portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 14:04:32 +00:00
Code Releases Activity

A2A Lepton-Meson Field contraction

Browse Source
This commit is contained in:
Vera Guelpers
2019-04-30 12:04:59 +01:00
parent c2cd0e15d7
commit 4f0631615f
5 changed files with 528 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

208
Grid/qcd/utils/A2Autils.h
View File

@@ -68,6 +68,16 @@ public:
const std::vector<ComplexField> &emB1,
int orthogdim, double *t_kernel = nullptr, double *t_gsum = nullptr);
template <typename TensorType>
static void LeptonField(TensorType &mat,
const FermionField *lhs_wi,
const FermionField *rhs_vj,
const std::vector<ComplexField> &lepton02,
const std::vector<ComplexField> &lepton03,
const std::vector<ComplexField> &lepton12,
const std::vector<ComplexField> &lepton13,
int orthogdim, double *t_kernel = nullptr, double *t_gsum = nullptr);
static void ContractWWVV(std::vector<PropagatorField> &WWVV,
const Eigen::Tensor<ComplexD,3> &WW_sd,
const FermionField *vs,
@@ -809,6 +819,204 @@ void A2Autils<FImpl>::AslashField(TensorType &mat,
if (t_gsum) *t_gsum += usecond();
}
// "Lepton" field w_i(x)^dag * L_mu * g^L_mu * v_j(x)
//
// With V-A current:
//
// g^L_mu = g_mu * (1 - g5)
//
// then in spin space
//
// ( 0 0 L_02 L_03 )
// L_mu * g^L_mu = ( 0 0 L_12 L_13 )
// ( 0 0 0 0 )
// ( 0 0 0 0 )
//
//
// With
//
// L_02 = 2*L3 + 2*i*L2
// L_03 = -2*L1 + 2*i*L0
// L_12 = 2*L1 + 2*i*L0
// L_13 = 2*L3 - 2*i*L2
//
// (where mu = (x,y,z,t) )
//
template <class FImpl>
template <typename TensorType>
void A2Autils<FImpl>::LeptonField(TensorType &mat,
const FermionField *lhs_wi,
const FermionField *rhs_vj,
const std::vector<ComplexField> &lepton02,
const std::vector<ComplexField> &lepton03,
const std::vector<ComplexField> &lepton12,
const std::vector<ComplexField> &lepton13,
int orthogdim, double *t_kernel, double *t_gsum)
{
typedef typename FermionField::vector_object vobj;
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
typedef iSpinMatrix<vector_type> SpinMatrix_v;
typedef iSpinMatrix<scalar_type> SpinMatrix_s;
typedef iSinglet<vector_type> Singlet_v;
typedef iSinglet<scalar_type> Singlet_s;
int Lblock = mat.dimension(3);
int Rblock = mat.dimension(4);
GridBase *grid = lhs_wi[0]._grid;
const int Nd = grid->_ndimension;
const int Nsimd = grid->Nsimd();
int Nt = grid->GlobalDimensions()[orthogdim];
int Nlep = lepton02.size();
assert(lepton03.size() == Nlep);
assert(lepton12.size() == Nlep);
assert(lepton13.size() == Nlep);
int fd=grid->_fdimensions[orthogdim];
int ld=grid->_ldimensions[orthogdim];
int rd=grid->_rdimensions[orthogdim];
// will locally sum vectors first
// sum across these down to scalars
// splitting the SIMD
int MFrvol = rd*Lblock*Rblock*Nlep;
int MFlvol = ld*Lblock*Rblock*Nlep;
