portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Merge branch 'develop' into feature/hmc_generalise

Browse Source
This commit is contained in:
Guido Cossu 2017-01-25 11:33:53 +00:00
commit 17629b8d9e
88 changed files with 7904 additions and 430 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -9,6 +9,7 @@
################
*~
*#
*.sublime-*
# Precompiled Headers #
#######################

2
Makefile.am
View File

@ -1,5 +1,5 @@
# additional include paths necessary to compile the C++ library
SUBDIRS = lib benchmarks tests
SUBDIRS = lib benchmarks tests extras
include $(top_srcdir)/doxygen.inc

41
benchmarks/Benchmark_mooee.cc
View File

@ -113,6 +113,36 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "Called " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
std::cout<<GridLogMessage << "******************"<<std::endl;
#define BENCH_ZDW(A,in,out) \
zDw.CayleyZeroCounters(); \
zDw. A (in,out); \
FGrid->Barrier(); \
t0=usecond(); \
for(int i=0;i<ncall;i++){ \
zDw. A (in,out); \
} \
t1=usecond(); \
FGrid->Barrier(); \
zDw.CayleyReport(); \
std::cout<<GridLogMessage << "Called ZDw " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
std::cout<<GridLogMessage << "******************"<<std::endl;
#define BENCH_DW_SSC(A,in,out) \
Dw.CayleyZeroCounters(); \
Dw. A (in,out); \
FGrid->Barrier(); \
t0=usecond(); \
for(int i=0;i<ncall;i++){ \
__SSC_START ; \
Dw. A (in,out); \
__SSC_STOP ; \
} \
t1=usecond(); \
FGrid->Barrier(); \
Dw.CayleyReport(); \
std::cout<<GridLogMessage << "Called " #A " "<< (t1-t0)/ncall<<" us"<<std::endl;\
std::cout<<GridLogMessage << "******************"<<std::endl;
#define BENCH_DW_MEO(A,in,out) \
Dw.CayleyZeroCounters(); \
Dw. A (in,out,0); \
@ -148,9 +178,15 @@ int main (int argc, char ** argv)
LatticeFermion sref(sFGrid);
LatticeFermion result(sFGrid);
std::cout<<GridLogMessage << "Constructing Vec5D Dw "<<std::endl;
DomainWallFermionVec5dR Dw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5);
RealD b=1.5;// Scale factor b+c=2, b-c=1
RealD c=0.5;
std::vector<ComplexD> gamma(Ls,std::complex<double>(1.0,0.0));
ZMobiusFermionVec5dR zDw(Umu,*sFGrid,*sFrbGrid,*sUGrid,*sUrbGrid,mass,M5,gamma,b,c);
std::cout<<GridLogMessage << "Calling Dhop "<<std::endl;
FGrid->Barrier();
@ -173,10 +209,13 @@ int main (int argc, char ** argv)
BENCH_DW_MEO(Dhop ,src,result);
BENCH_DW_MEO(DhopEO ,src_o,r_e);
BENCH_DW(Meooe ,src_o,r_e);
BENCH_DW_SSC(Meooe ,src_o,r_e);
BENCH_DW(Mooee ,src_o,r_o);
BENCH_DW(MooeeInv,src_o,r_o);
BENCH_ZDW(Mooee ,src_o,r_o);
BENCH_ZDW(MooeeInv,src_o,r_o);
}
Grid_finalize();

17
configure.ac
View File

@ -102,6 +102,13 @@ case ${ac_MKL} in
AC_DEFINE([USE_MKL], [1], [Define to 1 if you use the Intel MKL]);;
esac
############### HDF5
AC_ARG_WITH([hdf5],
[AS_HELP_STRING([--with-hdf5=prefix],
[try this for a non-standard install prefix of the HDF5 library])],
[AM_CXXFLAGS="-I$with_hdf5/include $AM_CXXFLAGS"]
[AM_LDFLAGS="-L$with_hdf5/lib $AM_LDFLAGS"])
############### first-touch
AC_ARG_ENABLE([numa],
[AC_HELP_STRING([--enable-numa=yes|no|prefix], [enable first touch numa opt])],
@ -148,6 +155,12 @@ AC_SEARCH_LIBS([fftw_execute], [fftw3],
[AC_DEFINE([HAVE_FFTW], [1], [Define to 1 if you have the `FFTW' library])]
[have_fftw=true])
AC_SEARCH_LIBS([H5Fopen], [hdf5_cpp],
[AC_DEFINE([HAVE_HDF5], [1], [Define to 1 if you have the `HDF5' library])]
[have_hdf5=true]
[LIBS="${LIBS} -lhdf5"], [], [-lhdf5])
AM_CONDITIONAL(BUILD_HDF5, [ test "${have_hdf5}X" == "trueX" ])
CXXFLAGS=$CXXFLAGS_CPY
LDFLAGS=$LDFLAGS_CPY
@ -393,10 +406,13 @@ AC_CONFIG_FILES(tests/IO/Makefile)
AC_CONFIG_FILES(tests/core/Makefile)
AC_CONFIG_FILES(tests/debug/Makefile)
AC_CONFIG_FILES(tests/forces/Makefile)
AC_CONFIG_FILES(tests/hadrons/Makefile)
AC_CONFIG_FILES(tests/hmc/Makefile)
AC_CONFIG_FILES(tests/solver/Makefile)
AC_CONFIG_FILES(tests/qdpxx/Makefile)
AC_CONFIG_FILES(benchmarks/Makefile)
AC_CONFIG_FILES(extras/Makefile)
AC_CONFIG_FILES(extras/Hadrons/Makefile)
AC_OUTPUT
git_commit=`cd $srcdir && ./scripts/configure.commit`
@ -425,6 +441,7 @@ GMP : `if test "x$have_gmp" = xtrue; then echo yes; else
LAPACK : ${ac_LAPACK}
FFTW : `if test "x$have_fftw" = xtrue; then echo yes; else echo no; fi`
LIME (ILDG support) : `if test "x$have_lime" = xtrue; then echo yes; else echo no; fi`
HDF5 : `if test "x$have_hdf5" = xtrue; then echo yes; else echo no; fi`
build DOXYGEN documentation : `if test "$DX_FLAG_doc" = '1'; then echo yes; else echo no; fi`
----- BUILD FLAGS -------------------------------------
CXXFLAGS:

317
extras/Hadrons/Application.cc Normal file
View File

@ -0,0 +1,317 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Application.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Application.hpp>
#include <Grid/Hadrons/GeneticScheduler.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
#define BIG_SEP "==============="
#define SEP "---------------"
/******************************************************************************
* Application implementation *
******************************************************************************/
// constructors ////////////////////////////////////////////////////////////////
Application::Application(void)
{
LOG(Message) << "Modules available:" << std::endl;
auto list = ModuleFactory::getInstance().getBuilderList();
for (auto &m: list)
{
LOG(Message) << " " << m << std::endl;
}
auto dim = GridDefaultLatt(), mpi = GridDefaultMpi(), loc(dim);
locVol_ = 1;
for (unsigned int d = 0; d < dim.size(); ++d)
{
loc[d] /= mpi[d];
locVol_ *= loc[d];
}
LOG(Message) << "Global lattice: " << dim << std::endl;
LOG(Message) << "MPI partition : " << mpi << std::endl;
LOG(Message) << "Local lattice : " << loc << std::endl;
}
Application::Application(const Application::GlobalPar &par)
: Application()
{
setPar(par);
}
Application::Application(const std::string parameterFileName)
: Application()
{
parameterFileName_ = parameterFileName;
}
// environment shortcut ////////////////////////////////////////////////////////
Environment & Application::env(void) const
{
return Environment::getInstance();
}
// access //////////////////////////////////////////////////////////////////////
void Application::setPar(const Application::GlobalPar &par)
{
par_ = par;
env().setSeed(strToVec<int>(par_.seed));
}
const Application::GlobalPar & Application::getPar(void)
{
return par_;
}
// execute /////////////////////////////////////////////////////////////////////
void Application::run(void)
{
if (!parameterFileName_.empty() and (env().getNModule() == 0))
{
parseParameterFile(parameterFileName_);
}
if (!scheduled_)
{
schedule();
}
printSchedule();
configLoop();
}
// parse parameter file ////////////////////////////////////////////////////////
class ObjectId: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(ObjectId,
std::string, name,
std::string, type);
};
void Application::parseParameterFile(const std::string parameterFileName)
{
XmlReader reader(parameterFileName);
GlobalPar par;
ObjectId id;
LOG(Message) << "Building application from '" << parameterFileName << "'..." << std::endl;
read(reader, "parameters", par);
setPar(par);
push(reader, "modules");
push(reader, "module");
do
{
read(reader, "id", id);
env().createModule(id.name, id.type, reader);
} while (reader.nextElement("module"));
pop(reader);
pop(reader);
}
void Application::saveParameterFile(const std::string parameterFileName)
{
XmlWriter writer(parameterFileName);
ObjectId id;
const unsigned int nMod = env().getNModule();
LOG(Message) << "Saving application to '" << parameterFileName << "'..." << std::endl;
write(writer, "parameters", getPar());
push(writer, "modules");
for (unsigned int i = 0; i < nMod; ++i)
{
push(writer, "module");
id.name = env().getModuleName(i);
id.type = env().getModule(i)->getRegisteredName();
write(writer, "id", id);
env().getModule(i)->saveParameters(writer, "options");
pop(writer);
}
pop(writer);
pop(writer);
}
// schedule computation ////////////////////////////////////////////////////////
#define MEM_MSG(size)\
sizeString((size)*locVol_) << " (" << sizeString(size) << "/site)"
#define DEFINE_MEMPEAK \
auto memPeak = [this](const std::vector<unsigned int> &program)\
{\
unsigned int memPeak;\
bool msg;\
\
msg = HadronsLogMessage.isActive();\
HadronsLogMessage.Active(false);\
env().dryRun(true);\
memPeak = env().executeProgram(program);\
env().dryRun(false);\
env().freeAll();\
HadronsLogMessage.Active(true);\
\
return memPeak;\
}
void Application::schedule(void)
{
DEFINE_MEMPEAK;
// build module dependency graph
LOG(Message) << "Building module graph..." << std::endl;
auto graph = env().makeModuleGraph();
auto con = graph.getConnectedComponents();
// constrained topological sort using a genetic algorithm
LOG(Message) << "Scheduling computation..." << std::endl;
LOG(Message) << " #module= " << graph.size() << std::endl;
LOG(Message) << " population size= " << par_.genetic.popSize << std::endl;
LOG(Message) << " max. generation= " << par_.genetic.maxGen << std::endl;
LOG(Message) << " max. cst. generation= " << par_.genetic.maxCstGen << std::endl;
LOG(Message) << " mutation rate= " << par_.genetic.mutationRate << std::endl;
unsigned int k = 0, gen, prevPeak, nCstPeak = 0;
std::random_device rd;
GeneticScheduler<unsigned int>::Parameters par;
par.popSize = par_.genetic.popSize;
par.mutationRate = par_.genetic.mutationRate;
par.seed = rd();
memPeak_ = 0;
CartesianCommunicator::BroadcastWorld(0, &(par.seed), sizeof(par.seed));
for (unsigned int i = 0; i < con.size(); ++i)
{
GeneticScheduler<unsigned int> scheduler(con[i], memPeak, par);
gen = 0;
do
{
LOG(Debug) << "Generation " << gen << ":" << std::endl;
scheduler.nextGeneration();
if (gen != 0)
{
if (prevPeak == scheduler.getMinValue())
{
nCstPeak++;
}
else
{
nCstPeak = 0;
}
}
prevPeak = scheduler.getMinValue();
if (gen % 10 == 0)
{
LOG(Iterative) << "Generation " << gen << ": "
<< MEM_MSG(scheduler.getMinValue()) << std::endl;
}
gen++;
} while ((gen < par_.genetic.maxGen)
and (nCstPeak < par_.genetic.maxCstGen));
auto &t = scheduler.getMinSchedule();
if (scheduler.getMinValue() > memPeak_)
{
memPeak_ = scheduler.getMinValue();
}
for (unsigned int j = 0; j < t.size(); ++j)
{
program_.push_back(t[j]);
}
}
scheduled_ = true;
}
void Application::saveSchedule(const std::string filename)
{
TextWriter writer(filename);
std::vector<std::string> program;
if (!scheduled_)
{
HADRON_ERROR("Computation not scheduled");
}
LOG(Message) << "Saving current schedule to '" << filename << "'..."
<< std::endl;
for (auto address: program_)
{
program.push_back(env().getModuleName(address));
}
write(writer, "schedule", program);
}
void Application::loadSchedule(const std::string filename)
{
DEFINE_MEMPEAK;
TextReader reader(filename);
std::vector<std::string> program;
LOG(Message) << "Loading schedule from '" << filename << "'..."
<< std::endl;
read(reader, "schedule", program);
program_.clear();
for (auto &name: program)
{
program_.push_back(env().getModuleAddress(name));
}
scheduled_ = true;
memPeak_ = memPeak(program_);
}
void Application::printSchedule(void)
{
if (!scheduled_)
{
HADRON_ERROR("Computation not scheduled");
}
LOG(Message) << "Schedule (memory peak: " << MEM_MSG(memPeak_) << "):"
<< std::endl;
for (unsigned int i = 0; i < program_.size(); ++i)
{
LOG(Message) << std::setw(4) << i + 1 << ": "
<< env().getModuleName(program_[i]) << std::endl;
}
}
// loop on configurations //////////////////////////////////////////////////////
void Application::configLoop(void)
{
auto range = par_.trajCounter;
for (unsigned int t = range.start; t < range.end; t += range.step)
{
LOG(Message) << BIG_SEP << " Starting measurement for trajectory " << t
<< " " << BIG_SEP << std::endl;
env().setTrajectory(t);
env().executeProgram(program_);
}
LOG(Message) << BIG_SEP << " End of measurement " << BIG_SEP << std::endl;
env().freeAll();
}

132
extras/Hadrons/Application.hpp Normal file
View File

@ -0,0 +1,132 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Application.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Application_hpp_
#define Hadrons_Application_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Environment.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
#include <Grid/Hadrons/Modules.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Main program manager *
******************************************************************************/
class Application
{
public:
class TrajRange: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(TrajRange,
unsigned int, start,
unsigned int, end,
unsigned int, step);
};
class GeneticPar: Serializable
{
public:
GeneticPar(void):
popSize{20}, maxGen{1000}, maxCstGen{100}, mutationRate{.1} {};
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(GeneticPar,
unsigned int, popSize,
unsigned int, maxGen,
unsigned int, maxCstGen,
double , mutationRate);
};
class GlobalPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(GlobalPar,
TrajRange, trajCounter,
GeneticPar, genetic,
std::string, seed);
};
public:
// constructors
Application(void);
Application(const GlobalPar &par);
Application(const std::string parameterFileName);
// destructor
virtual ~Application(void) = default;
// access
void setPar(const GlobalPar &par);
const GlobalPar & getPar(void);
// module creation
template <typename M>
void createModule(const std::string name);
template <typename M>
void createModule(const std::string name, const typename M::Par &par);
// execute
void run(void);
// XML parameter file I/O
void parseParameterFile(const std::string parameterFileName);
void saveParameterFile(const std::string parameterFileName);
// schedule computation
void schedule(void);
void saveSchedule(const std::string filename);
void loadSchedule(const std::string filename);
void printSchedule(void);
// loop on configurations
void configLoop(void);
private:
// environment shortcut
Environment & env(void) const;
private:
long unsigned int locVol_;
std::string parameterFileName_{""};
GlobalPar par_;
std::vector<unsigned int> program_;
Environment::Size memPeak_;
bool scheduled_{false};
};
/******************************************************************************
* Application template implementation *
******************************************************************************/
// module creation /////////////////////////////////////////////////////////////
template <typename M>
void Application::createModule(const std::string name)
{
env().createModule<M>(name);
}
template <typename M>
void Application::createModule(const std::string name,
const typename M::Par &par)
{
env().createModule<M>(name, par);
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Application_hpp_

743
extras/Hadrons/Environment.cc Normal file
View File

@ -0,0 +1,743 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Environment.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Environment.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
/******************************************************************************
* Environment implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
Environment::Environment(void)
{
nd_ = GridDefaultLatt().size();
grid4d_.reset(SpaceTimeGrid::makeFourDimGrid(
GridDefaultLatt(), GridDefaultSimd(nd_, vComplex::Nsimd()),
GridDefaultMpi()));
gridRb4d_.reset(SpaceTimeGrid::makeFourDimRedBlackGrid(grid4d_.get()));
auto loc = getGrid()->LocalDimensions();
locVol_ = 1;
for (unsigned int d = 0; d < loc.size(); ++d)
{
locVol_ *= loc[d];
}
rng4d_.reset(new GridParallelRNG(grid4d_.get()));
}
// dry run /////////////////////////////////////////////////////////////////////
void Environment::dryRun(const bool isDry)
{
dryRun_ = isDry;
}
bool Environment::isDryRun(void) const
{
return dryRun_;
}
// trajectory number ///////////////////////////////////////////////////////////
void Environment::setTrajectory(const unsigned int traj)
{
traj_ = traj;
}
unsigned int Environment::getTrajectory(void) const
{
return traj_;
}
// grids ///////////////////////////////////////////////////////////////////////
void Environment::createGrid(const unsigned int Ls)
{
if (grid5d_.find(Ls) == grid5d_.end())
{
auto g = getGrid();
grid5d_[Ls].reset(SpaceTimeGrid::makeFiveDimGrid(Ls, g));
gridRb5d_[Ls].reset(SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, g));
}
}
GridCartesian * Environment::getGrid(const unsigned int Ls) const
{
try
{
if (Ls == 1)
{
return grid4d_.get();
}
else
{
return grid5d_.at(Ls).get();
}
}
catch(std::out_of_range &)
{
HADRON_ERROR("no grid with Ls= " << Ls);
}
}
GridRedBlackCartesian * Environment::getRbGrid(const unsigned int Ls) const
{
try
{
if (Ls == 1)
{
return gridRb4d_.get();
}
else
{
return gridRb5d_.at(Ls).get();
}
}
catch(std::out_of_range &)
{
HADRON_ERROR("no red-black 5D grid with Ls= " << Ls);
}
}
unsigned int Environment::getNd(void) const
{
return nd_;
}
// random number generator /////////////////////////////////////////////////////
void Environment::setSeed(const std::vector<int> &seed)
{
rng4d_->SeedFixedIntegers(seed);
}
GridParallelRNG * Environment::get4dRng(void) const
{
return rng4d_.get();
}
// module management ///////////////////////////////////////////////////////////
void Environment::pushModule(Environment::ModPt &pt)
{
std::string name = pt->getName();
if (!hasModule(name))
{
std::vector<unsigned int> inputAddress;
unsigned int address;
ModuleInfo m;
m.data = std::move(pt);
m.type = typeIdPt(*m.data.get());
m.name = name;
auto input = m.data->getInput();
for (auto &in: input)
{
if (!hasObject(in))
{
addObject(in , -1);
}
m.input.push_back(objectAddress_[in]);
}
auto output = m.data->getOutput();
module_.push_back(std::move(m));
address = static_cast<unsigned int>(module_.size() - 1);
moduleAddress_[name] = address;
for (auto &out: output)
{
if (!hasObject(out))
{
addObject(out, address);
}
else
{
if (object_[objectAddress_[out]].module < 0)
{
object_[objectAddress_[out]].module = address;
}
else
{
HADRON_ERROR("object '" + out
+ "' is already produced by module '"
+ module_[object_[getObjectAddress(out)].module].name
+ "' (while pushing module '" + name + "')");
}
}
}
}
else
{
HADRON_ERROR("module '" + name + "' already exists");
}
}
unsigned int Environment::getNModule(void) const
{
return module_.size();
}
void Environment::createModule(const std::string name, const std::string type,
XmlReader &reader)
{
auto &factory = ModuleFactory::getInstance();
auto pt = factory.create(type, name);
pt->parseParameters(reader, "options");
pushModule(pt);
}
ModuleBase * Environment::getModule(const unsigned int address) const
{
if (hasModule(address))
{
return module_[address].data.get();
}
else
{
HADRON_ERROR("no module with address " + std::to_string(address));
}
}
ModuleBase * Environment::getModule(const std::string name) const
{
return getModule(getModuleAddress(name));
}
unsigned int Environment::getModuleAddress(const std::string name) const
{
if (hasModule(name))
{
return moduleAddress_.at(name);
}
else
{
HADRON_ERROR("no module with name '" + name + "'");
}
}
std::string Environment::getModuleName(const unsigned int address) const
{
if (hasModule(address))
{
return module_[address].name;
}
else
{
HADRON_ERROR("no module with address " + std::to_string(address));
}
}
std::string Environment::getModuleType(const unsigned int address) const
{
if (hasModule(address))
{
return typeName(module_[address].type);
}
else
{
HADRON_ERROR("no module with address " + std::to_string(address));
}
}
std::string Environment::getModuleType(const std::string name) const
{
return getModuleType(getModuleAddress(name));
}
bool Environment::hasModule(const unsigned int address) const
{
return (address < module_.size());
}
bool Environment::hasModule(const std::string name) const
{
return (moduleAddress_.find(name) != moduleAddress_.end());
}
Graph<unsigned int> Environment::makeModuleGraph(void) const
{
Graph<unsigned int> moduleGraph;
for (unsigned int i = 0; i < module_.size(); ++i)
{
moduleGraph.addVertex(i);
for (auto &j: module_[i].input)
{
moduleGraph.addEdge(object_[j].module, i);
}
}
return moduleGraph;
}
#define BIG_SEP "==============="
#define SEP "---------------"
#define MEM_MSG(size)\
sizeString((size)*locVol_) << " (" << sizeString(size) << "/site)"
Environment::Size
Environment::executeProgram(const std::vector<unsigned int> &p)
{
Size memPeak = 0, sizeBefore, sizeAfter;
std::vector<std::set<unsigned int>> freeProg;
bool continueCollect, nothingFreed;
// build garbage collection schedule
freeProg.resize(p.size());
for (unsigned int i = 0; i < object_.size(); ++i)
{
auto pred = [i, this](const unsigned int j)
{
auto &in = module_[j].input;
auto it = std::find(in.begin(), in.end(), i);
return (it != in.end()) or (j == object_[i].module);
};
auto it = std::find_if(p.rbegin(), p.rend(), pred);
if (it != p.rend())
{
freeProg[p.rend() - it - 1].insert(i);
}
}
// program execution
for (unsigned int i = 0; i < p.size(); ++i)
{
// execute module
if (!isDryRun())
{
LOG(Message) << SEP << " Measurement step " << i+1 << "/"
<< p.size() << " (module '" << module_[p[i]].name
<< "') " << SEP << std::endl;
}
(*module_[p[i]].data)();
sizeBefore = getTotalSize();
// print used memory after execution
if (!isDryRun())
{
LOG(Message) << "Allocated objects: " << MEM_MSG(sizeBefore)
<< std::endl;
}
if (sizeBefore > memPeak)
{
memPeak = sizeBefore;
}
// garbage collection for step i
if (!isDryRun())
{
LOG(Message) << "Garbage collection..." << std::endl;
}
nothingFreed = true;
do
{
continueCollect = false;
auto toFree = freeProg[i];
for (auto &j: toFree)
{
// continue garbage collection while there are still
// objects without owners
continueCollect = continueCollect or !hasOwners(j);
if(freeObject(j))
{
// if an object has been freed, remove it from
// the garbage collection schedule
freeProg[i].erase(j);
nothingFreed = false;
}
}
} while (continueCollect);
// any remaining objects in step i garbage collection schedule
// is scheduled for step i + 1
if (i + 1 < p.size())
{
for (auto &j: freeProg[i])
{
freeProg[i + 1].insert(j);
}
}
// print used memory after garbage collection if necessary
if (!isDryRun())
{
sizeAfter = getTotalSize();
if (sizeBefore != sizeAfter)
{
LOG(Message) << "Allocated objects: " << MEM_MSG(sizeAfter)
<< std::endl;
}
else
{
LOG(Message) << "Nothing to free" << std::endl;
}
}
}
return memPeak;
}
Environment::Size Environment::executeProgram(const std::vector<std::string> &p)
{
std::vector<unsigned int> pAddress;
for (auto &n: p)
{
pAddress.push_back(getModuleAddress(n));
}
return executeProgram(pAddress);
}
// general memory management ///////////////////////////////////////////////////
void Environment::addObject(const std::string name, const int moduleAddress)
{
if (!hasObject(name))
{
ObjInfo info;
info.name = name;
info.module = moduleAddress;
object_.push_back(std::move(info));
objectAddress_[name] = static_cast<unsigned int>(object_.size() - 1);
}
else
{
HADRON_ERROR("object '" + name + "' already exists");
}
}
void Environment::registerObject(const unsigned int address,
const unsigned int size, const unsigned int Ls)
{
if (!hasRegisteredObject(address))
{
if (hasObject(address))
{
object_[address].size = size;
object_[address].Ls = Ls;
object_[address].isRegistered = true;
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
else
{
HADRON_ERROR("object with address " + std::to_string(address)
+ " already registered");
}
}
void Environment::registerObject(const std::string name,
const unsigned int size, const unsigned int Ls)
{
if (!hasObject(name))
{
addObject(name);
}
registerObject(getObjectAddress(name), size, Ls);
}
unsigned int Environment::getObjectAddress(const std::string name) const
{
if (hasObject(name))
{
return objectAddress_.at(name);
}
else
{
HADRON_ERROR("no object with name '" + name + "'");
}
}
std::string Environment::getObjectName(const unsigned int address) const
{
if (hasObject(address))
{
return object_[address].name;
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
std::string Environment::getObjectType(const unsigned int address) const
{
if (hasRegisteredObject(address))
{
return typeName(object_[address].type);
}
else if (hasObject(address))
{
HADRON_ERROR("object with address " + std::to_string(address)
+ " exists but is not registered");
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
std::string Environment::getObjectType(const std::string name) const
{
return getObjectType(getObjectAddress(name));
}
Environment::Size Environment::getObjectSize(const unsigned int address) const
{
if (hasRegisteredObject(address))
{
return object_[address].size;
}
else if (hasObject(address))
{
HADRON_ERROR("object with address " + std::to_string(address)
+ " exists but is not registered");
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
Environment::Size Environment::getObjectSize(const std::string name) const
{
return getObjectSize(getObjectAddress(name));
}
unsigned int Environment::getObjectLs(const unsigned int address) const
{
if (hasRegisteredObject(address))
{
return object_[address].Ls;
}
else if (hasObject(address))
{
HADRON_ERROR("object with address " + std::to_string(address)
+ " exists but is not registered");
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
unsigned int Environment::getObjectLs(const std::string name) const
{
return getObjectLs(getObjectAddress(name));
}
bool Environment::hasObject(const unsigned int address) const
{
return (address < object_.size());
}
bool Environment::hasObject(const std::string name) const
{
auto it = objectAddress_.find(name);
return ((it != objectAddress_.end()) and hasObject(it->second));
}
bool Environment::hasRegisteredObject(const unsigned int address) const
{
if (hasObject(address))
{
return object_[address].isRegistered;
}
else
{
return false;
}
}
bool Environment::hasRegisteredObject(const std::string name) const
{
if (hasObject(name))
{
return hasRegisteredObject(getObjectAddress(name));
}
else
{
return false;
}
}
bool Environment::hasCreatedObject(const unsigned int address) const
{
if (hasObject(address))
{
return (object_[address].data != nullptr);
}
else
{
return false;
}
}
bool Environment::hasCreatedObject(const std::string name) const
{
if (hasObject(name))
{
return hasCreatedObject(getObjectAddress(name));
}
else
{
return false;
}
}
bool Environment::isObject5d(const unsigned int address) const
{
return (getObjectLs(address) > 1);
}
bool Environment::isObject5d(const std::string name) const
{
return (getObjectLs(name) > 1);
}
Environment::Size Environment::getTotalSize(void) const
{
Environment::Size size = 0;
for (auto &o: object_)
{
if (o.isRegistered)
{
size += o.size;
}
}
return size;
}
void Environment::addOwnership(const unsigned int owner,
const unsigned int property)
{
if (hasObject(property))
{
object_[property].owners.insert(owner);
}
else
{
HADRON_ERROR("no object with address " + std::to_string(property));
}
if (hasObject(owner))
{
object_[owner].properties.insert(property);
}
else
{
HADRON_ERROR("no object with address " + std::to_string(owner));
}
}
void Environment::addOwnership(const std::string owner,
const std::string property)
{
addOwnership(getObjectAddress(owner), getObjectAddress(property));
}
bool Environment::hasOwners(const unsigned int address) const
{
if (hasObject(address))
{
return (!object_[address].owners.empty());
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
bool Environment::hasOwners(const std::string name) const
{
return hasOwners(getObjectAddress(name));
}
bool Environment::freeObject(const unsigned int address)
{
if (!hasOwners(address))
{
if (!isDryRun() and object_[address].isRegistered)
{
LOG(Message) << "Destroying object '" << object_[address].name
<< "'" << std::endl;
}
for (auto &p: object_[address].properties)
{
object_[p].owners.erase(address);
}
object_[address].size = 0;
object_[address].Ls = 0;
object_[address].isRegistered = false;
object_[address].type = nullptr;
object_[address].owners.clear();
object_[address].properties.clear();
object_[address].data.reset(nullptr);
return true;
}
else
{
return false;
}
}
bool Environment::freeObject(const std::string name)
{
return freeObject(getObjectAddress(name));
}
void Environment::freeAll(void)
{
for (unsigned int i = 0; i < object_.size(); ++i)
{
freeObject(i);
}
}
void Environment::printContent(void)
{
LOG(Message) << "Modules: " << std::endl;
for (unsigned int i = 0; i < module_.size(); ++i)
{
LOG(Message) << std::setw(4) << i << ": "
<< getModuleName(i) << std::endl;
}
LOG(Message) << "Objects: " << std::endl;
for (unsigned int i = 0; i < object_.size(); ++i)
{
LOG(Message) << std::setw(4) << i << ": "
<< getObjectName(i) << std::endl;
}
}

