portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-16 06:47:06 +01:00
Code Releases Activity

Merge branch 'feature/dirichlet' of https://github.com/paboyle/Grid into feature/dirichlet

Browse Source
This commit is contained in:
Peter Boyle
2022-05-25 12:38:10 -04:00
parent 136d843ce7 18028f4309
commit 60f4cb0ffd
63 changed files with 2295 additions and 1120 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
Grid/DisableWarnings.h
View File

@ -44,14 +44,22 @@ directory
#ifdef __NVCC__
//disables nvcc specific warning in json.hpp
#pragma clang diagnostic ignored "-Wdeprecated-register"
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
//disables nvcc specific warning in json.hpp
#pragma nv_diag_suppress unsigned_compare_with_zero
#pragma nv_diag_suppress cast_to_qualified_type
//disables nvcc specific warning in many files
#pragma nv_diag_suppress esa_on_defaulted_function_ignored
#pragma nv_diag_suppress extra_semicolon
#else
//disables nvcc specific warning in json.hpp
#pragma diag_suppress unsigned_compare_with_zero
#pragma diag_suppress cast_to_qualified_type
//disables nvcc specific warning in many files
#pragma diag_suppress esa_on_defaulted_function_ignored
#pragma diag_suppress extra_semicolon
//Eigen only
#endif
#endif
// Disable vectorisation in Eigen on the Power8/9 and PowerPC

4
Grid/Grid_Eigen_Dense.h
View File

@ -14,7 +14,11 @@
/* NVCC save and restore compile environment*/
#ifdef __NVCC__
#pragma push
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
#pragma nv_diag_suppress code_is_unreachable
#else
#pragma diag_suppress code_is_unreachable
#endif
#pragma push_macro("__CUDA_ARCH__")
#pragma push_macro("__NVCC__")
#pragma push_macro("__CUDACC__")

38
Grid/algorithms/iterative/Deflation.h
View File

@ -113,7 +113,43 @@ public:
blockPromote(guess_coarse,guess,subspace);
guess.Checkerboard() = src.Checkerboard();
};
};
void operator()(const std::vector<FineField> &src,std::vector<FineField> &guess) {
int Nevec = (int)evec_coarse.size();
int Nsrc = (int)src.size();
// make temp variables
std::vector<CoarseField> src_coarse(Nsrc,evec_coarse[0].Grid());
std::vector<CoarseField> guess_coarse(Nsrc,evec_coarse[0].Grid());
//Preporcessing
std::cout << GridLogMessage << "Start BlockProject for loop" << std::endl;
for (int j=0;j<Nsrc;j++)
{
guess_coarse[j] = Zero();
std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
blockProject(src_coarse[j],src[j],subspace);
}
//deflation set up for eigen vector batchsize 1 and source batch size equal number of sources
std::cout << GridLogMessage << "Start ProjectAccum for loop" << std::endl;
for (int i=0;i<Nevec;i++)
{
std::cout << GridLogMessage << "ProjectAccum Nvec: " << i << std::endl;
const CoarseField & tmp = evec_coarse[i];
for (int j=0;j<Nsrc;j++)
{
axpy(guess_coarse[j],TensorRemove(innerProduct(tmp,src_coarse[j])) / eval_coarse[i],tmp,guess_coarse[j]);
}
}
//postprocessing
std::cout << GridLogMessage << "Start BlockPromote for loop" << std::endl;
for (int j=0;j<Nsrc;j++)
{
std::cout << GridLogMessage << "BlockProject iter: " << j << std::endl;
blockPromote(guess_coarse[j],guess[j],subspace);
guess[j].Checkerboard() = src[j].Checkerboard();
}
};
};

3
Grid/communicator/Communicator_mpi3.cc
View File

@ -390,16 +390,15 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
list.push_back(xrq);
off_node_bytes+=bytes;
} else {
// TODO : make a OMP loop on CPU, call threaded bcopy
void *shm = (void *) this->ShmBufferTranslate(dest,recv);
assert(shm!=NULL);
// std::cout <<"acceleratorCopyDeviceToDeviceAsynch"<< std::endl;
acceleratorCopyDeviceToDeviceAsynch(xmit,shm,bytes);
}
}
if ( CommunicatorPolicy == CommunicatorPolicySequential ) {
this->StencilSendToRecvFromComplete(list,dir);
list.resize(0);
}
return off_node_bytes;

9
Grid/parallelIO/IldgIO.h
View File

@ -31,6 +31,7 @@ directory
#include <fstream>
#include <iomanip>
#include <iostream>
#include <string>
#include <map>
#include <pwd.h>
@ -654,7 +655,8 @@ class IldgWriter : public ScidacWriter {
// Fill ILDG header data struct
//////////////////////////////////////////////////////
ildgFormat ildgfmt ;
ildgfmt.field = std::string("su3gauge");
const std::string stNC = std::to_string( Nc ) ;
ildgfmt.field = std::string("su"+stNC+"gauge");
if ( format == std::string("IEEE32BIG") ) {
ildgfmt.precision = 32;
@ -871,7 +873,8 @@ class IldgReader : public GridLimeReader {
} else {
assert(found_ildgFormat);
assert ( ildgFormat_.field == std::string("su3gauge") );
const std::string stNC = std::to_string( Nc ) ;
assert ( ildgFormat_.field == std::string("su"+stNC+"gauge") );
///////////////////////////////////////////////////////////////////////////////////////
// Populate our Grid metadata as best we can
@ -879,7 +882,7 @@ class IldgReader : public GridLimeReader {
std::ostringstream vers; vers << ildgFormat_.version;
FieldMetaData_.hdr_version = vers.str();
FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
FieldMetaData_.data_type = std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC);
FieldMetaData_.nd=4;
FieldMetaData_.dimension.resize(4);

33
Grid/parallelIO/MetaData.h
View File

@ -6,8 +6,8 @@
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
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
@ -182,8 +182,8 @@ class GaugeStatistics
public:
void operator()(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
{
header.link_trace=WilsonLoops<Impl>::linkTrace(data);
header.plaquette =WilsonLoops<Impl>::avgPlaquette(data);
header.link_trace = WilsonLoops<Impl>::linkTrace(data);
header.plaquette = WilsonLoops<Impl>::avgPlaquette(data);
}
};
typedef GaugeStatistics<PeriodicGimplD> PeriodicGaugeStatistics;
@ -203,20 +203,24 @@ template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzCo
//////////////////////////////////////////////////////////////////////
inline void reconstruct3(LorentzColourMatrix & cm)
{
const int x=0;
const int y=1;
const int z=2;
assert( Nc < 4 && Nc > 1 ) ;
for(int mu=0;mu<Nd;mu++){
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
#if Nc == 2
cm(mu)()(1,0) = -adj(cm(mu)()(0,y)) ;
cm(mu)()(1,1) = adj(cm(mu)()(0,x)) ;
#else
const int x=0 , y=1 , z=2 ; // a little disinenuous labelling
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
#endif
}
}
////////////////////////////////////////////////////////////////////////////////
// Some data types for intermediate storage
////////////////////////////////////////////////////////////////////////////////
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, Nc-1>, Nd >;
typedef iLorentzColour2x3<Complex> LorentzColour2x3;
typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
@ -278,7 +282,6 @@ struct GaugeSimpleMunger{
template <class fobj, class sobj>
struct GaugeSimpleUnmunger {
void operator()(sobj &in, fobj &out) {
for (int mu = 0; mu < Nd; mu++) {
for (int i = 0; i < Nc; i++) {
@ -317,8 +320,8 @@ template<class fobj,class sobj>
struct Gauge3x2munger{
void operator() (fobj &in,sobj &out){
for(int mu=0;mu<Nd;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
for(int i=0;i<Nc-1;i++){
for(int j=0;j<Nc;j++){
out(mu)()(i,j) = in(mu)(i)(j);
}}
}
@ -330,8 +333,8 @@ template<class fobj,class sobj>
struct Gauge3x2unmunger{
void operator() (sobj &in,fobj &out){
for(int mu=0;mu<Nd;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
for(int i=0;i<Nc-1;i++){
for(int j=0;j<Nc;j++){
out(mu)(i)(j) = in(mu)()(i,j);
}}
}

