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

Global edit with change to View usage. autoView() creates a wrapper object that closes the view when scope closes.

Browse Source
This commit is contained in:
Peter Boyle
2020-06-05 18:52:35 -04:00
parent f39c2a240b
commit 1a4c8c3387
78 changed files with 773 additions and 778 deletions
Download Patch File Download Diff File Expand all files Collapse all files

79
Grid/qcd/action/fermion/GparityWilsonImpl.h
View File

@ -232,15 +232,17 @@ public:
if ( Params.twists[mu] ) {
Uconj = where(coor==neglink,-Uconj,Uconj);
}
auto U_v = U.View(CpuRead);
auto Uds_v = Uds.View(CpuWrite);
auto Uconj_v = Uconj.View(CpuRead);
auto Utmp_v= Utmp.View(CpuWrite);
thread_foreach(ss,U_v,{
Uds_v[ss](0)(mu) = U_v[ss]();
Uds_v[ss](1)(mu) = Uconj_v[ss]();
});
{
autoView( U_v , U, CpuRead);
autoView( Uconj_v , Uconj, CpuRead);
autoView( Uds_v , Uds, CpuWrite);
autoView( Utmp_v, Utmp, CpuWrite);
thread_foreach(ss,U_v,{
Uds_v[ss](0)(mu) = U_v[ss]();
Uds_v[ss](1)(mu) = Uconj_v[ss]();
});
}
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
Uconj = adj(Cshift(Uconj,mu,-1));
@ -250,19 +252,25 @@ public:
Utmp = where(coor==0,Uconj,Utmp);
}
thread_foreach(ss,Utmp_v,{
Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
});
{
autoView( Uds_v , Uds, CpuWrite);
autoView( Utmp_v, Utmp, CpuWrite);
thread_foreach(ss,Utmp_v,{
Uds_v[ss](0)(mu+4) = Utmp_v[ss]();
});
}
Utmp = Uconj;
if ( Params.twists[mu] ) {
Utmp = where(coor==0,U,Utmp);
}
thread_foreach(ss,Utmp_v,{
Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
});
{
autoView( Uds_v , Uds, CpuWrite);
autoView( Utmp_v, Utmp, CpuWrite);
thread_foreach(ss,Utmp_v,{
Uds_v[ss](1)(mu+4) = Utmp_v[ss]();
});
}
}
}
@ -272,11 +280,14 @@ public:
GaugeLinkField link(mat.Grid());
// use lorentz for flavour as hack.
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
auto link_v = link.View(CpuWrite);
auto tmp_v = tmp.View(CpuRead);
thread_foreach(ss,tmp_v,{
link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
});
{
autoView( link_v , link, CpuWrite);
autoView( tmp_v , tmp, CpuRead);
thread_foreach(ss,tmp_v,{
link_v[ss]() = tmp_v[ss](0, 0) + conjugate(tmp_v[ss](1, 1));
});
}
PokeIndex<LorentzIndex>(mat, link, mu);
return;
}
@ -306,16 +317,18 @@ public:
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
auto tmp_v = tmp.View(CpuWrite);
auto Atilde_v = Atilde.View(CpuRead);
auto Btilde_v = Btilde.View(CpuRead);
thread_for(ss,tmp.Grid()->oSites(),{
for (int s = 0; s < Ls; s++) {
int sF = s + Ls * ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
}
});
{
autoView( tmp_v , tmp, CpuWrite);
autoView( Atilde_v , Atilde, CpuRead);
autoView( Btilde_v , Btilde, CpuRead);
thread_for(ss,tmp.Grid()->oSites(),{
for (int s = 0; s < Ls; s++) {
int sF = s + Ls * ss;
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
}
});
}
PokeIndex<LorentzIndex>(mat, tmp, mu);
return;
}

24
Grid/qcd/action/fermion/WilsonCloverFermion.h
View File

@ -264,8 +264,8 @@ private:
{
CloverFieldType T(F.Grid());
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView(T_v,T,CpuWrite);
autoView(F_v,F,CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 1) = timesMinusI(F_v[i]()());
@ -282,8 +282,8 @@ private:
CloverFieldType T(F.Grid());
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView(T_v, T,CpuWrite);
autoView(F_v, F,CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 1) = -F_v[i]()();
@ -300,8 +300,8 @@ private:
CloverFieldType T(F.Grid());
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView(T_v,T,CpuWrite);
autoView(F_v,F,CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 0) = timesMinusI(F_v[i]()());
@ -318,8 +318,8 @@ private:
CloverFieldType T(F.Grid());
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView( T_v , T, CpuWrite);
autoView( F_v , F, CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 1) = timesI(F_v[i]()());
@ -336,8 +336,8 @@ private:
CloverFieldType T(F.Grid());
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView( T_v ,T,CpuWrite);
autoView( F_v ,F,CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 1) = -(F_v[i]()());
@ -355,8 +355,8 @@ private:
T = Zero();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
autoView( T_v , T,CpuWrite);
autoView( F_v , F,CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 0) = timesI(F_v[i]()());

