portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-19 08:17:05 +01:00
Code Releases Activity

round-up after testing of prefetches in stencil close

Browse Source
This commit is contained in:
nmeyer-ur
2020-06-03 11:58:20 +02:00
parent 5050833b42
commit 5ee3ea2144
4 changed files with 41 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Grid/qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h
View File

@ -164,12 +164,7 @@ Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \
MULT_2SPIN_1(Dir); \
PREFETCH_CHIMU(base); \
/* PREFETCH_GAUGE_L1(NxtDir); */ \
MULT_2SPIN_2; \
if (s == 0) { \
if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
} \
RECON; \
nmu++; \
}
@ -180,12 +175,7 @@ Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \
MULT_2SPIN_1(Dir); \
PREFETCH_CHIMU(base); \
/* PREFETCH_GAUGE_L1(NxtDir); */ \
MULT_2SPIN_2; \
if (s == 0) { \
if ((Dir == 0) || (Dir == 4)) { PREFETCH_GAUGE_L2(Dir); } \
} \
RECON; \
nmu++; \
}

3
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -445,18 +445,21 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite); return;}
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSite); /* printf("."); */ return;}
//if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSite); printf("."); return;}
#endif
} else if( interior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALLNB(GenericDhopSiteInt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteInt); return;}
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); /* printf("-"); */ return;}
//if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); printf("-"); return;}
#endif
} else if( exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteExt); return;}
#ifndef GRID_NVCC
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt); return;}
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteExt); /* printf("+"); */ return;}
//if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteExt); printf("+"); return;}
#endif
}
assert(0 && " Kernel optimisation case not covered ");