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

Make view specify where and drive data motion - first cut.

Browse Source 
This is a compile tiime option --enable-unified=yes/no
This commit is contained in:
Peter Boyle
2020-05-21 16:13:16 -04:00
parent ebb60330c9
commit 7860a50f70
48 changed files with 688 additions and 718 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Grid/qcd/action/fermion/GparityWilsonImpl.h
View File

@ -233,10 +233,10 @@ public:
Uconj = where(coor==neglink,-Uconj,Uconj);
}
auto U_v = U.View();
auto Uds_v = Uds.View();
auto Uconj_v = Uconj.View();
auto Utmp_v= Utmp.View();
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]();
@ -272,8 +272,8 @@ public:
GaugeLinkField link(mat.Grid());
// use lorentz for flavour as hack.
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
auto link_v = link.View();
auto tmp_v = tmp.View();
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));
});
@ -306,9 +306,9 @@ public:
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
auto tmp_v = tmp.View();
auto Atilde_v = Atilde.View();
auto Btilde_v = Btilde.View();
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;

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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(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();
auto F_v = F.View();
auto T_v = T.View(CpuWrite);
auto F_v = F.View(CpuRead);
thread_for(i, CloverTerm.Grid()->oSites(),
{
T_v[i]()(0, 0) = timesI(F_v[i]()());

12
Grid/qcd/action/fermion/WilsonImpl.h
View File

@ -106,9 +106,9 @@ public:
const _SpinorField & phi,
int mu)
{
auto out_v= out.View();
auto phi_v= phi.View();
auto Umu_v= Umu.View();
auto out_v= out.View(CpuWrite);
auto phi_v= phi.View(CpuRead);
auto Umu_v= Umu.View(CpuRead);
thread_for(sss,out.Grid()->oSites(),{
multLink(out_v[sss],Umu_v[sss],phi_v[sss],mu);
});
@ -191,9 +191,9 @@ public:
int Ls=Btilde.Grid()->_fdimensions[0];
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
auto tmp_v = tmp.View();
auto Btilde_v = Btilde.View();
auto Atilde_v = Atilde.View();
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++){

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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(CpuRead);
auto phi = phi_i.View(CpuRead);
auto chi = chi_i.View(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();
auto phi=phi_i.View();
auto chi=chi_i.View();
auto psi=psi_i.View(CpuRead);
auto phi=phi_i.View(CpuRead);
auto chi=chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(CpuWrite);
SiteHalfSpinor BcastP;
SiteHalfSpinor BcastM;
@ -586,8 +586,8 @@ CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi_i, FermionField
}
#else
{
auto psi = psi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(CpuRead);
auto chi = chi_i.View(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();
auto psi = psi_i.View();
auto chi = chi_i.View();
auto phi = phi_i.View(AcceleratorRead);
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi=chi_i.View();
auto psi=psi_i.View(AcceleratorRead);
auto chi=chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
int Ls = this->Ls;
auto plee = & this->lee[0];

32
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();
auto UUUmu_v = UUUmu.View();
auto in_v = in.View();
auto out_v = out.View();
auto Umu_v = Umu.View(CpuRead);
auto UUUmu_v = UUUmu.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
thread_for( ss,Umu.Grid()->oSites(),{
for(int s=0;s<Ls;s++){
int sU=ss;
@ -339,10 +339,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl &
}
// do the compute
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
if (dag == DaggerYes) {
for (int ss = myblock; ss < myblock+myn; ++ss) {
@ -376,10 +376,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl &
DhopComputeTime2-=usecond();
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
if (dag == DaggerYes) {
int sz=st.surface_list.size();
thread_for( ss,sz,{
@ -418,10 +418,10 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st,
DhopComputeTime -= usecond();
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
if (dag == DaggerYes) {
thread_for( ss,U.Grid()->oSites(),{
int sU=ss;