29
Grid/qcd/action/fermion/WilsonImpl.h
View File

@ -106,9 +106,9 @@ public:
const _SpinorField & phi,
int mu)
{
auto out_v= out.View(CpuWrite);
auto phi_v= phi.View(CpuRead);
auto Umu_v= Umu.View(CpuRead);
autoView( out_v, out, CpuWrite);
autoView( phi_v, phi, CpuRead);
autoView( Umu_v, Umu, CpuRead);
thread_for(sss,out.Grid()->oSites(),{
multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
});
@ -191,18 +191,19 @@ public:
int Ls=Btilde.Grid()->_fdimensions[0];
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
auto tmp_v = tmp.View(CpuWrite);
auto Btilde_v = Btilde.View(CpuRead);
auto Atilde_v = Atilde.View(CpuRead);
thread_for(sss,tmp.Grid()->oSites(),{
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
}
});
{
autoView( tmp_v , tmp, CpuWrite);
autoView( Btilde_v , Btilde, CpuRead);
autoView( Atilde_v , Atilde, CpuRead);
thread_for(sss,tmp.Grid()->oSites(),{
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp_v[sU] = tmp_v[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde_v[sF],Atilde_v[sF])); // ordering here
}
});
}
PokeIndex<LorentzIndex>(mat,tmp,mu);
}
};

20
Grid/qcd/action/fermion/implementation/CayleyFermion5Dcache.h
View File

@ -50,9 +50,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
chi_i.Checkerboard()=psi_i.Checkerboard();
GridBase *grid=psi_i.Grid();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i,AcceleratorRead);
autoView(phi , phi_i,AcceleratorRead);
autoView(chi , chi_i,AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
auto pdiag = &diag[0];
@ -93,9 +93,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
{
chi_i.Checkerboard()=psi_i.Checkerboard();
GridBase *grid=psi_i.Grid();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i,AcceleratorRead);
autoView(phi , phi_i,AcceleratorRead);
autoView(chi , chi_i,AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
auto pdiag = &diag[0];
@ -131,8 +131,8 @@ CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi_i, FermionField &chi
chi_i.Checkerboard()=psi_i.Checkerboard();
GridBase *grid=psi_i.Grid();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i,AcceleratorRead);
autoView(chi , chi_i,AcceleratorWrite);
int Ls=this->Ls;
@ -193,8 +193,8 @@ CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi_i, FermionField &chi
GridBase *grid=psi_i.Grid();
int Ls=this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i,AcceleratorRead);
autoView(chi , chi_i,AcceleratorWrite);
auto plee = & lee [0];
auto pdee = & dee [0];

24
Grid/qcd/action/fermion/implementation/CayleyFermion5Dvec.h
View File

@ -65,9 +65,9 @@ CayleyFermion5D<Impl>::M5D(const FermionField &psi_i,
EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
chi_i.Checkerboard()=psi_i.Checkerboard();
GridBase *grid=psi_i.Grid();
auto psi = psi_i.View(CpuRead);
auto phi = phi_i.View(CpuRead);
auto chi = chi_i.View(CpuWrite);
autoView(psi, psi_i,CpuRead);
autoView(phi, phi_i,CpuRead);
autoView(chi, chi_i,CpuWrite);
int Ls = this->Ls;
int LLs = grid->_rdimensions[0];
const int nsimd= Simd::Nsimd();
@ -213,9 +213,9 @@ CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi_i,
EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
chi_i.Checkerboard()=psi_i.Checkerboard();
GridBase *grid=psi_i.Grid();
auto psi=psi_i.View(CpuRead);
auto phi=phi_i.View(CpuRead);
auto chi=chi_i.View(CpuWrite);
autoView(psi,psi_i,CpuRead);
autoView(phi,phi_i,CpuRead);
autoView(chi,chi_i,CpuWrite);
int Ls = this->Ls;
int LLs = grid->_rdimensions[0];
int nsimd= Simd::Nsimd();
@ -357,8 +357,8 @@ CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi_i, FermionField
Vector<iSinglet<Simd> > &Matm)
{
EnableIf<Impl::LsVectorised&&EnableBool,int> sfinae=0;
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(CpuWrite);
autoView(psi , psi_i,CpuRead);
autoView(chi , chi_i,CpuWrite);
#ifndef AVX512
{
SiteHalfSpinor BcastP;
@ -535,8 +535,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
EnableIf<Impl::LsVectorised,int> sfinae=0;
#ifndef AVX512
{
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(CpuWrite);
autoView(psi , psi_i,CpuRead);
autoView(chi , chi_i,CpuWrite);
SiteHalfSpinor BcastP;
SiteHalfSpinor BcastM;
@ -586,8 +586,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
}
#else
{
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(CpuWrite);
autoView(psi , psi_i,CpuRead);
autoView(chi , chi_i,CpuWrite);
// pointers
// MASK_REGS;
#define Chi_00 %zmm0

