portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-10 06:00:45 +01:00
Code Releases Activity

GPU version

Browse Source
This commit is contained in:
Peter Boyle 2018-05-14 19:41:47 -04:00
parent aab27a655a
commit eb7d34a4cc
1 changed files with 32 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -39,7 +39,7 @@ NAMESPACE_BEGIN(Grid);
//////////////////////////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
class WilsonKernelsStatic { class WilsonKernelsStatic {
public: public:
enum { OptGeneric, OptHandUnroll, OptInlineAsm }; enum { OptGeneric, OptHandUnroll, OptInlineAsm, OptGpu };
enum { CommsAndCompute, CommsThenCompute }; enum { CommsAndCompute, CommsThenCompute };
static int Opt; static int Opt;
static int Comms; static int Comms;
@ -59,6 +59,7 @@ public:
{ {
//bgq_l1p_optimisation(1); //bgq_l1p_optimisation(1);
switch(Opt) { switch(Opt) {
#if defined(AVX512) || defined (QPX) #if defined(AVX512) || defined (QPX)
case WilsonKernelsStatic::OptInlineAsm: case WilsonKernelsStatic::OptInlineAsm:
if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite (st,U,buf,sF,sU,Ls,Nsite,in,out); if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite (st,U,buf,sF,sU,Ls,Nsite,in,out);
@ -67,6 +68,18 @@ public:
else assert(0); else assert(0);
break; break;
#endif #endif
case WilsonKernelsStatic::OptGpu:
for (int site = 0; site < Nsite; site++) {
for (int s = 0; s < Ls; s++) {
if(interior&&exterior) WilsonKernels<Impl>::GpuDhopSite(st,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteInt(st,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteExt(st,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
}
break;
case WilsonKernelsStatic::OptHandUnroll: case WilsonKernelsStatic::OptHandUnroll:
for (int site = 0; site < Nsite; site++) { for (int site = 0; site < Nsite; site++) {
for (int s = 0; s < Ls; s++) { for (int s = 0; s < Ls; s++) {
@ -128,6 +141,18 @@ public:
else assert(0); else assert(0);
break; break;
#endif #endif
case WilsonKernelsStatic::OptGpu:
for (int site = 0; site < Nsite; site++) {
for (int s = 0; s < Ls; s++) {
if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteDagInt(st,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteDagExt(st,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
}
break;
case WilsonKernelsStatic::OptHandUnroll: case WilsonKernelsStatic::OptHandUnroll:
for (int site = 0; site < Nsite; site++) { for (int site = 0; site < Nsite; site++) {
for (int s = 0; s < Ls; s++) { for (int s = 0; s < Ls; s++) {
@ -215,6 +240,12 @@ public:
private: private:
// Specialised variants // Specialised variants
static accelerator void GpuDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
static accelerator void GpuDhopSiteDag(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionFieldView &in, FermionFieldView &out);
static accelerator void GenericDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor * buf, static accelerator void GenericDhopSite(StencilView &st, DoubledGaugeFieldView &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionFieldView &in, FermionFieldView &out); int sF, int sU, const FermionFieldView &in, FermionFieldView &out);