63
Grid/parallelIO/NerscIO.h
View File

@ -9,6 +9,7 @@
Author: Matt Spraggs <matthew.spraggs@gmail.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Jamie Hudspith <renwick.james.hudspth@gmail.com>
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
@ -30,6 +31,8 @@
#ifndef GRID_NERSC_IO_H
#define GRID_NERSC_IO_H
#include <string>
NAMESPACE_BEGIN(Grid);
using namespace Grid;
@ -145,15 +148,17 @@ public:
std::string format(header.floating_point);
int ieee32big = (format == std::string("IEEE32BIG"));
int ieee32 = (format == std::string("IEEE32"));
int ieee64big = (format == std::string("IEEE64BIG"));
int ieee64 = (format == std::string("IEEE64") || format == std::string("IEEE64LITTLE"));
const int ieee32big = (format == std::string("IEEE32BIG"));
const int ieee32 = (format == std::string("IEEE32"));
const int ieee64big = (format == std::string("IEEE64BIG"));
const int ieee64 = (format == std::string("IEEE64") || \
format == std::string("IEEE64LITTLE"));
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
const std::string stNC = std::to_string( Nc ) ;
if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE") ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<vLorentzColourMatrixD, LorentzColour2x3F>
(Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
@ -164,7 +169,7 @@ public:
(Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
} else if ( header.data_type == std::string("4D_SU"+stNC+"_GAUGE_"+stNC+"x"+stNC) ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<vLorentzColourMatrixD,LorentzColourMatrixF>
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
@ -209,27 +214,29 @@ public:
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
std::string file,
std::string ens_label = std::string("DWF"))
std::string ens_label = std::string("DWF"),
std::string ens_id = std::string("UKQCD"),
unsigned int sequence_number = 1)
{
writeConfiguration(Umu,file,0,1,ens_label);
writeConfiguration(Umu,file,0,1,ens_label,ens_id,sequence_number);
}
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
std::string file,
int two_row,
int bits32,
std::string ens_label = std::string("DWF"))
std::string ens_label = std::string("DWF"),
std::string ens_id = std::string("UKQCD"),
unsigned int sequence_number = 1)
{
typedef vLorentzColourMatrixD vobj;
typedef typename vobj::scalar_object sobj;
FieldMetaData header;
///////////////////////////////////////////
// Following should become arguments
///////////////////////////////////////////
header.sequence_number = 1;
header.ensemble_id = std::string("UKQCD");
header.sequence_number = sequence_number;
header.ensemble_id = ens_id;
header.ensemble_label = ens_label;
header.hdr_version = "1.0" ;
typedef LorentzColourMatrixD fobj3D;
typedef LorentzColour2x3D fobj2D;
@ -243,10 +250,14 @@ public:
uint64_t offset;
// Sod it -- always write 3x3 double
header.floating_point = std::string("IEEE64BIG");
header.data_type = std::string("4D_SU3_GAUGE_3x3");
GaugeSimpleUnmunger<fobj3D,sobj> munge;
// Sod it -- always write NcxNc double
header.floating_point = std::string("IEEE64BIG");
const std::string stNC = std::to_string( Nc ) ;
if( two_row ) {
header.data_type = std::string("4D_SU" + stNC + "_GAUGE" );
} else {
header.data_type = std::string("4D_SU" + stNC + "_GAUGE_" + stNC + "x" + stNC );
}
if ( grid->IsBoss() ) {
truncate(file);
offset = writeHeader(header,file);
@ -254,8 +265,15 @@ public:
grid->Broadcast(0,(void *)&offset,sizeof(offset));
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
nersc_csum,scidac_csuma,scidac_csumb);
if( two_row ) {
Gauge3x2unmunger<fobj2D,sobj> munge;
BinaryIO::writeLatticeObject<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point,
nersc_csum,scidac_csuma,scidac_csumb);
} else {
GaugeSimpleUnmunger<fobj3D,sobj> munge;
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
nersc_csum,scidac_csuma,scidac_csumb);
}
header.checksum = nersc_csum;
if ( grid->IsBoss() ) {
writeHeader(header,file);
@ -287,8 +305,7 @@ public:
header.plaquette=0.0;
MachineCharacteristics(header);
uint64_t offset;
uint64_t offset;
#ifdef RNG_RANLUX
header.floating_point = std::string("UINT64");
header.data_type = std::string("RANLUX48");
@ -328,7 +345,7 @@ public:
GridBase *grid = parallel.Grid();
uint64_t offset = readHeader(file,grid,header);
uint64_t offset = readHeader(file,grid,header);
FieldMetaData clone(header);

4
Grid/pugixml/pugixml.cc
View File

@ -16,8 +16,12 @@
#ifdef __NVCC__
#pragma push
#if (__CUDACC_VER_MAJOR__ >= 11) && (__CUDACC_VER_MINOR__ >= 5)
#pragma nv_diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
#else
#pragma diag_suppress declared_but_not_referenced // suppress "function was declared but never referenced warning"
#endif
#endif
#include "pugixml.h"

12
Grid/qcd/action/ActionParams.h
View File

@ -37,24 +37,32 @@ NAMESPACE_BEGIN(Grid);
// These can move into a params header and be given MacroMagic serialisation
struct GparityWilsonImplParams {
Coordinate twists;
GparityWilsonImplParams() : twists(Nd, 0) {};
Coordinate dirichlet; // Blocksize of dirichlet BCs
GparityWilsonImplParams() : twists(Nd, 0), dirichlet(Nd, 0) {};
};
struct WilsonImplParams {
bool overlapCommsCompute;
Coordinate dirichlet; // Blocksize of dirichlet BCs
AcceleratorVector<Real,Nd> twist_n_2pi_L;
AcceleratorVector<Complex,Nd> boundary_phases;
WilsonImplParams() {
dirichlet.resize(Nd,0);
boundary_phases.resize(Nd, 1.0);
twist_n_2pi_L.resize(Nd, 0.0);
};
WilsonImplParams(const AcceleratorVector<Complex,Nd> phi) : boundary_phases(phi), overlapCommsCompute(false) {
twist_n_2pi_L.resize(Nd, 0.0);
dirichlet.resize(Nd,0);
}
};
struct StaggeredImplParams {
StaggeredImplParams() {};
Coordinate dirichlet; // Blocksize of dirichlet BCs
StaggeredImplParams()
{
dirichlet.resize(Nd,0);
};
};
struct OneFlavourRationalParams : Serializable {

14
Grid/qcd/action/fermion/CayleyFermion5D.h
View File

@ -68,9 +68,17 @@ public:
///////////////////////////////////////////////////////////////
// Support for MADWF tricks
///////////////////////////////////////////////////////////////
virtual RealD Mass(void) { return mass; };
RealD Mass(void) { return (mass_plus + mass_minus) / 2.0; };
RealD MassPlus(void) { return mass_plus; };
RealD MassMinus(void) { return mass_minus; };
void SetMass(RealD _mass) {
mass=_mass;
mass_plus=mass_minus=_mass;
SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c); // Reset coeffs
} ;
void SetMass(RealD _mass_plus, RealD _mass_minus) {
mass_plus=_mass_plus;
mass_minus=_mass_minus;
SetCoefficientsInternal(_zolo_hi,_gamma,_b,_c); // Reset coeffs
} ;
void P(const FermionField &psi, FermionField &chi);
@ -108,7 +116,7 @@ public:
void MeooeDag5D (const FermionField &in, FermionField &out);
// protected:
RealD mass;
RealD mass_plus, mass_minus;
// Save arguments to SetCoefficientsInternal
Vector<Coeff_t> _gamma;