20
Grid/qcd/action/fermion/implementation/DomainWallEOFAFermionCache.h
View File

@ -46,9 +46,9 @@ void DomainWallEOFAFermion<Impl>::M5D(const FermionField& psi_i, const FermionFi
chi_i.Checkerboard() = psi_i.Checkerboard();
int Ls = this->Ls;
GridBase* grid = psi_i.Grid();
auto phi = phi_i.View(AcceleratorRead);
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView( phi , phi_i, AcceleratorRead);
autoView( psi , psi_i, AcceleratorRead);
autoView( chi , chi_i, AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
auto pdiag = &diag[0];
auto pupper = &upper[0];
@ -82,9 +82,9 @@ void DomainWallEOFAFermion<Impl>::M5Ddag(const FermionField& psi_i, const Fermio
GridBase* grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView( psi , psi_i, AcceleratorRead);
autoView( phi , phi_i, AcceleratorRead);
autoView( chi , chi_i, AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
auto pdiag = &diag[0];
auto pupper = &upper[0];
@ -116,8 +116,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInv(const FermionField& psi_i, FermionFie
{
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase* grid = psi_i.Grid();
auto psi=psi_i.View(AcceleratorRead);
auto chi=chi_i.View(AcceleratorWrite);
autoView( psi, psi_i, AcceleratorRead);
autoView( chi, chi_i, AcceleratorWrite);
int Ls = this->Ls;
auto plee = & this->lee[0];
@ -172,8 +172,8 @@ void DomainWallEOFAFermion<Impl>::MooeeInvDag(const FermionField& psi_i, Fermion
{
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase* grid = psi_i.Grid();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView( psi, psi_i, AcceleratorRead);
autoView( chi, chi_i, AcceleratorWrite);
int Ls = this->Ls;
auto plee = & this->lee[0];

8
Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermion5DImplementation.h
View File

@ -221,10 +221,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopDir(const FermionField &in, FermionFi
Compressor compressor;
Stencil.HaloExchange(in,compressor);
auto Umu_v = Umu.View(CpuRead);
auto UUUmu_v = UUUmu.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
autoView( Umu_v , Umu, CpuRead);
autoView( UUUmu_v , UUUmu, CpuRead);
autoView( in_v , in, CpuRead);
autoView( out_v , out, CpuWrite);
thread_for( ss,Umu.Grid()->oSites(),{
for(int s=0;s<Ls;s++){
int sU=ss;

16
Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
View File

@ -250,10 +250,10 @@ void ImprovedStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGauge
////////////////////////
// Call the single hop
////////////////////////
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto B_v = B.View(CpuWrite);
auto Btilde_v = Btilde.View(CpuWrite);
autoView( U_v , U, CpuRead);
autoView( UUU_v , UUU, CpuRead);
autoView( B_v , B, CpuWrite);
autoView( Btilde_v , Btilde, CpuWrite);
thread_for(sss,B.Grid()->oSites(),{
Kernels::DhopDirKernel(st, U_v, UUU_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
});
@ -378,10 +378,10 @@ void ImprovedStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionFiel
Compressor compressor;
Stencil.HaloExchange(in, compressor);
auto Umu_v = Umu.View(CpuRead);
auto UUUmu_v = UUUmu.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
autoView( Umu_v , Umu, CpuRead);
autoView( UUUmu_v , UUUmu, CpuRead);
autoView( in_v , in, CpuRead);
autoView( out_v , out, CpuWrite);
thread_for( sss, in.Grid()->oSites(),{
Kernels::DhopDirKernel(Stencil, Umu_v, UUUmu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
});

40
Grid/qcd/action/fermion/implementation/MobiusEOFAFermionCache.h
View File

@ -44,9 +44,9 @@ void MobiusEOFAFermion<Impl>::M5D(const FermionField &psi_i, const FermionField
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(phi , phi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
@ -84,9 +84,9 @@ void MobiusEOFAFermion<Impl>::M5D_shift(const FermionField &psi_i, const Fermion
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(phi , phi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
auto pm = this->pm;
int shift_s = (pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
@ -132,9 +132,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag(const FermionField &psi_i, const FermionFie
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(phi , phi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
@ -174,9 +174,9 @@ void MobiusEOFAFermion<Impl>::M5Ddag_shift(const FermionField &psi_i, const Ferm
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(phi , phi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
assert(phi.Checkerboard() == psi.Checkerboard());
@ -226,8 +226,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv(const FermionField &psi_i, FermionField &
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
auto plee = & this->lee [0];
auto pdee = & this->dee [0];
@ -286,8 +286,8 @@ void MobiusEOFAFermion<Impl>::MooeeInv_shift(const FermionField &psi_i, FermionF
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
auto pm = this->pm;
auto plee = & this->lee [0];
@ -354,8 +354,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag(const FermionField &psi_i, FermionFiel
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
int Ls = this->Ls;
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
auto plee = & this->lee [0];
auto pdee = & this->dee [0];
@ -410,8 +410,8 @@ void MobiusEOFAFermion<Impl>::MooeeInvDag_shift(const FermionField &psi_i, Fermi
{
chi_i.Checkerboard() = psi_i.Checkerboard();
GridBase *grid = psi_i.Grid();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
autoView(psi , psi_i, AcceleratorRead);
autoView(chi , chi_i, AcceleratorWrite);
int Ls = this->Ls;
auto pm = this->pm;