40
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();
auto UUU_v = UUU.View();
auto B_v = B.View();
auto Btilde_v = Btilde.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto B_v = B.View(CpuWrite);
auto Btilde_v = Btilde.View(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();
auto UUUmu_v = UUUmu.View();
auto in_v = in.View();
auto out_v = out.View();
auto Umu_v = Umu.View(CpuRead);
auto UUUmu_v = UUUmu.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
thread_for( sss, in.Grid()->oSites(),{
Kernels::DhopDirKernel(Stencil, Umu_v, UUUmu_v, Stencil.CommBuf(), sss, sss, in_v, out_v, dir, disp);
});
@ -449,10 +449,10 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
}
// do the compute
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
if (dag == DaggerYes) {
for (int ss = myblock; ss < myblock+myn; ++ss) {
int sU = ss;
@ -479,10 +479,10 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalOverlappedComms(StencilImpl &st
DhopComputeTime2 -= usecond();
{
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
if (dag == DaggerYes) {
int sz=st.surface_list.size();
thread_for(ss,sz,{
@ -520,10 +520,10 @@ void ImprovedStaggeredFermion<Impl>::DhopInternalSerialComms(StencilImpl &st, Le
st.HaloExchange(in, compressor);
DhopCommTime += usecond();
auto U_v = U.View();
auto UUU_v = UUU.View();
auto in_v = in.View();
auto out_v = out.View();
auto U_v = U.View(CpuRead);
auto UUU_v = UUU.View(CpuRead);
auto in_v = in.View(CpuRead);
auto out_v = out.View(CpuWrite);
DhopComputeTime -= usecond();
if (dag == DaggerYes) {
thread_for(sss, in.Grid()->oSites(),{

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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto phi = phi_i.View();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto phi = phi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(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();
auto chi = chi_i.View();
auto psi = psi_i.View(AcceleratorRead);
auto chi = chi_i.View(AcceleratorWrite);
int Ls = this->Ls;
auto pm = this->pm;

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();
auto tmp2_v = tmp2.View();
auto q_in_1_v=q_in_1.View();
auto q_in_2_v=q_in_2.View();
auto q_out_v = q_out.View();
auto Umu_v = Umu.View();
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);
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();
auto tmpFwd_v = tmpFwd.View();
auto tmpBwd_v = tmpBwd.View();
auto Umu_v = Umu.View();
auto q_out_v = q_out.View();
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);
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();
auto in_v = in.View();
auto st_v = st.View();
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto st_v = st.View(AcceleratorRead);
auto out_Xm = out[0].View();
auto out_Ym = out[1].View();
auto out_Zm = out[2].View();
auto out_Tm = out[3].View();
auto out_Xp = out[4].View();
auto out_Yp = out[5].View();
auto out_Zp = out[6].View();
auto out_Tp = out[7].View();
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);
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();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(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();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(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();
auto in_v = in.View();
auto out_v = out.View();
auto st_v = st.View();
auto U_v = U.View(AcceleratorRead);
auto in_v = in.View(AcceleratorRead);
auto out_v = out.View(AcceleratorWrite);
auto st_v = st.View(AcceleratorRead);
if( interior && exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDag); return;}

8
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();
auto W_v = W.View();
auto U_v = U.View(CpuWrite);
auto W_v = W.View(CpuRead);
thread_for( ss, U.Grid()->oSites(), {
U_v[ss](mu) = U_v[ss](mu) + W_v[ss]();
});
@ -131,8 +131,8 @@ public:
//static std::chrono::duration<double> diff;
//auto start = std::chrono::high_resolution_clock::now();
auto U_v = U.View();
auto P_v = P.View();
auto U_v = U.View(CpuWrite);
auto P_v = P.View(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));

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();
auto action_v = action.View();
auto p_v = p.View(CpuRead);
auto action_v = action.View(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();
auto force_v = force.View();
auto p_v = p.View(CpuRead);
auto force_v = force.View(CpuWrite);
int permute_type;
StencilEntry *SE;