385
extras/Hadrons/Environment.hpp Normal file
View File

@ -0,0 +1,385 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Environment.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Environment_hpp_
#define Hadrons_Environment_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Graph.hpp>
#ifndef SITE_SIZE_TYPE
#define SITE_SIZE_TYPE unsigned int
#endif
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Global environment *
******************************************************************************/
// forward declaration of Module
class ModuleBase;
class Object
{
public:
Object(void) = default;
virtual ~Object(void) = default;
};
template <typename T>
class Holder: public Object
{
public:
Holder(void) = default;
Holder(T *pt);
virtual ~Holder(void) = default;
T & get(void) const;
T * getPt(void) const;
void reset(T *pt);
private:
std::unique_ptr<T> objPt_{nullptr};
};
class Environment
{
SINGLETON(Environment);
public:
typedef SITE_SIZE_TYPE Size;
typedef std::unique_ptr<ModuleBase> ModPt;
typedef std::unique_ptr<GridCartesian> GridPt;
typedef std::unique_ptr<GridRedBlackCartesian> GridRbPt;
typedef std::unique_ptr<GridParallelRNG> RngPt;
typedef std::unique_ptr<LatticeBase> LatticePt;
private:
struct ModuleInfo
{
const std::type_info *type{nullptr};
std::string name;
ModPt data{nullptr};
std::vector<unsigned int> input;
};
struct ObjInfo
{
Size size{0};
unsigned int Ls{0};
bool isRegistered{false};
const std::type_info *type{nullptr};
std::string name;
int module{-1};
std::set<unsigned int> owners, properties;
std::unique_ptr<Object> data{nullptr};
};
public:
// dry run
void dryRun(const bool isDry);
bool isDryRun(void) const;
// trajectory number
void setTrajectory(const unsigned int traj);
unsigned int getTrajectory(void) const;
// grids
void createGrid(const unsigned int Ls);
GridCartesian * getGrid(const unsigned int Ls = 1) const;
GridRedBlackCartesian * getRbGrid(const unsigned int Ls = 1) const;
unsigned int getNd(void) const;
// random number generator
void setSeed(const std::vector<int> &seed);
GridParallelRNG * get4dRng(void) const;
// module management
void pushModule(ModPt &pt);
template <typename M>
void createModule(const std::string name);
template <typename M>
void createModule(const std::string name,
const typename M::Par &par);
void createModule(const std::string name,
const std::string type,
XmlReader &reader);
unsigned int getNModule(void) const;
ModuleBase * getModule(const unsigned int address) const;
ModuleBase * getModule(const std::string name) const;
template <typename M>
M * getModule(const unsigned int address) const;
template <typename M>
M * getModule(const std::string name) const;
unsigned int getModuleAddress(const std::string name) const;
std::string getModuleName(const unsigned int address) const;
std::string getModuleType(const unsigned int address) const;
std::string getModuleType(const std::string name) const;
bool hasModule(const unsigned int address) const;
bool hasModule(const std::string name) const;
Graph<unsigned int> makeModuleGraph(void) const;
Size executeProgram(const std::vector<unsigned int> &p);
Size executeProgram(const std::vector<std::string> &p);
// general memory management
void addObject(const std::string name,
const int moduleAddress = -1);
void registerObject(const unsigned int address,
const unsigned int size,
const unsigned int Ls = 1);
void registerObject(const std::string name,
const unsigned int size,
const unsigned int Ls = 1);
template <typename T>
unsigned int lattice4dSize(void) const;
template <typename T>
void registerLattice(const unsigned int address,
const unsigned int Ls = 1);
template <typename T>
void registerLattice(const std::string name,
const unsigned int Ls = 1);
template <typename T>
void setObject(const unsigned int address, T *object);
template <typename T>
void setObject(const std::string name, T *object);
template <typename T>
T * getObject(const unsigned int address) const;
template <typename T>
T * getObject(const std::string name) const;
template <typename T>
T * createLattice(const unsigned int address);
template <typename T>
T * createLattice(const std::string name);
unsigned int getObjectAddress(const std::string name) const;
std::string getObjectName(const unsigned int address) const;
std::string getObjectType(const unsigned int address) const;
std::string getObjectType(const std::string name) const;
Size getObjectSize(const unsigned int address) const;
Size getObjectSize(const std::string name) const;
unsigned int getObjectLs(const unsigned int address) const;
unsigned int getObjectLs(const std::string name) const;
bool hasObject(const unsigned int address) const;
bool hasObject(const std::string name) const;
bool hasRegisteredObject(const unsigned int address) const;
bool hasRegisteredObject(const std::string name) const;
bool hasCreatedObject(const unsigned int address) const;
bool hasCreatedObject(const std::string name) const;
bool isObject5d(const unsigned int address) const;
bool isObject5d(const std::string name) const;
Environment::Size getTotalSize(void) const;
void addOwnership(const unsigned int owner,
const unsigned int property);
void addOwnership(const std::string owner,
const std::string property);
bool hasOwners(const unsigned int address) const;
bool hasOwners(const std::string name) const;
bool freeObject(const unsigned int address);
bool freeObject(const std::string name);
void freeAll(void);
void printContent(void);
private:
// general
bool dryRun_{false};
unsigned int traj_, locVol_;
// grids
GridPt grid4d_;
std::map<unsigned int, GridPt> grid5d_;
GridRbPt gridRb4d_;
std::map<unsigned int, GridRbPt> gridRb5d_;
unsigned int nd_;
// random number generator
RngPt rng4d_;
// module and related maps
std::vector<ModuleInfo> module_;
std::map<std::string, unsigned int> moduleAddress_;
// lattice store
std::map<unsigned int, LatticePt> lattice_;
// object store
std::vector<ObjInfo> object_;
std::map<std::string, unsigned int> objectAddress_;
};
/******************************************************************************
* Holder template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename T>
Holder<T>::Holder(T *pt)
: objPt_(pt)
{}
// access //////////////////////////////////////////////////////////////////////
template <typename T>
T & Holder<T>::get(void) const
{
return &objPt_.get();
}
template <typename T>
T * Holder<T>::getPt(void) const
{
return objPt_.get();
}
template <typename T>
void Holder<T>::reset(T *pt)
{
objPt_.reset(pt);
}
/******************************************************************************
* Environment template implementation *
******************************************************************************/
// module management ///////////////////////////////////////////////////////////
template <typename M>
void Environment::createModule(const std::string name)
{
ModPt pt(new M(name));
pushModule(pt);
}
template <typename M>
void Environment::createModule(const std::string name,
const typename M::Par &par)
{
ModPt pt(new M(name));
static_cast<M *>(pt.get())->setPar(par);
pushModule(pt);
}
template <typename M>
M * Environment::getModule(const unsigned int address) const
{
if (auto *pt = dynamic_cast<M *>(getModule(address)))
{
return pt;
}
else
{
HADRON_ERROR("module '" + module_[address].name
+ "' does not have type " + typeid(M).name()
+ "(object type: " + getModuleType(address) + ")");
}
}
template <typename M>
M * Environment::getModule(const std::string name) const
{
return getModule<M>(getModuleAddress(name));
}
template <typename T>
unsigned int Environment::lattice4dSize(void) const
{
return sizeof(typename T::vector_object)/getGrid()->Nsimd();
}
template <typename T>
void Environment::registerLattice(const unsigned int address,
const unsigned int Ls)
{
createGrid(Ls);
registerObject(address, Ls*lattice4dSize<T>(), Ls);
}
template <typename T>
void Environment::registerLattice(const std::string name, const unsigned int Ls)
{
createGrid(Ls);
registerObject(name, Ls*lattice4dSize<T>(), Ls);
}
template <typename T>
void Environment::setObject(const unsigned int address, T *object)
{
if (hasRegisteredObject(address))
{
object_[address].data.reset(new Holder<T>(object));
object_[address].type = &typeid(T);
}
else if (hasObject(address))
{
HADRON_ERROR("object with address " + std::to_string(address) +
" exists but is not registered");
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
template <typename T>
void Environment::setObject(const std::string name, T *object)
{
setObject(getObjectAddress(name), object);
}
template <typename T>
T * Environment::getObject(const unsigned int address) const
{
if (hasRegisteredObject(address))
{
if (auto h = dynamic_cast<Holder<T> *>(object_[address].data.get()))
{
return h->getPt();
}
else
{
HADRON_ERROR("object with address " + std::to_string(address) +
" does not have type '" + typeid(T).name() +
"' (has type '" + getObjectType(address) + "')");
}
}
else if (hasObject(address))
{
HADRON_ERROR("object with address " + std::to_string(address) +
" exists but is not registered");
}
else
{
HADRON_ERROR("no object with address " + std::to_string(address));
}
}
template <typename T>
T * Environment::getObject(const std::string name) const
{
return getObject<T>(getObjectAddress(name));
}
template <typename T>
T * Environment::createLattice(const unsigned int address)
{
GridCartesian *g = getGrid(getObjectLs(address));
setObject(address, new T(g));
return getObject<T>(address);
}
template <typename T>
T * Environment::createLattice(const std::string name)
{
return createLattice<T>(getObjectAddress(name));
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Environment_hpp_

106
extras/Hadrons/Factory.hpp Normal file
View File

@ -0,0 +1,106 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Factory.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Factory_hpp_
#define Hadrons_Factory_hpp_
#include <Grid/Hadrons/Global.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* abstract factory class *
******************************************************************************/
template <typename T>
class Factory
{
public:
typedef std::function<std::unique_ptr<T>(const std::string)> Func;
public:
// constructor
Factory(void) = default;
// destructor
virtual ~Factory(void) = default;
// registration
void registerBuilder(const std::string type, const Func &f);
// get builder list
std::vector<std::string> getBuilderList(void) const;
// factory
std::unique_ptr<T> create(const std::string type,
const std::string name) const;
private:
std::map<std::string, Func> builder_;
};
/******************************************************************************
* template implementation *
******************************************************************************/
// registration ////////////////////////////////////////////////////////////////
template <typename T>
void Factory<T>::registerBuilder(const std::string type, const Func &f)
{
builder_[type] = f;
}
// get module list /////////////////////////////////////////////////////////////
template <typename T>
std::vector<std::string> Factory<T>::getBuilderList(void) const
{
std::vector<std::string> list;
for (auto &b: builder_)
{
list.push_back(b.first);
}
return list;
}
// factory /////////////////////////////////////////////////////////////////////
template <typename T>
std::unique_ptr<T> Factory<T>::create(const std::string type,
const std::string name) const
{
Func func;
try
{
func = builder_.at(type);
}
catch (std::out_of_range &)
{
HADRON_ERROR("object of type '" + type + "' unknown");
}
return func(name);
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Factory_hpp_

329
extras/Hadrons/GeneticScheduler.hpp Normal file
View File

@ -0,0 +1,329 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/GeneticScheduler.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_GeneticScheduler_hpp_
#define Hadrons_GeneticScheduler_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Graph.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Scheduler based on a genetic algorithm *
******************************************************************************/
template <typename T>
class GeneticScheduler
{
public:
typedef std::vector<T> Gene;
typedef std::pair<Gene *, Gene *> GenePair;
typedef std::function<int(const Gene &)> ObjFunc;
struct Parameters
{
double mutationRate;
unsigned int popSize, seed;
};
public:
// constructor
GeneticScheduler(Graph<T> &graph, const ObjFunc &func,
const Parameters &par);
// destructor
virtual ~GeneticScheduler(void) = default;
// access
const Gene & getMinSchedule(void);
int getMinValue(void);
// breed a new generation
void nextGeneration(void);
// heuristic benchmarks
void benchmarkCrossover(const unsigned int nIt);
// print population
friend std::ostream & operator<<(std::ostream &out,
const GeneticScheduler<T> &s)
{
out << "[";
for (auto &p: s.population_)
{
out << p.first << ", ";
}
out << "\b\b]";
return out;
}
private:
// evolution steps
void initPopulation(void);
void doCrossover(void);
void doMutation(void);
// genetic operators
GenePair selectPair(void);
void crossover(Gene &c1, Gene &c2, const Gene &p1, const Gene &p2);
void mutation(Gene &m, const Gene &c);
private:
Graph<T> &graph_;
const ObjFunc &func_;
const Parameters par_;
std::multimap<int, Gene> population_;
std::mt19937 gen_;
};
/******************************************************************************
* template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename T>
GeneticScheduler<T>::GeneticScheduler(Graph<T> &graph, const ObjFunc &func,
const Parameters &par)
: graph_(graph)
, func_(func)
, par_(par)
{
gen_.seed(par_.seed);
}
// access //////////////////////////////////////////////////////////////////////
template <typename T>
const typename GeneticScheduler<T>::Gene &
GeneticScheduler<T>::getMinSchedule(void)
{
return population_.begin()->second;
}
template <typename T>
int GeneticScheduler<T>::getMinValue(void)
{
return population_.begin()->first;
}
// breed a new generation //////////////////////////////////////////////////////
template <typename T>
void GeneticScheduler<T>::nextGeneration(void)
{
// random initialization of the population if necessary
if (population_.size() != par_.popSize)
{
initPopulation();
}
LOG(Debug) << "Starting population:\n" << *this << std::endl;
// random mutations
//PARALLEL_FOR_LOOP
for (unsigned int i = 0; i < par_.popSize; ++i)
{
doMutation();
}
LOG(Debug) << "After mutations:\n" << *this << std::endl;
// mating
//PARALLEL_FOR_LOOP
for (unsigned int i = 0; i < par_.popSize/2; ++i)
{
doCrossover();
}
LOG(Debug) << "After mating:\n" << *this << std::endl;
// grim reaper
auto it = population_.begin();
std::advance(it, par_.popSize);
population_.erase(it, population_.end());
LOG(Debug) << "After grim reaper:\n" << *this << std::endl;
}
// evolution steps /////////////////////////////////////////////////////////////
template <typename T>
void GeneticScheduler<T>::initPopulation(void)
{
population_.clear();
for (unsigned int i = 0; i < par_.popSize; ++i)
{
auto p = graph_.topoSort(gen_);
population_.insert(std::make_pair(func_(p), p));
}
}
template <typename T>
void GeneticScheduler<T>::doCrossover(void)
{
auto p = selectPair();
Gene &p1 = *(p.first), &p2 = *(p.second);
Gene c1, c2;
crossover(c1, c2, p1, p2);
PARALLEL_CRITICAL
{
population_.insert(std::make_pair(func_(c1), c1));
population_.insert(std::make_pair(func_(c2), c2));
}
}
template <typename T>
void GeneticScheduler<T>::doMutation(void)
{
std::uniform_real_distribution<double> mdis(0., 1.);
std::uniform_int_distribution<unsigned int> pdis(0, population_.size() - 1);
if (mdis(gen_) < par_.mutationRate)
{
Gene m;
auto it = population_.begin();
std::advance(it, pdis(gen_));
mutation(m, it->second);
PARALLEL_CRITICAL
{
population_.insert(std::make_pair(func_(m), m));
}
}
}
// genetic operators ///////////////////////////////////////////////////////////
template <typename T>
typename GeneticScheduler<T>::GenePair GeneticScheduler<T>::selectPair(void)
{
std::vector<double> prob;
unsigned int ind;
Gene *p1, *p2;
for (auto &c: population_)
{
prob.push_back(1./c.first);
}
do
{
double probCpy;
std::discrete_distribution<unsigned int> dis1(prob.begin(), prob.end());
auto rIt = population_.begin();
ind = dis1(gen_);
std::advance(rIt, ind);
p1 = &(rIt->second);
probCpy = prob[ind];
prob[ind] = 0.;
std::discrete_distribution<unsigned int> dis2(prob.begin(), prob.end());
rIt = population_.begin();
std::advance(rIt, dis2(gen_));
p2 = &(rIt->second);
prob[ind] = probCpy;
} while (p1 == p2);
return std::make_pair(p1, p2);
}
template <typename T>
void GeneticScheduler<T>::crossover(Gene &c1, Gene &c2, const Gene &p1,
const Gene &p2)
{
Gene buf;
std::uniform_int_distribution<unsigned int> dis(0, p1.size() - 1);
unsigned int cut = dis(gen_);
c1.clear();
buf = p2;
for (unsigned int i = 0; i < cut; ++i)
{
c1.push_back(p1[i]);
buf.erase(std::find(buf.begin(), buf.end(), p1[i]));
}
for (unsigned int i = 0; i < buf.size(); ++i)
{
c1.push_back(buf[i]);
}
c2.clear();
buf = p2;
for (unsigned int i = cut; i < p1.size(); ++i)
{
buf.erase(std::find(buf.begin(), buf.end(), p1[i]));
}
for (unsigned int i = 0; i < buf.size(); ++i)
{
c2.push_back(buf[i]);
}
for (unsigned int i = cut; i < p1.size(); ++i)
{
c2.push_back(p1[i]);
}
}
template <typename T>
void GeneticScheduler<T>::mutation(Gene &m, const Gene &c)
{
Gene buf;
std::uniform_int_distribution<unsigned int> dis(0, c.size() - 1);
unsigned int cut = dis(gen_);
Graph<T> g1 = graph_, g2 = graph_;
for (unsigned int i = 0; i < cut; ++i)
{
g1.removeVertex(c[i]);
}
for (unsigned int i = cut; i < c.size(); ++i)
{
g2.removeVertex(c[i]);
}
if (g1.size() > 0)
{
buf = g1.topoSort(gen_);
}
if (g2.size() > 0)
{
m = g2.topoSort(gen_);
}
for (unsigned int i = cut; i < c.size(); ++i)
{
m.push_back(buf[i - cut]);
}
}
template <typename T>
void GeneticScheduler<T>::benchmarkCrossover(const unsigned int nIt)
{
Gene p1, p2, c1, c2;
double neg = 0., eq = 0., pos = 0., total;
int improvement;
LOG(Message) << "Benchmarking crossover..." << std::endl;
for (unsigned int i = 0; i < nIt; ++i)
{
p1 = graph_.topoSort(gen_);
p2 = graph_.topoSort(gen_);
crossover(c1, c2, p1, p2);
improvement = (func_(c1) + func_(c2) - func_(p1) - func_(p2))/2;
if (improvement < 0) neg++; else if (improvement == 0) eq++; else pos++;
}
total = neg + eq + pos;
LOG(Message) << " -: " << neg/total << " =: " << eq/total
<< " +: " << pos/total << std::endl;
}
END_HADRONS_NAMESPACE
#endif // Hadrons_GeneticScheduler_hpp_

82
extras/Hadrons/Global.cc Normal file
View File

@ -0,0 +1,82 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Global.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Global.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
HadronsLogger Hadrons::HadronsLogError(1,"Error");
HadronsLogger Hadrons::HadronsLogWarning(1,"Warning");
HadronsLogger Hadrons::HadronsLogMessage(1,"Message");
HadronsLogger Hadrons::HadronsLogIterative(1,"Iterative");
HadronsLogger Hadrons::HadronsLogDebug(1,"Debug");
// pretty size formatting //////////////////////////////////////////////////////
std::string Hadrons::sizeString(long unsigned int bytes)
{
constexpr unsigned int bufSize = 256;
const char *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
char buf[256];
long unsigned int s = 0;
double count = bytes;
while (count >= 1024 && s < 7)
{
s++;
count /= 1024;
}
if (count - floor(count) == 0.0)
{
snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
}
else
{
snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
}
return std::string(buf);
}
// type utilities //////////////////////////////////////////////////////////////
constexpr unsigned int maxNameSize = 1024u;
std::string Hadrons::typeName(const std::type_info *info)
{
char *buf;
std::string name;
buf = abi::__cxa_demangle(info->name(), nullptr, nullptr, nullptr);
name = buf;
free(buf);
return name;
}

150
extras/Hadrons/Global.hpp Normal file
View File

@ -0,0 +1,150 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Global.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Global_hpp_
#define Hadrons_Global_hpp_
#include <set>
#include <stack>
#include <Grid/Grid.h>
#include <cxxabi.h>
#define BEGIN_HADRONS_NAMESPACE \
namespace Grid {\
using namespace QCD;\
namespace Hadrons {\
using Grid::operator<<;
#define END_HADRONS_NAMESPACE }}
#define BEGIN_MODULE_NAMESPACE(name)\
namespace name {\
using Grid::operator<<;
#define END_MODULE_NAMESPACE }
/* the 'using Grid::operator<<;' statement prevents a very nasty compilation
* error with GCC 5 (clang & GCC 6 compile fine without it).
*/
// FIXME: find a way to do that in a more general fashion
#ifndef FIMPL
#define FIMPL WilsonImplR
#endif
BEGIN_HADRONS_NAMESPACE
// type aliases
#define TYPE_ALIASES(FImpl, suffix)\
typedef FermionOperator<FImpl> FMat##suffix; \
typedef typename FImpl::FermionField FermionField##suffix; \
typedef typename FImpl::PropagatorField PropagatorField##suffix; \
typedef typename FImpl::SitePropagator SitePropagator##suffix; \
typedef typename FImpl::DoubledGaugeField DoubledGaugeField##suffix;\
typedef std::function<void(FermionField##suffix &, \
const FermionField##suffix &)> SolverFn##suffix;
// logger
class HadronsLogger: public Logger
{
public:
HadronsLogger(int on, std::string nm): Logger("Hadrons", on, nm,
GridLogColours, "BLACK"){};
};
#define LOG(channel) std::cout << HadronsLog##channel
#define HADRON_ERROR(msg)\
LOG(Error) << msg << " (" << __FUNCTION__ << " at " << __FILE__ << ":"\
<< __LINE__ << ")" << std::endl;\
abort();
#define DEBUG_VAR(var) LOG(Debug) << #var << "= " << (var) << std::endl;
extern HadronsLogger HadronsLogError;
extern HadronsLogger HadronsLogWarning;
extern HadronsLogger HadronsLogMessage;
extern HadronsLogger HadronsLogIterative;
extern HadronsLogger HadronsLogDebug;
// singleton pattern
#define SINGLETON(name)\
public:\
name(const name &e) = delete;\
void operator=(const name &e) = delete;\
static name & getInstance(void)\
{\
static name e;\
return e;\
}\
private:\
name(void);
#define SINGLETON_DEFCTOR(name)\
public:\
name(const name &e) = delete;\
void operator=(const name &e) = delete;\
static name & getInstance(void)\
{\
static name e;\
return e;\
}\
private:\
name(void) = default;
// pretty size formating
std::string sizeString(long unsigned int bytes);
// type utilities
template <typename T>
const std::type_info * typeIdPt(const T &x)
{
return &typeid(x);
}
std::string typeName(const std::type_info *info);
template <typename T>
const std::type_info * typeIdPt(void)
{
return &typeid(T);
}
template <typename T>
std::string typeName(const T &x)
{
return typeName(typeIdPt(x));
}
template <typename T>
std::string typeName(void)
{
return typeName(typeIdPt<T>());
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Global_hpp_