12
Grid/qcd/action/fermion/implementation/NaiveStaggeredFermionImplementation.h
View File

@ -208,9 +208,9 @@ void NaiveStaggeredFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeFie
////////////////////////
// Call the single hop
////////////////////////
auto U_v = U.View(CpuRead);
auto B_v = B.View(CpuWrite);
auto Btilde_v = Btilde.View(CpuWrite);
autoView( U_v , U, CpuRead);
autoView( B_v , B, CpuWrite);
autoView( Btilde_v , Btilde, CpuWrite);
thread_for(sss,B.Grid()->oSites(),{
Kernels::DhopDirKernel(st, U_v, U_v, st.CommBuf(), sss, sss, B_v, Btilde_v, mu,1);
});
@ -315,9 +315,9 @@ void NaiveStaggeredFermion<Impl>::DhopDir(const FermionField &in, FermionField &
Compressor compressor;
Stencil.HaloExchange(in, compressor);
auto Umu_v = Umu.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
autoView( Umu_v , Umu, CpuRead);
autoView( in_v , in, CpuRead);
autoView( out_v , out, CpuWrite);
// thread_for( sss, in.Grid()->oSites(),{
// Kernels::DhopDirKernel(Stencil, Umu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
// });

20
Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h
View File

@ -261,11 +261,11 @@ void StaggeredKernels<Impl>::DhopImproved(StencilImpl &st, LebesgueOrder &lo,
GridBase *FGrid=in.Grid();
GridBase *UGrid=U.Grid();
typedef StaggeredKernels<Impl> ThisKernel;
auto UUU_v = UUU.View(AcceleratorRead);
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
autoView( UUU_v , UUU, AcceleratorRead);
autoView( U_v , U, AcceleratorRead);
autoView( in_v , in, AcceleratorRead);
autoView( out_v , out, AcceleratorWrite);
autoView( st_v , st, AcceleratorRead);
SiteSpinor * buf = st.CommBuf();
int Ls=1;
@ -301,11 +301,11 @@ void StaggeredKernels<Impl>::DhopNaive(StencilImpl &st, LebesgueOrder &lo,
GridBase *FGrid=in.Grid();
GridBase *UGrid=U.Grid();
typedef StaggeredKernels<Impl> ThisKernel;
auto UUU_v= U.View(AcceleratorRead);
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
autoView( UUU_v , U, AcceleratorRead);
autoView( U_v , U, AcceleratorRead);
autoView( in_v , in, AcceleratorRead);
autoView( out_v , out, AcceleratorWrite);
autoView( st_v , st, AcceleratorRead);
SiteSpinor * buf = st.CommBuf();
int Ls=1;

22
Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
View File

@ -475,12 +475,12 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
// Inefficient comms method but not performance critical.
tmp1 = Cshift(q_in_1, mu, 1);
tmp2 = Cshift(q_in_2, mu, 1);
auto tmp1_v = tmp1.View(CpuWrite);
auto tmp2_v = tmp2.View(CpuWrite);
auto q_in_1_v=q_in_1.View(CpuRead);
auto q_in_2_v=q_in_2.View(CpuRead);
auto q_out_v = q_out.View(CpuRead);
auto Umu_v = Umu.View(CpuRead);
autoView( tmp1_v , tmp1, CpuWrite);
autoView( tmp2_v , tmp2, CpuWrite);
autoView( q_in_1_v,q_in_1, CpuRead);
autoView( q_in_2_v,q_in_2, CpuRead);
autoView( q_out_v , q_out, CpuRead);
autoView( Umu_v , Umu, CpuRead);
thread_for(sU, Umu.Grid()->oSites(),{
Kernels::ContractConservedCurrentSiteFwd(tmp1_v[sU],
q_in_2_v[sU],
@ -526,11 +526,11 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
tmp = lattice_cmplx*q_in;
tmpBwd = Cshift(tmp, mu, -1);
auto coords_v = coords.View(CpuRead);
auto tmpFwd_v = tmpFwd.View(CpuRead);
auto tmpBwd_v = tmpBwd.View(CpuRead);
auto Umu_v = Umu.View(CpuRead);
auto q_out_v = q_out.View(CpuWrite);
autoView( coords_v , coords, CpuRead);
autoView( tmpFwd_v , tmpFwd, CpuRead);
autoView( tmpBwd_v , tmpBwd, CpuRead);
autoView( Umu_v , Umu, CpuRead);
autoView( q_out_v , q_out, CpuWrite);
thread_for(sU, Umu.Grid()->oSites(), {