435
Grid/qcd/action/fermion/CloverHelpers.h Normal file
View File

@ -0,0 +1,435 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonCloverFermionImplementation.h
Copyright (C) 2017 - 2022
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
Author: Mattia Bruno <mattia.bruno@cern.ch>
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 */
#pragma once
#include <Grid/Grid.h>
#include <Grid/qcd/spin/Dirac.h>
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
////////////////////////////////////////////
// Standard Clover
// (4+m0) + csw * clover_term
// Exp Clover
// (4+m0) * exp(csw/(4+m0) clover_term)
// = (4+m0) + csw * clover_term + ...
////////////////////////////////////////////
NAMESPACE_BEGIN(Grid);
//////////////////////////////////
// Generic Standard Clover
//////////////////////////////////
template<class Impl>
class CloverHelpers: public WilsonCloverHelpers<Impl> {
public:
INHERIT_IMPL_TYPES(Impl);
INHERIT_CLOVER_TYPES(Impl);
typedef WilsonCloverHelpers<Impl> Helpers;
static void Instantiate(CloverField& CloverTerm, CloverField& CloverTermInv, RealD csw_t, RealD diag_mass) {
GridBase *grid = CloverTerm.Grid();
CloverTerm += diag_mass;
int lvol = grid->lSites();
int DimRep = Impl::Dimension;
{
autoView(CTv,CloverTerm,CpuRead);
autoView(CTIv,CloverTermInv,CpuWrite);
thread_for(site, lvol, {
Coordinate lcoor;
grid->LocalIndexToLocalCoor(site, lcoor);
Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
peekLocalSite(Qx, CTv, lcoor);
for (int j = 0; j < Ns; j++)
for (int k = 0; k < Ns; k++)
for (int a = 0; a < DimRep; a++)
for (int b = 0; b < DimRep; b++){
auto zz = Qx()(j, k)(a, b);
EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
}
EigenInvCloverOp = EigenCloverOp.inverse();
for (int j = 0; j < Ns; j++)
for (int k = 0; k < Ns; k++)
for (int a = 0; a < DimRep; a++)
for (int b = 0; b < DimRep; b++)
Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
pokeLocalSite(Qxinv, CTIv, lcoor);
});
}
}
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
return Helpers::Cmunu(U, lambda, mu, nu);
}
};
//////////////////////////////////
// Generic Exp Clover
//////////////////////////////////
template<class Impl>
class ExpCloverHelpers: public WilsonCloverHelpers<Impl> {
public:
INHERIT_IMPL_TYPES(Impl);
INHERIT_CLOVER_TYPES(Impl);
template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
typedef WilsonCloverHelpers<Impl> Helpers;
// Can this be avoided?
static void IdentityTimesC(const CloverField& in, RealD c) {
int DimRep = Impl::Dimension;
autoView(in_v, in, AcceleratorWrite);
accelerator_for(ss, in.Grid()->oSites(), 1, {
for (int sa=0; sa<Ns; sa++)
for (int ca=0; ca<DimRep; ca++)
in_v[ss]()(sa,sa)(ca,ca) = c;
});
}
static int getNMAX(RealD prec, RealD R) {
/* compute stop condition for exponential */
int NMAX=1;
RealD cond=R*R/2.;
while (cond*std::exp(R)>prec) {
NMAX++;
cond*=R/(double)(NMAX+1);
}
return NMAX;
}
static int getNMAX(Lattice<iImplClover<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
static int getNMAX(Lattice<iImplClover<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
static void Instantiate(CloverField& Clover, CloverField& CloverInv, RealD csw_t, RealD diag_mass) {
GridBase* grid = Clover.Grid();
CloverField ExpClover(grid);
int NMAX = getNMAX(Clover, 3.*csw_t/diag_mass);
Clover *= (1.0/diag_mass);
// Taylor expansion, slow but generic
// Horner scheme: a0 + a1 x + a2 x^2 + .. = a0 + x (a1 + x(...))
// qN = cN
// qn = cn + qn+1 X
std::vector<RealD> cn(NMAX+1);
cn[0] = 1.0;
for (int i=1; i<=NMAX; i++)
cn[i] = cn[i-1] / RealD(i);
ExpClover = Zero();
IdentityTimesC(ExpClover, cn[NMAX]);
for (int i=NMAX-1; i>=0; i--)
ExpClover = ExpClover * Clover + cn[i];
// prepare inverse
CloverInv = (-1.0)*Clover;
Clover = ExpClover * diag_mass;
ExpClover = Zero();
IdentityTimesC(ExpClover, cn[NMAX]);
for (int i=NMAX-1; i>=0; i--)
ExpClover = ExpClover * CloverInv + cn[i];
CloverInv = ExpClover * (1.0/diag_mass);
}
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
assert(0);
return lambda;
}
};
//////////////////////////////////
// Compact Standard Clover
//////////////////////////////////
template<class Impl>
class CompactCloverHelpers: public CompactWilsonCloverHelpers<Impl>,
public WilsonCloverHelpers<Impl> {
public:
INHERIT_IMPL_TYPES(Impl);
INHERIT_CLOVER_TYPES(Impl);
INHERIT_COMPACT_CLOVER_TYPES(Impl);
typedef WilsonCloverHelpers<Impl> Helpers;
typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
static void MassTerm(CloverField& Clover, RealD diag_mass) {
Clover += diag_mass;
}
static void Exponentiate_Clover(CloverDiagonalField& Diagonal,
CloverTriangleField& Triangle,
RealD csw_t, RealD diag_mass) {
// Do nothing
}
// TODO: implement Cmunu for better performances with compact layout, but don't do it
// here, but rather in WilsonCloverHelpers.h -> CompactWilsonCloverHelpers
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
return Helpers::Cmunu(U, lambda, mu, nu);
}
};
//////////////////////////////////
// Compact Exp Clover
//////////////////////////////////
template<class Impl>
class CompactExpCloverHelpers: public CompactWilsonCloverHelpers<Impl> {
public:
INHERIT_IMPL_TYPES(Impl);
INHERIT_CLOVER_TYPES(Impl);
INHERIT_COMPACT_CLOVER_TYPES(Impl);
template <typename vtype> using iImplClover = iScalar<iMatrix<iMatrix<vtype, Impl::Dimension>, Ns>>;
typedef CompactWilsonCloverHelpers<Impl> CompactHelpers;
static void MassTerm(CloverField& Clover, RealD diag_mass) {
// do nothing!
// mass term is multiplied to exp(Clover) below
}
static int getNMAX(RealD prec, RealD R) {
/* compute stop condition for exponential */
int NMAX=1;
RealD cond=R*R/2.;
while (cond*std::exp(R)>prec) {
NMAX++;
cond*=R/(double)(NMAX+1);
}
return NMAX;
}
static int getNMAX(Lattice<iImplCloverDiagonal<vComplexD>> &t, RealD R) {return getNMAX(1e-12,R);}
static int getNMAX(Lattice<iImplCloverDiagonal<vComplexF>> &t, RealD R) {return getNMAX(1e-6,R);}
static void ExponentiateHermitean6by6(const iMatrix<ComplexD,6> &arg, const RealD& alpha, const std::vector<RealD>& cN, const int Niter, iMatrix<ComplexD,6>& dest){
typedef iMatrix<ComplexD,6> mat;
RealD qn[6];
RealD qnold[6];
RealD p[5];
RealD trA2, trA3, trA4;
mat A2, A3, A4, A5;
A2 = alpha * alpha * arg * arg;
A3 = alpha * arg * A2;
A4 = A2 * A2;
A5 = A2 * A3;
trA2 = toReal( trace(A2) );
trA3 = toReal( trace(A3) );
trA4 = toReal( trace(A4));
p[0] = toReal( trace(A3 * A3)) / 6.0 - 0.125 * trA4 * trA2 - trA3 * trA3 / 18.0 + trA2 * trA2 * trA2/ 48.0;
p[1] = toReal( trace(A5)) / 5.0 - trA3 * trA2 / 6.0;
p[2] = toReal( trace(A4)) / 4.0 - 0.125 * trA2 * trA2;
p[3] = trA3 / 3.0;
p[4] = 0.5 * trA2;
qnold[0] = cN[Niter];
qnold[1] = 0.0;
qnold[2] = 0.0;
qnold[3] = 0.0;
qnold[4] = 0.0;
qnold[5] = 0.0;
for(int i = Niter-1; i >= 0; i--)
{
qn[0] = p[0] * qnold[5] + cN[i];
qn[1] = p[1] * qnold[5] + qnold[0];
qn[2] = p[2] * qnold[5] + qnold[1];
qn[3] = p[3] * qnold[5] + qnold[2];
qn[4] = p[4] * qnold[5] + qnold[3];
qn[5] = qnold[4];
qnold[0] = qn[0];
qnold[1] = qn[1];
qnold[2] = qn[2];
qnold[3] = qn[3];
qnold[4] = qn[4];
qnold[5] = qn[5];
}
mat unit(1.0);
dest = (qn[0] * unit + qn[1] * alpha * arg + qn[2] * A2 + qn[3] * A3 + qn[4] * A4 + qn[5] * A5);
}
static void Exponentiate_Clover(CloverDiagonalField& Diagonal, CloverTriangleField& Triangle, RealD csw_t, RealD diag_mass) {
GridBase* grid = Diagonal.Grid();
int NMAX = getNMAX(Diagonal, 3.*csw_t/diag_mass);
//
// Implementation completely in Daniel's layout
//
// Taylor expansion with Cayley-Hamilton recursion
// underlying Horner scheme as above
std::vector<RealD> cn(NMAX+1);
cn[0] = 1.0;
for (int i=1; i<=NMAX; i++){
cn[i] = cn[i-1] / RealD(i);
}
// Taken over from Daniel's implementation
conformable(Diagonal, Triangle);
long lsites = grid->lSites();
{
typedef typename SiteCloverDiagonal::scalar_object scalar_object_diagonal;
typedef typename SiteCloverTriangle::scalar_object scalar_object_triangle;
typedef iMatrix<ComplexD,6> mat;
autoView(diagonal_v, Diagonal, CpuRead);
autoView(triangle_v, Triangle, CpuRead);
autoView(diagonalExp_v, Diagonal, CpuWrite);
autoView(triangleExp_v, Triangle, CpuWrite);
thread_for(site, lsites, { // NOTE: Not on GPU because of (peek/poke)LocalSite
mat srcCloverOpUL(0.0); // upper left block
mat srcCloverOpLR(0.0); // lower right block
mat ExpCloverOp;
scalar_object_diagonal diagonal_tmp = Zero();
scalar_object_diagonal diagonal_exp_tmp = Zero();
scalar_object_triangle triangle_tmp = Zero();
scalar_object_triangle triangle_exp_tmp = Zero();
Coordinate lcoor;
grid->LocalIndexToLocalCoor(site, lcoor);
peekLocalSite(diagonal_tmp, diagonal_v, lcoor);
peekLocalSite(triangle_tmp, triangle_v, lcoor);
int block;
block = 0;
for(int i = 0; i < 6; i++){
for(int j = 0; j < 6; j++){
if (i == j){
srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
}
else{
srcCloverOpUL(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
}
}
}
block = 1;
for(int i = 0; i < 6; i++){
for(int j = 0; j < 6; j++){
if (i == j){
srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(diagonal_tmp()(block)(i)));
}
else{
srcCloverOpLR(i,j) = static_cast<ComplexD>(TensorRemove(CompactHelpers::triangle_elem(triangle_tmp, block, i, j)));
}
}
}
// exp(Clover)
ExponentiateHermitean6by6(srcCloverOpUL,1.0/diag_mass,cn,NMAX,ExpCloverOp);
block = 0;
for(int i = 0; i < 6; i++){
for(int j = 0; j < 6; j++){
if (i == j){
diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
}
else if(i < j){
triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
}
}
}
ExponentiateHermitean6by6(srcCloverOpLR,1.0/diag_mass,cn,NMAX,ExpCloverOp);
block = 1;
for(int i = 0; i < 6; i++){
for(int j = 0; j < 6; j++){
if (i == j){
diagonal_exp_tmp()(block)(i) = ExpCloverOp(i,j);
}
else if(i < j){
triangle_exp_tmp()(block)(CompactHelpers::triangle_index(i, j)) = ExpCloverOp(i,j);
}
}
}
pokeLocalSite(diagonal_exp_tmp, diagonalExp_v, lcoor);
pokeLocalSite(triangle_exp_tmp, triangleExp_v, lcoor);
});
}
Diagonal *= diag_mass;
Triangle *= diag_mass;
}
static GaugeLinkField Cmunu(std::vector<GaugeLinkField> &U, GaugeLinkField &lambda, int mu, int nu) {
assert(0);
return lambda;
}
};
NAMESPACE_END(Grid);

3
Grid/qcd/action/fermion/CompactWilsonCloverFermion.h
View File

@ -31,6 +31,7 @@
#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
#include <Grid/qcd/action/fermion/CloverHelpers.h>
NAMESPACE_BEGIN(Grid);
@ -85,7 +86,7 @@ NAMESPACE_BEGIN(Grid);
// + (2 * 1 + 4 * 1/2) triangle parts = 4 triangle parts = 60 complex words per site
// = 84 complex words per site
template<class Impl>
template<class Impl, class CloverHelpers>
class CompactWilsonCloverFermion : public WilsonFermion<Impl>,
public WilsonCloverHelpers<Impl>,
public CompactWilsonCloverHelpers<Impl> {

