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

Merge branch 'develop' into feature/staggered-comms-compute

Browse Source 
Conflicts:
	lib/qcd/action/fermion/ImprovedStaggeredFermion.cc
This commit is contained in:
Azusa Yamaguchi
2018-05-21 13:07:29 +01:00
parent 0e127b1fc7 a61e0df54b
commit 7fbbb31a50
238 changed files with 18155 additions and 7613 deletions
Download Patch File Download Diff File Expand all files Collapse all files

148
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -47,7 +47,8 @@ int WilsonFermionStatic::HandOptDslash;
template <class Impl>
WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
GridRedBlackCartesian &Hgrid, RealD _mass,
const ImplParams &p)
const ImplParams &p,
const WilsonAnisotropyCoefficients &anis)
: Kernels(p),
_grid(&Fgrid),
_cbgrid(&Hgrid),
@ -60,16 +61,41 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
Umu(&Fgrid),
UmuEven(&Hgrid),
UmuOdd(&Hgrid),
_tmp(&Hgrid)
_tmp(&Hgrid),
anisotropyCoeff(anis)
{
// Allocate the required comms buffer
ImportGauge(_Umu);
if (anisotropyCoeff.isAnisotropic){
diag_mass = mass + 1.0 + (Nd-1)*(anisotropyCoeff.nu / anisotropyCoeff.xi_0);
} else {
diag_mass = 4.0 + mass;
}
}
template <class Impl>
void WilsonFermion<Impl>::ImportGauge(const GaugeField &_Umu) {
GaugeField HUmu(_Umu._grid);
HUmu = _Umu * (-0.5);
//Here multiply the anisotropy coefficients
if (anisotropyCoeff.isAnisotropic)
{
for (int mu = 0; mu < Nd; mu++)
{
GaugeLinkField U_dir = (-0.5)*PeekIndex<LorentzIndex>(_Umu, mu);
if (mu != anisotropyCoeff.t_direction)
U_dir *= (anisotropyCoeff.nu / anisotropyCoeff.xi_0);
PokeIndex<LorentzIndex>(HUmu, U_dir, mu);
}
}
else
{
HUmu = _Umu * (-0.5);
}
Impl::DoubleStore(GaugeGrid(), Umu, HUmu);
pickCheckerboard(Even, UmuEven, Umu);
pickCheckerboard(Odd, UmuOdd, Umu);
@ -83,14 +109,14 @@ template <class Impl>
RealD WilsonFermion<Impl>::M(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Dhop(in, out, DaggerNo);
return axpy_norm(out, 4 + mass, in, out);
return axpy_norm(out, diag_mass, in, out);
}
template <class Impl>
RealD WilsonFermion<Impl>::Mdag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Dhop(in, out, DaggerYes);
return axpy_norm(out, 4 + mass, in, out);
return axpy_norm(out, diag_mass, in, out);
}
template <class Impl>
@ -114,7 +140,7 @@ void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
template <class Impl>
void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
typename FermionField::scalar_type scal(4.0 + mass);
typename FermionField::scalar_type scal(diag_mass);
out = scal * in;
}
@ -127,7 +153,7 @@ void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
template<class Impl>
void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
out = (1.0/(4.0+mass))*in;
out = (1.0/(diag_mass))*in;
}
template<class Impl>
@ -204,7 +230,7 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
FermionField Btilde(B._grid);
FermionField Atilde(B._grid);
Atilde = A;
Atilde = A;//redundant
st.HaloExchange(B, compressor);
@ -429,6 +455,112 @@ void WilsonFermion<Impl>::DhopInternalSerial(StencilImpl &st, LebesgueOrder &lo,
};
/*Change ends */
/*******************************************************************************
* Conserved current utilities for Wilson fermions, for contracting propagators
* to make a conserved current sink or inserting the conserved current
* sequentially.
******************************************************************************/
template <class Impl>
void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
PropagatorField &q_in_2,
PropagatorField &q_out,
Current curr_type,
unsigned int mu)
{
Gamma g5(Gamma::Algebra::Gamma5);
conformable(_grid, q_in_1._grid);
conformable(_grid, q_in_2._grid);
conformable(_grid, q_out._grid);
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);
parallel_for (unsigned int sU = 0; sU < Umu._grid->oSites(); ++sU)
{
Kernels::ContractConservedCurrentSiteFwd(tmp1._odata[sU],
q_in_2._odata[sU],
q_out._odata[sU],
Umu, sU, mu);
Kernels::ContractConservedCurrentSiteBwd(q_in_1._odata[sU],
tmp2._odata[sU],
q_out._odata[sU],
Umu, sU, mu);
}
}
template <class Impl>
void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
PropagatorField &q_out,
Current curr_type,
unsigned int mu,
std::vector<Real> mom,
unsigned int tmin,
unsigned int tmax)
{
conformable(_grid, q_in._grid);
conformable(_grid, q_out._grid);
Lattice<iSinglet<Simd>> ph(_grid), coor(_grid);
ComplexD i(0.0,1.0);
PropagatorField tmpFwd(_grid), tmpBwd(_grid), tmp(_grid);
unsigned int tshift = (mu == Tp) ? 1 : 0;
unsigned int LLt = GridDefaultLatt()[Tp];
// Momentum projection
ph = zero;
for(unsigned int mu = 0; mu < Nd - 1; mu++)
{
LatticeCoordinate(coor, mu);
ph = ph + mom[mu]*coor*((1./(_grid->_fdimensions[mu])));
}
ph = exp((RealD)(2*M_PI)*i*ph);
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*ph;
tmp = ph*q_in;
tmpBwd = Cshift(tmp, mu, -1);
parallel_for (unsigned int sU = 0; sU < Umu._grid->oSites(); ++sU)
{
// Compute the sequential conserved current insertion only if our simd
// object contains a timeslice we need.
vInteger t_mask = ((coords._odata[sU] >= tmin) &&
(coords._odata[sU] <= tmax));
Integer timeSlices = Reduce(t_mask);
if (timeSlices > 0)
{
Kernels::SeqConservedCurrentSiteFwd(tmpFwd._odata[sU],
q_out._odata[sU],
Umu, sU, mu, t_mask);
}
// Repeat for backward direction.
t_mask = ((coords._odata[sU] >= (tmin + tshift)) &&
(coords._odata[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._odata[sU] == t0 ));
timeSlices = Reduce(t_mask);
if (timeSlices > 0)
{
Kernels::SeqConservedCurrentSiteBwd(tmpBwd._odata[sU],
q_out._odata[sU],
Umu, sU, mu, t_mask);
}
}
}
FermOpTemplateInstantiate(WilsonFermion);
AdjointFermOpTemplateInstantiate(WilsonFermion);
TwoIndexFermOpTemplateInstantiate(WilsonFermion);