portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-16 23:07:05 +01:00
Code Releases Activity

Make view specify where and drive data motion - first cut.

Browse Source 
This is a compile tiime option --enable-unified=yes/no
This commit is contained in:
Peter Boyle
2020-05-21 16:13:16 -04:00
parent ebb60330c9
commit 7860a50f70
48 changed files with 688 additions and 718 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Grid/cshift/Cshift_common.h
View File

@ -52,7 +52,6 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
int stride=rhs.Grid()->_slice_stride[dimension];
auto rhs_v = rhs.View();
if ( cbmask == 0x3 ) {
for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
@ -73,6 +72,7 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimen
}
}
}
auto rhs_v = rhs.View(AcceleratorRead);
auto buffer_p = & buffer[0];
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
@ -100,7 +100,7 @@ Gather_plane_extract(const Lattice<vobj> &rhs,
int e2=rhs.Grid()->_slice_block[dimension];
int n1=rhs.Grid()->_slice_stride[dimension];
auto rhs_v = rhs.View();
auto rhs_v = rhs.View(AcceleratorRead);
if ( cbmask ==0x3){
accelerator_for2d(n,e1,b,e2,1,{
int o = n*n1;
@ -179,7 +179,7 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
}
}
auto rhs_v = rhs.View();
auto rhs_v = rhs.View(AcceleratorWrite);
auto buffer_p = & buffer[0];
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
@ -204,7 +204,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
int e2=rhs.Grid()->_slice_block[dimension];
if(cbmask ==0x3 ) {
auto rhs_v = rhs.View();
auto rhs_v = rhs.View(AcceleratorWrite);
accelerator_for2d(n,e1,b,e2,1,{
int o = n*rhs.Grid()->_slice_stride[dimension];
int offset = b+n*rhs.Grid()->_slice_block[dimension];
@ -216,7 +216,7 @@ template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,ExtractPointerA
// Test_cshift_red_black code.
// std::cout << "Scatter_plane merge assert(0); think this is buggy FIXME "<< std::endl;// think this is buggy FIXME
std::cout<<" Unthreaded warning -- buffer is not densely packed ??"<<std::endl;
auto rhs_v = rhs.View();
auto rhs_v = rhs.View(CpuWrite);
for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){
int o = n*rhs.Grid()->_slice_stride[dimension];
@ -272,8 +272,8 @@ template<class vobj> void Copy_plane(Lattice<vobj>& lhs,const Lattice<vobj> &rhs
}
}
auto rhs_v = rhs.View();
auto lhs_v = lhs.View();
auto rhs_v = rhs.View(AcceleratorRead);
auto lhs_v = lhs.View(AcceleratorWrite);
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
lhs_v[table[i].first]=rhs_v[table[i].second];
@ -315,8 +315,8 @@ template<class vobj> void Copy_plane_permute(Lattice<vobj>& lhs,const Lattice<vo
}}
}
auto rhs_v = rhs.View();
auto lhs_v = lhs.View();
auto rhs_v = rhs.View(AcceleratorRead);
auto lhs_v = lhs.View(AcceleratorWrite);
auto table = &Cshift_table[0];
accelerator_for(i,ent,1,{
permute(lhs_v[table[i].first],rhs_v[table[i].second],permute_type);