portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-11 03:46:55 +01:00
Code Releases Activity

OMP collapse changes to make NVCC happy

Browse Source
This commit is contained in:
paboyle
2018-01-28 01:21:53 +00:00
parent b6ebf35af5
commit e657f9a344
4 changed files with 25 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
lib/lattice/Lattice_matrix_reduction.h
View File

@ -51,12 +51,11 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
#pragma omp parallel
thread_region
{
std::vector<vobj> s_x(Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
thread_loop_collapse2( (int n=0;n<nblock;n++),{
for(int b=0;b<block;b++){
int o = n*stride + b;
@ -72,7 +71,7 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
}
R[o+i*ostride]=dot;
}
}}
}});
}
};
@ -101,12 +100,11 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
#pragma omp parallel
thread_region
{
std::vector<vobj> s_x(Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
thread_loop_collapse2( (int n=0;n<nblock;n++),{
for(int b=0;b<block;b++){
int o = n*stride + b;
@ -122,7 +120,7 @@ static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<
}
R[o+i*ostride]=dot;
}
}}
}});
}
};
@ -159,14 +157,12 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj>
typedef typename vobj::vector_typeD vector_typeD;
#pragma omp parallel
{
thread_region {
std::vector<vobj> Left(Nblock);
std::vector<vobj> Right(Nblock);
Eigen::MatrixXcd mat_thread = Eigen::MatrixXcd::Zero(Nblock,Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
thread_loop_collapse2((int n=0;n<nblock;n++),{
for(int b=0;b<block;b++){
int o = n*stride + b;
@ -182,9 +178,8 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj>
auto rtmp = TensorRemove(tmp);
mat_thread(i,j) += Reduce(rtmp);
}}
}}
#pragma omp critical
{
}});
thread_critical {
mat += mat_thread;
}
}

2
lib/lattice/Lattice_reduction.h
View File

@ -358,7 +358,7 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
tensor_reduced at; at=av;
thread_loop_collapse(2, (int n=0;n<e1;n++),{
thread_loop_collapse2( (int n=0;n<e1;n++),{
for(int b=0;b<e2;b++){
int ss= so+n*stride+b;
R[ss] = at*X[ss]+Y[ss];