portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

Avx512 changes for assembler kernels

Browse Source
This commit is contained in:
paboyle
2016-03-26 22:24:07 -06:00
parent 644fd6d32e
commit 165bffc2e7
14 changed files with 2806 additions and 1132 deletions
Download Patch File Download Diff File Expand all files Collapse all files

566
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@ -308,548 +308,11 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
namespace Grid {
namespace QCD {
#if 0
template<class Impl>
int WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
// std::cout << "Hand op Dhop "<<std::endl;
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
REGISTER Simd result_00 ; zeroit(result_00); // 12 regs on knc
REGISTER Simd result_01 ; zeroit(result_01); // 12 regs on knc
REGISTER Simd result_02 ; zeroit(result_02); // 12 regs on knc
REGISTER Simd result_10 ; zeroit(result_10); // 12 regs on knc
REGISTER Simd result_11 ; zeroit(result_11); // 12 regs on knc
REGISTER Simd result_12 ; zeroit(result_12); // 12 regs on knc
REGISTER Simd result_20 ; zeroit(result_20); // 12 regs on knc
REGISTER Simd result_21 ; zeroit(result_21); // 12 regs on knc
REGISTER Simd result_22 ; zeroit(result_22); // 12 regs on knc
REGISTER Simd result_30 ; zeroit(result_30); // 12 regs on knc
REGISTER Simd result_31 ; zeroit(result_31); // 12 regs on knc
REGISTER Simd result_32 ; zeroit(result_32); // 12 regs on knc
REGISTER Simd Chi_00; // two spinor; 6 regs
REGISTER Simd Chi_01;
REGISTER Simd Chi_02;
REGISTER Simd Chi_10;
REGISTER Simd Chi_11;
REGISTER Simd Chi_12; // 14 left
REGISTER Simd UChi_00; // two spinor; 6 regs
REGISTER Simd UChi_01;
REGISTER Simd UChi_02;
REGISTER Simd UChi_10;
REGISTER Simd UChi_11;
REGISTER Simd UChi_12; // 8 left
REGISTER Simd U_00; // two rows of U matrix
REGISTER Simd U_10;
REGISTER Simd U_20;
REGISTER Simd U_01;
REGISTER Simd U_11;
REGISTER Simd U_21; // 2 reg left.
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
#define Chimu_02 Chi_02
#define Chimu_10 Chi_10
#define Chimu_11 Chi_11
#define Chimu_12 Chi_12
#define Chimu_20 UChi_00
#define Chimu_21 UChi_01
#define Chimu_22 UChi_02
#define Chimu_30 UChi_10
#define Chimu_31 UChi_11
#define Chimu_32 UChi_12
StencilEntry *SE;
int offset, ptype;
int num = 0;
// Xp
SE=st.GetEntry(ptype,Xp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
XP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Xp);
XP_RECON_ACCUM;
num++;
}
// Yp
SE=st.GetEntry(ptype,Yp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
YP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Yp);
YP_RECON_ACCUM;
num++;
}
// Zp
SE=st.GetEntry(ptype,Zp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
ZP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Zp);
ZP_RECON_ACCUM;
num++;
}
// Tp
SE=st.GetEntry(ptype,Tp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
TP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Tp);
TP_RECON_ACCUM;
num++;
}
// Xm
SE=st.GetEntry(ptype,Xm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