760
extras/Hadrons/Graph.hpp Normal file
View File

@ -0,0 +1,760 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Graph.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Graph_hpp_
#define Hadrons_Graph_hpp_
#include <Grid/Hadrons/Global.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Oriented graph class *
******************************************************************************/
// I/O for edges
template <typename T>
std::ostream & operator<<(std::ostream &out, const std::pair<T, T> &e)
{
out << "\"" << e.first << "\" -> \"" << e.second << "\"";
return out;
}
// main class
template <typename T>
class Graph
{
public:
typedef std::pair<T, T> Edge;
public:
// constructor
Graph(void);
// destructor
virtual ~Graph(void) = default;
// access
void addVertex(const T &value);
void addEdge(const Edge &e);
void addEdge(const T &start, const T &end);
std::vector<T> getVertices(void) const;
void removeVertex(const T &value);
void removeEdge(const Edge &e);
void removeEdge(const T &start, const T &end);
unsigned int size(void) const;
// tests
bool gotValue(const T &value) const;
// graph topological manipulations
std::vector<T> getAdjacentVertices(const T &value) const;
std::vector<T> getChildren(const T &value) const;
std::vector<T> getParents(const T &value) const;
std::vector<T> getRoots(void) const;
std::vector<Graph<T>> getConnectedComponents(void) const;
std::vector<T> topoSort(void);
template <typename Gen>
std::vector<T> topoSort(Gen &gen);
std::vector<std::vector<T>> allTopoSort(void);
// I/O
friend std::ostream & operator<<(std::ostream &out, const Graph<T> &g)
{
out << "{";
for (auto &e: g.edgeSet_)
{
out << e << ", ";
}
if (g.edgeSet_.size() != 0)
{
out << "\b\b";
}
out << "}";
return out;
}
private:
// vertex marking
void mark(const T &value, const bool doMark = true);
void markAll(const bool doMark = true);
void unmark(const T &value);
void unmarkAll(void);
bool isMarked(const T &value) const;
const T * getFirstMarked(const bool isMarked = true) const;
template <typename Gen>
const T * getRandomMarked(const bool isMarked, Gen &gen);
const T * getFirstUnmarked(void) const;
template <typename Gen>
const T * getRandomUnmarked(Gen &gen);
// prune marked/unmarked vertices
void removeMarked(const bool isMarked = true);
void removeUnmarked(void);
// depth-first search marking
void depthFirstSearch(void);
void depthFirstSearch(const T &root);
private:
std::map<T, bool> isMarked_;
std::set<Edge> edgeSet_;
};
// build depedency matrix from topological sorts
template <typename T>
std::map<T, std::map<T, bool>>
makeDependencyMatrix(const std::vector<std::vector<T>> &topSort);
/******************************************************************************
* template implementation *
******************************************************************************
* in all the following V is the number of vertex and E is the number of edge
* in the worst case E = V^2
*/
// constructor /////////////////////////////////////////////////////////////////
template <typename T>
Graph<T>::Graph(void)
{}
// access //////////////////////////////////////////////////////////////////////
// complexity: log(V)
template <typename T>
void Graph<T>::addVertex(const T &value)
{
isMarked_[value] = false;
}
// complexity: O(log(V))
template <typename T>
void Graph<T>::addEdge(const Edge &e)
{
addVertex(e.first);
addVertex(e.second);
edgeSet_.insert(e);
}
// complexity: O(log(V))
template <typename T>
void Graph<T>::addEdge(const T &start, const T &end)
{
addEdge(Edge(start, end));
}
template <typename T>
std::vector<T> Graph<T>::getVertices(void) const
{
std::vector<T> vertex;
for (auto &v: isMarked_)
{
vertex.push_back(v.first);
}
return vertex;
}
// complexity: O(V*log(V))
template <typename T>
void Graph<T>::removeVertex(const T &value)
{
// remove vertex from the mark table
auto vIt = isMarked_.find(value);
if (vIt != isMarked_.end())
{
isMarked_.erase(vIt);
}
else
{
HADRON_ERROR("vertex " << value << " does not exists");
}
// remove all edges containing the vertex
auto pred = [&value](const Edge &e)
{
return ((e.first == value) or (e.second == value));
};
auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
while (eIt != edgeSet_.end())
{
edgeSet_.erase(eIt);
eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
}
}
// complexity: O(log(V))
template <typename T>
void Graph<T>::removeEdge(const Edge &e)
{
auto eIt = edgeSet_.find(e);
if (eIt != edgeSet_.end())
{
edgeSet_.erase(eIt);
}
else
{
HADRON_ERROR("edge " << e << " does not exists");
}
}
// complexity: O(log(V))
template <typename T>
void Graph<T>::removeEdge(const T &start, const T &end)
{
removeEdge(Edge(start, end));
}
// complexity: O(1)
template <typename T>
unsigned int Graph<T>::size(void) const
{
return isMarked_.size();
}
// tests ///////////////////////////////////////////////////////////////////////
// complexity: O(log(V))
template <typename T>
bool Graph<T>::gotValue(const T &value) const
{
auto it = isMarked_.find(value);
if (it == isMarked_.end())
{
return false;
}
else
{
return true;
}
}
// vertex marking //////////////////////////////////////////////////////////////
// complexity: O(log(V))
template <typename T>
void Graph<T>::mark(const T &value, const bool doMark)
{
if (gotValue(value))
{
isMarked_[value] = doMark;
}
else
{
HADRON_ERROR("vertex " << value << " does not exists");
}
}
// complexity: O(V*log(V))
template <typename T>
void Graph<T>::markAll(const bool doMark)
{
for (auto &v: isMarked_)
{
mark(v.first, doMark);
}
}
// complexity: O(log(V))
template <typename T>
void Graph<T>::unmark(const T &value)
{
mark(value, false);
}
// complexity: O(V*log(V))
template <typename T>
void Graph<T>::unmarkAll(void)
{
markAll(false);
}
// complexity: O(log(V))
template <typename T>
bool Graph<T>::isMarked(const T &value) const
{
if (gotValue(value))
{
return isMarked_.at(value);
}
else
{
HADRON_ERROR("vertex " << value << " does not exists");
return false;
}
}
// complexity: O(log(V))
template <typename T>
const T * Graph<T>::getFirstMarked(const bool isMarked) const
{
auto pred = [&isMarked](const std::pair<T, bool> &v)
{
return (v.second == isMarked);
};
auto vIt = std::find_if(isMarked_.begin(), isMarked_.end(), pred);
if (vIt != isMarked_.end())
{
return &(vIt->first);
}
else
{
return nullptr;
}
}
// complexity: O(log(V))
template <typename T>
template <typename Gen>
const T * Graph<T>::getRandomMarked(const bool isMarked, Gen &gen)
{
auto pred = [&isMarked](const std::pair<T, bool> &v)
{
return (v.second == isMarked);
};
std::uniform_int_distribution<unsigned int> dis(0, size() - 1);
auto rIt = isMarked_.begin();
std::advance(rIt, dis(gen));
auto vIt = std::find_if(rIt, isMarked_.end(), pred);
if (vIt != isMarked_.end())
{
return &(vIt->first);
}
else
{
vIt = std::find_if(isMarked_.begin(), rIt, pred);
if (vIt != rIt)
{
return &(vIt->first);
}
else
{
return nullptr;
}
}
}
// complexity: O(log(V))
template <typename T>
const T * Graph<T>::getFirstUnmarked(void) const
{
return getFirstMarked(false);
}
// complexity: O(log(V))
template <typename T>
template <typename Gen>
const T * Graph<T>::getRandomUnmarked(Gen &gen)
{
return getRandomMarked(false, gen);
}
// prune marked/unmarked vertices //////////////////////////////////////////////
// complexity: O(V^2*log(V))
template <typename T>
void Graph<T>::removeMarked(const bool isMarked)
{
auto isMarkedCopy = isMarked_;
for (auto &v: isMarkedCopy)
{
if (v.second == isMarked)
{
removeVertex(v.first);
}
}
}
// complexity: O(V^2*log(V))
template <typename T>
void Graph<T>::removeUnmarked(void)
{
removeMarked(false);
}
// depth-first search marking //////////////////////////////////////////////////
// complexity: O(V*log(V))
template <typename T>
void Graph<T>::depthFirstSearch(void)
{
depthFirstSearch(isMarked_.begin()->first);
}
// complexity: O(V*log(V))
template <typename T>
void Graph<T>::depthFirstSearch(const T &root)
{
std::vector<T> adjacentVertex;
mark(root);
adjacentVertex = getAdjacentVertices(root);
for (auto &v: adjacentVertex)
{
if (!isMarked(v))
{
depthFirstSearch(v);
}
}
}
// graph topological manipulations /////////////////////////////////////////////
// complexity: O(V*log(V))
template <typename T>
std::vector<T> Graph<T>::getAdjacentVertices(const T &value) const
{
std::vector<T> adjacentVertex;
auto pred = [&value](const Edge &e)
{
return ((e.first == value) or (e.second == value));
};
auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
while (eIt != edgeSet_.end())
{
if (eIt->first == value)
{
adjacentVertex.push_back((*eIt).second);
}
else if (eIt->second == value)
{
adjacentVertex.push_back((*eIt).first);
}
eIt = find_if(++eIt, edgeSet_.end(), pred);
}
return adjacentVertex;
}
// complexity: O(V*log(V))
template <typename T>
std::vector<T> Graph<T>::getChildren(const T &value) const
{
std::vector<T> child;
auto pred = [&value](const Edge &e)
{
return (e.first == value);
};
auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
while (eIt != edgeSet_.end())
{
child.push_back((*eIt).second);
eIt = find_if(++eIt, edgeSet_.end(), pred);
}
return child;
}
// complexity: O(V*log(V))
template <typename T>
std::vector<T> Graph<T>::getParents(const T &value) const
{
std::vector<T> parent;
auto pred = [&value](const Edge &e)
{
return (e.second == value);
};
auto eIt = find_if(edgeSet_.begin(), edgeSet_.end(), pred);
while (eIt != edgeSet_.end())
{
parent.push_back((*eIt).first);
eIt = find_if(++eIt, edgeSet_.end(), pred);
}
return parent;
}
// complexity: O(V^2*log(V))
template <typename T>
std::vector<T> Graph<T>::getRoots(void) const
{
std::vector<T> root;
for (auto &v: isMarked_)
{
auto parent = getParents(v.first);
if (parent.size() == 0)
{
root.push_back(v.first);
}
}
return root;
}
// complexity: O(V^2*log(V))
template <typename T>
std::vector<Graph<T>> Graph<T>::getConnectedComponents(void) const
{
std::vector<Graph<T>> res;
Graph<T> copy(*this);
while (copy.size() > 0)
{
copy.depthFirstSearch();
res.push_back(copy);
res.back().removeUnmarked();
res.back().unmarkAll();
copy.removeMarked();
copy.unmarkAll();
}
return res;
}
// topological sort using a directed DFS algorithm
// complexity: O(V*log(V))
template <typename T>
std::vector<T> Graph<T>::topoSort(void)
{
std::stack<T> buf;
std::vector<T> res;
const T *vPt;
std::map<T, bool> tmpMarked(isMarked_);
// visit function
std::function<void(const T &)> visit = [&](const T &v)
{
if (tmpMarked.at(v))
{
HADRON_ERROR("cannot topologically sort a cyclic graph");
}
if (!isMarked(v))
{
std::vector<T> child = getChildren(v);
tmpMarked[v] = true;
for (auto &c: child)
{
visit(c);
}
mark(v);
tmpMarked[v] = false;
buf.push(v);
}
};
// reset temporary marks
for (auto &v: tmpMarked)
{
tmpMarked.at(v.first) = false;
}
// loop on unmarked vertices
unmarkAll();
vPt = getFirstUnmarked();
while (vPt)
{
visit(*vPt);
vPt = getFirstUnmarked();
}
unmarkAll();
// create result vector
while (!buf.empty())
{
res.push_back(buf.top());
buf.pop();
}
return res;
}
// random version of the topological sort
// complexity: O(V*log(V))
template <typename T>
template <typename Gen>
std::vector<T> Graph<T>::topoSort(Gen &gen)
{
std::stack<T> buf;
std::vector<T> res;
const T *vPt;
std::map<T, bool> tmpMarked(isMarked_);
// visit function
std::function<void(const T &)> visit = [&](const T &v)
{
if (tmpMarked.at(v))
{
HADRON_ERROR("cannot topologically sort a cyclic graph");
}
if (!isMarked(v))
{
std::vector<T> child = getChildren(v);
tmpMarked[v] = true;
std::shuffle(child.begin(), child.end(), gen);
for (auto &c: child)
{
visit(c);
}
mark(v);
tmpMarked[v] = false;
buf.push(v);
}
};
// reset temporary marks
for (auto &v: tmpMarked)
{
tmpMarked.at(v.first) = false;
}
// loop on unmarked vertices
unmarkAll();
vPt = getRandomUnmarked(gen);
while (vPt)
{
visit(*vPt);
vPt = getRandomUnmarked(gen);
}
unmarkAll();
// create result vector
while (!buf.empty())
{
res.push_back(buf.top());
buf.pop();
}
return res;
}
// generate all possible topological sorts
// Y. L. Varol & D. Rotem, Comput. J. 24(1), pp. 83–84, 1981
// http://comjnl.oupjournals.org/cgi/doi/10.1093/comjnl/24.1.83
// complexity: O(V*log(V)) (from the paper, but really ?)
template <typename T>
std::vector<std::vector<T>> Graph<T>::allTopoSort(void)
{
std::vector<std::vector<T>> res;
std::map<T, std::map<T, bool>> iMat;
// create incidence matrix
for (auto &v1: isMarked_)
for (auto &v2: isMarked_)
{
iMat[v1.first][v2.first] = false;
}
for (auto &v: isMarked_)
{
auto cVec = getChildren(v.first);
for (auto &c: cVec)
{
iMat[v.first][c] = true;
}
}
// generate initial topological sort
res.push_back(topoSort());
// generate all other topological sorts by permutation
std::vector<T> p = res[0];
const unsigned int n = size();
std::vector<unsigned int> loc(n);
unsigned int i, k, k1;
T obj_k, obj_k1;
bool isFinal;
for (unsigned int j = 0; j < n; ++j)
{
loc[j] = j;
}
i = 0;
while (i < n-1)
{
k = loc[i];
k1 = k + 1;
obj_k = p[k];
if (k1 >= n)
{
isFinal = true;
obj_k1 = obj_k;
}
else
{
isFinal = false;
obj_k1 = p[k1];
}
if (iMat[res[0][i]][obj_k1] or isFinal)
{
for (unsigned int l = k; l >= i + 1; --l)
{
p[l] = p[l-1];
}
p[i] = obj_k;
loc[i] = i;
i++;
}
else
{
p[k] = obj_k1;
p[k1] = obj_k;
loc[i] = k1;
i = 0;
res.push_back(p);
}
}
return res;
}
// build depedency matrix from topological sorts ///////////////////////////////
// complexity: something like O(V^2*log(V!))
template <typename T>
std::map<T, std::map<T, bool>>
makeDependencyMatrix(const std::vector<std::vector<T>> &topSort)
{
std::map<T, std::map<T, bool>> m;
const std::vector<T> &vList = topSort[0];
for (auto &v1: vList)
for (auto &v2: vList)
{
bool dep = true;
for (auto &t: topSort)
{
auto i1 = std::find(t.begin(), t.end(), v1);
auto i2 = std::find(t.begin(), t.end(), v2);
dep = dep and (i1 - i2 > 0);
if (!dep) break;
}
m[v1][v2] = dep;
}
return m;
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Graph_hpp_

80
extras/Hadrons/HadronsXmlRun.cc Normal file
View File

@ -0,0 +1,80 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/HadronsXmlRun.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Application.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
int main(int argc, char *argv[])
{
// parse command line
std::string parameterFileName, scheduleFileName = "";
if (argc < 2)
{
std::cerr << "usage: " << argv[0] << " <parameter file> [<precomputed schedule>] [Grid options]";
std::cerr << std::endl;
std::exit(EXIT_FAILURE);
}
parameterFileName = argv[1];
if (argc > 2)
{
if (argv[2][0] != '-')
{
scheduleFileName = argv[2];
}
}
// initialization
Grid_init(&argc, &argv);
HadronsLogError.Active(GridLogError.isActive());
HadronsLogWarning.Active(GridLogWarning.isActive());
HadronsLogMessage.Active(GridLogMessage.isActive());
HadronsLogIterative.Active(GridLogIterative.isActive());
HadronsLogDebug.Active(GridLogDebug.isActive());
LOG(Message) << "Grid initialized" << std::endl;
// execution
Application application(parameterFileName);
application.parseParameterFile(parameterFileName);
if (!scheduleFileName.empty())
{
application.loadSchedule(scheduleFileName);
}
application.run();
// epilogue
LOG(Message) << "Grid is finalizing now" << std::endl;
Grid_finalize();
return EXIT_SUCCESS;
}

72
extras/Hadrons/HadronsXmlSchedule.cc Normal file
View File

@ -0,0 +1,72 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/HadronsXmlSchedule.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Application.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
int main(int argc, char *argv[])
{
// parse command line
std::string parameterFileName, scheduleFileName;
if (argc < 3)
{
std::cerr << "usage: " << argv[0] << " <parameter file> <schedule output> [Grid options]";
std::cerr << std::endl;
std::exit(EXIT_FAILURE);
}
parameterFileName = argv[1];
scheduleFileName = argv[2];
// initialization
Grid_init(&argc, &argv);
HadronsLogError.Active(GridLogError.isActive());
HadronsLogWarning.Active(GridLogWarning.isActive());
HadronsLogMessage.Active(GridLogMessage.isActive());
HadronsLogIterative.Active(GridLogIterative.isActive());
HadronsLogDebug.Active(GridLogDebug.isActive());
LOG(Message) << "Grid initialized" << std::endl;
// execution
Application application;
application.parseParameterFile(parameterFileName);
application.schedule();
application.printSchedule();
application.saveSchedule(scheduleFileName);
// epilogue
LOG(Message) << "Grid is finalizing now" << std::endl;
Grid_finalize();
return EXIT_SUCCESS;
}

29
extras/Hadrons/Makefile.am Normal file
View File

@ -0,0 +1,29 @@
lib_LIBRARIES = libHadrons.a
bin_PROGRAMS = HadronsXmlRun HadronsXmlSchedule
include modules.inc
libHadrons_a_SOURCES = \
$(modules_cc) \
Application.cc \
Environment.cc \
Global.cc \
Module.cc
libHadrons_adir = $(pkgincludedir)/Hadrons
nobase_libHadrons_a_HEADERS = \
$(modules_hpp) \
Application.hpp \
Environment.hpp \
Factory.hpp \
GeneticScheduler.hpp \
Global.hpp \
Graph.hpp \
Module.hpp \
Modules.hpp \
ModuleFactory.hpp
HadronsXmlRun_SOURCES = HadronsXmlRun.cc
HadronsXmlRun_LDADD = libHadrons.a -lGrid
HadronsXmlSchedule_SOURCES = HadronsXmlSchedule.cc
HadronsXmlSchedule_LDADD = libHadrons.a -lGrid

71
extras/Hadrons/Module.cc Normal file
View File

@ -0,0 +1,71 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Module.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Module.hpp>
using namespace Grid;
using namespace QCD;
using namespace Hadrons;
/******************************************************************************
* ModuleBase implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
ModuleBase::ModuleBase(const std::string name)
: name_(name)
, env_(Environment::getInstance())
{}
// access //////////////////////////////////////////////////////////////////////
std::string ModuleBase::getName(void) const
{
return name_;
}
Environment & ModuleBase::env(void) const
{
return env_;
}
// get factory registration name if available
std::string ModuleBase::getRegisteredName(void)
{
HADRON_ERROR("module '" + getName() + "' has a type not registered"
+ " in the factory");
}
// execution ///////////////////////////////////////////////////////////////////
void ModuleBase::operator()(void)
{
setup();
if (!env().isDryRun())
{
execute();
}
}

198
extras/Hadrons/Module.hpp Normal file
View File

@ -0,0 +1,198 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Module.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Module_hpp_
#define Hadrons_Module_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Environment.hpp>
BEGIN_HADRONS_NAMESPACE
// module registration macros
#define MODULE_REGISTER(mod, base)\
class mod: public base\
{\
public:\
typedef base Base;\
using Base::Base;\
virtual std::string getRegisteredName(void)\
{\
return std::string(#mod);\
}\
};\
class mod##ModuleRegistrar\
{\
public:\
mod##ModuleRegistrar(void)\
{\
ModuleFactory &modFac = ModuleFactory::getInstance();\
modFac.registerBuilder(#mod, [&](const std::string name)\
{\
return std::unique_ptr<mod>(new mod(name));\
});\
}\
};\
static mod##ModuleRegistrar mod##ModuleRegistrarInstance;
#define MODULE_REGISTER_NS(mod, base, ns)\
class mod: public base\
{\
public:\
typedef base Base;\
using Base::Base;\
virtual std::string getRegisteredName(void)\
{\
return std::string(#ns "::" #mod);\
}\
};\
class ns##mod##ModuleRegistrar\
{\
public:\
ns##mod##ModuleRegistrar(void)\
{\
ModuleFactory &modFac = ModuleFactory::getInstance();\
modFac.registerBuilder(#ns "::" #mod, [&](const std::string name)\
{\
return std::unique_ptr<ns::mod>(new ns::mod(name));\
});\
}\
};\
static ns##mod##ModuleRegistrar ns##mod##ModuleRegistrarInstance;
#define ARG(...) __VA_ARGS__
/******************************************************************************
* Module class *
******************************************************************************/
// base class
class ModuleBase
{
public:
// constructor
ModuleBase(const std::string name);
// destructor
virtual ~ModuleBase(void) = default;
// access
std::string getName(void) const;
Environment &env(void) const;
// get factory registration name if available
virtual std::string getRegisteredName(void);
// dependencies/products
virtual std::vector<std::string> getInput(void) = 0;
virtual std::vector<std::string> getOutput(void) = 0;
// parse parameters
virtual void parseParameters(XmlReader &reader, const std::string name) = 0;
virtual void saveParameters(XmlWriter &writer, const std::string name) = 0;
// setup
virtual void setup(void) {};
// execution
void operator()(void);
virtual void execute(void) = 0;
private:
std::string name_;
Environment &env_;
};
// derived class, templating the parameter class
template <typename P>
class Module: public ModuleBase
{
public:
typedef P Par;
public:
// constructor
Module(const std::string name);
// destructor
virtual ~Module(void) = default;
// parse parameters
virtual void parseParameters(XmlReader &reader, const std::string name);
virtual void saveParameters(XmlWriter &writer, const std::string name);
// parameter access
const P & par(void) const;
void setPar(const P &par);
private:
P par_;
};
// no parameter type
class NoPar {};
template <>
class Module<NoPar>: public ModuleBase
{
public:
// constructor
Module(const std::string name): ModuleBase(name) {};
// destructor
virtual ~Module(void) = default;
// parse parameters (do nothing)
virtual void parseParameters(XmlReader &reader, const std::string name) {};
virtual void saveParameters(XmlWriter &writer, const std::string name)
{
push(writer, "options");
pop(writer);
};
};
/******************************************************************************
* Template implementation *
******************************************************************************/
template <typename P>
Module<P>::Module(const std::string name)
: ModuleBase(name)
{}
template <typename P>
void Module<P>::parseParameters(XmlReader &reader, const std::string name)
{
read(reader, name, par_);
}
template <typename P>
void Module<P>::saveParameters(XmlWriter &writer, const std::string name)
{
write(writer, name, par_);
}
template <typename P>
const P & Module<P>::par(void) const
{
return par_;
}
template <typename P>
void Module<P>::setPar(const P &par)
{
par_ = par;
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Module_hpp_

49
extras/Hadrons/ModuleFactory.hpp Normal file
View File

@ -0,0 +1,49 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/ModuleFactory.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_ModuleFactory_hpp_
#define Hadrons_ModuleFactory_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Factory.hpp>
#include <Grid/Hadrons/Module.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* ModuleFactory *
******************************************************************************/
class ModuleFactory: public Factory<ModuleBase>
{
SINGLETON_DEFCTOR(ModuleFactory)
};
END_HADRONS_NAMESPACE
#endif // Hadrons_ModuleFactory_hpp_

40
extras/Hadrons/Modules.hpp Normal file
View File

@ -0,0 +1,40 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Modules/MAction/DWF.hpp>
#include <Grid/Hadrons/Modules/MAction/Wilson.hpp>
#include <Grid/Hadrons/Modules/MContraction/Baryon.hpp>
#include <Grid/Hadrons/Modules/MContraction/Meson.hpp>
#include <Grid/Hadrons/Modules/MGauge/Load.hpp>
#include <Grid/Hadrons/Modules/MGauge/Random.hpp>
#include <Grid/Hadrons/Modules/MGauge/Unit.hpp>
#include <Grid/Hadrons/Modules/MSolver/RBPrecCG.hpp>
#include <Grid/Hadrons/Modules/MSource/Point.hpp>
#include <Grid/Hadrons/Modules/MSource/SeqGamma.hpp>
#include <Grid/Hadrons/Modules/MSource/Z2.hpp>
#include <Grid/Hadrons/Modules/Quark.hpp>

134
extras/Hadrons/Modules/MAction/DWF.hpp Normal file
View File

@ -0,0 +1,134 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MAction/DWF.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_DWF_hpp_
#define Hadrons_DWF_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Domain wall quark action *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MAction)
class DWFPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(DWFPar,
std::string, gauge,
unsigned int, Ls,
double , mass,
double , M5);
};
template <typename FImpl>
class TDWF: public Module<DWFPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TDWF(const std::string name);
// destructor
virtual ~TDWF(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(DWF, TDWF<FIMPL>, MAction);
/******************************************************************************
* DWF template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TDWF<FImpl>::TDWF(const std::string name)
: Module<DWFPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TDWF<FImpl>::getInput(void)
{
std::vector<std::string> in = {par().gauge};
return in;
}
template <typename FImpl>
std::vector<std::string> TDWF<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TDWF<FImpl>::setup(void)
{
unsigned int size;
size = 2*env().template lattice4dSize<typename FImpl::DoubledGaugeField>();
env().registerObject(getName(), size, par().Ls);
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TDWF<FImpl>::execute(void)
{
LOG(Message) << "Setting up domain wall fermion matrix with m= "
<< par().mass << ", M5= " << par().M5 << " and Ls= "
<< par().Ls << " using gauge field '" << par().gauge << "'"
<< std::endl;
env().createGrid(par().Ls);
auto &U = *env().template getObject<LatticeGaugeField>(par().gauge);
auto &g4 = *env().getGrid();
auto &grb4 = *env().getRbGrid();
auto &g5 = *env().getGrid(par().Ls);
auto &grb5 = *env().getRbGrid(par().Ls);
FMat *fMatPt = new DomainWallFermion<FImpl>(U, g5, grb5, g4, grb4,
par().mass, par().M5);
env().setObject(getName(), fMatPt);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_DWF_hpp_

