portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-03 18:55:56 +01:00
Code Releases Activity

980 GiB/s Wilson; 680 GiB/s DW (DP)

Browse Source
This commit is contained in:
nils meyer 2020-04-15 10:55:06 +02:00
parent 581392f2f2
commit c12a67030a
8 changed files with 2787 additions and 318 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

409
Grid/qcd/action/fermion/implementation/WilsonKernelsAsmA64FX.h
View File

@ -82,7 +82,11 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
template<> void template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
@ -293,157 +297,6 @@ WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFie
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif #endif
//#undef MAYBEPERM
//#undef MULT_2SPIN
//#define MAYBEPERM(A,B)
//#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
/////////////////////////////////////////////////////////////////
// Ls vectorised, undag Kernel, single
/////////////////////////////////////////////////////////////////
#ifdef DWFVEC5D
#undef KERNEL_DAG
#define INTERIOR_AND_EXTERIOR
#undef INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
#undef MULT_2SPIN
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LSNOPF(ptr,pf)
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// Ls vectorised, dag Kernel, single
/////////////////////////////////////////////////////////////////
#define KERNEL_DAG
#define INTERIOR_AND_EXTERIOR
#undef INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif // VEC 5D
//#undef COMPLEX_SIGNS
//#undef MAYBEPERM
//#undef MULT_2SPIN
// undefine // undefine
#include <simd/Fujitsu_A64FX_undef.h> #include <simd/Fujitsu_A64FX_undef.h>
@ -478,20 +331,38 @@ WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilView &st, Double
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
#undef INTERIOR_AND_EXTERIOR #undef INTERIOR_AND_EXTERIOR
#define INTERIOR #define INTERIOR
@ -499,20 +370,38 @@ WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
#undef INTERIOR_AND_EXTERIOR #undef INTERIOR_AND_EXTERIOR
#undef INTERIOR #undef INTERIOR
@ -520,20 +409,38 @@ WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldV
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
///////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////
// XYZT vectorised, dag Kernel, double // XYZT vectorised, dag Kernel, double
@ -545,20 +452,38 @@ WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldV
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
#undef INTERIOR_AND_EXTERIOR #undef INTERIOR_AND_EXTERIOR
#define INTERIOR #define INTERIOR
@ -566,20 +491,38 @@ WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldV
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
#undef INTERIOR_AND_EXTERIOR #undef INTERIOR_AND_EXTERIOR
#undef INTERIOR #undef INTERIOR
@ -587,168 +530,40 @@ WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFie
template<> void template<> void
WilsonKernels<WilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
template<> void template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf, WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out) int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#if defined (WILSONKERNELSASMBODYA64FX)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h>
#else
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif
// KNL stuff
#undef MAYBEPERM
//#undef MULT_2SPIN
#define MAYBEPERM(A,B)
//#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LS(ptr,pf)
/////////////////////////////////////////////////////////////////
// Ls vectorised, undag Kernel, double
/////////////////////////////////////////////////////////////////
#ifdef DWFVEC5D
#undef KERNEL_DAG
#define INTERIOR_AND_EXTERIOR
#undef INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
#undef MULT_2SPIN
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN_LSNOPF(ptr,pf)
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// Ls vectorised, dag Kernel, double
/////////////////////////////////////////////////////////////////
#define KERNEL_DAG
#define INTERIOR_AND_EXTERIOR
#undef INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
template<> void
WilsonKernels<DomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilView &st, DoubledGaugeFieldView &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionFieldView &in, FermionFieldView &out)
#include <qcd/action/fermion/implementation/WilsonKernelsAsmBody.h>
#endif // VEC 5D
// undefs // undefs
#undef WILSONKERNELSASMBODYA64FX #undef WILSONKERNELSASMBODYA64FX