46
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -348,18 +348,18 @@ template <class Impl>
void WilsonKernels<Impl>::DhopDirAll( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int Ls,
int Nsite, const FermionField &in, std::vector<FermionField> &out)
{
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto st_v = st.View(AcceleratorRead);
autoView(U_v ,U,AcceleratorRead);
autoView(in_v ,in,AcceleratorRead);
autoView(st_v ,st,AcceleratorRead);
auto out_Xm = out[0].View(AcceleratorWrite);
auto out_Ym = out[1].View(AcceleratorWrite);
auto out_Zm = out[2].View(AcceleratorWrite);
auto out_Tm = out[3].View(AcceleratorWrite);
auto out_Xp = out[4].View(AcceleratorWrite);
auto out_Yp = out[5].View(AcceleratorWrite);
auto out_Zp = out[6].View(AcceleratorWrite);
auto out_Tp = out[7].View(AcceleratorWrite);
autoView(out_Xm,out[0],AcceleratorWrite);
autoView(out_Ym,out[1],AcceleratorWrite);
autoView(out_Zm,out[2],AcceleratorWrite);
autoView(out_Tm,out[3],AcceleratorWrite);
autoView(out_Xp,out[4],AcceleratorWrite);
autoView(out_Yp,out[5],AcceleratorWrite);
autoView(out_Zp,out[6],AcceleratorWrite);
autoView(out_Tp,out[7],AcceleratorWrite);
auto CBp=st.CommBuf();
accelerator_forNB(sss,Nsite*Ls,Simd::Nsimd(),{
int sU=sss/Ls;
@ -383,10 +383,10 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
assert(dirdisp<=7);
assert(dirdisp>=0);
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
autoView(U_v ,U ,AcceleratorRead);
autoView(in_v ,in ,AcceleratorRead);
autoView(out_v,out,AcceleratorWrite);
autoView(st_v ,st ,AcceleratorRead);
auto CBp=st.CommBuf();
#define LoopBody(Dir) \
case Dir : \
@ -438,10 +438,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
int Ls, int Nsite, const FermionField &in, FermionField &out,
int interior,int exterior)
{
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
autoView(U_v , U,AcceleratorRead);
autoView(in_v , in,AcceleratorRead);
autoView(out_v,out,AcceleratorWrite);
autoView(st_v , st,AcceleratorRead);
if( interior && exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSite); return;}
@ -469,10 +469,10 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
int Ls, int Nsite, const FermionField &in, FermionField &out,
int interior,int exterior)
{
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
autoView(U_v ,U,AcceleratorRead);
autoView(in_v ,in,AcceleratorRead);
autoView(out_v,out,AcceleratorWrite);
autoView(st_v ,st,AcceleratorRead);
if( interior && exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDag); return;}

10
Grid/qcd/action/gauge/GaugeImplTypes.h
View File

@ -86,8 +86,8 @@ public:
// Move this elsewhere? FIXME
static inline void AddLink(Field &U, LinkField &W, int mu) { // U[mu] += W
auto U_v = U.View(CpuWrite);
auto W_v = W.View(CpuRead);
autoView(U_v,U,CpuWrite);
autoView(W_v,W,CpuRead);
thread_for( ss, U.Grid()->oSites(), {
U_v[ss](mu) = U_v[ss](mu) + W_v[ss]();
});
@ -131,15 +131,15 @@ public:
//static std::chrono::duration<double> diff;
//auto start = std::chrono::high_resolution_clock::now();
auto U_v = U.View(CpuWrite);
auto P_v = P.View(CpuRead);
autoView(U_v,U,CpuWrite);
autoView(P_v,P,CpuRead);
thread_for(ss, P.Grid()->oSites(),{
for (int mu = 0; mu < Nd; mu++) {
U_v[ss](mu) = ProjectOnGroup(Exponentiate(P_v[ss](mu), ep, Nexp) * U_v[ss](mu));
}
});
//auto end = std::chrono::high_resolution_clock::now();
//auto end = std::chrono::high_resolution_clock::now();
// diff += end - start;
// std::cout << "Time to exponentiate matrix " << diff.count() << " s\n";
}

8
Grid/qcd/action/scalar/ScalarInteractionAction.h
View File

@ -89,8 +89,8 @@ public:
action = (2.0 * Ndim + mass_square) * phisquared - lambda * phisquared * phisquared;
auto p_v = p.View(CpuRead);
auto action_v = action.View(CpuWrite);
autoView( p_v , p, CpuRead);
autoView( action_v , action, CpuWrite);
for (int mu = 0; mu < Ndim; mu++)
{
// pshift = Cshift(p, mu, +1); // not efficient, implement with stencils
@ -146,8 +146,8 @@ public:
for (int point = 0; point < npoint; point++)
{
auto p_v = p.View(CpuRead);
auto force_v = force.View(CpuWrite);
autoView( p_v , p, CpuRead);
autoView( force_v , force, CpuWrite);
int permute_type;
StencilEntry *SE;

2
Grid/qcd/modules/Registration.h
View File

@ -81,7 +81,7 @@ static Registrar<OneFlavourRatioEOFModule<FermionImplementationPolicy>,
static Registrar< ConjugateGradientModule<WilsonFermionR::FermionField>,
HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("ConjugateGradient");
static Registrar< BiCGSTABModule<WilsonFermionR::FermionField>,
HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CGWFmodXMLInit("BiCGSTAB");
HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __BiCGWFmodXMLInit("BiCGSTAB");
static Registrar< ConjugateResidualModule<WilsonFermionR::FermionField>,
HMC_SolverModuleFactory<solver_string, WilsonFermionR::FermionField, Serialiser> > __CRWFmodXMLInit("ConjugateResidual");

