portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Overlap comms with comput now supported

Browse Source
This commit is contained in:
Peter Boyle 2019-07-12 09:09:40 +01:00
parent 91e2cf9b40
commit bd155ca5c0
1 changed files with 10 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -321,14 +321,17 @@ void WilsonKernels<Impl>::DhopDirK( StencilView &st, DoubledGaugeFieldView &U,Si
coalescedWrite(out[sF], result,lane);
}
#define KERNEL_CALL(A) \
#define KERNEL_CALLNB(A) \
const uint64_t NN = Nsite*Ls; \
accelerator_for( ss, NN, Simd::Nsimd(), { \
accelerator_forNB( ss, NN, Simd::Nsimd(), { \
int sF = ss; \
int sU = ss/Ls; \
WilsonKernels<Impl>::A(st_v,U_v,buf,sF,sU,in_v,out_v); \
});
#define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
#define ASM_CALL(A) \
thread_for( ss, Nsite, { \
int sU = ss; \
@ -353,9 +356,9 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSite); printf("."); return;}
#endif
} else if( interior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteInt); return;}
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALLNB(GenericDhopSiteInt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteInt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt); return;}
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); printf("-"); return;}
#endif
} else if( exterior ) {
@ -379,20 +382,20 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if( interior && exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDag); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDag); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDag); return;}
#endif
} else if( interior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagInt); return;}
#endif
} else if( exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDagExt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagExt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagExt); return;}
#endif
}