62
Grid/qcd/action/fermion/Fermion.h
View File

@ -138,38 +138,52 @@ typedef WilsonTMFermion<WilsonImplF> WilsonTMFermionF;
typedef WilsonTMFermion<WilsonImplD> WilsonTMFermionD;
// Clover fermions
typedef WilsonCloverFermion<WilsonImplR> WilsonCloverFermionR;
typedef WilsonCloverFermion<WilsonImplF> WilsonCloverFermionF;
typedef WilsonCloverFermion<WilsonImplD> WilsonCloverFermionD;
template <typename WImpl> using WilsonClover = WilsonCloverFermion<WImpl, CloverHelpers<WImpl>>;
template <typename WImpl> using WilsonExpClover = WilsonCloverFermion<WImpl, ExpCloverHelpers<WImpl>>;
typedef WilsonCloverFermion<WilsonAdjImplR> WilsonCloverAdjFermionR;
typedef WilsonCloverFermion<WilsonAdjImplF> WilsonCloverAdjFermionF;
typedef WilsonCloverFermion<WilsonAdjImplD> WilsonCloverAdjFermionD;
typedef WilsonClover<WilsonImplR> WilsonCloverFermionR;
typedef WilsonClover<WilsonImplF> WilsonCloverFermionF;
typedef WilsonClover<WilsonImplD> WilsonCloverFermionD;
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
typedef WilsonCloverFermion<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
typedef WilsonExpClover<WilsonImplR> WilsonExpCloverFermionR;
typedef WilsonExpClover<WilsonImplF> WilsonExpCloverFermionF;
typedef WilsonExpClover<WilsonImplD> WilsonExpCloverFermionD;
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
typedef WilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
typedef WilsonClover<WilsonAdjImplR> WilsonCloverAdjFermionR;
typedef WilsonClover<WilsonAdjImplF> WilsonCloverAdjFermionF;
typedef WilsonClover<WilsonAdjImplD> WilsonCloverAdjFermionD;
typedef WilsonClover<WilsonTwoIndexSymmetricImplR> WilsonCloverTwoIndexSymmetricFermionR;
typedef WilsonClover<WilsonTwoIndexSymmetricImplF> WilsonCloverTwoIndexSymmetricFermionF;
typedef WilsonClover<WilsonTwoIndexSymmetricImplD> WilsonCloverTwoIndexSymmetricFermionD;
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplR> WilsonCloverTwoIndexAntiSymmetricFermionR;
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplF> WilsonCloverTwoIndexAntiSymmetricFermionF;
typedef WilsonClover<WilsonTwoIndexAntiSymmetricImplD> WilsonCloverTwoIndexAntiSymmetricFermionD;
// Compact Clover fermions
typedef CompactWilsonCloverFermion<WilsonImplR> CompactWilsonCloverFermionR;
typedef CompactWilsonCloverFermion<WilsonImplF> CompactWilsonCloverFermionF;
typedef CompactWilsonCloverFermion<WilsonImplD> CompactWilsonCloverFermionD;
template <typename WImpl> using CompactWilsonClover = CompactWilsonCloverFermion<WImpl, CompactCloverHelpers<WImpl>>;
template <typename WImpl> using CompactWilsonExpClover = CompactWilsonCloverFermion<WImpl, CompactExpCloverHelpers<WImpl>>;
typedef CompactWilsonCloverFermion<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
typedef CompactWilsonCloverFermion<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
typedef CompactWilsonCloverFermion<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
typedef CompactWilsonClover<WilsonImplR> CompactWilsonCloverFermionR;
typedef CompactWilsonClover<WilsonImplF> CompactWilsonCloverFermionF;
typedef CompactWilsonClover<WilsonImplD> CompactWilsonCloverFermionD;
typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
typedef CompactWilsonCloverFermion<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
typedef CompactWilsonExpClover<WilsonImplR> CompactWilsonExpCloverFermionR;
typedef CompactWilsonExpClover<WilsonImplF> CompactWilsonExpCloverFermionF;
typedef CompactWilsonExpClover<WilsonImplD> CompactWilsonExpCloverFermionD;
typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
typedef CompactWilsonCloverFermion<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
typedef CompactWilsonClover<WilsonAdjImplR> CompactWilsonCloverAdjFermionR;
typedef CompactWilsonClover<WilsonAdjImplF> CompactWilsonCloverAdjFermionF;
typedef CompactWilsonClover<WilsonAdjImplD> CompactWilsonCloverAdjFermionD;
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplR> CompactWilsonCloverTwoIndexSymmetricFermionR;
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplF> CompactWilsonCloverTwoIndexSymmetricFermionF;
typedef CompactWilsonClover<WilsonTwoIndexSymmetricImplD> CompactWilsonCloverTwoIndexSymmetricFermionD;
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplR> CompactWilsonCloverTwoIndexAntiSymmetricFermionR;
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplF> CompactWilsonCloverTwoIndexAntiSymmetricFermionF;
typedef CompactWilsonClover<WilsonTwoIndexAntiSymmetricImplD> CompactWilsonCloverTwoIndexAntiSymmetricFermionD;
// Domain Wall fermions
typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;

3
Grid/qcd/action/fermion/WilsonCloverFermion.h
View File

@ -32,6 +32,7 @@
#include <Grid/qcd/action/fermion/WilsonCloverTypes.h>
#include <Grid/qcd/action/fermion/WilsonCloverHelpers.h>
#include <Grid/qcd/action/fermion/CloverHelpers.h>
NAMESPACE_BEGIN(Grid);
@ -51,7 +52,7 @@ NAMESPACE_BEGIN(Grid);
// csw_r = csw_t to recover the isotropic version
//////////////////////////////////////////////////////////////////
template <class Impl>
template<class Impl, class CloverHelpers>
class WilsonCloverFermion : public WilsonFermion<Impl>,
public WilsonCloverHelpers<Impl>
{

2
Grid/qcd/action/fermion/WilsonCloverHelpers.h
View File

@ -209,6 +209,8 @@ public:
};
////////////////////////////////////////////////////////
template<class Impl> class CompactWilsonCloverHelpers {
public:

2
Grid/qcd/action/fermion/WilsonCloverTypes.h
View File

@ -47,8 +47,6 @@ class CompactWilsonCloverTypes {
public:
INHERIT_IMPL_TYPES(Impl);
static_assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3, "Wrong dimensions");
static constexpr int Nred = Nc * Nhs; // 6
static constexpr int Nblock = Nhs; // 2
static constexpr int Ndiagonal = Nred; // 6

42
Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
View File

@ -47,7 +47,7 @@ CayleyFermion5D<Impl>::CayleyFermion5D(GaugeField &_Umu,
FiveDimRedBlackGrid,
FourDimGrid,
FourDimRedBlackGrid,_M5,p),
mass(_mass)
mass_plus(_mass), mass_minus(_mass)
{
}
@ -209,8 +209,8 @@ void CayleyFermion5D<Impl>::M5D (const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
Vector<Coeff_t> diag (Ls,1.0);
Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass;
Vector<Coeff_t> lower(Ls,-1.0); lower[0] =mass;
Vector<Coeff_t> upper(Ls,-1.0); upper[Ls-1]=mass_minus;
Vector<Coeff_t> lower(Ls,-1.0); lower[0] =mass_plus;
M5D(psi,chi,chi,lower,diag,upper);
}
template<class Impl>
@ -220,8 +220,8 @@ void CayleyFermion5D<Impl>::Meooe5D (const FermionField &psi, FermionField &D
Vector<Coeff_t> diag = bs;
Vector<Coeff_t> upper= cs;
Vector<Coeff_t> lower= cs;
upper[Ls-1]=-mass*upper[Ls-1];
lower[0] =-mass*lower[0];
upper[Ls-1]=-mass_minus*upper[Ls-1];
lower[0] =-mass_plus*lower[0];
M5D(psi,psi,Din,lower,diag,upper);
}
// FIXME Redunant with the above routine; check this and eliminate
@ -235,8 +235,8 @@ template<class Impl> void CayleyFermion5D<Impl>::Meo5D (const FermionField &
upper[i]=-ceo[i];
lower[i]=-ceo[i];
}
upper[Ls-1]=-mass*upper[Ls-1];
lower[0] =-mass*lower[0];
upper[Ls-1]=-mass_minus*upper[Ls-1];
lower[0] =-mass_plus*lower[0];
M5D(psi,psi,chi,lower,diag,upper);
}
template<class Impl>
@ -250,8 +250,8 @@ void CayleyFermion5D<Impl>::Mooee (const FermionField &psi, FermionField &
upper[i]=-cee[i];
lower[i]=-cee[i];
}
upper[Ls-1]=-mass*upper[Ls-1];
lower[0] =-mass*lower[0];
upper[Ls-1]=-mass_minus*upper[Ls-1];
lower[0] =-mass_plus*lower[0];
M5D(psi,psi,chi,lower,diag,upper);
}
template<class Impl>
@ -266,9 +266,9 @@ void CayleyFermion5D<Impl>::MooeeDag (const FermionField &psi, FermionField &
// Assemble the 5d matrix
if ( s==0 ) {
upper[s] = -cee[s+1] ;
lower[s] = mass*cee[Ls-1];
lower[s] = mass_minus*cee[Ls-1];
} else if ( s==(Ls-1)) {
upper[s] = mass*cee[0];
upper[s] = mass_plus*cee[0];
lower[s] = -cee[s-1];
} else {
upper[s]=-cee[s+1];
@ -291,8 +291,8 @@ void CayleyFermion5D<Impl>::M5Ddag (const FermionField &psi, FermionField &chi)
Vector<Coeff_t> diag(Ls,1.0);
Vector<Coeff_t> upper(Ls,-1.0);
Vector<Coeff_t> lower(Ls,-1.0);
upper[Ls-1]=-mass*upper[Ls-1];
lower[0] =-mass*lower[0];
upper[Ls-1]=-mass_plus*upper[Ls-1];
lower[0] =-mass_minus*lower[0];
M5Ddag(psi,chi,chi,lower,diag,upper);
}
@ -307,9 +307,9 @@ void CayleyFermion5D<Impl>::MeooeDag5D (const FermionField &psi, FermionField
for (int s=0;s<Ls;s++){
if ( s== 0 ) {
upper[s] = cs[s+1];
lower[s] =-mass*cs[Ls-1];
lower[s] =-mass_minus*cs[Ls-1];
} else if ( s==(Ls-1) ) {
upper[s] =-mass*cs[0];
upper[s] =-mass_plus*cs[0];
lower[s] = cs[s-1];
} else {
upper[s] = cs[s+1];
@ -552,7 +552,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
lee[i] =-cee[i+1]/bee[i]; // sub-diag entry on the ith column
leem[i]=mass*cee[Ls-1]/bee[0];
leem[i]=mass_minus*cee[Ls-1]/bee[0];
for(int j=0;j<i;j++) {
assert(bee[j+1]!=Coeff_t(0.0));
leem[i]*= aee[j]/bee[j+1];
@ -560,7 +560,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
uee[i] =-aee[i]/bee[i]; // up-diag entry on the ith row
ueem[i]=mass;
ueem[i]=mass_plus;
for(int j=1;j<=i;j++) ueem[i]*= cee[j]/bee[j];
ueem[i]*= aee[0]/bee[0];
@ -573,7 +573,7 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,Vector<Coeff_t
}
{
Coeff_t delta_d=mass*cee[Ls-1];
Coeff_t delta_d=mass_minus*cee[Ls-1];
for(int j=0;j<Ls-1;j++) {
assert(bee[j] != Coeff_t(0.0));
delta_d *= cee[j]/bee[j];
@ -642,6 +642,10 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
Current curr_type,
unsigned int mu)
{
assert(mass_plus == mass_minus);
RealD mass = mass_plus;
#if (!defined(GRID_HIP))
Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX,
@ -777,6 +781,8 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
assert(mu>=0);
assert(mu<Nd);
assert(mass_plus == mass_minus);
RealD mass = mass_plus;
#if 0
int tshift = (mu == Nd-1) ? 1 : 0;