156
Grid/qcd/action/fermion/implementation/WilsonKernelsAsmBodyA64FX.h
View File

@ -61,6 +61,29 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define DIR7_RECON TP_RECON_ACCUM #define DIR7_RECON TP_RECON_ACCUM
#endif #endif
//using namespace std;
#undef SHOW
//#define SHOW
#undef WHERE
#ifdef INTERIOR_AND_EXTERIOR
#define WHERE "INT_AND_EXT"
#endif
#ifdef INTERIOR
#define WHERE "INT"
#endif
#ifdef EXTERIOR
#define WHERE "EXT"
#endif
//#pragma message("here")
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
// Comms then compute kernel // Comms then compute kernel
//////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////
@ -69,16 +92,17 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
basep = st.GetPFInfo(nent,plocal); nent++; \ basep = st.GetPFInfo(nent,plocal); nent++; \
if ( local ) { \ if ( local ) { \
LOAD64(%r10,isigns); \ /* PREFETCH_GAUGE_L1(Dir); slightly worse performance */ \
PROJ(base); \ PROJ(base); \
/* PREFETCH_GAUGE_L1(Dir); slightly worse performance */ \
MAYBEPERM(PERMUTE_DIR,perm); \ MAYBEPERM(PERMUTE_DIR,perm); \
} else { \ } else { \
LOAD_CHI(base); \ LOAD_CHI(base); \
} \ } \
MULT_2SPIN_DIR_PF(Dir,basep); \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \ base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \ PREFETCH_CHIMU(base); \
LOAD64(%r10,isigns); \ MULT_2SPIN_DIR_PF(Dir,basep); \
PREFETCH_GAUGE_L1(NxtDir); \
RECON; \ RECON; \
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
@ -99,17 +123,15 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
basep = st.GetPFInfo(nent,plocal); nent++; \ basep = st.GetPFInfo(nent,plocal); nent++; \
if ( local ) { \ if ( local ) { \
LOAD64(%r10,isigns); \
PROJ(base); \ PROJ(base); \
MAYBEPERM(PERMUTE_DIR,perm); \ MAYBEPERM(PERMUTE_DIR,perm); \
}else if ( st.same_node[Dir] ) {LOAD_CHI(base);} \ }else if ( st.same_node[Dir] ) {LOAD_CHI(base);} \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
if ( local || st.same_node[Dir] ) { \ if ( local || st.same_node[Dir] ) { \
MULT_2SPIN_DIR_PF(Dir,basep); \ MULT_2SPIN_DIR_PF(Dir,basep); \
PREFETCH_CHIMU(base); \
LOAD64(%r10,isigns); \
RECON; \ RECON; \
} else { PREFETCH_CHIMU(base); } } \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \ #define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \ base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
@ -132,7 +154,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
if((!local)&&(!st.same_node[Dir]) ) { \ if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \ LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(Dir,base); \ MULT_2SPIN_DIR_PF(Dir,base); \
LOAD64(%r10,isigns); \
RECON; \ RECON; \
nmu++; \ nmu++; \
} }
@ -144,7 +165,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
if((!local)&&(!st.same_node[Dir]) ) { \ if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \ LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(Dir,base); \ MULT_2SPIN_DIR_PF(Dir,base); \
LOAD64(%r10,isigns); \
RECON; \ RECON; \
nmu++; \ nmu++; \
} }
@ -159,7 +179,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
uint64_t basep; uint64_t basep;
const uint64_t plocal =(uint64_t) & in[0]; const uint64_t plocal =(uint64_t) & in[0];
COMPLEX_SIGNS(isigns);
MASK_REGS; MASK_REGS;
int nmax=U.oSites(); int nmax=U.oSites();
for(int site=0;site<Ns;site++) { for(int site=0;site<Ns;site++) {
@ -181,22 +200,137 @@ Author: Nils Meyer <nils.meyer@ur.de>
int ent=ss*8;// 2*Ndim int ent=ss*8;// 2*Ndim
int nent=ssn*8; int nent=ssn*8;
uint64_t delta_base, delta_base_p;
ASM_LEG_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON); ASM_LEG_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON);
#ifdef SHOW
float rescale = 64. * 12.;
std::cout << "=================================================================" << std::endl;
std::cout << "ss = " << ss << " ssn = " << ssn << std::endl;
std::cout << "sU = " << sU << " ssU = " << ssU << std::endl;
std::cout << " " << std::endl;
std::cout << "Dir = " << Xp << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Xp] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON); ASM_LEG(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON);
#ifdef SHOW
std::cout << "Dir = " << Yp << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Yp] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON); ASM_LEG(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON);
#ifdef SHOW
std::cout << "Dir = " << Zp << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Zp] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON); ASM_LEG(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON);
#ifdef SHOW
std::cout << "Dir = " << Tp << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Tp] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON); ASM_LEG(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON);
#ifdef SHOW
std::cout << "Dir = " << Xm << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Xm] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON); ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON);
#ifdef SHOW
std::cout << "Dir = " << Ym << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Ym] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON); ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON);
#ifdef SHOW
std::cout << "Dir = " << Zm << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Zm] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON); ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
#ifdef SHOW
std::cout << "Dir = " << Tm << " " << WHERE<< std::endl;
std::cout << "ent nent local perm = " << ent << " " << nent << " " << local << " " << perm << std::endl;
std::cout << "st.same_node[Dir] = " << st.same_node[Tm] << std::endl;
std::cout << "base = " << (base - plocal)/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
#ifdef EXTERIOR #ifdef EXTERIOR
if (nmu==0) break; if (nmu==0) break;
// if (nmu!=0) std::cout << "EXT "<<sU<<std::endl; // if (nmu!=0) std::cout << "EXT "<<sU<<std::endl;
#endif #endif
base = (uint64_t) &out[ss]; base = (uint64_t) &out[ss];
basep= st.GetPFInfo(nent,plocal); nent++; basep= st.GetPFInfo(nent,plocal); nent++;
#ifdef SHOW
std::cout << "Dir = FINAL " << WHERE<< std::endl;;
base_ss = base;
std::cout << "base = " << (base - (uint64_t) &out[0])/rescale << std::endl;
std::cout << "Basep = " << (basep - plocal)/rescale << std::endl;
//printf("U = %llu\n", (uint64_t)&[sU](Dir));
std::cout << "----------------------------------------------------" << std::endl;
#endif
//basep = (uint64_t) &out[ssn];
RESULT(base,basep); RESULT(base,basep);
} }
ssU++; ssU++;

