencil, Umu, Stencil.CommBuf(), sss, sss, in, out, dirdisp, gamma);
}
};
@ -350,13 +348,11 @@ void WilsonFermion<Impl>::DhopInternal(StencilImpl &st, LebesgueOrder &lo,
st.HaloExchange(in, compressor);
if (dag == DaggerYes) {
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DhopSiteDag(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
}
} else {
PARALLEL_FOR_LOOP
for (int sss = 0; sss < in._grid->oSites(); sss++) {
parallel_for (int sss = 0; sss < in._grid->oSites(); sss++) {
Kernels::DhopSite(st, lo, U, st.CommBuf(), sss, sss, 1, 1, in, out);
}
}

65
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -275,8 +275,7 @@ void WilsonFermion5D<Impl>::DhopDir(const FermionField &in, FermionField &out,in
assert(dirdisp<=7);
assert(dirdisp>=0);
PARALLEL_FOR_LOOP
for(int ss=0;ss<Umu._grid->oSites();ss++){
parallel_for(int ss=0;ss<Umu._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
int sU=ss;
int sF = s+Ls*sU;
@ -323,8 +322,7 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
////////////////////////
DerivDhopComputeTime -= usecond();
PARALLEL_FOR_LOOP
for (int sss = 0; sss < U._grid->oSites(); sss++) {
parallel_for (int sss = 0; sss < U._grid->oSites(); sss++) {
for (int s = 0; s < Ls; s++) {
int sU = sss;
int sF = s + Ls * sU;
@ -493,73 +491,18 @@ void WilsonFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOr
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
if (dag == DaggerYes) {
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
}
} else {
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
}
}
/*
if (dag == DaggerYes) {
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
}
#ifdef AVX512_SWITCHOFF
} else if (stat.is_init() ) {
int nthreads;
stat.start();
#pragma omp parallel
{
#pragma omp master
nthreads = omp_get_num_threads();
int mythread = omp_get_thread_num();
stat.enter(mythread);
#pragma omp for nowait
for(int ss=0;ss<U._grid->oSites();ss++) {
int sU=ss;
int sF=LLs*sU;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
}
stat.exit(mythread);
}
stat.accum(nthreads);
#endif
} else {
#if 1
PARALLEL_FOR_LOOP
for (int ss = 0; ss < U._grid->oSites(); ss++) {
int sU = ss;
int sF = LLs * sU;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out);
}
#else
#ifdef GRID_OMP
#pragma omp parallel
#endif
{
int len = U._grid->oSites();
int me, myoff,mywork;
GridThread::GetWorkBarrier(len,me, mywork,myoff);
int sF = LLs * myoff;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out);
}
#endif
}
*/
DhopComputeTime+=usecond();
}

3
lib/qcd/action/gauge/GaugeImpl.h
View File

@ -66,8 +66,7 @@ public:
// Move this elsewhere? FIXME
static inline void AddGaugeLink(GaugeField &U, GaugeLinkField &W,
int mu) { // U[mu] += W
PARALLEL_FOR_LOOP
for (auto ss = 0; ss < U._grid->oSites(); ss++) {
parallel_for (auto ss = 0; ss < U._grid->oSites(); ss++) {
U._odata[ss]._internal[mu] =
U._odata[ss]._internal[mu] + W._odata[ss]._internal;
}

24
lib/qcd/utils/LinalgUtils.h
View File

@ -48,8 +48,7 @@ void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,Coeff a,Coeff b)
GridBase *grid=x._grid;
Gamma G5(Gamma::Gamma5);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss++){
parallel_for(int ss=0;ss<grid->oSites();ss++){
vobj tmp;
tmp = a*x._odata[ss];
tmp = tmp + G5*(b*timesI(x._odata[ss]));
@ -65,8 +64,7 @@ void axpby_ssp(Lattice<vobj> &z, Coeff a,const Lattice<vobj> &x,Coeff b,const La
conformable(x,z);
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp = a*x._odata[ss+s]+b*y._odata[ss+sp];
vstream(z._odata[ss+s],tmp);
}
@ -81,8 +79,7 @@ void ag5xpby_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Gamma5);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp;
tmp = G5*x._odata[ss+s]*a;
tmp = tmp + b*y._odata[ss+sp];
@ -99,8 +96,7 @@ void axpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Gamma5);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp;
tmp = G5*y._odata[ss+sp]*b;
tmp = tmp + a*x._odata[ss+s];
@ -117,8 +113,7 @@ void ag5xpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Gamma5);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp1;
vobj tmp2;
tmp1 = a*x._odata[ss+s]+b*y._odata[ss+sp];
@ -135,8 +130,7 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
conformable(x,z);
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp;
spProj5m(tmp,y._odata[ss+sp]);
tmp = a*x._odata[ss+s]+b*tmp;
@ -152,8 +146,7 @@ void axpby_ssp_pplus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,con
conformable(x,z);
GridBase *grid=x._grid;
int Ls = grid->_rdimensions[0];
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp;
spProj5p(tmp,y._odata[ss+sp]);
tmp = a*x._odata[ss+s]+b*tmp;
@ -169,8 +162,7 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
conformable(x,z);
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Gamma5);
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
parallel_for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
vobj tmp;
for(int s=0;s<Ls;s++){
int sp = Ls-1-s;

6
lib/qcd/utils/SUn.h
View File

@ -221,8 +221,7 @@ class SU {
int i0, i1;
su2SubGroupIndex(i0, i1, su2_index);
PARALLEL_FOR_LOOP
for (int ss = 0; ss < grid->oSites(); ss++) {
parallel_for (int ss = 0; ss < grid->oSites(); ss++) {
subgroup._odata[ss]()()(0, 0) = source._odata[ss]()()(i0, i0);
subgroup._odata[ss]()()(0, 1) = source._odata[ss]()()(i0, i1);
subgroup._odata[ss]()()(1, 0) = source._odata[ss]()()(i1, i0);
@ -252,8 +251,7 @@ class SU {
su2SubGroupIndex(i0, i1, su2_index);
dest = 1.0; // start out with identity
PARALLEL_FOR_LOOP
for (int ss = 0; ss < grid->oSites(); ss++) {
parallel_for (int ss = 0; ss < grid->oSites(); ss++) {
dest._odata[ss]()()(i0, i0) = subgroup._odata[ss]()()(0, 0);
dest._odata[ss]()()(i0, i1) = subgroup._odata[ss]()()(0, 1);
dest._odata[ss]()()(i1, i0) = subgroup._odata[ss]()()(1, 0);

2
tests/forces/Test_contfrac_force.cc
View File

@ -31,8 +31,6 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);

2
tests/forces/Test_dwf_force.cc
View File

@ -31,8 +31,6 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);

2
tests/forces/Test_dwf_gpforce.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_gp_rect_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_gpdwf_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_gpwilson_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_partfrac_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_rect_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_wilson_force.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_wilson_force_phiMdagMphi.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{

2
tests/forces/Test_wilson_force_phiMphi.cc
View File

@ -31,7 +31,7 @@ using namespace std;
using namespace Grid;
using namespace Grid::QCD;
#define parallel_for PARALLEL_FOR_LOOP for
int main (int argc, char ** argv)
{