portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-08-23 10:59:36 +01:00
Code Releases Activity

Bug fix for CPU build

Browse Source
This commit is contained in:
Peter Boyle
2026-08-14 17:53:35 -04:00
parent 76b4bd6d12
commit 06ce057920
1 changed files with 20 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Grid/algorithms/multigrid/GeneralCoarsenedMatrix.h
+20
View File
@@ -453,8 +453,18 @@ public:
autoView( A_v , _A[k], AcceleratorWrite); autoView( A_v , _A[k], AcceleratorWrite);
autoView( FT_v , FT[k], AcceleratorRead); autoView( FT_v , FT[k], AcceleratorRead);
accelerator_for(sss, osites, nbasis, { accelerator_for(sss, osites, nbasis, {
#ifdef GRID_SIMT
int j = acceleratorSIMTlane(nbasis); int j = acceleratorSIMTlane(nbasis);
A_v[sss](i,j) = FT_v[sss](j); A_v[sss](i,j) = FT_v[sss](j);
#else
// CPU build: acceleratorSIMTlane()==0 -- an un-looped SIMT tensor
// index writes ONLY j=0 and silently drops the other nbasis-1
// columns (caught by Test_schur_dense_coarse import certificate,
// 2026-08-14). Loop explicitly.
for(int j=0;j<nbasis;j++){
A_v[sss](i,j) = FT_v[sss](j);
}
#endif
}); });
} }
tinv+=usecond(); tinv+=usecond();
@@ -630,8 +640,18 @@ public:
autoView( A_v , _A[k], AcceleratorWrite); autoView( A_v , _A[k], AcceleratorWrite);
autoView( FT_v , FT[k], AcceleratorRead); autoView( FT_v , FT[k], AcceleratorRead);
accelerator_for(sss, osites, nbasis, { accelerator_for(sss, osites, nbasis, {
#ifdef GRID_SIMT
int j = acceleratorSIMTlane(nbasis); int j = acceleratorSIMTlane(nbasis);
A_v[sss](i,j) = FT_v[sss](j); A_v[sss](i,j) = FT_v[sss](j);
#else
// CPU build: acceleratorSIMTlane()==0 -- an un-looped SIMT tensor
// index writes ONLY j=0 and silently drops the other nbasis-1
// columns (caught by Test_schur_dense_coarse import certificate,
// 2026-08-14). Loop explicitly.
for(int j=0;j<nbasis;j++){
A_v[sss](i,j) = FT_v[sss](j);
}
#endif
}); });
} }
tinv+=usecond(); tinv+=usecond();