126
extras/Hadrons/Modules/MAction/Wilson.hpp Normal file
View File

@ -0,0 +1,126 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MAction/Wilson.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Wilson_hpp_
#define Hadrons_Wilson_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* TWilson quark action *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MAction)
class WilsonPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonPar,
std::string, gauge,
double , mass);
};
template <typename FImpl>
class TWilson: public Module<WilsonPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TWilson(const std::string name);
// destructor
virtual ~TWilson(void) = default;
// dependencies/products
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Wilson, TWilson<FIMPL>, MAction);
/******************************************************************************
* TWilson template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TWilson<FImpl>::TWilson(const std::string name)
: Module<WilsonPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TWilson<FImpl>::getInput(void)
{
std::vector<std::string> in = {par().gauge};
return in;
}
template <typename FImpl>
std::vector<std::string> TWilson<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TWilson<FImpl>::setup(void)
{
unsigned int size;
size = 2*env().template lattice4dSize<typename FImpl::DoubledGaugeField>();
env().registerObject(getName(), size);
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TWilson<FImpl>::execute()
{
LOG(Message) << "Setting up TWilson fermion matrix with m= " << par().mass
<< " using gauge field '" << par().gauge << "'" << std::endl;
auto &U = *env().template getObject<LatticeGaugeField>(par().gauge);
auto &grid = *env().getGrid();
auto &gridRb = *env().getRbGrid();
FMat *fMatPt = new WilsonFermion<FImpl>(U, grid, gridRb, par().mass);
env().setObject(getName(), fMatPt);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Wilson_hpp_

131
extras/Hadrons/Modules/MContraction/Baryon.hpp Normal file
View File

@ -0,0 +1,131 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MContraction/Baryon.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Baryon_hpp_
#define Hadrons_Baryon_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Baryon *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MContraction)
class BaryonPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(BaryonPar,
std::string, q1,
std::string, q2,
std::string, q3,
std::string, output);
};
template <typename FImpl1, typename FImpl2, typename FImpl3>
class TBaryon: public Module<BaryonPar>
{
public:
TYPE_ALIASES(FImpl1, 1);
TYPE_ALIASES(FImpl2, 2);
TYPE_ALIASES(FImpl3, 3);
class Result: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
std::vector<std::vector<std::vector<Complex>>>, corr);
};
public:
// constructor
TBaryon(const std::string name);
// destructor
virtual ~TBaryon(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Baryon, ARG(TBaryon<FIMPL, FIMPL, FIMPL>), MContraction);
/******************************************************************************
* TBaryon implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2, typename FImpl3>
TBaryon<FImpl1, FImpl2, FImpl3>::TBaryon(const std::string name)
: Module<BaryonPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2, typename FImpl3>
std::vector<std::string> TBaryon<FImpl1, FImpl2, FImpl3>::getInput(void)
{
std::vector<std::string> input = {par().q1, par().q2, par().q3};
return input;
}
template <typename FImpl1, typename FImpl2, typename FImpl3>
std::vector<std::string> TBaryon<FImpl1, FImpl2, FImpl3>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2, typename FImpl3>
void TBaryon<FImpl1, FImpl2, FImpl3>::execute(void)
{
LOG(Message) << "Computing baryon contractions '" << getName() << "' using"
<< " quarks '" << par().q1 << "', '" << par().q2 << "', and '"
<< par().q3 << "'" << std::endl;
XmlWriter writer(par().output);
PropagatorField1 &q1 = *env().template getObject<PropagatorField1>(par().q1);
PropagatorField2 &q2 = *env().template getObject<PropagatorField2>(par().q2);
PropagatorField3 &q3 = *env().template getObject<PropagatorField3>(par().q2);
LatticeComplex c(env().getGrid());
Result result;
// FIXME: do contractions
write(writer, "meson", result);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Baryon_hpp_

148
extras/Hadrons/Modules/MContraction/Meson.hpp Normal file
View File

@ -0,0 +1,148 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MContraction/Meson.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Meson_hpp_
#define Hadrons_Meson_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* TMeson *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MContraction)
class MesonPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(MesonPar,
std::string, q1,
std::string, q2,
std::string, output);
};
template <typename FImpl1, typename FImpl2>
class TMeson: public Module<MesonPar>
{
public:
TYPE_ALIASES(FImpl1, 1);
TYPE_ALIASES(FImpl2, 2);
class Result: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(Result,
std::vector<std::vector<std::vector<Complex>>>, corr);
};
public:
// constructor
TMeson(const std::string name);
// destructor
virtual ~TMeson(void) = default;
// dependencies/products
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Meson, ARG(TMeson<FIMPL, FIMPL>), MContraction);
/******************************************************************************
* TMeson implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2>
TMeson<FImpl1, FImpl2>::TMeson(const std::string name)
: Module<MesonPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2>
std::vector<std::string> TMeson<FImpl1, FImpl2>::getInput(void)
{
std::vector<std::string> input = {par().q1, par().q2};
return input;
}
template <typename FImpl1, typename FImpl2>
std::vector<std::string> TMeson<FImpl1, FImpl2>::getOutput(void)
{
std::vector<std::string> output = {getName()};
return output;
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl1, typename FImpl2>
void TMeson<FImpl1, FImpl2>::execute(void)
{
LOG(Message) << "Computing meson contractions '" << getName() << "' using"
<< " quarks '" << par().q1 << "' and '" << par().q2 << "'"
<< std::endl;
XmlWriter writer(par().output);
PropagatorField1 &q1 = *env().template getObject<PropagatorField1>(par().q1);
PropagatorField2 &q2 = *env().template getObject<PropagatorField2>(par().q2);
LatticeComplex c(env().getGrid());
SpinMatrix g[Ns*Ns], g5;
std::vector<TComplex> buf;
Result result;
g5 = makeGammaProd(Ns*Ns - 1);
result.corr.resize(Ns*Ns);
for (unsigned int i = 0; i < Ns*Ns; ++i)
{
g[i] = makeGammaProd(i);
}
for (unsigned int iSink = 0; iSink < Ns*Ns; ++iSink)
{
result.corr[iSink].resize(Ns*Ns);
for (unsigned int iSrc = 0; iSrc < Ns*Ns; ++iSrc)
{
c = trace(g[iSink]*q1*g[iSrc]*g5*adj(q2)*g5);
sliceSum(c, buf, Tp);
result.corr[iSink][iSrc].resize(buf.size());
for (unsigned int t = 0; t < buf.size(); ++t)
{
result.corr[iSink][iSrc][t] = TensorRemove(buf[t]);
}
}
}
write(writer, "meson", result);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Meson_hpp_

78
extras/Hadrons/Modules/MGauge/Load.cc Normal file
View File

@ -0,0 +1,78 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Load.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Modules/MGauge/Load.hpp>
using namespace Grid;
using namespace Hadrons;
using namespace MGauge;
/******************************************************************************
* TLoad implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
TLoad::TLoad(const std::string name)
: Module<LoadPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
std::vector<std::string> TLoad::getInput(void)
{
std::vector<std::string> in;
return in;
}
std::vector<std::string> TLoad::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
void TLoad::setup(void)
{
env().registerLattice<LatticeGaugeField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
void TLoad::execute(void)
{
NerscField header;
std::string fileName = par().file + "."
+ std::to_string(env().getTrajectory());
LOG(Message) << "Loading NERSC configuration from file '" << fileName
<< "'" << std::endl;
LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
NerscIO::readConfiguration(U, header, fileName);
LOG(Message) << "NERSC header:" << std::endl;
dump_nersc_header(header, LOG(Message));
}

73
extras/Hadrons/Modules/MGauge/Load.hpp Normal file
View File

@ -0,0 +1,73 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Load.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Load_hpp_
#define Hadrons_Load_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Load a NERSC configuration *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MGauge)
class LoadPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(LoadPar,
std::string, file);
};
class TLoad: public Module<LoadPar>
{
public:
// constructor
TLoad(const std::string name);
// destructor
virtual ~TLoad(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Load, TLoad, MGauge);
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Load_hpp_

69
extras/Hadrons/Modules/MGauge/Random.cc Normal file
View File

@ -0,0 +1,69 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Random.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Modules/MGauge/Random.hpp>
using namespace Grid;
using namespace Hadrons;
using namespace MGauge;
/******************************************************************************
* TRandom implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
TRandom::TRandom(const std::string name)
: Module<NoPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
std::vector<std::string> TRandom::getInput(void)
{
return std::vector<std::string>();
}
std::vector<std::string> TRandom::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
void TRandom::setup(void)
{
env().registerLattice<LatticeGaugeField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
void TRandom::execute(void)
{
LOG(Message) << "Generating random gauge configuration" << std::endl;
LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
SU3::HotConfiguration(*env().get4dRng(), U);
}

66
extras/Hadrons/Modules/MGauge/Random.hpp Normal file
View File

@ -0,0 +1,66 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Random.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Random_hpp_
#define Hadrons_Random_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Random gauge *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MGauge)
class TRandom: public Module<NoPar>
{
public:
// constructor
TRandom(const std::string name);
// destructor
virtual ~TRandom(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Random, TRandom, MGauge);
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Random_hpp_

69
extras/Hadrons/Modules/MGauge/Unit.cc Normal file
View File

@ -0,0 +1,69 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Unit.cc
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
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/Hadrons/Modules/MGauge/Unit.hpp>
using namespace Grid;
using namespace Hadrons;
using namespace MGauge;
/******************************************************************************
* TUnit implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
TUnit::TUnit(const std::string name)
: Module<NoPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
std::vector<std::string> TUnit::getInput(void)
{
return std::vector<std::string>();
}
std::vector<std::string> TUnit::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
void TUnit::setup(void)
{
env().registerLattice<LatticeGaugeField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
void TUnit::execute(void)
{
LOG(Message) << "Creating unit gauge configuration" << std::endl;
LatticeGaugeField &U = *env().createLattice<LatticeGaugeField>(getName());
SU3::ColdConfiguration(*env().get4dRng(), U);
}

66
extras/Hadrons/Modules/MGauge/Unit.hpp Normal file
View File

@ -0,0 +1,66 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MGauge/Unit.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Unit_hpp_
#define Hadrons_Unit_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Unit gauge *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MGauge)
class TUnit: public Module<NoPar>
{
public:
// constructor
TUnit(const std::string name);
// destructor
virtual ~TUnit(void) = default;
// dependencies/products
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Unit, TUnit, MGauge);
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Unit_hpp_

132
extras/Hadrons/Modules/MSolver/RBPrecCG.hpp Normal file
View File

@ -0,0 +1,132 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MSolver/RBPrecCG.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_RBPrecCG_hpp_
#define Hadrons_RBPrecCG_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* Schur red-black preconditioned CG *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MSolver)
class RBPrecCGPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(RBPrecCGPar,
std::string, action,
double , residual);
};
template <typename FImpl>
class TRBPrecCG: public Module<RBPrecCGPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TRBPrecCG(const std::string name);
// destructor
virtual ~TRBPrecCG(void) = default;
// dependencies/products
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(RBPrecCG, TRBPrecCG<FIMPL>, MSolver);
/******************************************************************************
* TRBPrecCG template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TRBPrecCG<FImpl>::TRBPrecCG(const std::string name)
: Module(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TRBPrecCG<FImpl>::getInput(void)
{
std::vector<std::string> in = {par().action};
return in;
}
template <typename FImpl>
std::vector<std::string> TRBPrecCG<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TRBPrecCG<FImpl>::setup(void)
{
auto Ls = env().getObjectLs(par().action);
env().registerObject(getName(), 0, Ls);
env().addOwnership(getName(), par().action);
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TRBPrecCG<FImpl>::execute(void)
{
auto &mat = *(env().template getObject<FMat>(par().action));
auto solver = [&mat, this](FermionField &sol, const FermionField &source)
{
ConjugateGradient<FermionField> cg(par().residual, 10000);
SchurRedBlackDiagMooeeSolve<FermionField> schurSolver(cg);
schurSolver(mat, source, sol);
};
LOG(Message) << "setting up Schur red-black preconditioned CG for"
<< " action '" << par().action << "' with residual "
<< par().residual << std::endl;
env().setObject(getName(), new SolverFn(solver));
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_RBPrecCG_hpp_

135
extras/Hadrons/Modules/MSource/Point.hpp Normal file
View File

@ -0,0 +1,135 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MSource/Point.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Point_hpp_
#define Hadrons_Point_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/*
Point source
------------
* src_x = delta_x,position
* options:
- position: space-separated integer sequence (e.g. "0 1 1 0")
*/
/******************************************************************************
* TPoint *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MSource)
class PointPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(PointPar,
std::string, position);
};
template <typename FImpl>
class TPoint: public Module<PointPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TPoint(const std::string name);
// destructor
virtual ~TPoint(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Point, TPoint<FIMPL>, MSource);
/******************************************************************************
* TPoint template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TPoint<FImpl>::TPoint(const std::string name)
: Module<PointPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TPoint<FImpl>::getInput(void)
{
std::vector<std::string> in;
return in;
}
template <typename FImpl>
std::vector<std::string> TPoint<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TPoint<FImpl>::setup(void)
{
env().template registerLattice<PropagatorField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TPoint<FImpl>::execute(void)
{
std::vector<int> position = strToVec<int>(par().position);
typename SitePropagator::scalar_object id;
LOG(Message) << "Creating point source at position [" << par().position
<< "]" << std::endl;
PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
id = 1.;
src = zero;
pokeSite(id, src, position);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Point_hpp_

164
extras/Hadrons/Modules/MSource/SeqGamma.hpp Normal file
View File

@ -0,0 +1,164 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MSource/SeqGamma.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_SeqGamma_hpp_
#define Hadrons_SeqGamma_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/*
Sequential source
-----------------------------
* src_x = q_x * theta(x_3 - tA) * theta(tB - x_3) * gamma * exp(i x.mom)
* options:
- q: input propagator (string)
- tA: begin timeslice (integer)
- tB: end timesilce (integer)
- gamma: gamma product to insert (integer)
- mom: momentum insertion, space-separated float sequence (e.g ".1 .2 1. 0.")
*/
/******************************************************************************
* SeqGamma *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MSource)
class SeqGammaPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(SeqGammaPar,
std::string, q,
unsigned int, tA,
unsigned int, tB,
unsigned int, gamma,
std::string, mom);
};
template <typename FImpl>
class TSeqGamma: public Module<SeqGammaPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TSeqGamma(const std::string name);
// destructor
virtual ~TSeqGamma(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(SeqGamma, TSeqGamma<FIMPL>, MSource);
/******************************************************************************
* TSeqGamma implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TSeqGamma<FImpl>::TSeqGamma(const std::string name)
: Module<SeqGammaPar>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TSeqGamma<FImpl>::getInput(void)
{
std::vector<std::string> in = {par().q};
return in;
}
template <typename FImpl>
std::vector<std::string> TSeqGamma<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TSeqGamma<FImpl>::setup(void)
{
env().template registerLattice<PropagatorField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TSeqGamma<FImpl>::execute(void)
{
if (par().tA == par().tB)
{
LOG(Message) << "Generating gamma_" << par().gamma
<< " sequential source at t= " << par().tA << std::endl;
}
else
{
LOG(Message) << "Generating gamma_" << par().gamma
<< " sequential source for "
<< par().tA << " <= t <= " << par().tB << std::endl;
}
PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
PropagatorField &q = *env().template getObject<PropagatorField>(par().q);
Lattice<iScalar<vInteger>> t(env().getGrid());
LatticeComplex ph(env().getGrid()), coor(env().getGrid());
SpinMatrix g;
std::vector<Real> p;
Complex i(0.0,1.0);
g = makeGammaProd(par().gamma);
p = strToVec<Real>(par().mom);
ph = zero;
for(unsigned int mu = 0; mu < env().getNd(); mu++)
{
LatticeCoordinate(coor, mu);
ph = ph + p[mu]*coor;
}
ph = exp(i*ph);
LatticeCoordinate(t, Tp);
src = where((t >= par().tA) and (t <= par().tB), g*ph*q, 0.*q);
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_SeqGamma_hpp_

151
extras/Hadrons/Modules/MSource/Z2.hpp Normal file
View File

@ -0,0 +1,151 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/MSource/Z2.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Z2_hpp_
#define Hadrons_Z2_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/*
Z_2 stochastic source
-----------------------------
* src_x = eta_x * theta(x_3 - tA) * theta(tB - x_3)
the eta_x are independent uniform random numbers in {+/- 1 +/- i}
* options:
- tA: begin timeslice (integer)
- tB: end timesilce (integer)
*/
/******************************************************************************
* Z2 stochastic source *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(MSource)
class Z2Par: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(Z2Par,
unsigned int, tA,
unsigned int, tB);
};
template <typename FImpl>
class TZ2: public Module<Z2Par>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TZ2(const std::string name);
// destructor
virtual ~TZ2(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(Z2, TZ2<FIMPL>, MSource);
/******************************************************************************
* TZ2 template implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TZ2<FImpl>::TZ2(const std::string name)
: Module<Z2Par>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TZ2<FImpl>::getInput(void)
{
std::vector<std::string> in;
return in;
}
template <typename FImpl>
std::vector<std::string> TZ2<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TZ2<FImpl>::setup(void)
{
env().template registerLattice<PropagatorField>(getName());
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TZ2<FImpl>::execute(void)
{
Lattice<iScalar<vInteger>> t(env().getGrid());
LatticeComplex eta(env().getGrid());
Complex shift(1., 1.);
if (par().tA == par().tB)
{
LOG(Message) << "Generating Z_2 wall source at t= " << par().tA
<< std::endl;
}
else
{
LOG(Message) << "Generating Z_2 band for " << par().tA << " <= t <= "
<< par().tB << std::endl;
}
PropagatorField &src = *env().template createLattice<PropagatorField>(getName());
LatticeCoordinate(t, Tp);
bernoulli(*env().get4dRng(), eta);
eta = (2.*eta - shift)*(1./::sqrt(2.));
eta = where((t >= par().tA) and (t <= par().tB), eta, 0.*eta);
src = 1.;
src = src*eta;
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons_Z2_hpp_

185
extras/Hadrons/Modules/Quark.hpp Normal file
View File

@ -0,0 +1,185 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: extras/Hadrons/Modules/Quark.hpp
Copyright (C) 2015
Copyright (C) 2016
Author: Antonin Portelli <antonin.portelli@me.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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef Hadrons_Quark_hpp_
#define Hadrons_Quark_hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* TQuark *
******************************************************************************/
class QuarkPar: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(QuarkPar,
std::string, source,
std::string, solver);
};
template <typename FImpl>
class TQuark: public Module<QuarkPar>
{
public:
TYPE_ALIASES(FImpl,);
public:
// constructor
TQuark(const std::string name);
// destructor
virtual ~TQuark(void) = default;
// dependencies/products
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
private:
unsigned int Ls_;
SolverFn *solver_{nullptr};
};
MODULE_REGISTER(Quark, TQuark<FIMPL>);
/******************************************************************************
* TQuark implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
TQuark<FImpl>::TQuark(const std::string name)
: Module(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> TQuark<FImpl>::getInput(void)
{
std::vector<std::string> in = {par().source, par().solver};
return in;
}
template <typename FImpl>
std::vector<std::string> TQuark<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName(), getName() + "_5d"};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void TQuark<FImpl>::setup(void)
{
Ls_ = env().getObjectLs(par().solver);
env().template registerLattice<PropagatorField>(getName());
if (Ls_ > 1)
{
env().template registerLattice<PropagatorField>(getName() + "_5d", Ls_);
}
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void TQuark<FImpl>::execute(void)
{
LOG(Message) << "Computing quark propagator '" << getName() << "'"
<< std::endl;
FermionField source(env().getGrid(Ls_)), sol(env().getGrid(Ls_)),
tmp(env().getGrid());
std::string propName = (Ls_ == 1) ? getName() : (getName() + "_5d");
PropagatorField &prop = *env().template createLattice<PropagatorField>(propName);
PropagatorField &fullSrc = *env().template getObject<PropagatorField>(par().source);
SolverFn &solver = *env().template getObject<SolverFn>(par().solver);
if (Ls_ > 1)
{
env().template createLattice<PropagatorField>(getName());
}
LOG(Message) << "Inverting using solver '" << par().solver
<< "' on source '" << par().source << "'" << std::endl;
for (unsigned int s = 0; s < Ns; ++s)
for (unsigned int c = 0; c < Nc; ++c)
{
LOG(Message) << "Inversion for spin= " << s << ", color= " << c
<< std::endl;
// source conversion for 4D sources
if (!env().isObject5d(par().source))
{
if (Ls_ == 1)
{
PropToFerm(source, fullSrc, s, c);
}
else
{
source = zero;
PropToFerm(tmp, fullSrc, s, c);
InsertSlice(tmp, source, 0, 0);
InsertSlice(tmp, source, Ls_-1, 0);
axpby_ssp_pplus(source, 0., source, 1., source, 0, 0);
axpby_ssp_pminus(source, 0., source, 1., source, Ls_-1, Ls_-1);
}
}
// source conversion for 5D sources
else
{
if (Ls_ != env().getObjectLs(par().source))
{
HADRON_ERROR("Ls mismatch between quark action and source");
}
else
{
PropToFerm(source, fullSrc, s, c);
}
}
sol = zero;
solver(sol, source);
FermToProp(prop, sol, s, c);
// create 4D propagators from 5D one if necessary
if (Ls_ > 1)
{
PropagatorField &p4d =
*env().template getObject<PropagatorField>(getName());
axpby_ssp_pminus(sol, 0., sol, 1., sol, 0, 0);
axpby_ssp_pplus(sol, 0., sol, 1., sol, 0, Ls_-1);
ExtractSlice(tmp, sol, 0, 0);
FermToProp(p4d, tmp, s, c);
}
}
}
END_HADRONS_NAMESPACE
#endif // Hadrons_Quark_hpp_

39
extras/Hadrons/Modules/templates/Module.cc.template Normal file
View File

@ -0,0 +1,39 @@
#include <Grid/Hadrons/Modules/___FILEBASENAME___.hpp>
using namespace Grid;
using namespace Hadrons;
/******************************************************************************
* T___FILEBASENAME___ implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
T___FILEBASENAME___::T___FILEBASENAME___(const std::string name)
: Module<___FILEBASENAME___Par>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
std::vector<std::string> T___FILEBASENAME___::getInput(void)
{
std::vector<std::string> in;
return in;
}
std::vector<std::string> T___FILEBASENAME___::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
void T___FILEBASENAME___::setup(void)
{
}
// execution ///////////////////////////////////////////////////////////////////
void T___FILEBASENAME___::execute(void)
{
}

40
extras/Hadrons/Modules/templates/Module.hpp.template Normal file
View File

@ -0,0 +1,40 @@
#ifndef Hadrons____FILEBASENAME____hpp_
#define Hadrons____FILEBASENAME____hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* ___FILEBASENAME___ *
******************************************************************************/
class ___FILEBASENAME___Par: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
unsigned int, i);
};
class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
{
public:
// constructor
T___FILEBASENAME___(const std::string name);
// destructor
virtual ~T___FILEBASENAME___(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER(___FILEBASENAME___, T___FILEBASENAME___);
END_HADRONS_NAMESPACE
#endif // Hadrons____FILEBASENAME____hpp_

40
extras/Hadrons/Modules/templates/Module_in_NS.cc.template Normal file
View File

@ -0,0 +1,40 @@
#include <Grid/Hadrons/Modules/___NAMESPACE___/___FILEBASENAME___.hpp>
using namespace Grid;
using namespace Hadrons;
using namespace ___NAMESPACE___;
/******************************************************************************
* T___FILEBASENAME___ implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
T___FILEBASENAME___::T___FILEBASENAME___(const std::string name)
: Module<___FILEBASENAME___Par>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
std::vector<std::string> T___FILEBASENAME___::getInput(void)
{
std::vector<std::string> in;
return in;
}
std::vector<std::string> T___FILEBASENAME___::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
void T___FILEBASENAME___::setup(void)
{
}
// execution ///////////////////////////////////////////////////////////////////
void T___FILEBASENAME___::execute(void)
{
}

44
extras/Hadrons/Modules/templates/Module_in_NS.hpp.template Normal file
View File

@ -0,0 +1,44 @@
#ifndef Hadrons____FILEBASENAME____hpp_
#define Hadrons____FILEBASENAME____hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* ___FILEBASENAME___ *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(___NAMESPACE___)
class ___FILEBASENAME___Par: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
unsigned int, i);
};
class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
{
public:
// constructor
T___FILEBASENAME___(const std::string name);
// destructor
virtual ~T___FILEBASENAME___(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(___FILEBASENAME___, T___FILEBASENAME___, ___NAMESPACE___);
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons____FILEBASENAME____hpp_

81
extras/Hadrons/Modules/templates/Module_tmp.hpp.template Normal file
View File

@ -0,0 +1,81 @@
#ifndef Hadrons____FILEBASENAME____hpp_
#define Hadrons____FILEBASENAME____hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* ___FILEBASENAME___ *
******************************************************************************/
class ___FILEBASENAME___Par: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
unsigned int, i);
};
template <typename FImpl>
class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
{
public:
// constructor
T___FILEBASENAME___(const std::string name);
// destructor
virtual ~T___FILEBASENAME___(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER(___FILEBASENAME___, T___FILEBASENAME___<FIMPL>);
/******************************************************************************
* T___FILEBASENAME___ implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
T___FILEBASENAME___<FImpl>::T___FILEBASENAME___(const std::string name)
: Module<___FILEBASENAME___Par>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> T___FILEBASENAME___<FImpl>::getInput(void)
{
std::vector<std::string> in;
return in;
}
template <typename FImpl>
std::vector<std::string> T___FILEBASENAME___<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void T___FILEBASENAME___<FImpl>::setup(void)
{
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void T___FILEBASENAME___<FImpl>::execute(void)
{
}
END_HADRONS_NAMESPACE
#endif // Hadrons____FILEBASENAME____hpp_

85
extras/Hadrons/Modules/templates/Module_tmp_in_NS.hpp.template Normal file
View File

