portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 07:47:06 +01:00
Code Releases Activity

Adding files for multiple implementations (cache opt) and Ls vectorisation

Browse Source 
of the 5D cayley form chiral fermions for the 5d matrix. With Ls entirely
in the vector direction, s-hopping terms involve rotations.

The serial dependence of the LDU inversion for Mobius and 4d even odd
checkerboarding is removed by simply applying Ls^2 operations (vectorised
many ways) as a dense matrix operation.

This should give similar throughput but high flops (non-compulsory flops)
but enable use of the KNL cache friendly kernels throughout the code.

Ls is still constrained to be a multiple of Nsimd, which is as much as 8 for AVX512
with single precision.
This commit is contained in:
paboyle
2016-07-14 22:59:21 +01:00
parent 62601bb649
commit adbc7c1188
6 changed files with 1395 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

206
lib/qcd/action/fermion/CayleyFermion5Dcache.cc Normal file
View File

@ -0,0 +1,206 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/CayleyFermion5D.cc
Copyright (C) 2015
Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
namespace Grid {
namespace QCD {
// FIXME -- make a version of these routines with site loop outermost for cache reuse.
// Pminus fowards
// Pplus backwards..
template<class Impl>
void CayleyFermion5D<Impl>::M5D(const FermionField &psi,
const FermionField &phi,
FermionField &chi,
std::vector<RealD> &lower,
std::vector<RealD> &diag,
std::vector<RealD> &upper)
{
int Ls =this->Ls;
GridBase *grid=psi._grid;
assert(phi.checkerboard == psi.checkerboard);
chi.checkerboard=psi.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
for(int s=0;s<Ls;s++){
auto tmp = psi._odata[0];
if ( s==0 ) {
spProj5m(tmp,psi._odata[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5p(tmp,psi._odata[ss+Ls-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else if ( s==(Ls-1)) {
spProj5m(tmp,psi._odata[ss+0]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5p(tmp,psi._odata[ss+s-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else {
spProj5m(tmp,psi._odata[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5p(tmp,psi._odata[ss+s-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
}
}
}
}
template<class Impl>
void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
const FermionField &phi,
FermionField &chi,
std::vector<RealD> &lower,
std::vector<RealD> &diag,
std::vector<RealD> &upper)
{
int Ls =this->Ls;
GridBase *grid=psi._grid;
assert(phi.checkerboard == psi.checkerboard);
chi.checkerboard=psi.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
for(int s=0;s<Ls;s++){
if ( s==0 ) {
spProj5p(tmp,psi._odata[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5m(tmp,psi._odata[ss+Ls-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else if ( s==(Ls-1)) {
spProj5p(tmp,psi._odata[ss+0]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5m(tmp,psi._odata[ss+s-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
} else {
spProj5p(tmp,psi._odata[ss+s+1]);
chi[ss+s]=diag[s]*phi[ss+s]+upper[s]*tmp;
spProj5m(tmp,psi._odata[ss+s-1]);
chi[ss+s]=chi[ss+s]+lower[s]*tmp;
}
}
}
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi, FermionField &chi)
{
GridBase *grid=psi._grid;
int Ls=this->Ls;
chi.checkerboard=psi.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
// Apply (L^{\prime})^{-1}
chi[ss]=psi[ss]; // chi[0]=psi[0]
for(int s=1;s<Ls;s++){
spProj5p(tmp,chi[ss+s-1]);
chi[ss+s] = psi[ss+s]-lee[s-1]*tmp;
}
// L_m^{-1}
for (int s=0;s<Ls-1;s++){ // Chi[ee] = 1 - sum[s<Ls-1] -leem[s]P_- chi
spProj5m(tmp,chi[ss+s]);
chi[ss+Ls-1] = chi[ss+Ls-1] - leem[s]*tmp;
}
// U_m^{-1} D^{-1}
for (int s=0;s<Ls-1;s++){
// Chi[s] + 1/d chi[s]
spProj5p(tmp,chi[ss+Ls-1]);
chi[ss+s] = (1.0/dee[s])*chi[ss+s]-(ueem[s]/dee[Ls-1])*tmp;
}
chi[ss+Ls-1]= (1.0/dee[Ls-1])*chi[ss+Ls-1];
// Apply U^{-1}
for (int s=Ls-2;s>=0;s--){
spProj5m(tmp,chi[ss+s+1]);
chi[ss+s] = chi[ss+s] - uee[s]*tmp;
}
}
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
{
GridBase *grid=psi._grid;
int Ls=this->Ls;
assert(psi.checkerboard == psi.checkerboard);
chi.checkerboard=psi.checkerboard;
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=Ls){ // adds Ls
auto tmp = psi._odata[0];
// Apply (U^{\prime})^{-dagger}
chi[ss]=psi[ss];
for (int s=1;s<Ls;s++){
spProj5m(tmp,chi[ss+s-1]);
chi[ss+s] = psi[ss+s]-uee[s-1]*tmp;
}
// U_m^{-\dagger}
for (int s=0;s<Ls-1;s++){
spProj5p(tmp,chi[ss+s]);
chi[ss+Ls-1] = chi[ss+Ls-1] - ueem[s]*tmp;
}
// L_m^{-\dagger} D^{-dagger}
for (int s=0;s<Ls-1;s++){
spProj5m(tmp,chi[ss+Ls-1]);
chi[ss+s] = (1.0/dee[s])*chi[ss+s]-(leem[s]/dee[Ls-1])*tmp;
}
chi[ss+Ls-1]= (1.0/dee[Ls-1])*chi[ss+Ls-1];
// Apply L^{-dagger}
for (int s=Ls-2;s>=0;s--){
spProj5p(tmp,chi[ss+s+1]);
chi[ss+s] = chi[ss+s] - lee[s]*tmp;
}
}
}
FermOp4dVecTemplateInstantiate(CayleyFermion5D);
GparityFermOpTemplateInstantiate(CayleyFermion5D);
}}