105
Grid/qcd/utils/A2Autils.h
View File

@ -185,13 +185,14 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
for(int i=0;i<Lblock;i++){
auto lhs_v = lhs_wi[i].View(CpuRead);
// Recreate view potentially expensive outside fo UVM mode
autoView(lhs_v,lhs_wi[i],CpuRead);
auto left = conjugate(lhs_v[ss]);
for(int j=0;j<Rblock;j++){
SpinMatrix_v vv;
auto rhs_v = rhs_vj[j].View(CpuRead);
// Recreate view potentially expensive outside fo UVM mode
autoView(rhs_v,rhs_vj[j],CpuRead);
auto right = rhs_v[ss];
for(int s1=0;s1<Ns;s1++){
for(int s2=0;s2<Ns;s2++){
@ -204,11 +205,10 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
for ( int m=0;m<Nmom;m++){
int idx = m+base;
auto mom_v = mom[m].View(CpuRead);
autoView(mom_v,mom[m],CpuRead);
auto phase = mom_v[ss];
mac(&lvSum[idx],&vv,&phase);
}
}
}
}
@ -371,7 +371,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
for(int i=0;i<Lblock;i++){
auto wi_v = wi[i].View(CpuRead);
autoView(wi_v,wi[i],CpuRead);
auto w = conjugate(wi_v[ss]);
if (g5) {
w()(2)(0) = - w()(2)(0);
@ -383,7 +383,7 @@ void A2Autils<FImpl>::PionFieldXX(Eigen::Tensor<ComplexD,3> &mat,
}
for(int j=0;j<Rblock;j++){
auto vj_v=vj[j].View(CpuRead);
autoView(vj_v,vj[j],CpuRead);
auto v = vj_v[ss];
auto vv = v()(0)(0);
@ -518,12 +518,12 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
for(int i=0;i<Lblock;i++){
auto wi_v = wi[i].View(CpuRead);
autoView(wi_v,wi[i],CpuRead);
auto w = conjugate(wi_v[ss]);
for(int j=0;j<Rblock;j++){
auto vj_v = vj[j].View(CpuRead);
autoView(vj_v,vj[j],CpuRead);
auto v = vj_v[ss];
auto vv = w()(0)(0) * v()(0)(0)// Gamma5 Dirac basis explicitly written out
@ -544,7 +544,7 @@ void A2Autils<FImpl>::PionFieldWVmom(Eigen::Tensor<ComplexD,4> &mat,
int base = Nmom*i+Nmom*Lblock*j+Nmom*Lblock*Rblock*r;
for ( int m=0;m<Nmom;m++){
int idx = m+base;
auto mom_v = mom[m].View(CpuRead);
autoView(mom_v,mom[m],CpuRead);
auto phase = mom_v[ss];
mac(&lvSum[idx],&vv,&phase()()());
}
@ -730,13 +730,13 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
for(int i=0;i<Lblock;i++)
{
auto wi_v = lhs_wi[i].View(CpuRead);
autoView(wi_v,lhs_wi[i],CpuRead);
auto left = conjugate(wi_v[ss]);
for(int j=0;j<Rblock;j++)
{
SpinMatrix_v vv;
auto vj_v = rhs_vj[j].View(CpuRead);
autoView(vj_v,rhs_vj[j],CpuRead);
auto right = vj_v[ss];
for(int s1=0;s1<Ns;s1++)
@ -752,8 +752,8 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
for ( int m=0;m<Nem;m++)
{
auto emB0_v = emB0[m].View(CpuRead);
auto emB1_v = emB1[m].View(CpuRead);
autoView(emB0_v,emB0[m],CpuRead);
autoView(emB1_v,emB1[m],CpuRead);
int idx = m+base;
auto b0 = emB0_v[ss];
auto b1 = emB1_v[ss];
@ -1014,21 +1014,21 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
for(int d_o=0;d_o<N_d;d_o+=d_unroll){
for(int t=0;t<N_t;t++){
for(int s=0;s<N_s;s++){
auto vs_v = vs[s].View(CpuRead);
auto tmp1 = vs_v[ss];
vobj tmp2 = Zero();
vobj tmp3 = Zero();
for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
auto vd_v = vd[d].View(CpuRead);
Scalar_v coeff = WW_sd(t,s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
}
autoView(vs_v,vs[s],CpuRead);
auto tmp1 = vs_v[ss];
vobj tmp2 = Zero();
vobj tmp3 = Zero();
for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
autoView(vd_v,vd[d],CpuRead);
Scalar_v coeff = WW_sd(t,s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
}
//////////////////////////
// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
//////////////////////////
OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
//////////////////////////
// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
//////////////////////////
OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
}}
}
@ -1067,21 +1067,20 @@ A2Autils<FImpl>::ContractWWVV(std::vector<PropagatorField> &WWVV,
thread_for(ss,grid->oSites(),{
for(int d_o=0;d_o<N_d;d_o+=d_unroll){
for(int s=0;s<N_s;s++){
auto vs_v = vs[s].View(CpuRead);
auto tmp1 = vs_v[ss];
vobj tmp2 = Zero();
vobj tmp3 = Zero();
for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
auto vd_v = vd[d].View(CpuRead);
Scalar_v coeff = buf(s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
}
//////////////////////////
// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
//////////////////////////
OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
autoView(vs_v,vs[s],CpuRead);
auto tmp1 = vs_v[ss];
vobj tmp2 = Zero();
vobj tmp3 = Zero();
for(int d=d_o;d<MIN(d_o+d_unroll,N_d);d++){
autoView(vd_v,vd[d],CpuRead);
Scalar_v coeff = buf(s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
}
//////////////////////////
// Fast outer product of tmp1 with a sum of terms suppressed by d_unroll
//////////////////////////
OuterProductWWVV(WWVV[t], tmp1, tmp2, Ns, ss);
}}
});
}
@ -1093,7 +1092,7 @@ inline void A2Autils<FImpl>::OuterProductWWVV(PropagatorField &WWVV,
const vobj &rhs,
const int Ns, const int ss)
{
auto WWVV_v = WWVV.View(CpuWrite);
autoView(WWVV_v,WWVV,CpuWrite);
for (int s1 = 0; s1 < Ns; s1++){
for (int s2 = 0; s2 < Ns; s2++){
WWVV_v[ss]()(s1,s2)(0, 0) += lhs()(s1)(0) * rhs()(s2)(0);
@ -1122,10 +1121,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourDiagonal(const PropagatorField &WWV
GridBase *grid = WWVV0.Grid();
auto WWVV0_v = WWVV0.View(CpuRead);
auto WWVV1_v = WWVV1.View(CpuRead);
auto O_trtr_v= O_trtr.View(CpuWrite);
auto O_fig8_v= O_fig8.View(CpuWrite);
autoView(WWVV0_v , WWVV0,CpuRead);
autoView(WWVV1_v , WWVV1,CpuRead);
autoView(O_trtr_v, O_trtr,CpuWrite);
autoView(O_fig8_v, O_fig8,CpuWrite);
thread_for(ss,grid->oSites(),{
typedef typename ComplexField::vector_object vobj;
@ -1166,10 +1165,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourMix(const PropagatorField &WWVV0,
GridBase *grid = WWVV0.Grid();
auto WWVV0_v = WWVV0.View(CpuRead);
auto WWVV1_v = WWVV1.View(CpuRead);
auto O_trtr_v= O_trtr.View(CpuWrite);
auto O_fig8_v= O_fig8.View(CpuWrite);
autoView( WWVV0_v , WWVV0,CpuRead);
autoView( WWVV1_v , WWVV1,CpuRead);
autoView( O_trtr_v, O_trtr,CpuWrite);
autoView( O_fig8_v, O_fig8,CpuWrite);
thread_for(ss,grid->oSites(),{