Vector<vector_type> lvSum(MFrvol);
parallel_for (int r = 0; r < MFrvol; r++)
{
lvSum[r] = zero;
}
Vector<scalar_type> lsSum(MFlvol);
parallel_for (int r = 0; r < MFlvol; r++)
{
lsSum[r] = scalar_type(0.0);
}
int e1= grid->_slice_nblock[orthogdim];
int e2= grid->_slice_block [orthogdim];
int stride=grid->_slice_stride[orthogdim];
// Nested parallelism would be ok
// Wasting cores here. Test case r
if (t_kernel) *t_kernel = -usecond();
parallel_for(int r=0;r<rd;r++)
{
int so=r*grid->_ostride[orthogdim]; // base offset for start of plane
for(int n=0;n<e1;n++)
for(int b=0;b<e2;b++)
{
int ss= so+n*stride+b;
for(int i=0;i<Lblock;i++)
{
auto left = conjugate(lhs_wi[i]._odata[ss]);
for(int j=0;j<Rblock;j++)
{
SpinMatrix_v vv;
auto right = rhs_vj[j]._odata[ss];
for(int s1=0;s1<Ns;s1++)
for(int s2=0;s2<Ns;s2++)
{
vv()(s1,s2)() = left()(s2)(0) * right()(s1)(0)
+ left()(s2)(1) * right()(s1)(1)
+ left()(s2)(2) * right()(s1)(2);
}
// After getting the sitewise product do the mom phase loop
int base = Nlep*i+Nlep*Lblock*j+Nlep*Lblock*Rblock*r;
for ( int m=0;m<Nlep;m++)
{
int idx = m+base;
auto lep02 = lepton02[m]._odata[ss];
auto lep12 = lepton12[m]._odata[ss];
auto lep03 = lepton03[m]._odata[ss];
auto lep13 = lepton13[m]._odata[ss];
lvSum[idx] += vv()(2,0)()*lep02()()() + vv()(3,0)()*lep03()()()
+ vv()(2,1)()*lep12()()() + vv()(3,1)()*lep13()()();
}
}
}
}
}
// Sum across simd lanes in the plane, breaking out orthog dir.
parallel_for(int rt=0;rt<rd;rt++)
{
std::vector<int> icoor(Nd);
std::vector<scalar_type> extracted(Nsimd);
for(int i=0;i<Lblock;i++)
for(int j=0;j<Rblock;j++)
for(int m=0;m<Nlep;m++)
{
int ij_rdx = m+Nlep*i+Nlep*Lblock*j+Nlep*Lblock*Rblock*rt;
extract<vector_type,scalar_type>(lvSum[ij_rdx],extracted);
for(int idx=0;idx<Nsimd;idx++)
{
grid->iCoorFromIindex(icoor,idx);
int ldx = rt+icoor[orthogdim]*rd;
int ij_ldx = m+Nlep*i+Nlep*Lblock*j+Nlep*Lblock*Rblock*ldx;
lsSum[ij_ldx]=lsSum[ij_ldx]+extracted[idx];
}
}
}
if (t_kernel) *t_kernel += usecond();
// ld loop and local only??
int pd = grid->_processors[orthogdim];
int pc = grid->_processor_coor[orthogdim];
parallel_for_nest2(int lt=0;lt<ld;lt++)
{
for(int pt=0;pt<pd;pt++)
{
int t = lt + pt*ld;
if (pt == pc)
{
for(int i=0;i<Lblock;i++)
for(int j=0;j<Rblock;j++)
for(int m=0;m<Nlep;m++)
{
int ij_dx = m+Nlep*i + Nlep*Lblock * j + Nlep*Lblock * Rblock * lt;
mat(m,0,t,i,j) = lsSum[ij_dx];
}
}
else
{
const scalar_type zz(0.0);
for(int i=0;i<Lblock;i++)
for(int j=0;j<Rblock;j++)
for(int m=0;m<Nlep;m++)
{
mat(m,0,t,i,j) = zz;
}
}
}
}
if (t_gsum) *t_gsum = -usecond();
grid->GlobalSumVector(&mat(0,0,0,0,0),Nlep*Nt*Lblock*Rblock);
if (t_gsum) *t_gsum += usecond();
}
////////////////////////////////////////////
// Schematic thoughts about more generalised four quark insertion
//