38
Grid/simd/Fujitsu_A64FX_asm_double.h
View File

@ -31,6 +31,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A) #define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A)
#define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A) #define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)
#define PF_GAUGE(A) #define PF_GAUGE(A)
#define PREFETCH_RESULT_L2_STORE(A) PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(A)
#define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define LOCK_GAUGE(A) #define LOCK_GAUGE(A)
@ -38,8 +39,11 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define MASK_REGS DECLARATIONS_A64FXd #define MASK_REGS DECLARATIONS_A64FXd
#define COMPLEX_SIGNS(A) #define COMPLEX_SIGNS(A)
#define LOAD64(A,B) #define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXd(A) #define SAVE_RESULT(A,B) RESULT_A64FXd(A); PREFETCH_CHIMU_L1(B);
#define MULT_2SPIN_DIR_PF(A,B) PREFETCH_GAUGE_L1(A); PREFETCH_CHIMU_L2(B); MULT_2SPIN_A64FXd(A); if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } #define MULT_2SPIN_DIR_PF(A,B) \
MULT_2SPIN_A64FXd(A); \
PREFETCH_CHIMU_L2(B); \
if (s == 0) { if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } }
#define MAYBEPERM(A,perm) { A ; } #define MAYBEPERM(A,perm) { A ; }
#define LOAD_CHI(base) LOAD_CHI_A64FXd(base) #define LOAD_CHI(base) LOAD_CHI_A64FXd(base)
#define ZERO_PSI #define ZERO_PSI
@ -321,12 +325,28 @@ asm ( \
// PERM3 // PERM3
#define PERM3_A64FXd #define PERM3_A64FXd
// LOAD_GAUGE
#define LOAD_GAUGE \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
{ \
asm ( \
"ptrue p5.d \n\t" \
"ld1d { z24.d }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1d { z25.d }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \
"ld1d { z26.d }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \
"ld1d { z27.d }, p5/z, [%[fetchptr], -5, mul vl] \n\t" \
"ld1d { z28.d }, p5/z, [%[fetchptr], -2, mul vl] \n\t" \
"ld1d { z29.d }, p5/z, [%[fetchptr], 1, mul vl] \n\t" \
: \
: [fetchptr] "r" (baseU + 2 * 3 * 64) \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
); \
}
// MULT_2SPIN // MULT_2SPIN
#define MULT_2SPIN_A64FXd(A) \ #define MULT_2SPIN_A64FXd(A) \
{ \ { \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \ const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
asm ( \ asm ( \
"ptrue p5.d \n\t" \
"ld1d { z24.d }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \ "ld1d { z24.d }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1d { z25.d }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \ "ld1d { z25.d }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \
"ld1d { z26.d }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \ "ld1d { z26.d }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \
@ -738,6 +758,18 @@ asm ( \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \ : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
); );
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(base) \
{ \
asm ( \
"prfd PSTL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
"prfd PSTL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
"prfd PSTL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
: \
: [fetchptr] "r" (base) \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
); \
}
// ADD_RESULT_INTERNAL // ADD_RESULT_INTERNAL
#define ADD_RESULT_INTERNAL_A64FXd \ #define ADD_RESULT_INTERNAL_A64FXd \
asm ( \ asm ( \

