portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

Fence changes and EXT kernel loop cout reduction

Browse Source
This commit is contained in:
Peter Boyle 2023-04-04 07:37:23 -07:00
parent b281b0166e
commit cc9d88ea1c
1 changed files with 9 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -428,9 +428,10 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
auto ptr = &st.surface_list[0]; \ auto ptr = &st.surface_list[0]; \
accelerator_forNB( ss, sz, Simd::Nsimd(), { \ accelerator_forNB( ss, sz, Simd::Nsimd(), { \
int sF = ptr[ss]; \ int sF = ptr[ss]; \
int sU = ss/Ls; \ int sU = sF/Ls; \
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v); \ WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v); \
}); }); \
accelerator_barrier();
#define ASM_CALL(A) \ #define ASM_CALL(A) \
thread_for( sss, Nsite, { \ thread_for( sss, Nsite, { \
@ -474,9 +475,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); return;} if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); return;}
#endif #endif
} else if( exterior ) { } else if( exterior ) {
// dependent on result of merge
acceleratorFenceComputeStream(); acceleratorFenceComputeStream();
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteExt); return;} if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL_EXT(GenericDhopSiteExt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt); return;} if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL_EXT(HandDhopSiteExt); return;}
#ifndef GRID_CUDA #ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteExt); return;} if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteExt); return;}
#endif #endif
@ -506,9 +508,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagInt); return;} if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagInt); return;}
#endif #endif
} else if( exterior ) { } else if( exterior ) {
// Dependent on result of merge
acceleratorFenceComputeStream(); acceleratorFenceComputeStream();
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDagExt); return;} if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL_EXT(GenericDhopSiteDagExt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt); return;} if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL_EXT(HandDhopSiteDagExt); return;}
#ifndef GRID_CUDA #ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagExt); return;} if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagExt); return;}
#endif #endif