1
Hadrons/Modules.hpp
View File

@@ -3,6 +3,7 @@
#include <Hadrons/Modules/MContraction/A2AAslashField.hpp>
#include <Hadrons/Modules/MContraction/A2ALoop.hpp>
#include <Hadrons/Modules/MContraction/A2AMesonField.hpp>
#include <Hadrons/Modules/MContraction/A2ALeptonField.hpp>
#include <Hadrons/Modules/MContraction/Meson.hpp>
#include <Hadrons/Modules/MContraction/DiscLoop.hpp>
#include <Hadrons/Modules/MContraction/Gamma3pt.hpp>

34
Hadrons/Modules/MContraction/A2ALeptonField.cc Normal file
View File

@@ -0,0 +1,34 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: Hadrons/Modules/MContraction/A2ALeptonField.cc
Copyright (C) 2015-2019
Author: Vera Guelpers <Vera.Guelpers@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 <Hadrons/Modules/MContraction/A2ALeptonField.hpp>
using namespace Grid;
using namespace Hadrons;
using namespace MContraction;
template class Grid::Hadrons::MContraction::TA2ALeptonField<FIMPL, PhotonR>;

283
Hadrons/Modules/MContraction/A2ALeptonField.hpp Normal file
View File

@@ -0,0 +1,283 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: Hadrons/Modules/MContraction/A2ALeptonField.cc
Copyright (C) 2015-2019
Author: Vera Guelpers <Vera.Guelpers@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 */
#ifndef Hadrons_MContraction_A2ALeptonField_hpp_
#define Hadrons_MContraction_A2ALeptonField_hpp_
#include <Hadrons/Global.hpp>
#include <Hadrons/Module.hpp>
#include <Hadrons/ModuleFactory.hpp>
#include <Hadrons/A2AMatrix.hpp>
#ifndef MLF_IO_TYPE
#define MLF_IO_TYPE ComplexF
#endif
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* A2ALeptonField *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MContraction)
class A2ALeptonFieldPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(A2ALeptonFieldPar,
int, cacheBlock,
int, block,
std::string, left,
std::string, right,
std::string, output,
std::vector<std::string>, lepton);
};
class A2ALeptonFieldMetadata: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(A2ALeptonFieldMetadata,
std::string, leptonName,
int, sidx1,
int, sidx2);
};
template <typename T, typename FImpl>
class LeptonFieldKernel: public A2AKernel<T, typename FImpl::FermionField>
{
public:
typedef typename FImpl::FermionField FermionField;
public:
LeptonFieldKernel(const std::vector<LatticeComplex> &lepton02,
const std::vector<LatticeComplex> &lepton03,
const std::vector<LatticeComplex> &lepton12,
const std::vector<LatticeComplex> &lepton13,
GridBase *grid)
: lepton02_(lepton02), lepton03_(lepton03), lepton12_(lepton12), lepton13_(lepton13), grid_(grid)
{
vol_ = 1.;
for (auto &d: grid_->GlobalDimensions())
{
vol_ *= d;
}
}
virtual ~LeptonFieldKernel(void) = default;
virtual void operator()(A2AMatrixSet<T> &m, const FermionField *left,
const FermionField *right,
const unsigned int orthogDim, double &t)
{
A2Autils<FImpl>::LeptonField(m, left, right, lepton02_, lepton03_, lepton12_, lepton13_, orthogDim, &t);
}
virtual double flops(const unsigned int blockSizei, const unsigned int blockSizej)
{
return 0.;
}
virtual double bytes(const unsigned int blockSizei, const unsigned int blockSizej)
{
return 0.;
}
private:
const std::vector<LatticeComplex> &lepton02_, &lepton03_, &lepton12_, &lepton13_;
GridBase *grid_;
double vol_;
};
template <typename FImpl, typename PhotonImpl>
class TA2ALeptonField: public Module<A2ALeptonFieldPar>
{
public:
FERM_TYPE_ALIASES(FImpl,);
typedef A2AMatrixBlockComputation<Complex,
FermionField,
A2ALeptonFieldMetadata,
MLF_IO_TYPE> Computation;
typedef LeptonFieldKernel<Complex, FImpl> Kernel;
public:
// constructor
TA2ALeptonField(const std::string name);
// destructor
virtual ~TA2ALeptonField(void) {};
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_TMP(A2ALeptonField, ARG(TA2ALeptonField<FIMPL, PhotonR>), MContraction);
/******************************************************************************
* TA2ALeptonField implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl, typename PhotonImpl>
TA2ALeptonField<FImpl, PhotonImpl>::TA2ALeptonField(const std::string name)
: Module<A2ALeptonFieldPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl, typename PhotonImpl>
std::vector<std::string> TA2ALeptonField<FImpl, PhotonImpl>::getInput(void)
{
std::vector<std::string> in = par().lepton;
in.push_back(par().left);
in.push_back(par().right);
return in;
}
template <typename FImpl, typename PhotonImpl>
std::vector<std::string> TA2ALeptonField<FImpl, PhotonImpl>::getOutput(void)
{
std::vector<std::string> out = {};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl, typename PhotonImpl>