154
Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h
View File

@ -32,17 +32,18 @@
#include <Grid/qcd/spin/Dirac.h>
#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
NAMESPACE_BEGIN(Grid);
template<class Impl>
CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
GridCartesian& Fgrid,
GridRedBlackCartesian& Hgrid,
const RealD _mass,
const RealD _csw_r,
const RealD _csw_t,
const RealD _cF,
const WilsonAnisotropyCoefficients& clover_anisotropy,
const ImplParams& impl_p)
template<class Impl, class CloverHelpers>
CompactWilsonCloverFermion<Impl, CloverHelpers>::CompactWilsonCloverFermion(GaugeField& _Umu,
GridCartesian& Fgrid,
GridRedBlackCartesian& Hgrid,
const RealD _mass,
const RealD _csw_r,
const RealD _csw_t,
const RealD _cF,
const WilsonAnisotropyCoefficients& clover_anisotropy,
const ImplParams& impl_p)
: WilsonBase(_Umu, Fgrid, Hgrid, _mass, impl_p, clover_anisotropy)
, csw_r(_csw_r)
, csw_t(_csw_t)
@ -58,50 +59,55 @@ CompactWilsonCloverFermion<Impl>::CompactWilsonCloverFermion(GaugeField& _Umu,
, BoundaryMask(&Fgrid)
, BoundaryMaskEven(&Hgrid), BoundaryMaskOdd(&Hgrid)
{
assert(Nd == 4 && Nc == 3 && Ns == 4 && Impl::Dimension == 3);
csw_r *= 0.5;
csw_t *= 0.5;
if (clover_anisotropy.isAnisotropic)
csw_r /= clover_anisotropy.xi_0;
ImportGauge(_Umu);
if (open_boundaries)
if (open_boundaries) {
this->BoundaryMaskEven.Checkerboard() = Even;
this->BoundaryMaskOdd.Checkerboard() = Odd;
CompactHelpers::SetupMasks(this->BoundaryMask, this->BoundaryMaskEven, this->BoundaryMaskOdd);
}
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::Dhop(const FermionField& in, FermionField& out, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Dhop(const FermionField& in, FermionField& out, int dag) {
WilsonBase::Dhop(in, out, dag);
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::DhopOE(const FermionField& in, FermionField& out, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopOE(const FermionField& in, FermionField& out, int dag) {
WilsonBase::DhopOE(in, out, dag);
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::DhopEO(const FermionField& in, FermionField& out, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopEO(const FermionField& in, FermionField& out, int dag) {
WilsonBase::DhopEO(in, out, dag);
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDir(const FermionField& in, FermionField& out, int dir, int disp) {
WilsonBase::DhopDir(in, out, dir, disp);
if(this->open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::DhopDirAll(const FermionField& in, std::vector<FermionField>& out) {
WilsonBase::DhopDirAll(in, out);
if(this->open_boundaries) {
for(auto& o : out) ApplyBoundaryMask(o);
}
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::M(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField& in, FermionField& out) {
out.Checkerboard() = in.Checkerboard();
WilsonBase::Dhop(in, out, DaggerNo); // call base to save applying bc
Mooee(in, Tmp);
@ -109,8 +115,8 @@ void CompactWilsonCloverFermion<Impl>::M(const FermionField& in, FermionField& o
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::Mdag(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField& in, FermionField& out) {
out.Checkerboard() = in.Checkerboard();
WilsonBase::Dhop(in, out, DaggerYes); // call base to save applying bc
MooeeDag(in, Tmp);
@ -118,20 +124,20 @@ void CompactWilsonCloverFermion<Impl>::Mdag(const FermionField& in, FermionField
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::Meooe(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Meooe(const FermionField& in, FermionField& out) {
WilsonBase::Meooe(in, out);
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MeooeDag(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeooeDag(const FermionField& in, FermionField& out) {
WilsonBase::MeooeDag(in, out);
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField& in, FermionField& out) {
if(in.Grid()->_isCheckerBoarded) {
if(in.Checkerboard() == Odd) {
MooeeInternal(in, out, DiagonalOdd, TriangleOdd);
@ -144,13 +150,13 @@ void CompactWilsonCloverFermion<Impl>::Mooee(const FermionField& in, FermionFiel
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MooeeDag(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField& in, FermionField& out) {
Mooee(in, out); // blocks are hermitian
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField& in, FermionField& out) {
if(in.Grid()->_isCheckerBoarded) {
if(in.Checkerboard() == Odd) {
MooeeInternal(in, out, DiagonalInvOdd, TriangleInvOdd);
@ -163,23 +169,23 @@ void CompactWilsonCloverFermion<Impl>::MooeeInv(const FermionField& in, FermionF
if(open_boundaries) ApplyBoundaryMask(out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MooeeInvDag(const FermionField& in, FermionField& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField& in, FermionField& out) {
MooeeInv(in, out); // blocks are hermitian
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::Mdir(const FermionField& in, FermionField& out, int dir, int disp) {
DhopDir(in, out, dir, disp);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MdirAll(const FermionField& in, std::vector<FermionField>& out) {
DhopDirAll(in, out);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField& force, const FermionField& X, const FermionField& Y, int dag) {
assert(!open_boundaries); // TODO check for changes required for open bc
// NOTE: code copied from original clover term
@ -251,7 +257,7 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
}
PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
Impl::TraceSpinImpl(lambda, Slambda); // traceSpin ok
force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu); // checked
force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu); // checked
count++;
}
@ -261,18 +267,18 @@ void CompactWilsonCloverFermion<Impl>::MDeriv(GaugeField& force, const FermionFi
force += clover_force;
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
assert(0);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag) {
assert(0);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField& in,
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField& in,
FermionField& out,
const CloverDiagonalField& diagonal,
const CloverTriangleField& triangle) {
@ -285,8 +291,8 @@ void CompactWilsonCloverFermion<Impl>::MooeeInternal(const FermionField&
CompactHelpers::MooeeKernel(diagonal.oSites(), 1, in, out, diagonal, triangle);
}
template<class Impl>
void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
template<class Impl, class CloverHelpers>
void CompactWilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField& _Umu) {
// NOTE: parts copied from original implementation
// Import gauge into base class
@ -318,22 +324,27 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
TmpOriginal += Helpers::fillCloverXT(Ex) * csw_t;
TmpOriginal += Helpers::fillCloverYT(Ey) * csw_t;
TmpOriginal += Helpers::fillCloverZT(Ez) * csw_t;
TmpOriginal += this->diag_mass;
// Handle mass term based on clover policy
CloverHelpers::MassTerm(TmpOriginal, this->diag_mass);
// Convert the data layout of the clover term
double t4 = usecond();
CompactHelpers::ConvertLayout(TmpOriginal, Diagonal, Triangle);
// Possible modify the boundary values
// Exponentiate the clover (nothing happens in case of the standard clover)
double t5 = usecond();
CloverHelpers::Exponentiate_Clover(Diagonal, Triangle, csw_t, this->diag_mass);
// Possible modify the boundary values
double t6 = usecond();
if(open_boundaries) CompactHelpers::ModifyBoundaries(Diagonal, Triangle, csw_t, cF, this->diag_mass);
// Invert the clover term in the improved layout
double t6 = usecond();
// Invert the Clover term (explicit inversion needed for the improvement in case of open boundary conditions)
double t7 = usecond();
CompactHelpers::Invert(Diagonal, Triangle, DiagonalInv, TriangleInv);
// Fill the remaining clover fields
double t7 = usecond();
double t8 = usecond();
pickCheckerboard(Even, DiagonalEven, Diagonal);
pickCheckerboard(Even, TriangleEven, Triangle);
pickCheckerboard(Odd, DiagonalOdd, Diagonal);
@ -344,20 +355,19 @@ void CompactWilsonCloverFermion<Impl>::ImportGauge(const GaugeField& _Umu) {
pickCheckerboard(Odd, TriangleInvOdd, TriangleInv);
// Report timings
double t8 = usecond();
#if 0
std::cout << GridLogMessage << "CompactWilsonCloverFermion::ImportGauge timings:"
<< " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
<< ", allocations = " << (t2 - t1) / 1e6
<< ", field strength = " << (t3 - t2) / 1e6
<< ", fill clover = " << (t4 - t3) / 1e6
<< ", convert = " << (t5 - t4) / 1e6
<< ", boundaries = " << (t6 - t5) / 1e6
<< ", inversions = " << (t7 - t6) / 1e6
<< ", pick cbs = " << (t8 - t7) / 1e6
<< ", total = " << (t8 - t0) / 1e6
<< std::endl;
#endif
double t9 = usecond();
std::cout << GridLogDebug << "CompactWilsonCloverFermion::ImportGauge timings:" << std::endl;
std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
std::cout << GridLogDebug << "allocations = " << (t2 - t1) / 1e6 << std::endl;
std::cout << GridLogDebug << "field strength = " << (t3 - t2) / 1e6 << std::endl;
std::cout << GridLogDebug << "fill clover = " << (t4 - t3) / 1e6 << std::endl;
std::cout << GridLogDebug << "convert = " << (t5 - t4) / 1e6 << std::endl;
std::cout << GridLogDebug << "exponentiation = " << (t6 - t5) / 1e6 << std::endl;
std::cout << GridLogDebug << "boundaries = " << (t7 - t6) / 1e6 << std::endl;
std::cout << GridLogDebug << "inversions = " << (t8 - t7) / 1e6 << std::endl;
std::cout << GridLogDebug << "pick cbs = " << (t9 - t8) / 1e6 << std::endl;
std::cout << GridLogDebug << "total = " << (t9 - t0) / 1e6 << std::endl;
}
NAMESPACE_END(Grid);

