portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-25 19:22:03 +01:00
Code Releases Activity

Offload more loops

Browse Source
This commit is contained in:
Peter Boyle
2020-06-10 12:57:55 -04:00
parent e97f3688db
commit 8720aecb80
3 changed files with 29 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
View File

@ -467,32 +467,7 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
conformable(_grid, q_in_1.Grid());
conformable(_grid, q_in_2.Grid());
conformable(_grid, q_out.Grid());
#if 0
PropagatorField tmp1(_grid), tmp2(_grid);
q_out = Zero();
// Forward, need q1(x + mu), q2(x). Backward, need q1(x), q2(x + mu).
// Inefficient comms method but not performance critical.
tmp1 = Cshift(q_in_1, mu, 1);
tmp2 = Cshift(q_in_2, mu, 1);
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],
q_out_v[sU],
Umu_v, sU, mu);
Kernels::ContractConservedCurrentSiteBwd(q_in_1_v[sU],
tmp2_v[sU],
q_out_v[sU],
Umu_v, sU, mu);
});
#else
#endif
assert(0);
}
@ -508,62 +483,7 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
{
conformable(_grid, q_in.Grid());
conformable(_grid, q_out.Grid());
#if 0
// Lattice<iSinglet<Simd>> ph(_grid), coor(_grid);
Complex i(0.0,1.0);
PropagatorField tmpFwd(_grid), tmpBwd(_grid), tmp(_grid);
unsigned int tshift = (mu == Tp) ? 1 : 0;
unsigned int LLt = GridDefaultLatt()[Tp];
q_out = Zero();
LatticeInteger coords(_grid);
LatticeCoordinate(coords, Tp);
// Need q(x + mu) and q(x - mu).
tmp = Cshift(q_in, mu, 1);
tmpFwd = tmp*lattice_cmplx;
tmp = lattice_cmplx*q_in;
tmpBwd = Cshift(tmp, mu, -1);
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(), {
// Compute the sequential conserved current insertion only if our simd
// object contains a timeslice we need.
vPredicate t_mask;
t_mask() = ((coords_v[sU] >= tmin) && (coords_v[sU] <= tmax));
Integer timeSlices = Reduce(t_mask());
if (timeSlices > 0) {
Kernels::SeqConservedCurrentSiteFwd(tmpFwd_v[sU],
q_out_v[sU],
Umu_v, sU, mu, t_mask);
}
// Repeat for backward direction.
t_mask() = ((coords_v[sU] >= (tmin + tshift)) &&
(coords_v[sU] <= (tmax + tshift)));
//if tmax = LLt-1 (last timeslice) include timeslice 0 if the time is shifted (mu=3)
unsigned int t0 = 0;
if((tmax==LLt-1) && (tshift==1)) t_mask() = (t_mask() || (coords_v[sU] == t0 ));
timeSlices = Reduce(t_mask());
if (timeSlices > 0) {
Kernels::SeqConservedCurrentSiteBwd(tmpBwd_v[sU],
q_out_v[sU],
Umu_v, sU, mu, t_mask);
}
});
#else
#endif
assert(0);
}
NAMESPACE_END(Grid);