38
Grid/simd/Fujitsu_A64FX_asm_single.h
View File

@ -31,6 +31,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A) #define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A)
#define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A) #define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)
#define PF_GAUGE(A) #define PF_GAUGE(A)
#define PREFETCH_RESULT_L2_STORE(A) PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(A)
#define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define LOCK_GAUGE(A) #define LOCK_GAUGE(A)
@ -38,8 +39,11 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define MASK_REGS DECLARATIONS_A64FXf #define MASK_REGS DECLARATIONS_A64FXf
#define COMPLEX_SIGNS(A) #define COMPLEX_SIGNS(A)
#define LOAD64(A,B) #define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXf(A) #define SAVE_RESULT(A,B) RESULT_A64FXf(A); PREFETCH_CHIMU_L1(B);
#define MULT_2SPIN_DIR_PF(A,B) PREFETCH_GAUGE_L1(A); PREFETCH_CHIMU_L2(B); MULT_2SPIN_A64FXf(A); if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } #define MULT_2SPIN_DIR_PF(A,B) \
MULT_2SPIN_A64FXf(A); \
PREFETCH_CHIMU_L2(B); \
if (s == 0) { if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } }
#define MAYBEPERM(A,perm) { A ; } #define MAYBEPERM(A,perm) { A ; }
#define LOAD_CHI(base) LOAD_CHI_A64FXf(base) #define LOAD_CHI(base) LOAD_CHI_A64FXf(base)
#define ZERO_PSI #define ZERO_PSI
@ -332,12 +336,28 @@ asm ( \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \ : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
); );
// LOAD_GAUGE
#define LOAD_GAUGE \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
{ \
asm ( \
"ptrue p5.s \n\t" \
"ld1w { z24.s }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1w { z25.s }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \
"ld1w { z26.s }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \
"ld1w { z27.s }, p5/z, [%[fetchptr], -5, mul vl] \n\t" \
"ld1w { z28.s }, p5/z, [%[fetchptr], -2, mul vl] \n\t" \
"ld1w { z29.s }, p5/z, [%[fetchptr], 1, mul vl] \n\t" \
: \
: [fetchptr] "r" (baseU + 2 * 3 * 64) \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
); \
}
// MULT_2SPIN // MULT_2SPIN
#define MULT_2SPIN_A64FXf(A) \ #define MULT_2SPIN_A64FXf(A) \
{ \ { \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \ const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
asm ( \ asm ( \
"ptrue p5.s \n\t" \
"ld1w { z24.s }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \ "ld1w { z24.s }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1w { z25.s }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \ "ld1w { z25.s }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \
"ld1w { z26.s }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \ "ld1w { z26.s }, p5/z, [%[fetchptr], 0, mul vl] \n\t" \
@ -749,6 +769,18 @@ asm ( \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \ : "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31" \
); );
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(base) \
{ \
asm ( \
"prfd PSTL2STRM, p5, [%[fetchptr], 0, mul vl] \n\t" \
"prfd PSTL2STRM, p5, [%[fetchptr], 4, mul vl] \n\t" \
"prfd PSTL2STRM, p5, [%[fetchptr], 8, mul vl] \n\t" \
: \
: [fetchptr] "r" (base) \
: "p5","cc","z0","z1","z2","z3","z4","z5","z6","z7","z8","z9","z10","z11","z12","z13","z14","z15","z16","z17","z18","z19","z20","z21","z22","z23","z24","z25","z26","z27","z28","z29","z30","z31","memory" \
); \
}
// ADD_RESULT_INTERNAL // ADD_RESULT_INTERNAL
#define ADD_RESULT_INTERNAL_A64FXf \ #define ADD_RESULT_INTERNAL_A64FXf \
asm ( \ asm ( \