112
Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h
View File

@ -34,8 +34,8 @@
NAMESPACE_BEGIN(Grid);
template<class Impl>
WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField& _Umu,
template<class Impl, class CloverHelpers>
WilsonCloverFermion<Impl, CloverHelpers>::WilsonCloverFermion(GaugeField& _Umu,
GridCartesian& Fgrid,
GridRedBlackCartesian& Hgrid,
const RealD _mass,
@ -74,8 +74,8 @@ WilsonCloverFermion<Impl>::WilsonCloverFermion(GaugeField&
}
// *NOT* EO
template <class Impl>
void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::M(const FermionField &in, FermionField &out)
{
FermionField temp(out.Grid());
@ -89,8 +89,8 @@ void WilsonCloverFermion<Impl>::M(const FermionField &in, FermionField &out)
out += temp;
}
template <class Impl>
void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::Mdag(const FermionField &in, FermionField &out)
{
FermionField temp(out.Grid());
@ -104,8 +104,8 @@ void WilsonCloverFermion<Impl>::Mdag(const FermionField &in, FermionField &out)
out += temp;
}
template <class Impl>
void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::ImportGauge(const GaugeField &_Umu)
{
double t0 = usecond();
WilsonFermion<Impl>::ImportGauge(_Umu);
@ -131,47 +131,11 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
CloverTerm += Helpers::fillCloverXT(Ex) * csw_t;
CloverTerm += Helpers::fillCloverYT(Ey) * csw_t;
CloverTerm += Helpers::fillCloverZT(Ez) * csw_t;
CloverTerm += diag_mass;
double t4 = usecond();
int lvol = _Umu.Grid()->lSites();
int DimRep = Impl::Dimension;
CloverHelpers::Instantiate(CloverTerm, CloverTermInv, csw_t, this->diag_mass);
double t5 = usecond();
{
autoView(CTv,CloverTerm,CpuRead);
autoView(CTIv,CloverTermInv,CpuWrite);
thread_for(site, lvol, {
Coordinate lcoor;
grid->LocalIndexToLocalCoor(site, lcoor);
Eigen::MatrixXcd EigenCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
Eigen::MatrixXcd EigenInvCloverOp = Eigen::MatrixXcd::Zero(Ns * DimRep, Ns * DimRep);
typename SiteClover::scalar_object Qx = Zero(), Qxinv = Zero();
peekLocalSite(Qx, CTv, lcoor);
//if (csw!=0){
for (int j = 0; j < Ns; j++)
for (int k = 0; k < Ns; k++)
for (int a = 0; a < DimRep; a++)
for (int b = 0; b < DimRep; b++){
auto zz = Qx()(j, k)(a, b);
EigenCloverOp(a + j * DimRep, b + k * DimRep) = std::complex<double>(zz);
}
// if (site==0) std::cout << "site =" << site << "\n" << EigenCloverOp << std::endl;
EigenInvCloverOp = EigenCloverOp.inverse();
//std::cout << EigenInvCloverOp << std::endl;
for (int j = 0; j < Ns; j++)
for (int k = 0; k < Ns; k++)
for (int a = 0; a < DimRep; a++)
for (int b = 0; b < DimRep; b++)
Qxinv()(j, k)(a, b) = EigenInvCloverOp(a + j * DimRep, b + k * DimRep);
// if (site==0) std::cout << "site =" << site << "\n" << EigenInvCloverOp << std::endl;
// }
pokeLocalSite(Qxinv, CTIv, lcoor);
});
}
double t6 = usecond();
// Separate the even and odd parts
pickCheckerboard(Even, CloverTermEven, CloverTerm);
pickCheckerboard(Odd, CloverTermOdd, CloverTerm);
@ -184,48 +148,44 @@ void WilsonCloverFermion<Impl>::ImportGauge(const GaugeField &_Umu)
pickCheckerboard(Even, CloverTermInvDagEven, adj(CloverTermInv));
pickCheckerboard(Odd, CloverTermInvDagOdd, adj(CloverTermInv));
double t7 = usecond();
double t6 = usecond();
#if 0
std::cout << GridLogMessage << "WilsonCloverFermion::ImportGauge timings:"
<< " WilsonFermion::Importgauge = " << (t1 - t0) / 1e6
<< ", allocations = " << (t2 - t1) / 1e6
<< ", field strength = " << (t3 - t2) / 1e6
<< ", fill clover = " << (t4 - t3) / 1e6
<< ", misc = " << (t5 - t4) / 1e6
<< ", inversions = " << (t6 - t5) / 1e6
<< ", pick cbs = " << (t7 - t6) / 1e6
<< ", total = " << (t7 - t0) / 1e6
<< std::endl;
#endif
std::cout << GridLogDebug << "WilsonCloverFermion::ImportGauge timings:" << std::endl;
std::cout << GridLogDebug << "WilsonFermion::Importgauge = " << (t1 - t0) / 1e6 << std::endl;
std::cout << GridLogDebug << "allocations = " << (t2 - t1) / 1e6 << std::endl;
std::cout << GridLogDebug << "field strength = " << (t3 - t2) / 1e6 << std::endl;
std::cout << GridLogDebug << "fill clover = " << (t4 - t3) / 1e6 << std::endl;
std::cout << GridLogDebug << "instantiation = " << (t5 - t4) / 1e6 << std::endl;
std::cout << GridLogDebug << "pick cbs = " << (t6 - t5) / 1e6 << std::endl;
std::cout << GridLogDebug << "total = " << (t6 - t0) / 1e6 << std::endl;
}
template <class Impl>
void WilsonCloverFermion<Impl>::Mooee(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::Mooee(const FermionField &in, FermionField &out)
{
this->MooeeInternal(in, out, DaggerNo, InverseNo);
}
template <class Impl>
void WilsonCloverFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeDag(const FermionField &in, FermionField &out)
{
this->MooeeInternal(in, out, DaggerYes, InverseNo);
}
template <class Impl>
void WilsonCloverFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInv(const FermionField &in, FermionField &out)
{
this->MooeeInternal(in, out, DaggerNo, InverseYes);
}
template <class Impl>
void WilsonCloverFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInvDag(const FermionField &in, FermionField &out)
{
this->MooeeInternal(in, out, DaggerYes, InverseYes);
}
template <class Impl>
void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MooeeInternal(const FermionField &in, FermionField &out, int dag, int inv)
{
out.Checkerboard() = in.Checkerboard();
CloverField *Clover;
@ -278,8 +238,8 @@ void WilsonCloverFermion<Impl>::MooeeInternal(const FermionField &in, FermionFie
} // MooeeInternal
// Derivative parts unpreconditioned pseudofermions
template <class Impl>
void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MDeriv(GaugeField &force, const FermionField &X, const FermionField &Y, int dag)
{
conformable(X.Grid(), Y.Grid());
conformable(X.Grid(), force.Grid());
@ -349,7 +309,7 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
}
PropagatorField Slambda = Gamma(sigma[count]) * Lambda; // sigma checked
Impl::TraceSpinImpl(lambda, Slambda); // traceSpin ok
force_mu -= factor*Helpers::Cmunu(U, lambda, mu, nu); // checked
force_mu -= factor*CloverHelpers::Cmunu(U, lambda, mu, nu); // checked
count++;
}
@ -360,15 +320,15 @@ void WilsonCloverFermion<Impl>::MDeriv(GaugeField &force, const FermionField &X,
}
// Derivative parts
template <class Impl>
void WilsonCloverFermion<Impl>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MooDeriv(GaugeField &mat, const FermionField &X, const FermionField &Y, int dag)
{
assert(0);
}
// Derivative parts
template <class Impl>
void WilsonCloverFermion<Impl>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
template<class Impl, class CloverHelpers>
void WilsonCloverFermion<Impl, CloverHelpers>::MeeDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag)
{
assert(0); // not implemented yet
}

85
Grid/qcd/action/fermion/implementation/WilsonFermionImplementation.h
View File

@ -4,12 +4,13 @@ Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonFermion.cc
Copyright (C) 2015
Copyright (C) 2022
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>
Author: Fabian Joswig <fabian.joswig@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
@ -599,11 +600,47 @@ void WilsonFermion<Impl>::ContractConservedCurrent(PropagatorField &q_in_1,
Current curr_type,
unsigned int mu)
{
if(curr_type != Current::Vector)
{
std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
exit(1);
}
Gamma g5(Gamma::Algebra::Gamma5);
conformable(_grid, q_in_1.Grid());
conformable(_grid, q_in_2.Grid());
conformable(_grid, q_out.Grid());
assert(0);
auto UGrid= this->GaugeGrid();
PropagatorField tmp_shifted(UGrid);
PropagatorField g5Lg5(UGrid);
PropagatorField R(UGrid);
PropagatorField gmuR(UGrid);
Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX,
Gamma::Algebra::GammaY,
Gamma::Algebra::GammaZ,
Gamma::Algebra::GammaT,
};
Gamma gmu=Gamma(Gmu[mu]);
g5Lg5=g5*q_in_1*g5;
tmp_shifted=Cshift(q_in_2,mu,1);
Impl::multLinkField(R,this->Umu,tmp_shifted,mu);
gmuR=gmu*R;
q_out=adj(g5Lg5)*R;
q_out-=adj(g5Lg5)*gmuR;
tmp_shifted=Cshift(q_in_1,mu,1);
Impl::multLinkField(g5Lg5,this->Umu,tmp_shifted,mu);
g5Lg5=g5*g5Lg5*g5;
R=q_in_2;
gmuR=gmu*R;
q_out-=adj(g5Lg5)*R;
q_out-=adj(g5Lg5)*gmuR;
}
@ -617,9 +654,51 @@ void WilsonFermion<Impl>::SeqConservedCurrent(PropagatorField &q_in,
unsigned int tmax,
ComplexField &lattice_cmplx)
{
if(curr_type != Current::Vector)
{
std::cout << GridLogError << "Only the conserved vector current is implemented so far." << std::endl;
exit(1);
}
int tshift = (mu == Nd-1) ? 1 : 0;
unsigned int LLt = GridDefaultLatt()[Tp];
conformable(_grid, q_in.Grid());
conformable(_grid, q_out.Grid());
assert(0);
auto UGrid= this->GaugeGrid();
PropagatorField tmp(UGrid);
PropagatorField Utmp(UGrid);
PropagatorField L(UGrid);
PropagatorField zz (UGrid);
zz=Zero();
LatticeInteger lcoor(UGrid); LatticeCoordinate(lcoor,Nd-1);
Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX,
Gamma::Algebra::GammaY,
Gamma::Algebra::GammaZ,
Gamma::Algebra::GammaT,
};
Gamma gmu=Gamma(Gmu[mu]);
tmp = Cshift(q_in,mu,1);
Impl::multLinkField(Utmp,this->Umu,tmp,mu);
tmp = ( Utmp*lattice_cmplx - gmu*Utmp*lattice_cmplx ); // Forward hop
tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
q_out = where((lcoor<=tmax),tmp,zz); // Position of current complicated
tmp = q_in *lattice_cmplx;
tmp = Cshift(tmp,mu,-1);
Impl::multLinkField(Utmp,this->Umu,tmp,mu+Nd); // Adjoint link
tmp = -( Utmp + gmu*Utmp );
// Mask the time
if (tmax == LLt - 1 && tshift == 1){ // quick fix to include timeslice 0 if tmax + tshift is over the last timeslice
unsigned int t0 = 0;
tmp = where(((lcoor==t0) || (lcoor>=tmin+tshift)),tmp,zz);
} else {
tmp = where((lcoor>=tmin+tshift),tmp,zz);
}
q_out+= where((lcoor<=tmax+tshift),tmp,zz); // Position of current complicated
}
NAMESPACE_END(Grid);