20
Grid/qcd/smearing/GaugeConfiguration.h
View File

@ -49,7 +49,7 @@ public:
private:
const unsigned int smearingLevels;
Smear_Stout<Gimpl> StoutSmearing;
Smear_Stout<Gimpl> *StoutSmearing;
std::vector<GaugeField> SmearedSet;
// Member functions
@ -72,7 +72,7 @@ private:
previous_u = *ThinLinks;
for (int smearLvl = 0; smearLvl < smearingLevels; ++smearLvl)
{
StoutSmearing.smear(SmearedSet[smearLvl], previous_u);
StoutSmearing->smear(SmearedSet[smearLvl], previous_u);
previous_u = SmearedSet[smearLvl];
// For debug purposes
@ -93,7 +93,7 @@ private:
GaugeLinkField SigmaKPrime_mu(grid);
GaugeLinkField GaugeKmu(grid), Cmu(grid);
StoutSmearing.BaseSmear(C, GaugeK);
StoutSmearing->BaseSmear(C, GaugeK);
SigmaK = Zero();
iLambda = Zero();
@ -107,7 +107,7 @@ private:
pokeLorentz(SigmaK, SigmaKPrime_mu * e_iQ + adj(Cmu) * iLambda_mu, mu);
pokeLorentz(iLambda, iLambda_mu, mu);
}
StoutSmearing.derivative(SigmaK, iLambda,
StoutSmearing->derivative(SigmaK, iLambda,
GaugeK); // derivative of SmearBase
return SigmaK;
}
@ -144,14 +144,14 @@ private:
// Exponential
iQ2 = iQ * iQ;
iQ3 = iQ * iQ2;
StoutSmearing.set_uw(u, w, iQ2, iQ3);
StoutSmearing.set_fj(f0, f1, f2, u, w);
StoutSmearing->set_uw(u, w, iQ2, iQ3);
StoutSmearing->set_fj(f0, f1, f2, u, w);
e_iQ = f0 * unity + timesMinusI(f1) * iQ - f2 * iQ2;
// Getting B1, B2, Gamma and Lambda
// simplify this part, reduntant calculations in set_fj
xi0 = StoutSmearing.func_xi0(w);
xi1 = StoutSmearing.func_xi1(w);
xi0 = StoutSmearing->func_xi0(w);
xi1 = StoutSmearing->func_xi1(w);
u2 = u * u;
w2 = w * w;
cosw = cos(w);
@ -219,7 +219,7 @@ public:
/* Standard constructor */
SmearedConfiguration(GridCartesian* UGrid, unsigned int Nsmear,
Smear_Stout<Gimpl>& Stout)
: smearingLevels(Nsmear), StoutSmearing(Stout), ThinLinks(NULL)
: smearingLevels(Nsmear), StoutSmearing(&Stout), ThinLinks(NULL)
{
for (unsigned int i = 0; i < smearingLevels; ++i)
SmearedSet.push_back(*(new GaugeField(UGrid)));
@ -227,7 +227,7 @@ public:
/*! For just thin links */
SmearedConfiguration()
: smearingLevels(0), StoutSmearing(), SmearedSet(), ThinLinks(NULL) {}
: smearingLevels(0), StoutSmearing(nullptr), SmearedSet(), ThinLinks(NULL) {}
// attach the smeared routines to the thin links U and fill the smeared set
void set_Field(GaugeField &U)