@ -0,0 +1,85 @@
#ifndef Hadrons____FILEBASENAME____hpp_
#define Hadrons____FILEBASENAME____hpp_
#include <Grid/Hadrons/Global.hpp>
#include <Grid/Hadrons/Module.hpp>
#include <Grid/Hadrons/ModuleFactory.hpp>
BEGIN_HADRONS_NAMESPACE
/******************************************************************************
* ___FILEBASENAME___ *
******************************************************************************/
BEGIN_MODULE_NAMESPACE(___NAMESPACE___)
class ___FILEBASENAME___Par: Serializable
{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(___FILEBASENAME___Par,
unsigned int, i);
};
template <typename FImpl>
class T___FILEBASENAME___: public Module<___FILEBASENAME___Par>
{
public:
// constructor
T___FILEBASENAME___(const std::string name);
// destructor
virtual ~T___FILEBASENAME___(void) = default;
// dependency relation
virtual std::vector<std::string> getInput(void);
virtual std::vector<std::string> getOutput(void);
// setup
virtual void setup(void);
// execution
virtual void execute(void);
};
MODULE_REGISTER_NS(___FILEBASENAME___, T___FILEBASENAME___<FIMPL>, ___NAMESPACE___);
/******************************************************************************
* T___FILEBASENAME___ implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
template <typename FImpl>
T___FILEBASENAME___<FImpl>::T___FILEBASENAME___(const std::string name)
: Module<___FILEBASENAME___Par>(name)
{}
// dependencies/products ///////////////////////////////////////////////////////
template <typename FImpl>
std::vector<std::string> T___FILEBASENAME___<FImpl>::getInput(void)
{
std::vector<std::string> in;
return in;
}
template <typename FImpl>
std::vector<std::string> T___FILEBASENAME___<FImpl>::getOutput(void)
{
std::vector<std::string> out = {getName()};
return out;
}
// setup ///////////////////////////////////////////////////////////////////////
template <typename FImpl>
void T___FILEBASENAME___<FImpl>::setup(void)
{
}
// execution ///////////////////////////////////////////////////////////////////
template <typename FImpl>
void T___FILEBASENAME___<FImpl>::execute(void)
{
}
END_MODULE_NAMESPACE
END_HADRONS_NAMESPACE
#endif // Hadrons____FILEBASENAME____hpp_

31
extras/Hadrons/add_module.sh Executable file
View File

@ -0,0 +1,31 @@
#!/usr/bin/env bash
if (( $# != 1 && $# != 2)); then
echo "usage: `basename $0` <module name> [<namespace>]" 1>&2
exit 1
fi
NAME=$1
NS=$2
if (( $# == 1 )); then
if [ -e "Modules/${NAME}.cc" ] || [ -e "Modules/${NAME}.hpp" ]; then
echo "error: files Modules/${NAME}.* already exists" 1>&2
exit 1
fi
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module.cc.template > Modules/${NAME}.cc
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module.hpp.template > Modules/${NAME}.hpp
elif (( $# == 2 )); then
mkdir -p Modules/${NS}
if [ -e "Modules/${NS}/${NAME}.cc" ] || [ -e "Modules/${NS}/${NAME}.hpp" ]; then
echo "error: files Modules/${NS}/${NAME}.* already exists" 1>&2
exit 1
fi
TMPCC=".${NS}.${NAME}.tmp.cc"
TMPHPP=".${NS}.${NAME}.tmp.hpp"
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_in_NS.cc.template > ${TMPCC}
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_in_NS.hpp.template > ${TMPHPP}
sed "s/___NAMESPACE___/${NS}/g" ${TMPCC} > Modules/${NS}/${NAME}.cc
sed "s/___NAMESPACE___/${NS}/g" ${TMPHPP} > Modules/${NS}/${NAME}.hpp
rm -f ${TMPCC} ${TMPHPP}
fi
./make_module_list.sh

28
extras/Hadrons/add_module_template.sh Executable file
View File

@ -0,0 +1,28 @@
#!/usr/bin/env bash
if (( $# != 1 && $# != 2)); then
echo "usage: `basename $0` <module name> [<namespace>]" 1>&2
exit 1
fi
NAME=$1
NS=$2
if (( $# == 1 )); then
if [ -e "Modules/${NAME}.cc" ] || [ -e "Modules/${NAME}.hpp" ]; then
echo "error: files Modules/${NAME}.* already exists" 1>&2
exit 1
fi
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_tmp.hpp.template > Modules/${NAME}.hpp
elif (( $# == 2 )); then
mkdir -p Modules/${NS}
if [ -e "Modules/${NS}/${NAME}.cc" ] || [ -e "Modules/${NS}/${NAME}.hpp" ]; then
echo "error: files Modules/${NS}/${NAME}.* already exists" 1>&2
exit 1
fi
TMPCC=".${NS}.${NAME}.tmp.cc"
TMPHPP=".${NS}.${NAME}.tmp.hpp"
sed "s/___FILEBASENAME___/${NAME}/g" Modules/templates/Module_tmp_in_NS.hpp.template > ${TMPHPP}
sed "s/___NAMESPACE___/${NS}/g" ${TMPHPP} > Modules/${NS}/${NAME}.hpp
rm -f ${TMPCC} ${TMPHPP}
fi
./make_module_list.sh

12
extras/Hadrons/make_module_list.sh Executable file
View File

@ -0,0 +1,12 @@
#!/usr/bin/env bash
echo 'modules_cc =\' > modules.inc
find Modules -name '*.cc' -type f -print | sed 's/^/ /;$q;s/$/ \\/' >> modules.inc
echo '' >> modules.inc
echo 'modules_hpp =\' >> modules.inc
find Modules -name '*.hpp' -type f -print | sed 's/^/ /;$q;s/$/ \\/' >> modules.inc
echo '' >> modules.inc
rm -f Modules.hpp
for f in `find Modules -name '*.hpp'`; do
echo "#include <Grid/Hadrons/${f}>" >> Modules.hpp
done

19
extras/Hadrons/modules.inc Normal file
View File

@ -0,0 +1,19 @@
modules_cc =\
Modules/MGauge/Load.cc \
Modules/MGauge/Random.cc \
Modules/MGauge/Unit.cc
modules_hpp =\
Modules/MAction/DWF.hpp \
Modules/MAction/Wilson.hpp \
Modules/MContraction/Baryon.hpp \
Modules/MContraction/Meson.hpp \
Modules/MGauge/Load.hpp \
Modules/MGauge/Random.hpp \
Modules/MGauge/Unit.hpp \
Modules/MSolver/RBPrecCG.hpp \
Modules/MSource/Point.hpp \
Modules/MSource/SeqGamma.hpp \
Modules/MSource/Z2.hpp \
Modules/Quark.hpp

1
extras/Makefile.am Normal file
View File

@ -0,0 +1 @@
SUBDIRS = Hadrons

65
lib/AlignedAllocator.cc Normal file
View File

@ -0,0 +1,65 @@
#include <Grid/Grid.h>
namespace Grid {
int PointerCache::victim;
PointerCache::PointerCacheEntry PointerCache::Entries[PointerCache::Ncache];
void *PointerCache::Insert(void *ptr,size_t bytes) {
if (bytes < 4096 ) return NULL;
#ifdef _OPENMP
assert(omp_in_parallel()==0);
#endif
void * ret = NULL;
int v = -1;
for(int e=0;e<Ncache;e++) {
if ( Entries[e].valid==0 ) {
v=e;
break;
}
}
if ( v==-1 ) {
v=victim;
victim = (victim+1)%Ncache;
}
if ( Entries[v].valid ) {
ret = Entries[v].address;
Entries[v].valid = 0;
Entries[v].address = NULL;
Entries[v].bytes = 0;
}
Entries[v].address=ptr;
Entries[v].bytes =bytes;
Entries[v].valid =1;
return ret;
}
void *PointerCache::Lookup(size_t bytes) {
if (bytes < 4096 ) return NULL;
#ifdef _OPENMP
assert(omp_in_parallel()==0);
#endif
for(int e=0;e<Ncache;e++){
if ( Entries[e].valid && ( Entries[e].bytes == bytes ) ) {
Entries[e].valid = 0;
return Entries[e].address;
}
}
return NULL;
}
}

49
lib/AlignedAllocator.h
View File

@ -1,4 +1,4 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -42,9 +42,32 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
namespace Grid {
class PointerCache {
private:
static const int Ncache=8;
static int victim;
typedef struct {
void *address;
size_t bytes;
int valid;
} PointerCacheEntry;
static PointerCacheEntry Entries[Ncache];
public:
static void *Insert(void *ptr,size_t bytes) ;
static void *Lookup(size_t bytes) ;
};
////////////////////////////////////////////////////////////////////
// A lattice of something, but assume the something is SIMDized.
////////////////////////////////////////////////////////////////////
template<typename _Tp>
class alignedAllocator {
public:
@ -66,27 +89,27 @@ public:
pointer allocate(size_type __n, const void* _p= 0)
{
size_type bytes = __n*sizeof(_Tp);
_Tp *ptr = (_Tp *) PointerCache::Lookup(bytes);
#ifdef HAVE_MM_MALLOC_H
_Tp * ptr = (_Tp *) _mm_malloc(__n*sizeof(_Tp),128);
if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) _mm_malloc(bytes,128);
#else
_Tp * ptr = (_Tp *) memalign(128,__n*sizeof(_Tp));
if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) memalign(128,bytes);
#endif
_Tp tmp;
#ifdef GRID_NUMA
#pragma omp parallel for schedule(static)
for(int i=0;i<__n;i++){
ptr[i]=tmp;
}
#endif
return ptr;
}
void deallocate(pointer __p, size_type) {
void deallocate(pointer __p, size_type __n) {
size_type bytes = __n * sizeof(_Tp);
pointer __freeme = (pointer)PointerCache::Insert((void *)__p,bytes);
#ifdef HAVE_MM_MALLOC_H
_mm_free((void *)__p);
if ( __freeme ) _mm_free((void *)__freeme);
#else
free((void *)__p);
if ( __freeme ) free((void *)__freeme);
#endif
}
void construct(pointer __p, const _Tp& __val) { };

2
lib/Grid.h
View File

@ -59,13 +59,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
///////////////////
// Grid headers
///////////////////
#include <Grid/serialisation/Serialisation.h>
#include "Config.h"
#include <Grid/Timer.h>
#include <Grid/PerfCount.h>
#include <Grid/Log.h>
#include <Grid/AlignedAllocator.h>
#include <Grid/Simd.h>
#include <Grid/serialisation/Serialisation.h>
#include <Grid/Threads.h>
#include <Grid/Lexicographic.h>
#include <Grid/Init.h>

1
lib/Hadrons Symbolic link
View File

@ -0,0 +1 @@
../extras/Hadrons

4
lib/Log.h
View File

@ -110,8 +110,8 @@ public:
friend std::ostream& operator<< (std::ostream& stream, Logger& log){
if ( log.active ) {
stream << log.background()<< log.topName << log.background()<< " : ";
stream << log.colour() <<std::setw(14) << std::left << log.name << log.background() << " : ";
stream << log.background()<< std::setw(10) << std::left << log.topName << log.background()<< " : ";
stream << log.colour() << std::setw(14) << std::left << log.name << log.background() << " : ";
if ( log.timestamp ) {
StopWatch.Stop();
GridTime now = StopWatch.Elapsed();

12
lib/Makefile.am
View File

@ -1,4 +1,5 @@
extra_sources=
extra_headers=
if BUILD_COMMS_MPI
extra_sources+=communicator/Communicator_mpi.cc
extra_sources+=communicator/Communicator_base.cc
@ -24,6 +25,12 @@ if BUILD_COMMS_NONE
extra_sources+=communicator/Communicator_base.cc
endif
if BUILD_HDF5
extra_sources+=serialisation/Hdf5IO.cc
extra_headers+=serialisation/Hdf5IO.h
extra_headers+=serialisation/Hdf5Type.h
endif
#
# Libraries
#
@ -32,6 +39,9 @@ include Eigen.inc
lib_LIBRARIES = libGrid.a
libGrid_a_SOURCES = $(CCFILES) $(extra_sources)
CCFILES += $(extra_sources)
HFILES += $(extra_headers)
libGrid_a_SOURCES = $(CCFILES)
libGrid_adir = $(pkgincludedir)
nobase_dist_pkginclude_HEADERS = $(HFILES) $(eigen_files) Config.h

5
lib/PerfCount.h
View File

@ -205,12 +205,13 @@ public:
void Stop(void) {
count=0;
cycles=0;
size_t ign;
#ifdef __linux__
if ( fd!= -1) {
::ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
::ioctl(cyclefd, PERF_EVENT_IOC_DISABLE, 0);
::read(fd, &count, sizeof(long long));
::read(cyclefd, &cycles, sizeof(long long));
ign=::read(fd, &count, sizeof(long long));
ign=::read(cyclefd, &cycles, sizeof(long long));
}
elapsed = cyclecount() - begin;
#else

2
lib/Stencil.h
View File

@ -113,7 +113,7 @@ Gather_plane_simple_table (std::vector<std::pair<int,int> >& table,const Lattice
{
PARALLEL_FOR_LOOP
for(int i=0;i<table.size();i++){
buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
vstream(buffer[off+table[i].first],compress(rhs._odata[so+table[i].second]));
}
}

2
lib/Threads.h
View File

@ -46,11 +46,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#endif
#define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for collapse(2)")
#define PARALLEL_REGION _Pragma("omp parallel")
#define PARALLEL_CRITICAL _Pragma("omp critical")
#else
#define PARALLEL_FOR_LOOP
#define PARALLEL_FOR_LOOP_INTERN
#define PARALLEL_NESTED_LOOP2
#define PARALLEL_REGION
#define PARALLEL_CRITICAL
#endif
namespace Grid {

4
lib/lattice/Lattice_transfer.h
View File

@ -386,7 +386,7 @@ void InsertSlice(Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int
}
// the above should guarantee that the operations are local
//PARALLEL_FOR_LOOP
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -428,7 +428,7 @@ void ExtractSlice(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice, in
}
}
// the above should guarantee that the operations are local
//PARALLEL_FOR_LOOP
PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);

120
lib/qcd/action/fermion/CayleyFermion5D.cc
View File

@ -29,6 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*************************************************************************************/
/* END LEGAL */
#include <Grid/Eigen/Dense>
#include <Grid.h>
@ -48,18 +49,18 @@ namespace QCD {
FourDimGrid,
FourDimRedBlackGrid,_M5,p),
mass(_mass)
{ }
{
}
template<class Impl>
void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
FermionField tmp(psi._grid);
this->DW(psi,tmp,DaggerNo);
this->DW(psi,this->tmp(),DaggerNo);
for(int s=0;s<Ls;s++){
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],this->tmp(),s,s);// chi = (1-c[s] D_W) psi
}
}
@ -87,8 +88,8 @@ template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
std::cout << GridLogMessage << "CayleyFermion5D Number of MooeeInv Calls : " << MooeeInvCalls << std::endl;
std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls : " << MooeeInvTime / MooeeInvCalls << " us" << std::endl;
// Flops = 9*12*Ls*vol/2
RealD mflops = 9.0*12*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
// Flops = MADD * Ls *Ls *4dvol * spin/colour/complex
RealD mflops = 2.0*24*this->Ls*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per rank : " << mflops/NP << std::endl;
}
@ -110,12 +111,11 @@ template<class Impl>
void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
FermionField tmp(psi._grid);
this->DW(psi,tmp,DaggerYes);
this->DW(psi,this->tmp(),DaggerYes);
for(int s=0;s<Ls;s++){
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],this->tmp(),s,s);// chi = (1-c[s] D_W) psi
}
}
template<class Impl>
@ -138,6 +138,7 @@ void CayleyFermion5D<Impl>::Meooe5D (const FermionField &psi, FermionField &D
lower[0] =-mass*lower[0];
M5D(psi,psi,Din,lower,diag,upper);
}
// FIXME Redunant with the above routine; check this and eliminate
template<class Impl> void CayleyFermion5D<Impl>::Meo5D (const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
@ -259,36 +260,33 @@ template<class Impl>
void CayleyFermion5D<Impl>::Meooe (const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
FermionField tmp(psi._grid);
Meooe5D(psi,tmp);
Meooe5D(psi,this->tmp());
if ( psi.checkerboard == Odd ) {
this->DhopEO(tmp,chi,DaggerNo);
this->DhopEO(this->tmp(),chi,DaggerNo);
} else {
this->DhopOE(tmp,chi,DaggerNo);
this->DhopOE(this->tmp(),chi,DaggerNo);
}
}
template<class Impl>
void CayleyFermion5D<Impl>::MeooeDag (const FermionField &psi, FermionField &chi)
{
FermionField tmp(psi._grid);
// Apply 4d dslash
if ( psi.checkerboard == Odd ) {
this->DhopEO(psi,tmp,DaggerYes);
this->DhopEO(psi,this->tmp(),DaggerYes);
} else {
this->DhopOE(psi,tmp,DaggerYes);
this->DhopOE(psi,this->tmp(),DaggerYes);
}
MeooeDag5D(tmp,chi);
MeooeDag5D(this->tmp(),chi);
}
template<class Impl>
void CayleyFermion5D<Impl>::Mdir (const FermionField &psi, FermionField &chi,int dir,int disp){
FermionField tmp(psi._grid);
Meo5D(psi,tmp);
Meo5D(psi,this->tmp());
// Apply 4d dslash fragment
this->DhopDir(tmp,chi,dir,disp);
this->DhopDir(this->tmp(),chi,dir,disp);
}
// force terms; five routines; default to Dhop on diagonal
template<class Impl>
@ -459,9 +457,91 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
for(int j=0;j<Ls-1;j++) delta_d *= cee[j]/bee[j];
dee[Ls-1] += delta_d;
}
int inv=1;
this->MooeeInternalCompute(0,inv,MatpInv,MatmInv);
this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
Vector<iSinglet<Simd> > & Matp,
Vector<iSinglet<Simd> > & Matm)
{
int Ls=this->Ls;
GridBase *grid = this->FermionRedBlackGrid();
int LLs = grid->_rdimensions[0];
if ( LLs == Ls ) return; // Not vectorised in 5th direction
Eigen::MatrixXcd Pplus = Eigen::MatrixXcd::Zero(Ls,Ls);
Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
for(int s=0;s<Ls;s++){
Pplus(s,s) = bee[s];
Pminus(s,s)= bee[s];
}
for(int s=0;s<Ls-1;s++){
Pminus(s,s+1) = -cee[s];
}
for(int s=0;s<Ls-1;s++){
Pplus(s+1,s) = -cee[s+1];
}
Pplus (0,Ls-1) = mass*cee[0];
Pminus(Ls-1,0) = mass*cee[Ls-1];
Eigen::MatrixXcd PplusMat ;
Eigen::MatrixXcd PminusMat;
if ( inv ) {
PplusMat =Pplus.inverse();
PminusMat=Pminus.inverse();
} else {
PplusMat =Pplus;
PminusMat=Pminus;
}
if(dag){
PplusMat.adjointInPlace();
PminusMat.adjointInPlace();
}
typedef typename SiteHalfSpinor::scalar_type scalar_type;
const int Nsimd=Simd::Nsimd();
Matp.resize(Ls*LLs);
Matm.resize(Ls*LLs);
for(int s2=0;s2<Ls;s2++){
for(int s1=0;s1<LLs;s1++){
int istride = LLs;
int ostride = 1;
Simd Vp;
Simd Vm;
scalar_type *sp = (scalar_type *)&Vp;
scalar_type *sm = (scalar_type *)&Vm;
for(int l=0;l<Nsimd;l++){
if ( switcheroo<Coeff_t>::iscomplex() ) {
sp[l] = PplusMat (l*istride+s1*ostride,s2);
sm[l] = PminusMat(l*istride+s1*ostride,s2);
} else {
// if real
scalar_type tmp;
tmp = PplusMat (l*istride+s1*ostride,s2);
sp[l] = scalar_type(tmp.real(),tmp.real());
tmp = PminusMat(l*istride+s1*ostride,s2);
sm[l] = scalar_type(tmp.real(),tmp.real());
}
}
Matp[LLs*s2+s1] = Vp;
Matm[LLs*s2+s1] = Vm;
}}
}
FermOpTemplateInstantiate(CayleyFermion5D);
GparityFermOpTemplateInstantiate(CayleyFermion5D);

43
lib/qcd/action/fermion/CayleyFermion5D.h
View File

@ -33,6 +33,31 @@ namespace Grid {
namespace QCD {
template<typename T> struct switcheroo {
static inline int iscomplex() { return 0; }
template<class vec>
static inline vec mult(vec a, vec b) {
return real_mult(a,b);
}
};
template<> struct switcheroo<ComplexD> {
static inline int iscomplex() { return 1; }
template<class vec>
static inline vec mult(vec a, vec b) {
return a*b;
}
};
template<> struct switcheroo<ComplexF> {
static inline int iscomplex() { return 1; }
template<class vec>
static inline vec mult(vec a, vec b) {
return a*b;
}
};
template<class Impl>
class CayleyFermion5D : public WilsonFermion5D<Impl>
{
@ -75,7 +100,19 @@ namespace Grid {
std::vector<Coeff_t> &lower,
std::vector<Coeff_t> &diag,
std::vector<Coeff_t> &upper);
void MooeeInternal(const FermionField &in, FermionField &out,int dag,int inv);
void MooeeInternalCompute(int dag, int inv, Vector<iSinglet<Simd> > & Matp, Vector<iSinglet<Simd> > & Matm);
void MooeeInternalAsm(const FermionField &in, FermionField &out,
int LLs, int site,
Vector<iSinglet<Simd> > &Matp,
Vector<iSinglet<Simd> > &Matm);
void MooeeInternalZAsm(const FermionField &in, FermionField &out,
int LLs, int site,
Vector<iSinglet<Simd> > &Matp,
Vector<iSinglet<Simd> > &Matm);
virtual void Instantiatable(void)=0;
@ -112,6 +149,12 @@ namespace Grid {
std::vector<Coeff_t> ueem;
std::vector<Coeff_t> dee;
// Matrices of 5d ee inverse params
Vector<iSinglet<Simd> > MatpInv;
Vector<iSinglet<Simd> > MatmInv;
Vector<iSinglet<Simd> > MatpInvDag;
Vector<iSinglet<Simd> > MatmInvDag;
// Constructors
CayleyFermion5D(GaugeField &_Umu,
GridCartesian &FiveDimGrid,