27
Grid/qcd/utils/BaryonUtils.h
View File

@ -273,10 +273,10 @@ void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
for (int ie=0; ie < 6 ; ie++)
wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
auto vbaryon_corr= baryon_corr.View(CpuWrite);
auto v1 = q1_left.View(CpuRead);
auto v2 = q2_left.View(CpuRead);
auto v3 = q3_left.View(CpuRead);
autoView(vbaryon_corr, baryon_corr,CpuWrite);
autoView( v1 , q1_left, CpuRead);
autoView( v2 , q2_left, CpuRead);
autoView( v3 , q3_left, CpuRead);
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{
@ -560,10 +560,10 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_Eye(const PropagatorField &qq_loop,
{
GridBase *grid = qs_ti.Grid();
auto vcorr= stn_corr.View(CpuWrite);
auto vq_loop = qq_loop.View(CpuRead);
auto vd_tf = qd_tf.View(CpuRead);
auto vs_ti = qs_ti.View(CpuRead);
autoView( vcorr, stn_corr, CpuWrite);
autoView( vq_loop , qq_loop, CpuRead);
autoView( vd_tf , qd_tf, CpuRead);
autoView( vs_ti , qs_ti, CpuRead);
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{
@ -597,12 +597,11 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_NonEye(const PropagatorField &qq_ti,
{
GridBase *grid = qs_ti.Grid();
auto vcorr= stn_corr.View(CpuWrite);
auto vq_ti = qq_ti.View(CpuRead);
auto vq_tf = qq_tf.View(CpuRead);
auto vd_tf = qd_tf.View(CpuRead);
auto vs_ti = qs_ti.View(CpuRead);
autoView( vcorr , stn_corr, CpuWrite);
autoView( vq_ti , qq_ti, CpuRead);
autoView( vq_tf , qq_tf, CpuRead);
autoView( vd_tf , qd_tf, CpuRead);
autoView( vs_ti , qs_ti, CpuRead);
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{
auto Dq_ti = vq_ti[ss];