XM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Xm);
XM_RECON_ACCUM;
num++;
}
// Ym
SE=st.GetEntry(ptype,Ym,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
YM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Ym);
YM_RECON_ACCUM;
num++;
}
// Zm
SE=st.GetEntry(ptype,Zm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
ZM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Zm);
ZM_RECON_ACCUM;
num++;
}
// Tm
SE=st.GetEntry(ptype,Tm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
TM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Tm);
TM_RECON_ACCUM;
num++;
}
SiteSpinor & ref (out._odata[ss]);
if ( Local ) {
vstream(ref()(0)(0),result_00);
vstream(ref()(0)(1),result_01);
vstream(ref()(0)(2),result_02);
vstream(ref()(1)(0),result_10);
vstream(ref()(1)(1),result_11);
vstream(ref()(1)(2),result_12);
vstream(ref()(2)(0),result_20);
vstream(ref()(2)(1),result_21);
vstream(ref()(2)(2),result_22);
vstream(ref()(3)(0),result_30);
vstream(ref()(3)(1),result_31);
vstream(ref()(3)(2),result_32);
return 1;
} else if ( num ) {
vstream(ref()(0)(0),ref()(0)(0)+result_00);
vstream(ref()(0)(1),ref()(0)(1)+result_01);
vstream(ref()(0)(2),ref()(0)(2)+result_02);
vstream(ref()(1)(0),ref()(1)(0)+result_10);
vstream(ref()(1)(1),ref()(1)(1)+result_11);
vstream(ref()(1)(2),ref()(1)(2)+result_12);
vstream(ref()(2)(0),ref()(2)(0)+result_20);
vstream(ref()(2)(1),ref()(2)(1)+result_21);
vstream(ref()(2)(2),ref()(2)(2)+result_22);
vstream(ref()(3)(0),ref()(3)(0)+result_30);
vstream(ref()(3)(1),ref()(3)(1)+result_31);
vstream(ref()(3)(2),ref()(3)(2)+result_32);
return 1;
}
return 0;
}
template<class Impl>
int WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
{
// std::cout << "Hand op Dhop "<<std::endl;
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
REGISTER Simd result_00 ; zeroit(result_00); // 12 regs on knc
REGISTER Simd result_01 ; zeroit(result_01); // 12 regs on knc
REGISTER Simd result_02 ; zeroit(result_02); // 12 regs on knc
REGISTER Simd result_10 ; zeroit(result_10); // 12 regs on knc
REGISTER Simd result_11 ; zeroit(result_11); // 12 regs on knc
REGISTER Simd result_12 ; zeroit(result_12); // 12 regs on knc
REGISTER Simd result_20 ; zeroit(result_20); // 12 regs on knc
REGISTER Simd result_21 ; zeroit(result_21); // 12 regs on knc
REGISTER Simd result_22 ; zeroit(result_22); // 12 regs on knc
REGISTER Simd result_30 ; zeroit(result_30); // 12 regs on knc
REGISTER Simd result_31 ; zeroit(result_31); // 12 regs on knc
REGISTER Simd result_32 ; zeroit(result_32); // 12 regs on knc
REGISTER Simd Chi_00; // two spinor; 6 regs
REGISTER Simd Chi_01;
REGISTER Simd Chi_02;
REGISTER Simd Chi_10;
REGISTER Simd Chi_11;
REGISTER Simd Chi_12; // 14 left
REGISTER Simd UChi_00; // two spinor; 6 regs
REGISTER Simd UChi_01;
REGISTER Simd UChi_02;
REGISTER Simd UChi_10;
REGISTER Simd UChi_11;
REGISTER Simd UChi_12; // 8 left
REGISTER Simd U_00; // two rows of U matrix
REGISTER Simd U_10;
REGISTER Simd U_20;
REGISTER Simd U_01;
REGISTER Simd U_11;
REGISTER Simd U_21; // 2 reg left.