50
Grid/qcd/utils/A2Autils.h
View File

@ -185,13 +185,13 @@ void A2Autils<FImpl>::MesonField(TensorType &mat,
for(int i=0;i<Lblock;i++){
auto lhs_v = lhs_wi[i].View();
auto lhs_v = lhs_wi[i].View(CpuRead);
auto left = conjugate(lhs_v[ss]);
for(int j=0;j<Rblock;j++){
SpinMatrix_v vv;
auto rhs_v = rhs_vj[j].View();
auto rhs_v = rhs_vj[j].View(CpuRead);
auto right = rhs_v[ss];
for(int s1=0;s1<Ns;s1++){
for(int s2=0;s2<Ns;s2++){
@ -204,7 +204,7 @@ 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();
auto mom_v = mom[m].View(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();
auto wi_v = wi[i].View(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();
auto vj_v=vj[j].View(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();
auto wi_v = wi[i].View(CpuRead);
auto w = conjugate(wi_v[ss]);
for(int j=0;j<Rblock;j++){
auto vj_v = vj[j].View();
auto vj_v = vj[j].View(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();
auto mom_v = mom[m].View(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();
auto wi_v = lhs_wi[i].View(CpuRead);
auto left = conjugate(wi_v[ss]);
for(int j=0;j<Rblock;j++)
{
SpinMatrix_v vv;
auto vj_v = rhs_vj[j].View();
auto vj_v = rhs_vj[j].View(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();
auto emB1_v = emB1[m].View();
auto emB0_v = emB0[m].View(CpuRead);
auto emB1_v = emB1[m].View(CpuRead);
int idx = m+base;
auto b0 = emB0_v[ss];
auto b1 = emB1_v[ss];
@ -1014,12 +1014,12 @@ 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();
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();
auto vd_v = vd[d].View(CpuRead);
Scalar_v coeff = WW_sd(t,s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
@ -1067,12 +1067,12 @@ 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();
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();
auto vd_v = vd[d].View(CpuRead);
Scalar_v coeff = buf(s,d);
tmp3 = conjugate(vd_v[ss]);
mac(&tmp2, &coeff, &tmp3);
@ -1093,7 +1093,7 @@ inline void A2Autils<FImpl>::OuterProductWWVV(PropagatorField &WWVV,
const vobj &rhs,
const int Ns, const int ss)
{
auto WWVV_v = WWVV.View();
auto WWVV_v = WWVV.View(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 +1122,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourDiagonal(const PropagatorField &WWV
GridBase *grid = WWVV0.Grid();
auto WWVV0_v = WWVV0.View();
auto WWVV1_v = WWVV1.View();
auto O_trtr_v= O_trtr.View();
auto O_fig8_v= O_fig8.View();
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);
thread_for(ss,grid->oSites(),{
typedef typename ComplexField::vector_object vobj;
@ -1166,10 +1166,10 @@ void A2Autils<FImpl>::ContractFourQuarkColourMix(const PropagatorField &WWVV0,
GridBase *grid = WWVV0.Grid();
auto WWVV0_v = WWVV0.View();
auto WWVV1_v = WWVV1.View();
auto O_trtr_v= O_trtr.View();
auto O_fig8_v= O_fig8.View();
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);
thread_for(ss,grid->oSites(),{

26
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();
auto v1 = q1_left.View();
auto v2 = q2_left.View();
auto v3 = q3_left.View();
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);
// 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();
auto vq_loop = qq_loop.View();
auto vd_tf = qd_tf.View();
auto vs_ti = qs_ti.View();
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);
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{
@ -597,11 +597,11 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_NonEye(const PropagatorField &qq_ti,
{
GridBase *grid = qs_ti.Grid();
auto vcorr= stn_corr.View();
auto vq_ti = qq_ti.View();
auto vq_tf = qq_tf.View();
auto vd_tf = qd_tf.View();
auto vs_ti = qs_ti.View();
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);
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{

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();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto y_v = y.View();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto y_v = y.View(AcceleratorRead);
auto z_v = z.View(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();
auto z_v = z.View();
auto x_v = x.View(AcceleratorRead);
auto z_v = z.View(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();
auto x_v = x.View();
auto z_v = z.View(AcceleratorWrite);
auto x_v = x.View(AcceleratorRead);
accelerator_for(ss,grid->oSites(),CComplex::Nsimd(),
{
for(int n = 0; n < nb; ++n) {

10
Grid/qcd/utils/SUn.h
View File

@ -222,9 +222,9 @@ public:
conformable(subgroup, Determinant);
int i0, i1;
su2SubGroupIndex(i0, i1, su2_index);
auto subgroup_v = subgroup.View();
auto source_v = source.View();
auto Determinant_v = Determinant.View();
auto subgroup_v = subgroup.View(CpuWrite);
auto source_v = source.View(CpuRead);
auto Determinant_v = Determinant.View(CpuWrite);
thread_for(ss, grid->oSites(), {
@ -257,8 +257,8 @@ public:
su2SubGroupIndex(i0, i1, su2_index);
dest = 1.0; // start out with identity
auto dest_v = dest.View();
auto subgroup_v = subgroup.View();
auto dest_v = dest.View(CpuWrite);
auto subgroup_v = subgroup.View(CpuRead);
thread_for(ss, grid->oSites(),
{
dest_v[ss]()()(i0, i0) = subgroup_v[ss]()()(0, 0);