48
Grid/qcd/utils/LinalgUtils.h
View File

@ -47,8 +47,8 @@ void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,Coeff a,Coeff b)
GridBase *grid=x.Grid();
Gamma G5(Gamma::Algebra::Gamma5);
auto x_v = x.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView(x_v, x, AcceleratorRead);
autoView(z_v, z, AcceleratorWrite);
accelerator_for( ss, x_v.size(),vobj::Nsimd(), {
auto tmp = a*x_v(ss) + G5*(b*timesI(x_v(ss)));
coalescedWrite(z_v[ss],tmp);
@ -63,9 +63,9 @@ void axpby_ssp(Lattice<vobj> &z, Coeff a,const Lattice<vobj> &x,Coeff b,const La
conformable(x,z);
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
// FIXME -- need a new class of accelerator_loop to implement this
//
uint64_t nloop = grid->oSites()/Ls;
@ -85,9 +85,9 @@ void ag5xpby_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Algebra::Gamma5);
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
uint64_t ss = sss*Ls;
@ -104,9 +104,9 @@ void axpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const L
conformable(x,z);
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
Gamma G5(Gamma::Algebra::Gamma5);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
@ -125,9 +125,9 @@ void ag5xpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
Gamma G5(Gamma::Algebra::Gamma5);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
@ -147,9 +147,9 @@ void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,co
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
uint64_t ss = sss*Ls;
@ -168,9 +168,9 @@ void axpby_ssp_pplus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,con
conformable(x,z);
GridBase *grid=x.Grid();
int Ls = grid->_rdimensions[0];
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( y_v, y, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
uint64_t ss = sss*Ls;
@ -189,8 +189,8 @@ void G5R5(Lattice<vobj> &z,const Lattice<vobj> &x)
conformable(x,z);
int Ls = grid->_rdimensions[0];
Gamma G5(Gamma::Algebra::Gamma5);
auto x_v = x.View(AcceleratorRead);
auto z_v = z.View(AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
uint64_t nloop = grid->oSites()/Ls;
accelerator_for(sss,nloop,vobj::Nsimd(),{
uint64_t ss = sss*Ls;
@ -222,8 +222,8 @@ void G5C(Lattice<iVector<CComplex, nbasis>> &z, const Lattice<iVector<CComplex,
static_assert(nbasis % 2 == 0, "");
int nb = nbasis / 2;
auto z_v = z.View(AcceleratorWrite);
auto x_v = x.View(AcceleratorRead);
autoView( z_v, z, AcceleratorWrite);
autoView( x_v, x, AcceleratorRead);
accelerator_for(ss,grid->oSites(),CComplex::Nsimd(),
{
for(int n = 0; n < nb; ++n) {

20
Grid/qcd/utils/SUn.h
View File

@ -222,10 +222,10 @@ public:
conformable(subgroup, Determinant);
int i0, i1;
su2SubGroupIndex(i0, i1, su2_index);
auto subgroup_v = subgroup.View(CpuWrite);
auto source_v = source.View(CpuRead);
auto Determinant_v = Determinant.View(CpuWrite);
autoView( subgroup_v , subgroup,CpuWrite);
autoView( source_v , source,CpuRead);
autoView( Determinant_v , Determinant,CpuWrite);
thread_for(ss, grid->oSites(), {
subgroup_v[ss]()()(0, 0) = source_v[ss]()()(i0, i0);
@ -257,8 +257,8 @@ public:
su2SubGroupIndex(i0, i1, su2_index);
dest = 1.0; // start out with identity
auto dest_v = dest.View(CpuWrite);
auto subgroup_v = subgroup.View(CpuRead);
autoView( dest_v , dest, CpuWrite);
autoView( subgroup_v, subgroup, CpuRead);
thread_for(ss, grid->oSites(),
{
dest_v[ss]()()(i0, i0) = subgroup_v[ss]()()(0, 0);
@ -266,6 +266,7 @@ public:
dest_v[ss]()()(i1, i0) = subgroup_v[ss]()()(1, 0);
dest_v[ss]()()(i1, i1) = subgroup_v[ss]()()(1, 1);
});
}
///////////////////////////////////////////////
@ -608,8 +609,8 @@ public:
// reunitarise??
template <typename LatticeMatrixType>
static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out,
double scale = 1.0) {
static void LieRandomize(GridParallelRNG &pRNG, LatticeMatrixType &out, double scale = 1.0)
{
GridBase *grid = out.Grid();
typedef typename LatticeMatrixType::vector_type vector_type;
@ -618,8 +619,7 @@ public:
typedef iSinglet<vector_type> vTComplexType;
typedef Lattice<vTComplexType> LatticeComplexType;
typedef typename GridTypeMapper<
typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
typedef typename GridTypeMapper<typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
LatticeComplexType ca(grid);
LatticeMatrixType lie(grid);
@ -629,6 +629,7 @@ public:
MatrixType ta;
lie = Zero();
for (int a = 0; a < AdjointDimension; a++) {
random(pRNG, ca);
@ -640,6 +641,7 @@ public:
la = ci * ca * ta;
lie = lie + la; // e^{i la ta}
}
taExp(lie, out);
}