672
lib/qcd/action/fermion/CayleyFermion5Dvec.cc
View File

@ -29,13 +29,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*************************************************************************************/
/* END LEGAL */
#include <Grid/Eigen/Dense>
#include <Grid.h>
namespace Grid {
namespace QCD {
/*
namespace QCD { /*
* Dense matrix versions of routines
*/
template<class Impl>
@ -126,7 +125,6 @@ PARALLEL_FOR_LOOP
for(int v=0;v<LLs;v++){
vprefetch(psi[ss+v+LLs]);
// vprefetch(phi[ss+v+LLs]);
int vp= (v==LLs-1) ? 0 : v+1;
int vm= (v==0 ) ? LLs-1 : v-1;
@ -145,9 +143,6 @@ PARALLEL_FOR_LOOP
Simd hm_11 = psi[ss+vm]()(1)(1);
Simd hm_12 = psi[ss+vm]()(1)(2);
// if ( ss==0) std::cout << " hp_00 " <<hp_00<<std::endl;
// if ( ss==0) std::cout << " hm_00 " <<hm_00<<std::endl;
if ( vp<=v ) {
hp_00.v = Optimization::Rotate::tRotate<2>(hp_00.v);
hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v);
@ -165,42 +160,20 @@ PARALLEL_FOR_LOOP
hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v);
}
/*
if ( ss==0) std::cout << " dphi_00 " <<d[v]()()() * phi[ss+v]()(0)(0) <<std::endl;
if ( ss==0) std::cout << " dphi_10 " <<d[v]()()() * phi[ss+v]()(1)(0) <<std::endl;
if ( ss==0) std::cout << " dphi_20 " <<d[v]()()() * phi[ss+v]()(2)(0) <<std::endl;
if ( ss==0) std::cout << " dphi_30 " <<d[v]()()() * phi[ss+v]()(3)(0) <<std::endl;
*/
Simd p_00 = d[v]()()() * phi[ss+v]()(0)(0) + l[v]()()()*hm_00;
Simd p_01 = d[v]()()() * phi[ss+v]()(0)(1) + l[v]()()()*hm_01;
Simd p_02 = d[v]()()() * phi[ss+v]()(0)(2) + l[v]()()()*hm_02;
Simd p_10 = d[v]()()() * phi[ss+v]()(1)(0) + l[v]()()()*hm_10;
Simd p_11 = d[v]()()() * phi[ss+v]()(1)(1) + l[v]()()()*hm_11;
Simd p_12 = d[v]()()() * phi[ss+v]()(1)(2) + l[v]()()()*hm_12;
Simd p_20 = d[v]()()() * phi[ss+v]()(2)(0) + u[v]()()()*hp_00;
Simd p_21 = d[v]()()() * phi[ss+v]()(2)(1) + u[v]()()()*hp_01;
Simd p_22 = d[v]()()() * phi[ss+v]()(2)(2) + u[v]()()()*hp_02;
Simd p_30 = d[v]()()() * phi[ss+v]()(3)(0) + u[v]()()()*hp_10;
Simd p_31 = d[v]()()() * phi[ss+v]()(3)(1) + u[v]()()()*hp_11;
Simd p_32 = d[v]()()() * phi[ss+v]()(3)(2) + u[v]()()()*hp_12;
// Can force these to real arithmetic and save 2x.
Simd p_00 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00);
Simd p_01 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01);
Simd p_02 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02);
Simd p_10 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10);
Simd p_11 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11);
Simd p_12 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12);
Simd p_20 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00);
Simd p_21 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01);
Simd p_22 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02);
Simd p_30 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10);
Simd p_31 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11);
Simd p_32 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_12);
// if ( ss==0){
/*
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(0) << " bad "<<p_00<<" diff "<<chi[ss+v]()(0)(0)-p_00<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(1) << " bad "<<p_01<<" diff "<<chi[ss+v]()(0)(1)-p_01<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(0)(2) << " bad "<<p_02<<" diff "<<chi[ss+v]()(0)(2)-p_02<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(0) << " bad "<<p_10<<" diff "<<chi[ss+v]()(1)(0)-p_10<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(1) << " bad "<<p_11<<" diff "<<chi[ss+v]()(1)(1)-p_11<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(1)(2) << " bad "<<p_12<<" diff "<<chi[ss+v]()(1)(2)-p_12<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(0) << " bad "<<p_20<<" diff "<<chi[ss+v]()(2)(0)-p_20<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(1) << " bad "<<p_21<<" diff "<<chi[ss+v]()(2)(1)-p_21<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(2)(2) << " bad "<<p_22<<" diff "<<chi[ss+v]()(2)(2)-p_22<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(0) << " bad "<<p_30<<" diff "<<chi[ss+v]()(3)(0)-p_30<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(1) << " bad "<<p_31<<" diff "<<chi[ss+v]()(3)(1)-p_31<<std::endl;
std::cout << ss<<" "<< v<< " good "<< chi[ss+v]()(3)(2) << " bad "<<p_32<<" diff "<<chi[ss+v]()(3)(2)-p_32<<std::endl;
}
*/
vstream(chi[ss+v]()(0)(0),p_00);
vstream(chi[ss+v]()(0)(1),p_01);
vstream(chi[ss+v]()(0)(2),p_02);
@ -261,7 +234,7 @@ void CayleyFermion5D<Impl>::M5Ddag(const FermionField &psi,
M5Dtime-=usecond();
PARALLEL_FOR_LOOP
for(int ss=0;ss<grid->oSites();ss+=LLs){ // adds LLs
#if 0
alignas(64) SiteHalfSpinor hp;
alignas(64) SiteHalfSpinor hm;
alignas(64) SiteSpinor fp;
@ -287,9 +260,504 @@ PARALLEL_FOR_LOOP
chi[ss+v] = chi[ss+v] +l[v]*fm;
}
#else
for(int v=0;v<LLs;v++){
vprefetch(psi[ss+v+LLs]);
int vp= (v==LLs-1) ? 0 : v+1;
int vm= (v==0 ) ? LLs-1 : v-1;
Simd hp_00 = psi[ss+vp]()(0)(0);
Simd hp_01 = psi[ss+vp]()(0)(1);
Simd hp_02 = psi[ss+vp]()(0)(2);
Simd hp_10 = psi[ss+vp]()(1)(0);
Simd hp_11 = psi[ss+vp]()(1)(1);
Simd hp_12 = psi[ss+vp]()(1)(2);
Simd hm_00 = psi[ss+vm]()(2)(0);
Simd hm_01 = psi[ss+vm]()(2)(1);
Simd hm_02 = psi[ss+vm]()(2)(2);
Simd hm_10 = psi[ss+vm]()(3)(0);
Simd hm_11 = psi[ss+vm]()(3)(1);
Simd hm_12 = psi[ss+vm]()(3)(2);
if ( vp<=v ) {
hp_00.v = Optimization::Rotate::tRotate<2>(hp_00.v);
hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v);
hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v);
hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v);
hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v);
hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v);
}
if ( vm>=v ) {
hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v);
hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v);
hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v);
hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v);
hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v);
hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v);
}
Simd p_00 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_00);
Simd p_01 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_01);
Simd p_02 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_02);
Simd p_10 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_10);
Simd p_11 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_11);
Simd p_12 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo<Coeff_t>::mult(u[v]()()(),hp_12);
Simd p_20 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_00);
Simd p_21 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_01);
Simd p_22 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_02);
Simd p_30 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_10);
Simd p_31 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_11);
Simd p_32 = switcheroo<Coeff_t>::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo<Coeff_t>::mult(l[v]()()(),hm_12);
vstream(chi[ss+v]()(0)(0),p_00);
vstream(chi[ss+v]()(0)(1),p_01);
vstream(chi[ss+v]()(0)(2),p_02);
vstream(chi[ss+v]()(1)(0),p_10);
vstream(chi[ss+v]()(1)(1),p_11);
vstream(chi[ss+v]()(1)(2),p_12);
vstream(chi[ss+v]()(2)(0),p_20);
vstream(chi[ss+v]()(2)(1),p_21);
vstream(chi[ss+v]()(2)(2),p_22);
vstream(chi[ss+v]()(3)(0),p_30);
vstream(chi[ss+v]()(3)(1),p_31);
vstream(chi[ss+v]()(3)(2),p_32);
}
#endif
}
M5Dtime+=usecond();
}
#ifdef AVX512
#include <simd/Intel512common.h>
#include <simd/Intel512avx.h>
#include <simd/Intel512single.h>
#endif
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInternalAsm(const FermionField &psi, FermionField &chi,
int LLs, int site,
Vector<iSinglet<Simd> > &Matp,
Vector<iSinglet<Simd> > &Matm)
{
#ifndef AVX512
{
SiteHalfSpinor BcastP;
SiteHalfSpinor BcastM;
SiteHalfSpinor SiteChiP;
SiteHalfSpinor SiteChiM;
// Ls*Ls * 2 * 12 * vol flops
for(int s1=0;s1<LLs;s1++){
for(int s2=0;s2<LLs;s2++){
for(int l=0; l<Simd::Nsimd();l++){ // simd lane
int s=s2+l*LLs;
int lex=s2+LLs*site;
if ( s2==0 && l==0) {
SiteChiP=zero;
SiteChiM=zero;
}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vbroadcast(BcastP()(sp )(co),psi[lex]()(sp)(co),l);
}}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vbroadcast(BcastM()(sp )(co),psi[lex]()(sp+2)(co),l);
}}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
SiteChiP()(sp)(co)=real_madd(Matp[LLs*s+s1]()()(),BcastP()(sp)(co),SiteChiP()(sp)(co)); // 1100 us.
SiteChiM()(sp)(co)=real_madd(Matm[LLs*s+s1]()()(),BcastM()(sp)(co),SiteChiM()(sp)(co)); // each found by commenting out
}}
}}
{
int lex = s1+LLs*site;
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
}}
}
}
}
#else
{
// pointers
// MASK_REGS;
#define Chi_00 %%zmm1
#define Chi_01 %%zmm2
#define Chi_02 %%zmm3
#define Chi_10 %%zmm4
#define Chi_11 %%zmm5
#define Chi_12 %%zmm6
#define Chi_20 %%zmm7
#define Chi_21 %%zmm8
#define Chi_22 %%zmm9
#define Chi_30 %%zmm10
#define Chi_31 %%zmm11
#define Chi_32 %%zmm12
#define BCAST0 %%zmm13
#define BCAST1 %%zmm14
#define BCAST2 %%zmm15
#define BCAST3 %%zmm16
#define BCAST4 %%zmm17
#define BCAST5 %%zmm18
#define BCAST6 %%zmm19
#define BCAST7 %%zmm20
#define BCAST8 %%zmm21
#define BCAST9 %%zmm22
#define BCAST10 %%zmm23
#define BCAST11 %%zmm24
int incr=LLs*LLs*sizeof(iSinglet<Simd>);
for(int s1=0;s1<LLs;s1++){
for(int s2=0;s2<LLs;s2++){
int lex=s2+LLs*site;
uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
uint64_t a2 = (uint64_t)&psi[lex];
for(int l=0; l<Simd::Nsimd();l++){ // simd lane
if ( (s2+l)==0 ) {
asm (
VPREFETCH1(0,%2) VPREFETCH1(0,%1)
VPREFETCH1(12,%2) VPREFETCH1(13,%2)
VPREFETCH1(14,%2) VPREFETCH1(15,%2)
VBCASTCDUP(0,%2,BCAST0)
VBCASTCDUP(1,%2,BCAST1)
VBCASTCDUP(2,%2,BCAST2)
VBCASTCDUP(3,%2,BCAST3)
VBCASTCDUP(4,%2,BCAST4) VMULMEM (0,%0,BCAST0,Chi_00)
VBCASTCDUP(5,%2,BCAST5) VMULMEM (0,%0,BCAST1,Chi_01)
VBCASTCDUP(6,%2,BCAST6) VMULMEM (0,%0,BCAST2,Chi_02)
VBCASTCDUP(7,%2,BCAST7) VMULMEM (0,%0,BCAST3,Chi_10)
VBCASTCDUP(8,%2,BCAST8) VMULMEM (0,%0,BCAST4,Chi_11)
VBCASTCDUP(9,%2,BCAST9) VMULMEM (0,%0,BCAST5,Chi_12)
VBCASTCDUP(10,%2,BCAST10) VMULMEM (0,%1,BCAST6,Chi_20)
VBCASTCDUP(11,%2,BCAST11) VMULMEM (0,%1,BCAST7,Chi_21)
VMULMEM (0,%1,BCAST8,Chi_22)
VMULMEM (0,%1,BCAST9,Chi_30)
VMULMEM (0,%1,BCAST10,Chi_31)
VMULMEM (0,%1,BCAST11,Chi_32)
: : "r" (a0), "r" (a1), "r" (a2) );
} else {
asm (
VBCASTCDUP(0,%2,BCAST0) VMADDMEM (0,%0,BCAST0,Chi_00)
VBCASTCDUP(1,%2,BCAST1) VMADDMEM (0,%0,BCAST1,Chi_01)
VBCASTCDUP(2,%2,BCAST2) VMADDMEM (0,%0,BCAST2,Chi_02)
VBCASTCDUP(3,%2,BCAST3) VMADDMEM (0,%0,BCAST3,Chi_10)
VBCASTCDUP(4,%2,BCAST4) VMADDMEM (0,%0,BCAST4,Chi_11)
VBCASTCDUP(5,%2,BCAST5) VMADDMEM (0,%0,BCAST5,Chi_12)
VBCASTCDUP(6,%2,BCAST6) VMADDMEM (0,%1,BCAST6,Chi_20)
VBCASTCDUP(7,%2,BCAST7) VMADDMEM (0,%1,BCAST7,Chi_21)
VBCASTCDUP(8,%2,BCAST8) VMADDMEM (0,%1,BCAST8,Chi_22)
VBCASTCDUP(9,%2,BCAST9) VMADDMEM (0,%1,BCAST9,Chi_30)
VBCASTCDUP(10,%2,BCAST10) VMADDMEM (0,%1,BCAST10,Chi_31)
VBCASTCDUP(11,%2,BCAST11) VMADDMEM (0,%1,BCAST11,Chi_32)
: : "r" (a0), "r" (a1), "r" (a2) );
}
a0 = a0+incr;
a1 = a1+incr;
a2 = a2+sizeof(Simd::scalar_type);
}}
{
int lexa = s1+LLs*site;
asm (
VSTORE(0,%0,Chi_00) VSTORE(1 ,%0,Chi_01) VSTORE(2 ,%0,Chi_02)
VSTORE(3,%0,Chi_10) VSTORE(4 ,%0,Chi_11) VSTORE(5 ,%0,Chi_12)
VSTORE(6,%0,Chi_20) VSTORE(7 ,%0,Chi_21) VSTORE(8 ,%0,Chi_22)
VSTORE(9,%0,Chi_30) VSTORE(10,%0,Chi_31) VSTORE(11,%0,Chi_32)
: : "r" ((uint64_t)&chi[lexa]) : "memory" );
}
}
}
#undef Chi_00
#undef Chi_01
#undef Chi_02
#undef Chi_10
#undef Chi_11
#undef Chi_12
#undef Chi_20
#undef Chi_21
#undef Chi_22
#undef Chi_30
#undef Chi_31
#undef Chi_32
#undef BCAST0
#undef BCAST1
#undef BCAST2
#undef BCAST3
#undef BCAST4
#undef BCAST5
#undef BCAST6
#undef BCAST7
#undef BCAST8
#undef BCAST9
#undef BCAST10
#undef BCAST11
#endif
};
// Z-mobius version
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInternalZAsm(const FermionField &psi, FermionField &chi,
int LLs, int site, Vector<iSinglet<Simd> > &Matp, Vector<iSinglet<Simd> > &Matm)
{
#ifndef AVX512
{
SiteHalfSpinor BcastP;
SiteHalfSpinor BcastM;
SiteHalfSpinor SiteChiP;
SiteHalfSpinor SiteChiM;
// Ls*Ls * 2 * 12 * vol flops
for(int s1=0;s1<LLs;s1++){
for(int s2=0;s2<LLs;s2++){
for(int l=0; l<Simd::Nsimd();l++){ // simd lane
int s=s2+l*LLs;
int lex=s2+LLs*site;
if ( s2==0 && l==0) {
SiteChiP=zero;
SiteChiM=zero;
}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vbroadcast(BcastP()(sp )(co),psi[lex]()(sp)(co),l);
}}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vbroadcast(BcastM()(sp )(co),psi[lex]()(sp+2)(co),l);
}}
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
SiteChiP()(sp)(co)=SiteChiP()(sp)(co)+ Matp[LLs*s+s1]()()()*BcastP()(sp)(co);
SiteChiM()(sp)(co)=SiteChiM()(sp)(co)+ Matm[LLs*s+s1]()()()*BcastM()(sp)(co);
}}
}}
{
int lex = s1+LLs*site;
for(int sp=0;sp<2;sp++){
for(int co=0;co<Nc;co++){
vstream(chi[lex]()(sp)(co), SiteChiP()(sp)(co));
vstream(chi[lex]()(sp+2)(co), SiteChiM()(sp)(co));
}}
}
}
}
#else
{
// pointers
// MASK_REGS;
#define Chi_00 %zmm0
#define Chi_01 %zmm1
#define Chi_02 %zmm2
#define Chi_10 %zmm3
#define Chi_11 %zmm4
#define Chi_12 %zmm5
#define Chi_20 %zmm6
#define Chi_21 %zmm7
#define Chi_22 %zmm8
#define Chi_30 %zmm9
#define Chi_31 %zmm10
#define Chi_32 %zmm11
#define pChi_00 %%zmm0
#define pChi_01 %%zmm1
#define pChi_02 %%zmm2
#define pChi_10 %%zmm3
#define pChi_11 %%zmm4
#define pChi_12 %%zmm5
#define pChi_20 %%zmm6
#define pChi_21 %%zmm7
#define pChi_22 %%zmm8
#define pChi_30 %%zmm9
#define pChi_31 %%zmm10
#define pChi_32 %%zmm11
#define BCAST_00 %zmm12
#define SHUF_00 %zmm13
#define BCAST_01 %zmm14
#define SHUF_01 %zmm15
#define BCAST_02 %zmm16
#define SHUF_02 %zmm17
#define BCAST_10 %zmm18
#define SHUF_10 %zmm19
#define BCAST_11 %zmm20
#define SHUF_11 %zmm21
#define BCAST_12 %zmm22
#define SHUF_12 %zmm23
#define Mp %zmm24
#define Mps %zmm25
#define Mm %zmm26
#define Mms %zmm27
#define N 8
int incr=LLs*LLs*sizeof(iSinglet<Simd>);
for(int s1=0;s1<LLs;s1++){
for(int s2=0;s2<LLs;s2++){
int lex=s2+LLs*site;
uint64_t a0 = (uint64_t)&Matp[LLs*s2+s1]; // should be cacheable
uint64_t a1 = (uint64_t)&Matm[LLs*s2+s1];
uint64_t a2 = (uint64_t)&psi[lex];
for(int l=0; l<Simd::Nsimd();l++){ // simd lane
if ( (s2+l)==0 ) {
LOAD64(%r8,a0);
LOAD64(%r9,a1);
LOAD64(%r10,a2);
asm (
VLOAD(0,%r8,Mp)// i r
VLOAD(0,%r9,Mm)
VSHUF(Mp,Mps) // r i
VSHUF(Mm,Mms)
VPREFETCH1(12,%r10) VPREFETCH1(13,%r10)
VPREFETCH1(14,%r10) VPREFETCH1(15,%r10)
VMULIDUP(0*N,%r10,Mps,Chi_00)
VMULIDUP(1*N,%r10,Mps,Chi_01)
VMULIDUP(2*N,%r10,Mps,Chi_02)
VMULIDUP(3*N,%r10,Mps,Chi_10)
VMULIDUP(4*N,%r10,Mps,Chi_11)
VMULIDUP(5*N,%r10,Mps,Chi_12)
VMULIDUP(6*N ,%r10,Mms,Chi_20)
VMULIDUP(7*N ,%r10,Mms,Chi_21)
VMULIDUP(8*N ,%r10,Mms,Chi_22)
VMULIDUP(9*N ,%r10,Mms,Chi_30)
VMULIDUP(10*N,%r10,Mms,Chi_31)
VMULIDUP(11*N,%r10,Mms,Chi_32)
VMADDSUBRDUP(0*N,%r10,Mp,Chi_00)
VMADDSUBRDUP(1*N,%r10,Mp,Chi_01)
VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
VMADDSUBRDUP(6*N ,%r10,Mm,Chi_20)
VMADDSUBRDUP(7*N ,%r10,Mm,Chi_21)
VMADDSUBRDUP(8*N ,%r10,Mm,Chi_22)
VMADDSUBRDUP(9*N ,%r10,Mm,Chi_30)
VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
);
} else {
LOAD64(%r8,a0);
LOAD64(%r9,a1);
LOAD64(%r10,a2);
asm (
VLOAD(0,%r8,Mp)
VSHUF(Mp,Mps)
VLOAD(0,%r9,Mm)
VSHUF(Mm,Mms)
VMADDSUBIDUP(0*N,%r10,Mps,Chi_00) // Mri * Pii +- Cir
VMADDSUBIDUP(1*N,%r10,Mps,Chi_01)
VMADDSUBIDUP(2*N,%r10,Mps,Chi_02)
VMADDSUBIDUP(3*N,%r10,Mps,Chi_10)
VMADDSUBIDUP(4*N,%r10,Mps,Chi_11)
VMADDSUBIDUP(5*N,%r10,Mps,Chi_12)
VMADDSUBIDUP(6 *N,%r10,Mms,Chi_20)
VMADDSUBIDUP(7 *N,%r10,Mms,Chi_21)
VMADDSUBIDUP(8 *N,%r10,Mms,Chi_22)
VMADDSUBIDUP(9 *N,%r10,Mms,Chi_30)
VMADDSUBIDUP(10*N,%r10,Mms,Chi_31)
VMADDSUBIDUP(11*N,%r10,Mms,Chi_32)
VMADDSUBRDUP(0*N,%r10,Mp,Chi_00) // Cir = Mir * Prr +- ( Mri * Pii +- Cir)
VMADDSUBRDUP(1*N,%r10,Mp,Chi_01) // Ci = MiPr + Ci + MrPi ; Cr = MrPr - ( MiPi - Cr)
VMADDSUBRDUP(2*N,%r10,Mp,Chi_02)
VMADDSUBRDUP(3*N,%r10,Mp,Chi_10)
VMADDSUBRDUP(4*N,%r10,Mp,Chi_11)
VMADDSUBRDUP(5*N,%r10,Mp,Chi_12)
VMADDSUBRDUP(6 *N,%r10,Mm,Chi_20)
VMADDSUBRDUP(7 *N,%r10,Mm,Chi_21)
VMADDSUBRDUP(8 *N,%r10,Mm,Chi_22)
VMADDSUBRDUP(9 *N,%r10,Mm,Chi_30)
VMADDSUBRDUP(10*N,%r10,Mm,Chi_31)
VMADDSUBRDUP(11*N,%r10,Mm,Chi_32)
);
}
a0 = a0+incr;
a1 = a1+incr;
a2 = a2+sizeof(Simd::scalar_type);
}}
{
int lexa = s1+LLs*site;
/*
SiteSpinor tmp;
asm (
VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01) VSTORE(2 ,%0,pChi_02)
VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11) VSTORE(5 ,%0,pChi_12)
VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21) VSTORE(8 ,%0,pChi_22)
VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31) VSTORE(11,%0,pChi_32)
: : "r" ((uint64_t)&tmp) : "memory" );
*/
asm (
VSTORE(0,%0,pChi_00) VSTORE(1 ,%0,pChi_01) VSTORE(2 ,%0,pChi_02)
VSTORE(3,%0,pChi_10) VSTORE(4 ,%0,pChi_11) VSTORE(5 ,%0,pChi_12)
VSTORE(6,%0,pChi_20) VSTORE(7 ,%0,pChi_21) VSTORE(8 ,%0,pChi_22)
VSTORE(9,%0,pChi_30) VSTORE(10,%0,pChi_31) VSTORE(11,%0,pChi_32)
: : "r" ((uint64_t)&chi[lexa]) : "memory" );
// if ( 1 || (site==0) ) {
// std::cout<<site << " s1 "<<s1<<"\n\t"<<tmp << "\n't" << chi[lexa] <<"\n\t"<<tmp-chi[lexa]<<std::endl;
// }
}
}
}
#undef Chi_00
#undef Chi_01
#undef Chi_02
#undef Chi_10
#undef Chi_11
#undef Chi_12
#undef Chi_20
#undef Chi_21
#undef Chi_22
#undef Chi_30
#undef Chi_31
#undef Chi_32
#undef BCAST0
#undef BCAST1
#undef BCAST2
#undef BCAST3
#undef BCAST4
#undef BCAST5
#undef BCAST6
#undef BCAST7
#undef BCAST8
#undef BCAST9
#undef BCAST10
#undef BCAST11
#endif
};
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv)
{
@ -299,108 +767,41 @@ void CayleyFermion5D<Impl>::MooeeInternal(const FermionField &psi, FermionField
chi.checkerboard=psi.checkerboard;
Eigen::MatrixXcd Pplus = Eigen::MatrixXcd::Zero(Ls,Ls);
Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);
Vector<iSinglet<Simd> > Matp;
Vector<iSinglet<Simd> > Matm;
Vector<iSinglet<Simd> > *_Matp;
Vector<iSinglet<Simd> > *_Matm;
for(int s=0;s<Ls;s++){
Pplus(s,s) = bee[s];
Pminus(s,s)= bee[s];
// MooeeInternalCompute(dag,inv,Matp,Matm);
if ( inv && dag ) {
_Matp = &MatpInvDag;
_Matm = &MatmInvDag;
}
for(int s=0;s<Ls-1;s++){
Pminus(s,s+1) = -cee[s];
if ( inv && (!dag) ) {
_Matp = &MatpInv;
_Matm = &MatmInv;
}
if ( !inv ) {
MooeeInternalCompute(dag,inv,Matp,Matm);
_Matp = &Matp;
_Matm = &Matm;
}
for(int s=0;s<Ls-1;s++){
Pplus(s+1,s) = -cee[s+1];
}
Pplus (0,Ls-1) = mass*cee[0];
Pminus(Ls-1,0) = mass*cee[Ls-1];
Eigen::MatrixXcd PplusMat ;
Eigen::MatrixXcd PminusMat;
if ( inv ) {
PplusMat =Pplus.inverse();
PminusMat=Pminus.inverse();
} else {
PplusMat =Pplus;
PminusMat=Pminus;
}
if(dag){
PplusMat.adjointInPlace();
PminusMat.adjointInPlace();
}
typedef typename SiteHalfSpinor::scalar_type scalar_type;
const int Nsimd=Simd::Nsimd();
Vector<iSinglet<Simd> > Matp(Ls*LLs);
Vector<iSinglet<Simd> > Matm(Ls*LLs);
assert(_Matp->size()==Ls*LLs);
for(int s2=0;s2<Ls;s2++){
for(int s1=0;s1<LLs;s1++){
int istride = LLs;
int ostride = 1;
Simd Vp;
Simd Vm;
scalar_type *sp = (scalar_type *)&Vp;
scalar_type *sm = (scalar_type *)&Vm;
for(int l=0;l<Nsimd;l++){
sp[l] = PplusMat (l*istride+s1*ostride ,s2);
sm[l] = PminusMat(l*istride+s1*ostride,s2);
}
Matp[LLs*s2+s1] = Vp;
Matm[LLs*s2+s1] = Vm;
}
}
MooeeInvCalls++;
MooeeInvTime-=usecond();
// Dynamic allocate on stack to get per thread without serialised heap acces
#pragma omp parallel
{
Vector<SiteHalfSpinor> SitePplus(LLs);
Vector<SiteHalfSpinor> SitePminus(LLs);
Vector<SiteHalfSpinor> SiteChiP(LLs);
Vector<SiteHalfSpinor> SiteChiM(LLs);
Vector<SiteSpinor> SiteChi(LLs);
SiteHalfSpinor BcastP;
SiteHalfSpinor BcastM;
#pragma omp for
for(auto site=0;site<vol;site++){
for(int s=0;s<LLs;s++){
int lex = s+LLs*site;
spProj5p(SitePplus[s] ,psi[lex]);
spProj5m(SitePminus[s],psi[lex]);
SiteChiP[s]=zero;
SiteChiM[s]=zero;
if ( switcheroo<Coeff_t>::iscomplex() ) {
PARALLEL_FOR_LOOP
for(auto site=0;site<vol;site++){
MooeeInternalZAsm(psi,chi,LLs,site,*_Matp,*_Matm);
}
int s=0;
for(int l=0; l<Simd::Nsimd();l++){ // simd lane
for(int s2=0;s2<LLs;s2++){ // Column loop of right hand side
vbroadcast(BcastP,SitePplus [s2],l);
vbroadcast(BcastM,SitePminus[s2],l);
for(int s1=0;s1<LLs;s1++){ // Column loop of reduction variables
SiteChiP[s1]=SiteChiP[s1]+Matp[LLs*s+s1]*BcastP;
SiteChiM[s1]=SiteChiM[s1]+Matm[LLs*s+s1]*BcastM;
}
s++;
}}
for(int s=0;s<LLs;s++){
int lex = s+LLs*site;
spRecon5p(SiteChi[s],SiteChiP[s]);
accumRecon5m(SiteChi[s],SiteChiM[s]);
chi[lex] = SiteChi[s]*0.5;
} else {
PARALLEL_FOR_LOOP
for(auto site=0;site<vol;site++){
MooeeInternalAsm(psi,chi,LLs,site,*_Matp,*_Matm);
}
}
}
MooeeInvTime+=usecond();
}
@ -414,4 +815,5 @@ template void CayleyFermion5D<DomainWallVec5dImplD>::MooeeInternal(const Fermion
template void CayleyFermion5D<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
}}

2
lib/qcd/action/fermion/FermionOperator.h
View File

@ -48,6 +48,8 @@ namespace Grid {
FermionOperator(const ImplParams &p= ImplParams()) : Impl(p) {};
virtual FermionField &tmp(void) = 0;
GridBase * Grid(void) { return FermionGrid(); }; // this is all the linalg routines need to know
GridBase * RedBlackGrid(void) { return FermionRedBlackGrid(); };