void TA2ALeptonField<FImpl, PhotonImpl>::setup(void)
{
envTmp(Computation, "computation", 1, envGetGrid(FermionField),
env().getNd() - 1, par().lepton.size()*8, 1, par().block,
par().cacheBlock, this);
envTmp(std::vector<LatticeComplex>, "Lmu", 1,
4, envGetGrid(LatticeComplex));
envTmp(std::vector<LatticeComplex>, "L02", 1,
par().lepton.size()*8, envGetGrid(LatticeComplex));
envTmp(std::vector<LatticeComplex>, "L03", 1,
par().lepton.size()*8, envGetGrid(LatticeComplex));
envTmp(std::vector<LatticeComplex>, "L12", 1,
par().lepton.size()*8, envGetGrid(LatticeComplex));
envTmp(std::vector<LatticeComplex>, "L13", 1,
par().lepton.size()*8, envGetGrid(LatticeComplex));
envTmpLat(PropagatorField, "prop_buf");
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl, typename PhotonImpl>
void TA2ALeptonField<FImpl, PhotonImpl>::execute(void)
{
auto &left = envGet(std::vector<FermionField>, par().left);
auto &right = envGet(std::vector<FermionField>, par().right);
int nt = env().getDim().back();
int N_i = left.size();
int N_j = right.size();
int block = par().block;
int cacheBlock = par().cacheBlock;
LOG(Message) << "Computing all-to-all lepton insertion fields" << std::endl;
LOG(Message) << "Left: '" << par().left << "' Right: '" << par().right << "'" << std::endl;
LOG(Message) << "leptons: " << std::endl;
for (auto &name: par().lepton)
{
LOG(Message) << " " << name << std::endl;
}
LOG(Message) << "lepton insertion field size: " << nt << "*" << N_i << "*" << N_j
<< " (filesize " << sizeString(nt*N_i*N_j*sizeof(MLF_IO_TYPE))
<< "/spin index/lepton)" << std::endl;
envGetTmp(std::vector<ComplexField>, L02);
envGetTmp(std::vector<ComplexField>, L03);
envGetTmp(std::vector<ComplexField>, L12);
envGetTmp(std::vector<ComplexField>, L13);
unsigned int ii = 0;
envGetTmp(PropagatorField, prop_buf);
envGetTmp(std::vector<LatticeComplex>, Lmu);
for (unsigned int i = 0; i < par().lepton.size(); ++i)
{
auto &lepton = envGet(PropagatorField, par().lepton[i]);
// need only half of the spin indices (s1s2) of the lepton field
// the other ones are zero because of the left-handed current
for (unsigned int s1 = 0; s1 < 2; ++s1)
for (unsigned int s2 = 0; s2 < 4; ++s2)
{
for (unsigned int mu = 0; mu < 4; ++mu)
{
prop_buf = GammaL(Gamma::gmu[mu]) * lepton;
//lepon is unit matix in color space, so just pick one diagonal enty
Lmu[mu] = peekColour(peekSpin(prop_buf,s1,s2),0,0);
}
// build the required combinations of lepton fields for the a2a kernel
// [g^L_mu * L_mu]_{ab} = L_{ab} (with mu = (x,y,z,t))
L02[ii] = 2.0*Lmu[3] + 2.0*timesI(Lmu[2]);
L03[ii] = 2.0*timesI(Lmu[0]) - 2.0*Lmu[1];
L12[ii] = 2.0*timesI(Lmu[0]) + 2.0*Lmu[1];
L13[ii] = 2.0*Lmu[3] - 2.0*timesI(Lmu[2]);
ii++;
}
}
auto ionameFn = [this](const unsigned int index, const unsigned int dummy)
{
//index = 8*l + 4*sindex1 + sindex2
unsigned int sindex = index % 8;
unsigned int sindex2 = sindex % 4;
unsigned int sindex1 = (sindex - sindex2)/4;
unsigned int l = (index - sindex)/8;
return par().lepton[l] + "_" + std::to_string(sindex1) + std::to_string(sindex2);
};
auto filenameFn = [this, &ionameFn](const unsigned int index, const unsigned int dummy)
{
return par().output + "." + std::to_string(vm().getTrajectory())
+ "/" + ionameFn(index, dummy) + ".h5";
};
auto metadataFn = [this](const unsigned int index, const unsigned int dummy)
{
A2ALeptonFieldMetadata md;
unsigned int sindex = index % 8;
unsigned int sindex2 = sindex % 4;
unsigned int sindex1 = (sindex - sindex2)/4;
unsigned int l = (index - sindex)/8;
md.leptonName = par().lepton[l];
md.sidx1 = sindex1;
md.sidx2 = sindex2;
return md;
};
// executing computation
Kernel kernel(L02, L03, L12, L13, envGetGrid(FermionField));
envGetTmp(Computation, computation);
computation.execute(left, right, kernel, ionameFn, filenameFn, metadataFn);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_MContraction_A2ALeptonField_hpp_

2
Hadrons/modules.inc
View File

@@ -7,6 +7,7 @@ modules_cc =\
Modules/MContraction/WeakNonEye3pt.cc \
Modules/MContraction/A2AAslashField.cc \
Modules/MContraction/A2AMesonField.cc \
Modules/MContraction/A2ALeptonField.cc \
Modules/MContraction/DiscLoop.cc \
Modules/MContraction/Gamma3pt.cc \
Modules/MFermion/FreeProp.cc \
@@ -72,6 +73,7 @@ modules_hpp =\
Modules/MContraction/A2AAslashField.hpp \
Modules/MContraction/A2ALoop.hpp \
Modules/MContraction/A2AMesonField.hpp \
Modules/MContraction/A2ALeptonField.hpp \
Modules/MContraction/Meson.hpp \
Modules/MContraction/DiscLoop.hpp \
Modules/MContraction/Gamma3pt.hpp \