portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-25 13:15:55 +01:00
Code Releases Activity

Bug fix for spread out in x (EO) direction.

Browse Source 
This is really annoying -- it is very hard to thread the loops with the index
recursion on buffer offset in the red-black case. Must think of a good threading
solution here.
This commit is contained in:
Peter Boyle 2015-11-04 09:57:57 +00:00
parent ec5af35166
commit 1271508ca2
1 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
lib/cshift/Cshift_common.h
View File

@ -29,14 +29,15 @@ Gather_plane_simple (const Lattice<vobj> &rhs,std::vector<cobj,alignedAllocator<
int e1=rhs._grid->_slice_nblock[dimension]; int e1=rhs._grid->_slice_nblock[dimension];
int e2=rhs._grid->_slice_block[dimension]; int e2=rhs._grid->_slice_block[dimension];
PARALLEL_NESTED_LOOP2 int bo=0;
//PARALLEL_NESTED_LOOP21
for(int n=0;n<e1;n++){ for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){ for(int b=0;b<e2;b++){
int o = n*rhs._grid->_slice_stride[dimension]; int o = n*rhs._grid->_slice_stride[dimension];
int bo = n*rhs._grid->_slice_block[dimension]; // int bo = n*rhs._grid->_slice_block[dimension];
int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup
if ( ocb &cbmask ) { if ( ocb &cbmask ) {
buffer[bo+b]=compress(rhs._odata[so+o+b],dimension,plane,so+o+b,rhs._grid); buffer[bo++]=compress(rhs._odata[so+o+b],dimension,plane,so+o+b,rhs._grid);
} }
} }
} }
@ -59,7 +60,7 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
int e1=rhs._grid->_slice_nblock[dimension]; int e1=rhs._grid->_slice_nblock[dimension];
int e2=rhs._grid->_slice_block[dimension]; int e2=rhs._grid->_slice_block[dimension];
PARALLEL_NESTED_LOOP2 //PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){ for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){ for(int b=0;b<e2;b++){
@ -109,14 +110,15 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,std::vector<v
int e1=rhs._grid->_slice_nblock[dimension]; int e1=rhs._grid->_slice_nblock[dimension];
int e2=rhs._grid->_slice_block[dimension]; int e2=rhs._grid->_slice_block[dimension];
PARALLEL_NESTED_LOOP2 int bo=0;
//PARALLEL_NESTED_LOOP2
for(int n=0;n<e1;n++){ for(int n=0;n<e1;n++){
for(int b=0;b<e2;b++){ for(int b=0;b<e2;b++){
int o =n*rhs._grid->_slice_stride[dimension]; int o =n*rhs._grid->_slice_stride[dimension];
int bo =n*rhs._grid->_slice_block[dimension]; // int bo =n*rhs._grid->_slice_block[dimension];
int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup int ocb=1<<rhs._grid->CheckerBoardFromOindex(o+b);// Could easily be a table lookup
if ( ocb & cbmask ) { if ( ocb & cbmask ) {
rhs._odata[so+o+b]=buffer[bo+b]; rhs._odata[so+o+b]=buffer[bo++];
} }
} }
} }