33
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -48,10 +48,12 @@ namespace QCD {
// typedef typename XXX GaugeField;
// typedef typename XXX GaugeActField;
// typedef typename XXX FermionField;
// typedef typename XXX PropagatorField;
// typedef typename XXX DoubledGaugeField;
// typedef typename XXX SiteSpinor;
// typedef typename XXX SiteHalfSpinor;
// typedef typename XXX Compressor;
// typedef typename XXX SitePropagator;
// typedef typename XXX SiteHalfSpinor;
// typedef typename XXX Compressor;
//
// and Methods:
// void ImportGauge(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
@ -94,14 +96,16 @@ namespace QCD {
////////////////////////////////////////////////////////////////////////
#define INHERIT_FIMPL_TYPES(Impl)\
typedef typename Impl::FermionField FermionField; \
typedef typename Impl::DoubledGaugeField DoubledGaugeField; \
typedef typename Impl::SiteSpinor SiteSpinor; \
typedef typename Impl::SiteHalfSpinor SiteHalfSpinor; \
typedef typename Impl::Compressor Compressor; \
typedef typename Impl::StencilImpl StencilImpl; \
typedef typename Impl::ImplParams ImplParams; \
typedef typename Impl::Coeff_t Coeff_t;
typedef typename Impl::FermionField FermionField; \
typedef typename Impl::PropagatorField PropagatorField; \
typedef typename Impl::DoubledGaugeField DoubledGaugeField; \
typedef typename Impl::SiteSpinor SiteSpinor; \
typedef typename Impl::SitePropagator SitePropagator; \
typedef typename Impl::SiteHalfSpinor SiteHalfSpinor; \
typedef typename Impl::Compressor Compressor; \
typedef typename Impl::StencilImpl StencilImpl; \
typedef typename Impl::ImplParams ImplParams; \
typedef typename Impl::Coeff_t Coeff_t; \
#define INHERIT_IMPL_TYPES(Base) \
INHERIT_GIMPL_TYPES(Base) \
@ -127,14 +131,17 @@ namespace QCD {
INHERIT_GIMPL_TYPES(Gimpl);
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
template <typename vtype> using iImplPropagator = iScalar<iMatrix<iMatrix<vtype, Dimension>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
@ -216,14 +223,17 @@ class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
INHERIT_GIMPL_TYPES(Gimpl);
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
template <typename vtype> using iImplPropagator = iScalar<iMatrix<iMatrix<vtype, Nrepresentation>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>;
template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
template <typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
// Make the doubled gauge field a *scalar*
typedef iImplDoubledGaugeField<typename Simd::scalar_type> SiteDoubledGaugeField; // This is a scalar
@ -352,14 +362,17 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
INHERIT_GIMPL_TYPES(Gimpl);
template <typename vtype> using iImplSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Ns>, Ngp>;
template <typename vtype> using iImplPropagator = iVector<iMatrix<iMatrix<vtype, Nrepresentation>, Ns>, Ngp >;
template <typename vtype> using iImplHalfSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, Ngp>;
template <typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>, Ngp>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;

4
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -61,7 +61,9 @@ WilsonFermion<Impl>::WilsonFermion(GaugeField &_Umu, GridCartesian &Fgrid,
LebesgueEvenOdd(_cbgrid),
Umu(&Fgrid),
UmuEven(&Hgrid),
UmuOdd(&Hgrid) {
UmuOdd(&Hgrid),
_tmp(&Hgrid)
{
// Allocate the required comms buffer
ImportGauge(_Umu);
}

3
lib/qcd/action/fermion/WilsonFermion.h
View File

@ -58,6 +58,9 @@ class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic {
GridBase *FermionGrid(void) { return _grid; }
GridBase *FermionRedBlackGrid(void) { return _cbgrid; }
FermionField _tmp;
FermionField &tmp(void) { return _tmp; }
//////////////////////////////////////////////////////////////////
// override multiply; cut number routines if pass dagger argument
// and also make interface more uniformly consistent

3
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -61,7 +61,8 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
UmuEven(_FourDimRedBlackGrid),
UmuOdd (_FourDimRedBlackGrid),
Lebesgue(_FourDimGrid),
LebesgueEvenOdd(_FourDimRedBlackGrid)
LebesgueEvenOdd(_FourDimRedBlackGrid),
_tmp(&FiveDimRedBlackGrid)
{
if (Impl::LsVectorised) {

3
lib/qcd/action/fermion/WilsonFermion5D.h
View File

@ -74,6 +74,9 @@ namespace QCD {
typedef WilsonKernels<Impl> Kernels;
PmuStat stat;
FermionField _tmp;
FermionField &tmp(void) { return _tmp; }
void Report(void);
void ZeroCounters(void);
double DhopCalls;

269
lib/serialisation/BaseIO.h
View File

@ -32,6 +32,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#include <type_traits>
namespace Grid {
// Vector IO utilities ///////////////////////////////////////////////////////
// helper function to read space-separated values
template <typename T>
std::vector<T> strToVec(const std::string s)
@ -67,12 +68,78 @@ namespace Grid {
return os;
}
class Serializable {};
// Vector element trait //////////////////////////////////////////////////////
template <typename T>
struct element
{
typedef T type;
static constexpr bool is_number = false;
};
template <typename T>
struct element<std::vector<T>>
{
typedef typename element<T>::type type;
static constexpr bool is_number = std::is_arithmetic<T>::value
or is_complex<T>::value
or element<T>::is_number;
};
// Vector flatening utility class ////////////////////////////////////////////
// Class to flatten a multidimensional std::vector
template <typename V>
class Flatten
{
public:
typedef typename element<V>::type Element;
public:
explicit Flatten(const V &vector);
const V & getVector(void);
const std::vector<Element> & getFlatVector(void);
const std::vector<size_t> & getDim(void);
private:
void accumulate(const Element &e);
template <typename W>
void accumulate(const W &v);
void accumulateDim(const Element &e);
template <typename W>
void accumulateDim(const W &v);
private:
const V &vector_;
std::vector<Element> flatVector_;
std::vector<size_t> dim_;
};
// Class to reconstruct a multidimensional std::vector
template <typename V>
class Reconstruct
{
public:
typedef typename element<V>::type Element;
public:
Reconstruct(const std::vector<Element> &flatVector,
const std::vector<size_t> &dim);
const V & getVector(void);
const std::vector<Element> & getFlatVector(void);
const std::vector<size_t> & getDim(void);
private:
void fill(std::vector<Element> &v);
template <typename W>
void fill(W &v);
void resize(std::vector<Element> &v, const unsigned int dim);
template <typename W>
void resize(W &v, const unsigned int dim);
private:
V vector_;
const std::vector<Element> &flatVector_;
std::vector<size_t> dim_;
size_t ind_{0};
unsigned int dimInd_{0};
};
// Abstract writer/reader classes ////////////////////////////////////////////
// static polymorphism implemented using CRTP idiom
class Serializable;
// Static abstract writer
template <typename T>
@ -87,12 +154,7 @@ namespace Grid {
typename std::enable_if<std::is_base_of<Serializable, U>::value, void>::type
write(const std::string& s, const U &output);
template <typename U>
typename std::enable_if<std::is_enum<U>::value, void>::type
write(const std::string& s, const U &output);
template <typename U>
typename std::enable_if<
!(std::is_base_of<Serializable, U>::value or std::is_enum<U>::value),
void>::type
typename std::enable_if<!std::is_base_of<Serializable, U>::value, void>::type
write(const std::string& s, const U &output);
private:
T *upcast;
@ -111,12 +173,7 @@ namespace Grid {
typename std::enable_if<std::is_base_of<Serializable, U>::value, void>::type
read(const std::string& s, U &output);
template <typename U>
typename std::enable_if<std::is_enum<U>::value, void>::type
read(const std::string& s, U &output);
template <typename U>
typename std::enable_if<
!(std::is_base_of<Serializable, U>::value or std::is_enum<U>::value),
void>::type
typename std::enable_if<!std::is_base_of<Serializable, U>::value, void>::type
read(const std::string& s, U &output);
protected:
template <typename U>
@ -125,16 +182,151 @@ namespace Grid {
T *upcast;
};
// type traits
// What is the vtype
template<typename T> struct isReader {
static const bool value = false;
};
template<typename T> struct isWriter {
static const bool value = false;
};
// Generic writer interface
// serializable base class
class Serializable
{
public:
template <typename T>
static inline void write(Writer<T> &WR,const std::string &s,
const Serializable &obj)
{}
template <typename T>
static inline void read(Reader<T> &RD,const std::string &s,
Serializable &obj)
{}
friend inline std::ostream & operator<<(std::ostream &os,
const Serializable &obj)
{
return os;
}
};
// Flatten class template implementation /////////////////////////////////////
template <typename V>
void Flatten<V>::accumulate(const Element &e)
{
flatVector_.push_back(e);
}
template <typename V>
template <typename W>
void Flatten<V>::accumulate(const W &v)
{
for (auto &e: v)
{
accumulate(e);
}
}
template <typename V>
void Flatten<V>::accumulateDim(const Element &e) {};
template <typename V>
template <typename W>
void Flatten<V>::accumulateDim(const W &v)
{
dim_.push_back(v.size());
accumulateDim(v[0]);
}
template <typename V>
Flatten<V>::Flatten(const V &vector)
: vector_(vector)
{
accumulate(vector_);
accumulateDim(vector_);
}
template <typename V>
const V & Flatten<V>::getVector(void)
{
return vector_;
}
template <typename V>
const std::vector<typename Flatten<V>::Element> &
Flatten<V>::getFlatVector(void)
{
return flatVector_;
}
template <typename V>
const std::vector<size_t> & Flatten<V>::getDim(void)
{
return dim_;
}
// Reconstruct class template implementation /////////////////////////////////
template <typename V>
void Reconstruct<V>::fill(std::vector<Element> &v)
{
for (auto &e: v)
{
e = flatVector_[ind_++];
}
}
template <typename V>
template <typename W>
void Reconstruct<V>::fill(W &v)
{
for (auto &e: v)
{
fill(e);
}
}
template <typename V>
void Reconstruct<V>::resize(std::vector<Element> &v, const unsigned int dim)
{
v.resize(dim_[dim]);
}
template <typename V>
template <typename W>
void Reconstruct<V>::resize(W &v, const unsigned int dim)
{
v.resize(dim_[dim]);
for (auto &e: v)
{
resize(e, dim + 1);
}
}
template <typename V>
Reconstruct<V>::Reconstruct(const std::vector<Element> &flatVector,
const std::vector<size_t> &dim)
: flatVector_(flatVector)
, dim_(dim)
{
resize(vector_, 0);
fill(vector_);
}
template <typename V>
const V & Reconstruct<V>::getVector(void)
{
return vector_;
}
template <typename V>
const std::vector<typename Reconstruct<V>::Element> &
Reconstruct<V>::getFlatVector(void)
{
return flatVector_;
}
template <typename V>
const std::vector<size_t> & Reconstruct<V>::getDim(void)
{
return dim_;
}
// Generic writer interface //////////////////////////////////////////////////
>>>>>>> develop
template <typename T>
inline void push(Writer<T> &w, const std::string &s) {
w.push(s);
@ -212,23 +404,13 @@ namespace Grid {
template <typename T>
template <typename U>
typename std::enable_if<std::is_enum<U>::value, void>::type
Writer<T>::write(const std::string &s, const U &output)
{
EnumIO<U>::write(*this, s, output);
}
template <typename T>
template <typename U>
typename std::enable_if<
!(std::is_base_of<Serializable, U>::value or std::is_enum<U>::value),
void>::type
typename std::enable_if<!std::is_base_of<Serializable, U>::value, void>::type
Writer<T>::write(const std::string &s, const U &output)
{
upcast->writeDefault(s, output);
}
// Reader template implementation ////////////////////////////////////////////
// Reader template implementation
template <typename T>
Reader<T>::Reader(void)
{
@ -257,17 +439,7 @@ namespace Grid {
template <typename T>
template <typename U>
typename std::enable_if<std::is_enum<U>::value, void>::type
Reader<T>::read(const std::string &s, U &output)
{
EnumIO<U>::read(*this, s, output);
}
template <typename T>
template <typename U>
typename std::enable_if<
!(std::is_base_of<Serializable, U>::value or std::is_enum<U>::value),
void>::type
typename std::enable_if<!std::is_base_of<Serializable, U>::value, void>::type
Reader<T>::read(const std::string &s, U &output)
{
upcast->readDefault(s, output);
@ -291,7 +463,6 @@ namespace Grid {
abort();
}
}
}
#endif

103
lib/serialisation/Hdf5IO.cc Normal file
View File

@ -0,0 +1,103 @@
#include <Grid.h>
using namespace Grid;
#ifndef H5_NO_NAMESPACE
using namespace H5NS;
#endif
// Writer implementation ///////////////////////////////////////////////////////
Hdf5Writer::Hdf5Writer(const std::string &fileName)
: fileName_(fileName)
, file_(fileName.c_str(), H5F_ACC_TRUNC)
{
group_ = file_.openGroup("/");
writeSingleAttribute(dataSetThres_, HDF5_GRID_GUARD "dataset_threshold",
Hdf5Type<unsigned int>::type());
}
void Hdf5Writer::push(const std::string &s)
{
group_ = group_.createGroup(s);
path_.push_back(s);
}
void Hdf5Writer::pop(void)
{
path_.pop_back();
if (path_.empty())
{
group_ = file_.openGroup("/");
}
else
{
auto binOp = [](const std::string &a, const std::string &b)->std::string
{
return a + "/" + b;
};
group_ = group_.openGroup(std::accumulate(path_.begin(), path_.end(),
std::string(""), binOp));
}
}
template <>
void Hdf5Writer::writeDefault(const std::string &s, const std::string &x)
{
StrType strType(PredType::C_S1, x.size());
writeSingleAttribute(*(x.data()), s, strType);
}
void Hdf5Writer::writeDefault(const std::string &s, const char *x)
{
std::string sx(x);
writeDefault(s, sx);
}
// Reader implementation ///////////////////////////////////////////////////////
Hdf5Reader::Hdf5Reader(const std::string &fileName)
: fileName_(fileName)
, file_(fileName.c_str(), H5F_ACC_RDONLY)
{
group_ = file_.openGroup("/");
readSingleAttribute(dataSetThres_, HDF5_GRID_GUARD "dataset_threshold",
Hdf5Type<unsigned int>::type());
}
void Hdf5Reader::push(const std::string &s)
{
group_ = group_.openGroup(s);
path_.push_back(s);
}
void Hdf5Reader::pop(void)
{
path_.pop_back();
if (path_.empty())
{
group_ = file_.openGroup("/");
}
else
{
auto binOp = [](const std::string &a, const std::string &b)->std::string
{
return a + "/" + b;
};
group_ = group_.openGroup(std::accumulate(path_.begin(), path_.end(),
std::string(""), binOp));
}
}
template <>
void Hdf5Reader::readDefault(const std::string &s, std::string &x)
{
Attribute attribute;
attribute = group_.openAttribute(s);
StrType strType = attribute.getStrType();
x.resize(strType.getSize());
attribute.read(strType, &(x[0]));
}

242
lib/serialisation/Hdf5IO.h Normal file
View File

@ -0,0 +1,242 @@
#ifndef GRID_SERIALISATION_HDF5_H
#define GRID_SERIALISATION_HDF5_H
#include <stack>
#include <string>
#include <vector>
#include <H5Cpp.h>
#include "Hdf5Type.h"
#ifndef H5_NO_NAMESPACE
#define H5NS H5
#endif
// default thresold above which datasets are used instead of attributes
#ifndef HDF5_DEF_DATASET_THRES
#define HDF5_DEF_DATASET_THRES 6u
#endif
// name guard for Grid metadata
#define HDF5_GRID_GUARD "_Grid_"
namespace Grid
{
class Hdf5Writer: public Writer<Hdf5Writer>
{
public:
Hdf5Writer(const std::string &fileName);
virtual ~Hdf5Writer(void) = default;
void push(const std::string &s);
void pop(void);
void writeDefault(const std::string &s, const char *x);
template <typename U>
void writeDefault(const std::string &s, const U &x);
template <typename U>
typename std::enable_if<element<std::vector<U>>::is_number, void>::type
writeDefault(const std::string &s, const std::vector<U> &x);
template <typename U>
typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
writeDefault(const std::string &s, const std::vector<U> &x);
private:
template <typename U>
void writeSingleAttribute(const U &x, const std::string &name,
const H5NS::DataType &type);
private:
std::string fileName_;
std::vector<std::string> path_;
H5NS::H5File file_;
H5NS::Group group_;
unsigned int dataSetThres_{HDF5_DEF_DATASET_THRES};
};
class Hdf5Reader: public Reader<Hdf5Reader>
{
public:
Hdf5Reader(const std::string &fileName);
virtual ~Hdf5Reader(void) = default;
void push(const std::string &s);
void pop(void);
template <typename U>
void readDefault(const std::string &s, U &output);
template <typename U>
typename std::enable_if<element<std::vector<U>>::is_number, void>::type
readDefault(const std::string &s, std::vector<U> &x);
template <typename U>
typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
readDefault(const std::string &s, std::vector<U> &x);
private:
template <typename U>
void readSingleAttribute(U &x, const std::string &name,
const H5NS::DataType &type);
private:
std::string fileName_;
std::vector<std::string> path_;
H5NS::H5File file_;
H5NS::Group group_;
unsigned int dataSetThres_;
};
// Writer template implementation ////////////////////////////////////////////
template <typename U>
void Hdf5Writer::writeSingleAttribute(const U &x, const std::string &name,
const H5NS::DataType &type)
{
H5NS::Attribute attribute;
hsize_t attrDim = 1;
H5NS::DataSpace attrSpace(1, &attrDim);
attribute = group_.createAttribute(name, type, attrSpace);
attribute.write(type, &x);
}
template <typename U>
void Hdf5Writer::writeDefault(const std::string &s, const U &x)
{
writeSingleAttribute(x, s, Hdf5Type<U>::type());
}
template <>
void Hdf5Writer::writeDefault(const std::string &s, const std::string &x);
template <typename U>
typename std::enable_if<element<std::vector<U>>::is_number, void>::type
Hdf5Writer::writeDefault(const std::string &s, const std::vector<U> &x)
{
// alias to element type
typedef typename element<std::vector<U>>::type Element;
// flatten the vector and getting dimensions
Flatten<std::vector<U>> flat(x);
std::vector<hsize_t> dim;
const auto &flatx = flat.getFlatVector();
for (auto &d: flat.getDim())
{
dim.push_back(d);
}
// write to file
H5NS::DataSpace dataSpace(dim.size(), dim.data());
if (flatx.size() > dataSetThres_)
{
H5NS::DataSet dataSet;
dataSet = group_.createDataSet(s, Hdf5Type<Element>::type(), dataSpace);
dataSet.write(flatx.data(), Hdf5Type<Element>::type());
}
else
{
H5NS::Attribute attribute;
attribute = group_.createAttribute(s, Hdf5Type<Element>::type(), dataSpace);
attribute.write(Hdf5Type<Element>::type(), flatx.data());
}
}
template <typename U>
typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
Hdf5Writer::writeDefault(const std::string &s, const std::vector<U> &x)
{
push(s);
writeSingleAttribute(x.size(), HDF5_GRID_GUARD "vector_size",
Hdf5Type<uint64_t>::type());
for (hsize_t i = 0; i < x.size(); ++i)
{
write(s + "_" + std::to_string(i), x[i]);
}
pop();
}
// Reader template implementation ////////////////////////////////////////////
template <typename U>
void Hdf5Reader::readSingleAttribute(U &x, const std::string &name,
const H5NS::DataType &type)
{
H5NS::Attribute attribute;
attribute = group_.openAttribute(name);
attribute.read(type, &x);
}
template <typename U>
void Hdf5Reader::readDefault(const std::string &s, U &output)
{
readSingleAttribute(output, s, Hdf5Type<U>::type());
}
template <>
void Hdf5Reader::readDefault(const std::string &s, std::string &x);
template <typename U>
typename std::enable_if<element<std::vector<U>>::is_number, void>::type
Hdf5Reader::readDefault(const std::string &s, std::vector<U> &x)
{
// alias to element type
typedef typename element<std::vector<U>>::type Element;
// read the dimensions
H5NS::DataSpace dataSpace;
std::vector<hsize_t> hdim;
std::vector<size_t> dim;
hsize_t size = 1;
if (group_.attrExists(s))
{
dataSpace = group_.openAttribute(s).getSpace();
}
else
{
dataSpace = group_.openDataSet(s).getSpace();
}
hdim.resize(dataSpace.getSimpleExtentNdims());
dataSpace.getSimpleExtentDims(hdim.data());
for (auto &d: hdim)
{
dim.push_back(d);
size *= d;
}
// read the flat vector
std::vector<Element> buf(size);
if (size > dataSetThres_)
{
H5NS::DataSet dataSet;
dataSet = group_.openDataSet(s);
dataSet.read(buf.data(), Hdf5Type<Element>::type());
}
else
{
H5NS::Attribute attribute;
attribute = group_.openAttribute(s);
attribute.read(Hdf5Type<Element>::type(), buf.data());
}
// reconstruct the multidimensional vector
Reconstruct<std::vector<U>> r(buf, dim);
x = r.getVector();
}
template <typename U>
typename std::enable_if<!element<std::vector<U>>::is_number, void>::type
Hdf5Reader::readDefault(const std::string &s, std::vector<U> &x)
{
uint64_t size;
push(s);
readSingleAttribute(size, HDF5_GRID_GUARD "vector_size",
Hdf5Type<uint64_t>::type());
x.resize(size);
for (hsize_t i = 0; i < x.size(); ++i)
{
read(s + "_" + std::to_string(i), x[i]);
}
pop();
}
}
#endif

77
lib/serialisation/Hdf5Type.h Normal file
View File

@ -0,0 +1,77 @@
#ifndef GRID_SERIALISATION_HDF5_TYPE_H
#define GRID_SERIALISATION_HDF5_TYPE_H
#include <H5Cpp.h>
#include <complex>
#include <memory>
#ifndef H5_NO_NAMESPACE
#define H5NS H5
#endif
#define HDF5_NATIVE_TYPE(predType, cType)\
template <>\
class Hdf5Type<cType>\
{\
public:\
static inline const H5NS::DataType & type(void)\
{\
return H5NS::PredType::predType;\
}\
static constexpr bool isNative = true;\
};
#define DEFINE_HDF5_NATIVE_TYPES \
HDF5_NATIVE_TYPE(NATIVE_B8, bool);\
HDF5_NATIVE_TYPE(NATIVE_CHAR, char);\
HDF5_NATIVE_TYPE(NATIVE_SCHAR, signed char);\
HDF5_NATIVE_TYPE(NATIVE_UCHAR, unsigned char);\
HDF5_NATIVE_TYPE(NATIVE_SHORT, short);\
HDF5_NATIVE_TYPE(NATIVE_USHORT, unsigned short);\
HDF5_NATIVE_TYPE(NATIVE_INT, int);\
HDF5_NATIVE_TYPE(NATIVE_UINT, unsigned int);\
HDF5_NATIVE_TYPE(NATIVE_LONG, long);\
HDF5_NATIVE_TYPE(NATIVE_ULONG, unsigned long);\
HDF5_NATIVE_TYPE(NATIVE_LLONG, long long);\
HDF5_NATIVE_TYPE(NATIVE_ULLONG, unsigned long long);\
HDF5_NATIVE_TYPE(NATIVE_FLOAT, float);\
HDF5_NATIVE_TYPE(NATIVE_DOUBLE, double);\
HDF5_NATIVE_TYPE(NATIVE_LDOUBLE, long double);
namespace Grid
{
template <typename T> class Hdf5Type
{
public:
static constexpr bool isNative = false;
};
DEFINE_HDF5_NATIVE_TYPES;
template <typename R>
class Hdf5Type<std::complex<R>>
{
public:
static inline const H5NS::DataType & type(void)
{
if (typePtr_ == nullptr)
{
typePtr_.reset(new H5NS::CompType(sizeof(std::complex<R>)));
typePtr_->insertMember("re", 0, Hdf5Type<R>::type());
typePtr_->insertMember("im", sizeof(R), Hdf5Type<R>::type());
}
return *typePtr_;
}
static constexpr bool isNative = false;
private:
static std::unique_ptr<H5NS::CompType> typePtr_;
};
template <typename R>
std::unique_ptr<H5NS::CompType> Hdf5Type<std::complex<R>>::typePtr_ = nullptr;
}
#undef HDF5_NATIVE_TYPE
#endif /* GRID_SERIALISATION_HDF5_TYPE_H */

130
lib/serialisation/MacroMagic.h
View File

@ -109,40 +109,36 @@ THE SOFTWARE.
//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
#define GRID_MACRO_MEMBER(A,B) A B;
#define GRID_MACRO_COMP_MEMBER(A,B) result = (result and (lhs. B == rhs. B));
#define GRID_MACRO_OS_WRITE_MEMBER(A,B) os<< #A <<" "#B <<" = "<< obj. B <<" ; " <<std::endl;
#define GRID_MACRO_READ_MEMBER(A,B) Grid::read(RD,#B,obj. B);
#define GRID_MACRO_WRITE_MEMBER(A,B) Grid::write(WR,#B,obj. B);
#define GRID_SERIALIZABLE_CLASS_MEMBERS(cname,...) \
\
\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_MEMBER,__VA_ARGS__)) \
\
\
template <typename T>\
static inline void write(Writer<T> &WR,const std::string &s, const cname &obj){ \
push(WR,s);\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_WRITE_MEMBER,__VA_ARGS__)) \
pop(WR);\
} \
\
\
template <typename T>\
static inline void read(Reader<T> &RD,const std::string &s, cname &obj){ \
push(RD,s);\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_READ_MEMBER,__VA_ARGS__)) \
pop(RD);\
} \
\
\
friend inline std::ostream & operator << (std::ostream &os, const cname &obj ) { \
os<<"class "<<#cname<<" {"<<std::endl;\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_OS_WRITE_MEMBER,__VA_ARGS__)) \
os<<"}"; \
return os;\
};
#define GRID_SERIALIZABLE_CLASS_MEMBERS(cname,...)\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_MEMBER,__VA_ARGS__))\
template <typename T>\
static inline void write(Writer<T> &WR,const std::string &s, const cname &obj){ \
push(WR,s);\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_WRITE_MEMBER,__VA_ARGS__)) \
pop(WR);\
}\
template <typename T>\
static inline void read(Reader<T> &RD,const std::string &s, cname &obj){ \
push(RD,s);\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_READ_MEMBER,__VA_ARGS__)) \
pop(RD);\
}\
friend inline std::ostream & operator << (std::ostream &os, const cname &obj ) { \
os<<"class "<<#cname<<" {"<<std::endl;\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_OS_WRITE_MEMBER,__VA_ARGS__)) \
os<<"}"; \
return os;\
}\
friend inline bool operator==(const cname &lhs, const cname &rhs) {\
bool result = true;\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_COMP_MEMBER,__VA_ARGS__))\
return result;\
}
#define GRID_ENUM_TYPE(obj) std::remove_reference<decltype(obj)>::type
#define GRID_MACRO_ENUMVAL(A,B) A = B,
@ -150,44 +146,52 @@ THE SOFTWARE.
#define GRID_MACRO_ENUMTEST(A,B) else if (buf == #A) {obj = GRID_ENUM_TYPE(obj)::A;}
#define GRID_MACRO_ENUMCASEIO(A,B) case GRID_ENUM_TYPE(obj)::A: os << #A; break;
namespace Grid {
template <typename U>
class EnumIO {};
}
#define GRID_SERIALIZABLE_ENUM(name,undefname,...)\
enum class name {\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMVAL,__VA_ARGS__))\
undefname = -1\
class name: public Grid::Serializable\
{\
public:\
enum EnumType\
{\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMVAL,__VA_ARGS__))\
undefname = -1\
};\
public:\
name(void): value_(undefname) {};\
name(EnumType value): value_(value) {};\
template <typename T>\
static inline void write(Grid::Writer<T> &WR,const std::string &s, const name &obj)\
{\
switch (obj.value_)\
{\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMCASE,__VA_ARGS__))\
default: Grid::write(WR,s,#undefname); break;\
}\
}\
\
template<>\
class EnumIO<name> {\
public:\
template <typename T>\
static inline void write(Writer<T> &WR,const std::string &s, const name &obj){ \
switch (obj) {\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMCASE,__VA_ARGS__))\
default: Grid::write(WR,s,#undefname); break;\
}\
}\
\
template <typename T>\
static inline void read(Reader<T> &RD,const std::string &s, name &obj){ \
std::string buf;\
Grid::read(RD, s, buf);\
if (buf == #undefname) {obj = name::undefname;}\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMTEST,__VA_ARGS__))\
else {obj = name::undefname;}\
}\
};\
\
inline std::ostream & operator << (std::ostream &os, const name &obj ) { \
template <typename T>\
static inline void read(Grid::Reader<T> &RD,const std::string &s, name &obj)\
{\
std::string buf;\
Grid::read(RD, s, buf);\
if (buf == #undefname) {obj = name::undefname;}\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMTEST,__VA_ARGS__))\
else {obj = name::undefname;}\
}\
inline operator EnumType(void) const\
{\
return value_;\
}\
inline friend std::ostream & operator<<(std::ostream &os, const name &obj)\
{\
switch (obj) {\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMCASEIO,__VA_ARGS__))\
default: os << #undefname; break;\
GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_ENUMCASEIO,__VA_ARGS__))\
default: os << #undefname; break;\
}\
return os;\
};
}\
private:\
EnumType value_;\
};
#endif