26
Grid/simd/Fujitsu_A64FX_intrin_double.h
View File

@ -31,6 +31,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A) #define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXd(A)
#define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A) #define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A)
#define PF_GAUGE(A) #define PF_GAUGE(A)
#define PREFETCH_RESULT_L2_STORE(A) PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(A)
#define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define LOCK_GAUGE(A) #define LOCK_GAUGE(A)
@ -38,8 +39,11 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define MASK_REGS DECLARATIONS_A64FXd #define MASK_REGS DECLARATIONS_A64FXd
#define COMPLEX_SIGNS(A) #define COMPLEX_SIGNS(A)
#define LOAD64(A,B) #define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXd(A) #define SAVE_RESULT(A,B) RESULT_A64FXd(A); PREFETCH_CHIMU_L1(B);
#define MULT_2SPIN_DIR_PF(A,B) PREFETCH_GAUGE_L1(A); PREFETCH_CHIMU_L2(B); MULT_2SPIN_A64FXd(A); if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } #define MULT_2SPIN_DIR_PF(A,B) \
MULT_2SPIN_A64FXd(A); \
PREFETCH_CHIMU_L2(B); \
if (s == 0) { if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } }
#define MAYBEPERM(A,perm) { A ; } #define MAYBEPERM(A,perm) { A ; }
#define LOAD_CHI(base) LOAD_CHI_A64FXd(base) #define LOAD_CHI(base) LOAD_CHI_A64FXd(base)
#define ZERO_PSI #define ZERO_PSI
@ -279,6 +283,17 @@ Author: Nils Meyer <nils.meyer@ur.de>
// PERM3 // PERM3
#define PERM3_A64FXd #define PERM3_A64FXd
// LOAD_GAUGE
#define LOAD_GAUGE \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
{ \
U_00 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -6 * 64)); \
U_10 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -3 * 64)); \
U_20 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 0 * 64)); \
U_01 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -5 * 64)); \
U_11 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + -2 * 64)); \
U_21 = svld1(pg1, (float64_t*)(baseU + 2 * 3 * 64 + 1 * 64)); \
}
// MULT_2SPIN // MULT_2SPIN
#define MULT_2SPIN_A64FXd(A) \ #define MULT_2SPIN_A64FXd(A) \
{ \ { \
@ -574,6 +589,13 @@ Author: Nils Meyer <nils.meyer@ur.de>
result_31 = __svzero(result_31); \ result_31 = __svzero(result_31); \
result_32 = __svzero(result_32); result_32 = __svzero(result_32);
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(base) \
{ \
svprfd(pg1, (int64_t*)(base + 0), SV_PSTL2STRM); \
svprfd(pg1, (int64_t*)(base + 256), SV_PSTL2STRM); \
svprfd(pg1, (int64_t*)(base + 512), SV_PSTL2STRM); \
}
// ADD_RESULT_INTERNAL // ADD_RESULT_INTERNAL
#define ADD_RESULT_INTERNAL_A64FXd \ #define ADD_RESULT_INTERNAL_A64FXd \
result_00 = svadd_x(pg1, result_00, Chimu_00); \ result_00 = svadd_x(pg1, result_00, Chimu_00); \