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
#define Chimu_02 Chi_02
#define Chimu_10 Chi_10
#define Chimu_11 Chi_11
#define Chimu_12 Chi_12
#define Chimu_20 UChi_00
#define Chimu_21 UChi_01
#define Chimu_22 UChi_02
#define Chimu_30 UChi_10
#define Chimu_31 UChi_11
#define Chimu_32 UChi_12
StencilEntry *SE;
int offset, ptype;
int num = 0;
// Xp
SE=st.GetEntry(ptype,Xp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
XM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Xp);
XM_RECON_ACCUM;
num++;
}
// Yp
SE=st.GetEntry(ptype,Yp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
YM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Yp);
YM_RECON_ACCUM;
num++;
}
// Zp
SE=st.GetEntry(ptype,Zp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
ZM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Zp);
ZM_RECON_ACCUM;
num++;
}
// Tp
SE=st.GetEntry(ptype,Tp,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
TM_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Tp);
TM_RECON_ACCUM;
num++;
}
// Xm
SE=st.GetEntry(ptype,Xm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
XP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Xm);
XP_RECON_ACCUM;
num++;
}
// Ym
SE=st.GetEntry(ptype,Ym,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
YP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Ym);
YP_RECON_ACCUM;
num++;
}
// Zm
SE=st.GetEntry(ptype,Zm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
ZP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Zm);
ZP_RECON_ACCUM;
num++;
}
// Tm
SE=st.GetEntry(ptype,Tm,ss);
offset = SE->_offset;
if (Local && SE->_is_local ) {
LOAD_CHIMU;
TP_PROJ;
if ( SE->_permute ) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
}
if ( Nonlocal && (!SE->_is_local) ) {
LOAD_CHI;
}
if ( (Local && SE->_is_local) || ( Nonlocal && (!SE->_is_local)) ) {
MULT_2SPIN(Tm);
TP_RECON_ACCUM;
num++;
}
SiteSpinor & ref (out._odata[ss]);
if ( Local ) {
vstream(ref()(0)(0),result_00);
vstream(ref()(0)(1),result_01);
vstream(ref()(0)(2),result_02);
vstream(ref()(1)(0),result_10);
vstream(ref()(1)(1),result_11);
vstream(ref()(1)(2),result_12);
vstream(ref()(2)(0),result_20);
vstream(ref()(2)(1),result_21);
vstream(ref()(2)(2),result_22);
vstream(ref()(3)(0),result_30);
vstream(ref()(3)(1),result_31);
vstream(ref()(3)(2),result_32);
return 1;
} else if ( num ) {
vstream(ref()(0)(0),ref()(0)(0)+result_00);
vstream(ref()(0)(1),ref()(0)(1)+result_01);
vstream(ref()(0)(2),ref()(0)(2)+result_02);
vstream(ref()(1)(0),ref()(1)(0)+result_10);
vstream(ref()(1)(1),ref()(1)(1)+result_11);
vstream(ref()(1)(2),ref()(1)(2)+result_12);
vstream(ref()(2)(0),ref()(2)(0)+result_20);
vstream(ref()(2)(1),ref()(2)(1)+result_21);
vstream(ref()(2)(2),ref()(2)(2)+result_22);
vstream(ref()(3)(0),ref()(3)(0)+result_30);
vstream(ref()(3)(1),ref()(3)(1)+result_31);
vstream(ref()(3)(2),ref()(3)(2)+result_32);
return 1;
}
return 0;
}
#else
template<class Impl>
int WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
@ -1094,7 +557,7 @@ int WilsonKernels<Impl >::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField
template<class Impl>
int WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out,bool l, bool nl)
int ss,int sU,const FermionField &in, FermionField &out)
{
// std::cout << "Hand op Dhop "<<std::endl;
typedef typename Simd::scalar_type S;
@ -1337,14 +800,13 @@ int WilsonKernels<Impl >::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeFi
}
#endif
////////////////////////////////////////////////
// Specialise Gparity to simple implementation
////////////////////////////////////////////////
template<>
int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
int sF,int sU,const FermionField &in, FermionField &out)
{
DiracOptDhopSite(st,U,buf,sF,sU,in,out); // returns void, will template override for Wilson Nc=3
//check consistency of return types between these functions and the ones in WilsonKernels.cc
@ -1355,7 +817,7 @@ int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,Doub
template<>
int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
int sF,int sU,const FermionField &in, FermionField &out)
{
DiracOptDhopSiteDag(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
return 0;
@ -1364,7 +826,7 @@ int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,D
template<>
int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
int sF,int sU,const FermionField &in, FermionField &out)
{
DiracOptDhopSite(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
return 0;
@ -1373,7 +835,7 @@ int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,Doub
template<>
int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out, bool Local, bool Nonlocal)
int sF,int sU,const FermionField &in, FermionField &out)
{
DiracOptDhopSiteDag(st,U,buf,sF,sU,in,out); // will template override for Wilson Nc=3
return 0;
@ -1383,29 +845,29 @@ int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,D
template int WilsonKernels<WilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out,bool l,bool n);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<WilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<WilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<WilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool n);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<GparityWilsonImplF>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
int ss,int sU,const FermionField &in, FermionField &out);
template int WilsonKernels<GparityWilsonImplD>::DiracOptHandDhopSiteDag(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int ss,int sU,const FermionField &in, FermionField &out, bool l, bool nl);
int ss,int sU,const FermionField &in, FermionField &out);
}}