portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-19 00:07:05 +01:00
Code Releases Activity

Performance regressed and is OK in icpx 2023.2

Browse Source
This commit is contained in:
Peter Boyle
2023-10-03 15:53:14 +00:00
parent afc316f501
commit d93eac7b1c
8 changed files with 1063 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
Grid/qcd/utils/WilsonLoops.h
View File

@ -464,7 +464,8 @@ public:
//U_padded: the gauge link fields padded out using the PaddedCell class
//Cell: the padded cell class
//gStencil: the precomputed generalized local stencil for the staple
static void StaplePaddedAll(std::vector<GaugeMat> &staple, const std::vector<GaugeMat> &U_padded, const PaddedCell &Cell, const GeneralLocalStencil &gStencil) {
static void StaplePaddedAll(std::vector<GaugeMat> &staple, const std::vector<GaugeMat> &U_padded, const PaddedCell &Cell, const GeneralLocalStencil &gStencil)
{
double t0 = usecond();
assert(U_padded.size() == Nd); assert(staple.size() == Nd);
assert(U_padded[0].Grid() == (GridBase*)Cell.grids.back());
@ -489,7 +490,7 @@ public:
autoView( gStaple_v , gStaple, AcceleratorWrite);
auto gStencil_v = gStencil.View();
accelerator_for(ss, ggrid->oSites(), ggrid->Nsimd(), {
accelerator_for(ss, ggrid->oSites(), (size_t)ggrid->Nsimd(), {
decltype(coalescedRead(Ug_dirs_v[0][0])) stencil_ss;
stencil_ss = Zero();
int off = outer_off;
@ -1201,7 +1202,7 @@ public:
autoView( gStaple_v , gStaple, AcceleratorWrite);
auto gStencil_v = gStencil.View();
accelerator_for(ss, ggrid->oSites(), ggrid->Nsimd(), {
accelerator_for(ss, ggrid->oSites(), (size_t)ggrid->Nsimd(), {
decltype(coalescedRead(Ug_dirs_v[0][0])) stencil_ss;
stencil_ss = Zero();
int s=offset;

2
Grid/stencil/GeneralLocalStencil.h
View File

@ -43,7 +43,7 @@ class GeneralLocalStencilView {
int _npoints; // Move to template param?
GeneralStencilEntry* _entries_p;
accelerator_inline GeneralStencilEntry * GetEntry(int point,int osite) {
accelerator_inline GeneralStencilEntry * GetEntry(int point,int osite) const {
return & this->_entries_p[point+this->_npoints*osite];
}