5
Grid/qcd/action/fermion/instantiation/CompactWilsonCloverFermionInstantiation.cc.master
View File

@ -9,6 +9,7 @@
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@ed.ac.uk>
Author: Daniel Richtmann <daniel.richtmann@gmail.com>
Author: Mattia Bruno <mattia.bruno@cern.ch>
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
@ -32,10 +33,12 @@
#include <Grid/qcd/spin/Dirac.h>
#include <Grid/qcd/action/fermion/CompactWilsonCloverFermion.h>
#include <Grid/qcd/action/fermion/implementation/CompactWilsonCloverFermionImplementation.h>
#include <Grid/qcd/action/fermion/CloverHelpers.h>
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class CompactWilsonCloverFermion<IMPLEMENTATION>;
template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactCloverHelpers<IMPLEMENTATION>>;
template class CompactWilsonCloverFermion<IMPLEMENTATION, CompactExpCloverHelpers<IMPLEMENTATION>>;
NAMESPACE_END(Grid);

51
Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonAdjImplD/WilsonKernelsInstantiationWilsonAdjImplD.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonAdjImplF/WilsonKernelsInstantiationWilsonAdjImplF.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

7
Grid/qcd/action/fermion/instantiation/WilsonCloverFermionInstantiation.cc.master
View File

@ -8,7 +8,8 @@
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@ed.ac.uk>
Author: Mattia Bruno <mattia.bruno@cern.ch>
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
@ -31,10 +32,12 @@
#include <Grid/qcd/spin/Dirac.h>
#include <Grid/qcd/action/fermion/WilsonCloverFermion.h>
#include <Grid/qcd/action/fermion/implementation/WilsonCloverFermionImplementation.h>
#include <Grid/qcd/action/fermion/CloverHelpers.h>
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonCloverFermion<IMPLEMENTATION>;
template class WilsonCloverFermion<IMPLEMENTATION, CloverHelpers<IMPLEMENTATION>>;
template class WilsonCloverFermion<IMPLEMENTATION, ExpCloverHelpers<IMPLEMENTATION>>;
NAMESPACE_END(Grid);

51
Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonImplD/WilsonKernelsInstantiationWilsonImplD.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonImplF/WilsonKernelsInstantiationWilsonImplF.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplD.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexAntiSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexAntiSymmetricImplF.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplD/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplD.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/WilsonTwoIndexSymmetricImplF/WilsonKernelsInstantiationWilsonTwoIndexSymmetricImplF.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD/WilsonKernelsInstantiationZWilsonImplD.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

51
Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc
View File

@ -1,51 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
Copyright (C) 2015, 2020
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
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/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
#ifndef AVX512
#ifndef QPX
#ifndef A64FX
#ifndef A64FXFIXEDSIZE
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
#endif
#endif
#endif
#endif
NAMESPACE_BEGIN(Grid);
#include "impl.h"
template class WilsonKernels<IMPLEMENTATION>;
NAMESPACE_END(Grid);

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplF/WilsonKernelsInstantiationZWilsonImplF.cc Symbolic link
View File

@ -0,0 +1 @@
../WilsonKernelsInstantiation.cc.master

24
Grid/qcd/action/fermion/instantiation/generate_instantiations.sh
View File

@ -18,6 +18,10 @@ WILSON_IMPL_LIST=" \
GparityWilsonImplF \
GparityWilsonImplD "
COMPACT_WILSON_IMPL_LIST=" \
WilsonImplF \
WilsonImplD "
DWF_IMPL_LIST=" \
WilsonImplF \
WilsonImplD \
@ -40,13 +44,23 @@ EOF
done
CC_LIST="WilsonCloverFermionInstantiation CompactWilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
CC_LIST="WilsonCloverFermionInstantiation WilsonFermionInstantiation WilsonKernelsInstantiation WilsonTMFermionInstantiation"
for impl in $WILSON_IMPL_LIST
do
for f in $CC_LIST
do
ln -f -s ../$f.cc.master $impl/$f$impl.cc
ln -f -s ../$f.cc.master $impl/$f$impl.cc
done
done
CC_LIST="CompactWilsonCloverFermionInstantiation"
for impl in $COMPACT_WILSON_IMPL_LIST
do
for f in $CC_LIST
do
ln -f -s ../$f.cc.master $impl/$f$impl.cc
done
done
@ -63,14 +77,14 @@ for impl in $DWF_IMPL_LIST $GDWF_IMPL_LIST
do
for f in $CC_LIST
do
ln -f -s ../$f.cc.master $impl/$f$impl.cc
ln -f -s ../$f.cc.master $impl/$f$impl.cc
done
done
# overwrite the .cc file in Gparity directories
for impl in $GDWF_IMPL_LIST
do
ln -f -s ../WilsonKernelsInstantiationGparity.cc.master $impl/WilsonKernelsInstantiation$impl.cc
ln -f -s ../WilsonKernelsInstantiationGparity.cc.master $impl/WilsonKernelsInstantiation$impl.cc
done
@ -84,7 +98,7 @@ for impl in $STAG_IMPL_LIST
do
for f in $CC_LIST
do
ln -f -s ../$f.cc.master $impl/$f$impl.cc
ln -f -s ../$f.cc.master $impl/$f$impl.cc
done
done

2
Grid/qcd/action/filters/DirichletFilter.h
View File

@ -55,7 +55,7 @@ struct DirichletFilter: public MomentumFilterBase<MomentaField>
for(int mu=0;mu<Nd;mu++) {
if ( (Block[mu]) && (Block[mu] < grid->GlobalDimensions()[mu] ) ) {
// If costly could provide Grid earlier and precompute masks
std::cout << " Dirichlet in mu="<<mu<<std::endl;
std::cout << GridLogMessage << " Dirichlet in mu="<<mu<<std::endl;
LatticeCoordinate(coor,mu);
auto P_mu = PeekIndex<LorentzIndex>(P, mu);
P_mu = where(mod(coor,Block[mu])==Integer(Block[mu]-1),zz,P_mu);

8
Grid/qcd/utils/GaugeFix.h
View File

@ -55,12 +55,12 @@ public:
}
}
static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
static void SteepestDescentGaugeFix(GaugeLorentz &Umu,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
GridBase *grid = Umu.Grid();
GaugeMat xform(grid);
SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog);
SteepestDescentGaugeFix(Umu,xform,alpha,maxiter,Omega_tol,Phi_tol,Fourier,orthog,err_on_no_converge);
}
static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1) {
static void SteepestDescentGaugeFix(GaugeLorentz &Umu,GaugeMat &xform,Real & alpha,int maxiter,Real Omega_tol, Real Phi_tol,bool Fourier=false,int orthog=-1,bool err_on_no_converge=true) {
GridBase *grid = Umu.Grid();
@ -122,6 +122,8 @@ public:
}
}
std::cout << GridLogError << "Gauge fixing did not converge in " << maxiter << " iterations." << std::endl;
if (err_on_no_converge) assert(0);
};
static Real SteepestDescentStep(std::vector<GaugeMat> &U,GaugeMat &xform,Real & alpha, GaugeMat & dmuAmu,int orthog) {
GridBase *grid = U[0].Grid();

3
Grid/qcd/utils/WilsonLoops.h
View File

@ -125,7 +125,6 @@ public:
return sumplaq / vol / faces / Nc; // Nd , Nc dependent... FIXME
}
//////////////////////////////////////////////////
// average over all x,y,z the temporal loop
//////////////////////////////////////////////////
@ -165,7 +164,7 @@ public:
double vol = Umu.Grid()->gSites();
return p.real() / vol / 4.0 / 3.0;
return p.real() / vol / (4.0 * Nc ) ;
};
//////////////////////////////////////////////////

5
Grid/stencil/SimpleCompressor.h
View File

@ -52,6 +52,11 @@ public:
return arg;
}
};
class SimpleStencilParams{
public:
Coordinate dirichlet;
SimpleStencilParams() {};
};
NAMESPACE_END(Grid);