26
Grid/simd/Fujitsu_A64FX_intrin_single.h
View File

@ -31,6 +31,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A) #define PREFETCH_CHIMU_L2(A) PREFETCH_CHIMU_L2_INTERNAL_A64FXf(A)
#define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A) #define PREFETCH_GAUGE_L2(A) PREFETCH_GAUGE_L2_INTERNAL_A64FXf(A)
#define PF_GAUGE(A) #define PF_GAUGE(A)
#define PREFETCH_RESULT_L2_STORE(A) PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(A)
#define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH1_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A) #define PREFETCH_CHIMU(A) PREFETCH_CHIMU_L1(A)
#define LOCK_GAUGE(A) #define LOCK_GAUGE(A)
@ -38,8 +39,11 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define MASK_REGS DECLARATIONS_A64FXf #define MASK_REGS DECLARATIONS_A64FXf
#define COMPLEX_SIGNS(A) #define COMPLEX_SIGNS(A)
#define LOAD64(A,B) #define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXf(A) #define SAVE_RESULT(A,B) RESULT_A64FXf(A); PREFETCH_CHIMU_L1(B);
#define MULT_2SPIN_DIR_PF(A,B) PREFETCH_GAUGE_L1(A); PREFETCH_CHIMU_L2(B); MULT_2SPIN_A64FXf(A); if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } #define MULT_2SPIN_DIR_PF(A,B) \
MULT_2SPIN_A64FXf(A); \
PREFETCH_CHIMU_L2(B); \
if (s == 0) { if ((A == 0) || (A == 4)) { PREFETCH_GAUGE_L2(A); } }
#define MAYBEPERM(A,perm) { A ; } #define MAYBEPERM(A,perm) { A ; }
#define LOAD_CHI(base) LOAD_CHI_A64FXf(base) #define LOAD_CHI(base) LOAD_CHI_A64FXf(base)
#define ZERO_PSI #define ZERO_PSI
@ -285,6 +289,17 @@ Author: Nils Meyer <nils.meyer@ur.de>
Chi_11 = svtbl(Chi_11, table0); \ Chi_11 = svtbl(Chi_11, table0); \
Chi_12 = svtbl(Chi_12, table0); Chi_12 = svtbl(Chi_12, table0);
// LOAD_GAUGE
#define LOAD_GAUGE \
const auto & ref(U[sU](A)); uint64_t baseU = (uint64_t)&ref; \
{ \
U_00 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -6 * 64)); \
U_10 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -3 * 64)); \
U_20 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 0 * 64)); \
U_01 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -5 * 64)); \
U_11 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + -2 * 64)); \
U_21 = svld1(pg1, (float32_t*)(baseU + 2 * 3 * 64 + 1 * 64)); \
}
// MULT_2SPIN // MULT_2SPIN
#define MULT_2SPIN_A64FXf(A) \ #define MULT_2SPIN_A64FXf(A) \
{ \ { \
@ -580,6 +595,13 @@ Author: Nils Meyer <nils.meyer@ur.de>
result_31 = __svzero(result_31); \ result_31 = __svzero(result_31); \
result_32 = __svzero(result_32); result_32 = __svzero(result_32);
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(base) \
{ \
svprfd(pg1, (int64_t*)(base + 0), SV_PSTL2STRM); \
svprfd(pg1, (int64_t*)(base + 256), SV_PSTL2STRM); \
svprfd(pg1, (int64_t*)(base + 512), SV_PSTL2STRM); \
}
// ADD_RESULT_INTERNAL // ADD_RESULT_INTERNAL
#define ADD_RESULT_INTERNAL_A64FXf \ #define ADD_RESULT_INTERNAL_A64FXf \
result_00 = svadd_x(pg1, result_00, Chimu_00); \ result_00 = svadd_x(pg1, result_00, Chimu_00); \

2
Grid/simd/Fujitsu_A64FX_undef.h
View File

@ -36,6 +36,8 @@ Author: Nils Meyer <nils.meyer@ur.de>
#undef PF_GAUGE #undef PF_GAUGE
#undef PREFETCH1_CHIMU #undef PREFETCH1_CHIMU
#undef PREFETCH_CHIMU #undef PREFETCH_CHIMU
#undef PREFETCH_RESULT_L2_STORE
#undef LOAD_GAUGE
#undef LOCK_GAUGE #undef LOCK_GAUGE
#undef UNLOCK_GAUGE #undef UNLOCK_GAUGE
#undef MASK_REGS #undef MASK_REGS

2410
Grid/simd/gridverter.py Executable file
View File

File diff suppressed because it is too large Load Diff