147
Grid/stencil/Stencil.h
View File

@ -131,7 +131,6 @@ class CartesianStencilAccelerator {
int _checkerboard;
int _npoints; // Move to template param?
int _osites;
int _dirichlet;
StencilVector _directions;
StencilVector _distances;
StencilVector _comms_send;
@ -503,7 +502,6 @@ public:
}
void AddCopy(void *from,void * to, Integer bytes)
{
// std::cout << "Adding CopyReceiveBuffer "<<std::hex<<from<<" "<<to<<std::dec<<" "<<bytes<<std::endl;
CopyReceiveBuffer obj;
obj.from_p = from;
obj.to_p = to;
@ -517,7 +515,7 @@ public:
cobj *from=(cobj *)CopyReceiveBuffers[i].from_p;
cobj *to =(cobj *)CopyReceiveBuffers[i].to_p;
Integer words = CopyReceiveBuffers[i].bytes/sizeof(cobj);
// std::cout << "CopyReceiveBuffer "<<std::hex<<from<<" "<<to<<std::dec<<" "<<words*sizeof(cobj)<<std::endl;
accelerator_forNB(j, words, cobj::Nsimd(), {
coalescedWrite(to[j] ,coalescedRead(from [j]));
});
@ -543,13 +541,12 @@ public:
&&(CachedTransfers[i].lane ==lane)
&&(CachedTransfers[i].cb ==cb)
){
// std::cout << "Found duplicate plane dir "<<direction<<" plane "<< OrthogPlane<< " simd "<<lane << " relproc "<<DestProc<< " bytes "<<bytes <<std::endl;
AddCopy(CachedTransfers[i].recv_buf,recv_buf,bytes);
return 1;
}
}
// std::cout << "No duplicate plane dir "<<direction<<" plane "<< OrthogPlane<< " simd "<<lane << " relproc "<<DestProc<<" bytes "<<bytes<<std::endl;
CachedTransfers.push_back(obj);
return 0;
}
@ -650,16 +647,14 @@ public:
/// Introduce a block structure and switch off comms on boundaries
void DirichletBlock(const Coordinate &dirichlet_block)
{
this->_dirichlet = 1;
for(int ii=0;ii<this->_npoints;ii++){
int dimension = this->_directions[ii];
int displacement = this->_distances[ii];
int shift = displacement;
int gd = _grid->_gdimensions[dimension];
int fd = _grid->_fdimensions[dimension];
int pd = _grid->_processors [dimension];
int ld = gd/pd;
int pc = _grid->_processor_coor[dimension];
int ld = fd/pd;
///////////////////////////////////////////
// Figure out dirichlet send and receive
// on this leg of stencil.
@ -668,7 +663,7 @@ public:
int block = dirichlet_block[dimension];
this->_comms_send[ii] = comm_dim;
this->_comms_recv[ii] = comm_dim;
if ( block ) {
if ( block && comm_dim ) {
assert(abs(displacement) < ld );
if( displacement > 0 ) {
@ -677,16 +672,16 @@ public:
// | | |
// noR
// noS
if ( (ld*(pc+1) ) % block == 0 ) this->_comms_recv[ii] = 0;
if ( ( ld*pc ) % block == 0 ) this->_comms_send[ii] = 0;
if ( ( (ld*(pc+1) ) % block ) == 0 ) this->_comms_recv[ii] = 0;
if ( ( (ld*pc ) % block ) == 0 ) this->_comms_send[ii] = 0;
} else {
// High side, low side
// | <--B--->|
// | | |
// noS
// noR
if ( (ld*(pc+1) ) % block == 0 ) this->_comms_send[ii] = 0;
if ( ( ld*pc ) % block == 0 ) this->_comms_recv[ii] = 0;
if ( ( (ld*(pc+1) ) % block ) == 0 ) this->_comms_send[ii] = 0;
if ( ( (ld*pc ) % block ) == 0 ) this->_comms_recv[ii] = 0;
}
}
}
@ -698,7 +693,6 @@ public:
const std::vector<int> &distances,
Parameters p)
{
this->_dirichlet = 0;
face_table_computed=0;
_grid = grid;
this->parameters=p;
@ -715,6 +709,8 @@ public:
this->_comms_recv.resize(npoints);
this->same_node.resize(npoints);
if ( p.dirichlet.size() ) DirichletBlock(p.dirichlet); // comms send/recv set up
_unified_buffer_size=0;
surface_list.resize(0);
@ -746,9 +742,6 @@ public:
int splice_dim = _grid->_simd_layout[dimension]>1 && (comm_dim);
int rotate_dim = _grid->_simd_layout[dimension]>2;
this->_comms_send[ii] = comm_dim;
this->_comms_recv[ii] = comm_dim;
assert ( (rotate_dim && comm_dim) == false) ; // Do not think spread out is supported
int sshift[2];
@ -911,25 +904,30 @@ public:
}
// Wrap locally dirichlet support case OR node local
if ( (offnode==0) || (comms_recv==0) ) {
if ( offnode==0 ) {
int permute_slice=0;
CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
} else {
if ( comms_recv==0 ) {
int permute_slice=1;
CopyPlane(point,dimension,x,sx,cbmask,permute_slice,wraparound);
} else {
ScatterPlane(point,dimension,x,cbmask,_unified_buffer_size,wraparound); // permute/extract/merge is done in comms phase
}
}
if ( offnode ) {
int words = buffer_size;
if (cbmask != 0x3) words=words>>1;
// int rank = grid->_processor;
// int recv_from_rank;
// int xmit_to_rank;
int unified_buffer_offset = _unified_buffer_size;
_unified_buffer_size += words;
ScatterPlane(point,dimension,x,cbmask,unified_buffer_offset,wraparound); // permute/extract/merge is done in comms phase
}
}
}
@ -1060,8 +1058,6 @@ public:
int comm_proc = ((x+sshift)/rd)%pd;
if (comm_proc) {
int words = buffer_size;
if (cbmask != 0x3) words=words>>1;
@ -1069,64 +1065,70 @@ public:
int bytes = words * compress.CommDatumSize();
int so = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
if ( !face_table_computed ) {
face_table.resize(face_idx+1);
std::vector<std::pair<int,int> > face_table_host ;
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
face_table[face_idx].resize(face_table_host.size());
acceleratorCopyToDevice(&face_table_host[0],
&face_table[face_idx][0],
face_table[face_idx].size()*sizeof(face_table_host[0]));
}
int comm_off = u_comm_offset;
// int rank = _grid->_processor;
int recv_from_rank;
int xmit_to_rank;
cobj *recv_buf;
cobj *send_buf;
_grid->ShiftedRanks(dimension,comm_proc,xmit_to_rank,recv_from_rank);
assert (xmit_to_rank != _grid->ThisRank());
assert (recv_from_rank != _grid->ThisRank());
cobj *recv_buf;
if ( compress.DecompressionStep() ) {
recv_buf=u_simd_recv_buf[0];
} else {
recv_buf=this->u_recv_buf_p;
if( comms_send ) {
if ( !face_table_computed ) {
face_table.resize(face_idx+1);
std::vector<std::pair<int,int> > face_table_host ;
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,comm_off,face_table_host);
face_table[face_idx].resize(face_table_host.size());
acceleratorCopyToDevice(&face_table_host[0],
&face_table[face_idx][0],
face_table[face_idx].size()*sizeof(face_table_host[0]));
}
if ( compress.DecompressionStep() ) {
recv_buf=u_simd_recv_buf[0];
} else {
recv_buf=this->u_recv_buf_p;
}
send_buf = this->u_send_buf_p; // Gather locally, must send
////////////////////////////////////////////////////////
// Gather locally
////////////////////////////////////////////////////////
assert(send_buf!=NULL);
Gather_plane_simple_table(face_table[face_idx],rhs,send_buf,compress,comm_off,so);
}
cobj *send_buf;
send_buf = this->u_send_buf_p; // Gather locally, must send
////////////////////////////////////////////////////////
// Gather locally
////////////////////////////////////////////////////////
assert(send_buf!=NULL);
if ( comms_send )
Gather_plane_simple_table(face_table[face_idx],rhs,send_buf,compress,u_comm_offset,so);
face_idx++;
int duplicate = CheckForDuplicate(dimension,sx,comm_proc,(void *)&recv_buf[u_comm_offset],0,bytes,cbmask);
int duplicate = CheckForDuplicate(dimension,sx,comm_proc,(void *)&recv_buf[comm_off],0,bytes,cbmask);
if ( (!duplicate) ) { // Force comms for now
///////////////////////////////////////////////////////////
// Build a list of things to do after we synchronise GPUs
// Start comms now???
///////////////////////////////////////////////////////////
AddPacket((void *)&send_buf[u_comm_offset],
(void *)&recv_buf[u_comm_offset],
AddPacket((void *)&send_buf[comm_off],
(void *)&recv_buf[comm_off],
xmit_to_rank, comms_send,
recv_from_rank, comms_recv,
bytes);
}
if ( compress.DecompressionStep() ) {
AddDecompress(&this->u_recv_buf_p[u_comm_offset],
&recv_buf[u_comm_offset],
if ( compress.DecompressionStep() && comms_recv ) {
AddDecompress(&this->u_recv_buf_p[comm_off],
&recv_buf[comm_off],
words,Decompressions);
}
u_comm_offset+=words;
}
face_idx++;
}
}
return 0;
}
@ -1155,7 +1157,6 @@ public:
int permute_type=_grid->PermuteType(dimension);
// std::cout << "SimdNew permute type "<<permute_type<<std::endl;
///////////////////////////////////////////////
// Simd direction uses an extract/merge pair
@ -1189,8 +1190,9 @@ public:
if ( any_offnode ) {
int comm_off = u_comm_offset;
for(int i=0;i<maxl;i++){
spointers[i] = (cobj *) &u_simd_send_buf[i][u_comm_offset];
spointers[i] = (cobj *) &u_simd_send_buf[i][comm_off];
}
int sx = (x+sshift)%rd;
@ -1199,15 +1201,15 @@ public:
face_table.resize(face_idx+1);
std::vector<std::pair<int,int> > face_table_host ;
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,comm_off,face_table_host);
face_table[face_idx].resize(face_table_host.size());
acceleratorCopyToDevice(&face_table_host[0],
&face_table[face_idx][0],
face_table[face_idx].size()*sizeof(face_table_host[0]));
}
// if ( comms_send )
Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type);
if ( comms_send )
Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type);
face_idx++;
//spointers[0] -- low
@ -1226,8 +1228,8 @@ public:
int nbr_plane = nbr_ic;
assert (sx == nbr_ox);
auto rp = &u_simd_recv_buf[i ][u_comm_offset];
auto sp = &u_simd_send_buf[nbr_plane][u_comm_offset];
auto rp = &u_simd_recv_buf[i ][comm_off];
auto sp = &u_simd_send_buf[nbr_plane][comm_off];
if(nbr_proc){
@ -1253,9 +1255,10 @@ public:
}
}
AddMerge(&this->u_recv_buf_p[u_comm_offset],rpointers,reduced_buffer_size,permute_type,Mergers);
AddMerge(&this->u_recv_buf_p[comm_off],rpointers,reduced_buffer_size,permute_type,Mergers);
u_comm_offset +=buffer_size;
}
}
return 0;