5
lib/serialisation/Serialisation.h
View File

@ -37,6 +37,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#include "TextIO.h"
#include "XmlIO.h"
#include "JSON_IO.h"
#ifdef HAVE_HDF5
#include "Hdf5IO.h"
#endif
//////////////////////////////////////////
// Todo:
//////////////////////////////////////////

27
lib/simd/Grid_avx.h
View File

@ -204,6 +204,29 @@ namespace Optimization {
}
};
struct MultRealPart{
inline __m256 operator()(__m256 a, __m256 b){
__m256 ymm0;
ymm0 = _mm256_shuffle_ps(a,a,_MM_SELECT_FOUR_FOUR(2,2,0,0)); // ymm0 <- ar ar,
return _mm256_mul_ps(ymm0,b); // ymm0 <- ar bi, ar br
}
inline __m256d operator()(__m256d a, __m256d b){
__m256d ymm0;
ymm0 = _mm256_shuffle_pd(a,a,0x0); // ymm0 <- ar ar, ar,ar b'00,00
return _mm256_mul_pd(ymm0,b); // ymm0 <- ar bi, ar br
}
};
struct MaddRealPart{
inline __m256 operator()(__m256 a, __m256 b, __m256 c){
__m256 ymm0 = _mm256_moveldup_ps(a); // ymm0 <- ar ar,
return _mm256_add_ps(_mm256_mul_ps( ymm0, b),c);
}
inline __m256d operator()(__m256d a, __m256d b, __m256d c){
__m256d ymm0 = _mm256_shuffle_pd( a, a, 0x0 );
return _mm256_add_pd(_mm256_mul_pd( ymm0, b),c);
}
};
struct MultComplex{
// Complex float
inline __m256 operator()(__m256 a, __m256 b){
@ -618,7 +641,9 @@ namespace Optimization {
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultRealPart MultRealPartSIMD;
typedef Optimization::MaddRealPart MaddRealPartSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;

25
lib/simd/Grid_avx512.h
View File

@ -189,6 +189,29 @@ namespace Optimization {
// 2mul,4 mac +add+sub = 8 flop type insns
// 3shuf + 2 (+shuf) = 5/6 simd perm and 1/2 the load.
struct MultRealPart{
inline __m512 operator()(__m512 a, __m512 b){
__m512 ymm0;
ymm0 = _mm512_moveldup_ps(a); // ymm0 <- ar ar,
return _mm512_mul_ps(ymm0,b); // ymm0 <- ar bi, ar br
}
inline __m512d operator()(__m512d a, __m512d b){
__m512d ymm0;
ymm0 = _mm512_shuffle_pd(a,a,0x00); // ymm0 <- ar ar, ar,ar b'00,00
return _mm512_mul_pd(ymm0,b); // ymm0 <- ar bi, ar br
}
};
struct MaddRealPart{
inline __m512 operator()(__m512 a, __m512 b, __m512 c){
__m512 ymm0 = _mm512_moveldup_ps(a); // ymm0 <- ar ar,
return _mm512_fmadd_ps( ymm0, b, c);
}
inline __m512d operator()(__m512d a, __m512d b, __m512d c){
__m512d ymm0 = _mm512_shuffle_pd( a, a, 0x00 );
return _mm512_fmadd_pd( ymm0, b, c);
}
};
struct MultComplex{
// Complex float
inline __m512 operator()(__m512 a, __m512 b){
@ -501,6 +524,8 @@ namespace Optimization {
typedef Optimization::Mult MultSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultRealPart MultRealPartSIMD;
typedef Optimization::MaddRealPart MaddRealPartSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;

16
lib/simd/Grid_generic.h
View File

@ -224,6 +224,21 @@ namespace Optimization {
#define cmul(a, b, c, i)\
c[i] = a[i]*b[i] - a[i+1]*b[i+1];\
c[i+1] = a[i]*b[i+1] + a[i+1]*b[i];
struct MultRealPart{
template <typename T>
inline vec<T> operator()(vec<T> a, vec<T> b){
vec<T> out;
VECTOR_FOR(i, W<T>::c, 1)
{
out.v[2*i] = a[2*i]*b[2*i];
out.v[2*i+1] = a[2*i]*b[2*i+1];
}
return out;
};
};
struct MultComplex{
// Complex
@ -456,6 +471,7 @@ namespace Optimization {
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultRealPart MultRealPartSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;

9
lib/simd/Grid_qpx.h
View File

@ -220,6 +220,14 @@ namespace Optimization {
}
};
struct MultRealPart{
// Complex double
inline vector4double operator()(vector4double a, vector4double b){
// return vec_xmul(b, a);
return vec_xmul(a, b);
}
FLOAT_WRAP_2(operator(), inline)
};
struct MultComplex{
// Complex double
inline vector4double operator()(vector4double a, vector4double b){
@ -430,6 +438,7 @@ typedef Optimization::Sub SubSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultRealPart MultRealPartSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;

29
lib/simd/Grid_sse4.h
View File

@ -177,6 +177,29 @@ namespace Optimization {
}
};
struct MultRealPart{
inline __m128 operator()(__m128 a, __m128 b){
__m128 ymm0;
ymm0 = _mm_shuffle_ps(a,a,_MM_SELECT_FOUR_FOUR(2,2,0,0)); // ymm0 <- ar ar,
return _mm_mul_ps(ymm0,b); // ymm0 <- ar bi, ar br
}
inline __m128d operator()(__m128d a, __m128d b){
__m128d ymm0;
ymm0 = _mm_shuffle_pd(a,a,0x0); // ymm0 <- ar ar, ar,ar b'00,00
return _mm_mul_pd(ymm0,b); // ymm0 <- ar bi, ar br
}
};
struct MaddRealPart{
inline __m128 operator()(__m128 a, __m128 b, __m128 c){
__m128 ymm0 = _mm_shuffle_ps(a,a,_MM_SELECT_FOUR_FOUR(2,2,0,0)); // ymm0 <- ar ar,
return _mm_add_ps(_mm_mul_ps( ymm0, b),c);
}
inline __m128d operator()(__m128d a, __m128d b, __m128d c){
__m128d ymm0 = _mm_shuffle_pd( a, a, 0x0 );
return _mm_add_pd(_mm_mul_pd( ymm0, b),c);
}
};
struct MultComplex{
// Complex float
inline __m128 operator()(__m128 a, __m128 b){
@ -325,9 +348,11 @@ namespace Optimization {
}
}
#ifndef _mm_alignr_epi64
#define _mm_alignr_epi32(a,b,n) _mm_alignr_epi8(a,b,(n*4)%16)
#define _mm_alignr_epi64(a,b,n) _mm_alignr_epi8(a,b,(n*8)%16)
#endif
template<int n> static inline __m128 tRotate(__m128 in){ return (__m128)_mm_alignr_epi32((__m128i)in,(__m128i)in,n); };
template<int n> static inline __m128d tRotate(__m128d in){ return (__m128d)_mm_alignr_epi64((__m128i)in,(__m128i)in,n); };
@ -415,6 +440,8 @@ namespace Optimization {
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::MultRealPart MultRealPartSIMD;
typedef Optimization::MaddRealPart MaddRealPartSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;

33
lib/simd/Grid_vector_types.h
View File

@ -101,6 +101,11 @@ template <typename T> using IfNotInteger = Invoke<std::enable_if<!std::is_integr
// general forms to allow for vsplat syntax
// need explicit declaration of types when used since
// clang cannot automatically determine the output type sometimes
template <class Out, class Input1, class Input2, class Input3, class Operation>
Out trinary(Input1 src_1, Input2 src_2, Input3 src_3, Operation op) {
return op(src_1, src_2, src_3);
}
template <class Out, class Input1, class Input2, class Operation>
Out binary(Input1 src_1, Input2 src_2, Operation op) {
return op(src_1, src_2);
@ -178,6 +183,7 @@ class Grid_simd {
const Grid_simd *__restrict__ r) {
*y = (*l) * (*r);
}
friend inline void sub(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Grid_simd *__restrict__ r) {
@ -188,7 +194,6 @@ class Grid_simd {
const Grid_simd *__restrict__ r) {
*y = (*l) + (*r);
}
friend inline void mac(Grid_simd *__restrict__ y,
const Scalar_type *__restrict__ a,
const Grid_simd *__restrict__ x) {
@ -260,7 +265,7 @@ class Grid_simd {
}
////////////////////////////
// opreator scalar * simd
// operator scalar * simd
////////////////////////////
friend inline Grid_simd operator*(const Scalar_type &a, Grid_simd b) {
Grid_simd va;
@ -446,6 +451,11 @@ inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
S* typepun =(S*) &src;
vsplat(ret,typepun[lane]);
}
template <class S, class V, IfComplex<S> =0>
inline void rbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
S* typepun =(S*) &src;
ret.v = unary<V>(real(typepun[lane]), VsplatSIMD());
}
///////////////////////
// Splat
@ -462,6 +472,10 @@ template <class S, class V>
inline void vsplat(Grid_simd<S, V> &ret, EnableIf<is_complex<S>, S> c) {
vsplat(ret, real(c), imag(c));
}
template <class S, class V>
inline void rsplat(Grid_simd<S, V> &ret, EnableIf<is_complex<S>, S> c) {
vsplat(ret, real(c), real(c));
}
// if real fill with a, if complex fill with a in the real part (first function
// above)
@ -563,6 +577,21 @@ inline Grid_simd<S, V> operator-(Grid_simd<S, V> a, Grid_simd<S, V> b) {
return ret;
};
// Distinguish between complex types and others
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> real_mult(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, MultRealPartSIMD());
return ret;
};
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> real_madd(Grid_simd<S, V> a, Grid_simd<S, V> b, Grid_simd<S,V> c) {
Grid_simd<S, V> ret;
ret.v = trinary<V>(a.v, b.v, c.v, MaddRealPartSIMD());
return ret;
};
// Distinguish between complex types and others
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V> b) {

16
lib/simd/Intel512avx.h
View File

@ -95,10 +95,14 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VIDUPd(SRC,DEST) "vpshufd $0xee," #SRC"," #DEST ";\n" // 32 bit level: 3,2,3,2
#define VIDUPf(SRC,DEST) "vmovshdup " #SRC ", " #DEST ";\n"
#define VBCASTRDUPd(OFF,A,DEST) "vbroadcastsd (" #OFF "*16+0)(" #A ")," #DEST ";\n"
#define VBCASTIDUPd(OFF,A,DEST) "vbroadcastsd (" #OFF "*16+8)(" #A ")," #DEST ";\n"
#define VBCASTRDUPf(OFF,PTR,DEST) "vbroadcastss (" #OFF "*8 +0)(" #PTR "), " #DEST ";\n"
#define VBCASTIDUPf(OFF,PTR,DEST) "vbroadcastss (" #OFF "*8 +4)(" #PTR "), " #DEST ";\n"
#define VBCASTRDUPd(OFF,A,DEST) "vbroadcastsd (" #OFF "*16+0)(" #A ")," #DEST ";\n"
#define VBCASTIDUPd(OFF,A,DEST) "vbroadcastsd (" #OFF "*16+8)(" #A ")," #DEST ";\n"
#define VBCASTRDUPf(OFF,PTR,DEST) "vbroadcastss (" #OFF "*8 +0)(" #PTR "), " #DEST ";\n"
#define VBCASTIDUPf(OFF,PTR,DEST) "vbroadcastss (" #OFF "*8 +4)(" #PTR "), " #DEST ";\n"
#define VBCASTCDUPf(OFF,A,DEST) "vbroadcastsd (" #OFF "*64 )(" #A ")," #DEST ";\n"
#define VBCASTZDUPf(OFF,A,DEST) "vbroadcastf32x4 (" #OFF "*64 )(" #A ")," #DEST ";\n"
#define VBCASTCDUP(OFF,A,DEST) VBCASTCDUPf(OFF,A,DEST)
#define VBCASTZDUP(OFF,A,DEST) VBCASTZDUPf(OFF,A,DEST)
#define VMADDSUBf(A,B,accum) "vfmaddsub231ps " #A "," #B "," #accum ";\n"
#define VMADDSUBd(A,B,accum) "vfmaddsub231pd " #A "," #B "," #accum ";\n"
@ -106,11 +110,15 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VMADDSUBMEMd(O,P,B,accum) "vfmaddsub231pd " #O"*64("#P "),"#B "," #accum ";\n"
#define VMADDRDUPf(O,P,B,accum) "vfmadd231ps (" #O"*8+0)("#P "){1to16},"#B "," #accum ";\n"
#define VMADDIDUPf(O,P,B,accum) "vfmadd231ps (" #O"*8+4)("#P "){1to16},"#B "," #accum ";\n"
#define VMADDSUBRDUPf(O,P,B,accum) "vfmaddsub231ps (" #O"*8+0)("#P "){1to16},"#B "," #accum ";\n"
#define VMADDSUBIDUPf(O,P,B,accum) "vfmaddsub231ps (" #O"*8+4)("#P "){1to16},"#B "," #accum ";\n"
#define VMULRDUPf(O,P,B,accum) "vmulps (" #O"*8+0)("#P "){1to16},"#B "," #accum ";\n"
#define VMULIDUPf(O,P,B,accum) "vmulps (" #O"*8+4)("#P "){1to16},"#B "," #accum ";\n"
#define VMADDRDUPd(O,P,B,accum) "vfmadd231pd (" #O"*16+0)("#P "){1to8},"#B "," #accum ";\n"
#define VMADDIDUPd(O,P,B,accum) "vfmadd231pd (" #O"*16+8)("#P "){1to8},"#B "," #accum ";\n"
#define VMADDSUBRDUPd(O,P,B,accum) "vfmaddsub231pd (" #O"*16+0)("#P "){1to8},"#B "," #accum ";\n"
#define VMADDSUBIDUPd(O,P,B,accum) "vfmaddsub231pd (" #O"*16+8)("#P "){1to8},"#B "," #accum ";\n"
#define VMULRDUPd(O,P,B,accum) "vmulpd (" #O"*16+0)("#P "){1to8},"#B "," #accum ";\n"

11
lib/simd/Intel512common.h
View File

@ -87,7 +87,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
VACCTIMESMINUSI1d(A,ACC,tmp) \
VACCTIMESMINUSI2d(A,ACC,tmp)
#define LOAD64i(A,ptr) __asm__ ( "movq %0, %" #A : : "r"(ptr) : #A );
#define LOAD64a(A,ptr) "movq %0, %" #A : : "r"(ptr) : #A
#define LOAD64i(A,ptr) __asm__ ( LOAD64a(A,ptr));
#define LOAD64(A,ptr) LOAD64i(A,ptr)
#define VMOVf(A,DEST) "vmovaps " #A ", " #DEST ";\n"
@ -108,8 +109,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
//"vprefetche0 "#O"*64("#A");\n" "vprefetche1 ("#O"+12)*64("#A");\n"
// "clevict0 "#O"*64("#A");\n"
#define VLOADf(OFF,PTR,DEST) "vmovaps " #OFF "*64(" #PTR "), " #DEST ";\n"
#define VLOADd(OFF,PTR,DEST) "vmovapd " #OFF "*64(" #PTR "), " #DEST ";\n"
#define VLOADf(OFF,PTR,DEST) "vmovups " #OFF "*64(" #PTR "), " #DEST ";\n"
#define VLOADd(OFF,PTR,DEST) "vmovupd " #OFF "*64(" #PTR "), " #DEST ";\n"
#define VADDf(A,B,DEST) "vaddps " #A "," #B "," #DEST ";\n"
#define VADDd(A,B,DEST) "vaddpd " #A "," #B "," #DEST ";\n"
@ -143,8 +144,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VSTOREf(OFF,PTR,SRC) "vmovntps " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#define VSTOREd(OFF,PTR,SRC) "vmovntpd " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#else
#define VSTOREf(OFF,PTR,SRC) "vmovaps " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#define VSTOREd(OFF,PTR,SRC) "vmovapd " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#define VSTOREf(OFF,PTR,SRC) "vmovups " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#define VSTOREd(OFF,PTR,SRC) "vmovupd " #SRC "," #OFF "*64(" #PTR ")" ";\n"
#endif
// Swaps Re/Im ; could unify this with IMCI

2
lib/simd/Intel512double.h
View File

@ -144,10 +144,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VMADDSUBMEM(O,P,B,accum) VMADDSUBMEMd(O,P,B,accum)
#define VMADDMEM(O,P,B,accum) VMADDMEMd(O,P,B,accum)
#define VMULMEM(O,P,B,accum) VMULMEMd(O,P,B,accum)
#undef VMADDRDUP
#undef VMADDSUBRDUP
#undef VMADDSUBIDUP
#undef VMULRDUP
#undef VMULIDUP
#define VMADDRDUP(O,P,B,accum) VMADDRDUPd(O,P,B,accum)
#define VMADDSUBRDUP(O,P,B,accum) VMADDSUBRDUPd(O,P,B,accum)
#define VMADDSUBIDUP(O,P,B,accum) VMADDSUBIDUPd(O,P,B,accum)
#define VMULRDUP(O,P,B,accum) VMULRDUPd(O,P,B,accum)

2
lib/simd/Intel512single.h
View File

@ -144,10 +144,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VMADDMEM(O,P,B,accum) VMADDMEMf(O,P,B,accum)
#define VMULMEM(O,P,B,accum) VMULMEMf(O,P,B,accum)
#undef VMADDRDUP
#undef VMADDSUBRDUP
#undef VMADDSUBIDUP
#undef VMULRDUP
#undef VMULIDUP
#define VMADDRDUP(O,P,B,accum) VMADDRDUPf(O,P,B,accum)
#define VMADDSUBRDUP(O,P,B,accum) VMADDSUBRDUPf(O,P,B,accum)
#define VMADDSUBIDUP(O,P,B,accum) VMADDSUBIDUPf(O,P,B,accum)
#define VMULRDUP(O,P,B,accum) VMULRDUPf(O,P,B,accum)

2
lib/stencil/Lebesgue.cc
View File

@ -65,7 +65,7 @@ void LebesgueOrder::CartesianBlocking(void)
{
_LebesgueReorder.resize(0);
std::cout << GridLogDebug << " CartesianBlocking ";
// std::cout << GridLogDebug << " CartesianBlocking ";
// for(int d=0;d<Block.size();d++) std::cout <<Block[d]<<" ";
// std::cout<<std::endl;

10
scripts/filelist
View File

@ -4,9 +4,8 @@ home=`pwd`
# library Make.inc
cd $home/lib
HFILES=`find . -type f -name '*.h' -not -path '*/Old/*' -not -path '*/Eigen/*'`
HFILES="$HFILES"
CCFILES=`find . -type f -name '*.cc' -not -name '*ommunicator*.cc'`
HFILES=`find . -type f -name '*.h' -not -name '*Hdf5*' -not -path '*/Old/*' -not -path '*/Eigen/*'`
CCFILES=`find . -type f -name '*.cc' -not -name '*Communicator*.cc' -not -name '*Hdf5*'`
echo HFILES=$HFILES > Make.inc
echo >> Make.inc
echo CCFILES=$CCFILES >> Make.inc
@ -24,10 +23,11 @@ for subdir in $dirs; do
echo "tests: ${TESTLIST} ${SUB}" > Make.inc
echo ${PREF}_PROGRAMS = ${TESTLIST} >> Make.inc
echo >> Make.inc
HADLINK=`[ $subdir = './hadrons' ] && echo '-lHadrons '`
for f in $TESTS; do
BNAME=`basename $f .cc`
echo ${BNAME}_SOURCES=$f >> Make.inc
echo ${BNAME}_LDADD=-lGrid>> Make.inc
echo ${BNAME}_SOURCES=$f >> Make.inc
echo ${BNAME}_LDADD=${HADLINK}-lGrid >> Make.inc
echo >> Make.inc
done
if [ $subdir != '.' ]; then

241
tests/IO/Test_serialisation.cc
View File

@ -29,132 +29,155 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
/* END LEGAL */
#include <Grid/Grid.h>
namespace Grid {
GRID_SERIALIZABLE_ENUM(myenum, undef, red, 1, blue, 2, green, 3);
class myclass: Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(myclass,
myenum, e,
std::vector<myenum>, ve,
std::string, name,
int, x,
double, y,
bool , b,
std::vector<double>, array,
std::vector<std::vector<double>>, twodimarray,
);
myclass() {}
myclass(int i)
: array(4,5.1), twodimarray(3,std::vector<double>(2,1.23456)), ve({myenum::blue, myenum::red})
{
e=myenum::red;
x=i;
y=2*i;
b=true;
name="bother said pooh";
}
};
}
using namespace Grid;
int16_t i16 = 1;
GRID_SERIALIZABLE_ENUM(myenum, undef, red, 1, blue, 2, green, 3);
class myclass: Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(myclass,
myenum, e,
std::vector<myenum>, ve,
std::string, name,
int, x,
double, y,
bool , b,
std::vector<double>, array,
std::vector<std::vector<double>>, twodimarray,
std::vector<std::vector<std::vector<Complex>>>, cmplx3darray
);
myclass() {}
myclass(int i)
: array(4,5.1)
, twodimarray(3,std::vector<double>(5, 1.23456))
, cmplx3darray(3,std::vector<std::vector<Complex>>(5, std::vector<Complex>(7, Complex(1.2, 3.4))))
, ve(2, myenum::blue)
{
e=myenum::red;
x=i;
y=2*i;
b=true;
name="bother said pooh";
}
};
int16_t i16 = 1;
uint16_t u16 = 2;
int32_t i32 = 3;
int32_t i32 = 3;
uint32_t u32 = 4;
int64_t i64 = 5;
int64_t i64 = 5;
uint64_t u64 = 6;
float f = M_PI;
double d = 2*M_PI;
bool b = false;
float f = M_PI;
double d = 2*M_PI;
bool b = false;
template <typename W, typename R, typename O>
void ioTest(const std::string &filename, const O &object, const std::string &name)
{
// writer needs to be destroyed so that writing physically happens
{
W writer(filename);
write(writer, "testobject", object);
}
R reader(filename);
O buf;
bool good;
read(reader, "testobject", buf);
good = (object == buf);
std::cout << name << " IO test: " << (good ? "success" : "failure");
std::cout << std::endl;
if (!good) exit(EXIT_FAILURE);
}
int main(int argc,char **argv)
{
{
XmlWriter WR("bother.xml");
// test basic type writing
push(WR,"BasicTypes");
write(WR,std::string("i16"),i16);
write(WR,"u16",u16);
write(WR,"i32",i32);
write(WR,"u32",u32);
write(WR,"i64",i64);
write(WR,"u64",u64);
write(WR,"f",f);
write(WR,"d",d);
write(WR,"b",b);
pop(WR);
// test serializable class writing
myclass obj(1234); // non-trivial constructor
write(WR,"obj",obj);
WR.write("obj2", obj);
std::cout << obj << std::endl;
std::vector<myclass> vec;
vec.push_back(myclass(1234));
vec.push_back(myclass(5678));
vec.push_back(myclass(3838));
write(WR, "objvec", vec);
};
std::cout << "==== basic IO" << std::endl;
XmlWriter WR("bother.xml");
// test basic type writing
std::cout << "-- basic writing to 'bother.xml'..." << std::endl;
push(WR,"BasicTypes");
write(WR,std::string("i16"),i16);
write(WR,"u16",u16);
write(WR,"i32",i32);
write(WR,"u32",u32);
write(WR,"i64",i64);
write(WR,"u64",u64);
write(WR,"f",f);
write(WR,"d",d);
write(WR,"b",b);
pop(WR);
// test serializable class writing
myclass obj(1234); // non-trivial constructor
std::vector<myclass> vec;
std::cout << "-- serialisable class writing to 'bother.xml'..." << std::endl;
write(WR,"obj",obj);
WR.write("obj2", obj);
vec.push_back(myclass(1234));
vec.push_back(myclass(5678));
vec.push_back(myclass(3838));
write(WR, "objvec", vec);
std::cout << "-- serialisable class writing to std::cout:" << std::endl;
std::cout << obj << std::endl;
std::cout << "-- serialisable class comparison:" << std::endl;
std::cout << "vec[0] == obj: " << ((vec[0] == obj) ? "true" : "false") << std::endl;
std::cout << "vec[1] == obj: " << ((vec[1] == obj) ? "true" : "false") << std::endl;
// read tests
myclass copy1, copy2, copy3;
std::vector<myclass> veccopy1, veccopy2, veccopy3;
std::cout << "\n==== IO self-consistency tests" << std::endl;
//// XML
{
XmlReader RD("bother.xml");
read(RD,"obj",copy1);
read(RD,"objvec", veccopy1);
std::cout << "Loaded (XML) -----------------" << std::endl;
std::cout << copy1 << std::endl << veccopy1 << std::endl;
}
ioTest<XmlWriter, XmlReader>("iotest.xml", obj, "XML (object) ");
ioTest<XmlWriter, XmlReader>("iotest.xml", vec, "XML (vector of objects)");
//// binary
{
BinaryWriter BWR("bother.bin");
write(BWR,"discard",copy1 );
write(BWR,"discard",veccopy1 );
}
{
BinaryReader BRD("bother.bin");
read (BRD,"discard",copy2 );
read (BRD,"discard",veccopy2 );
std::cout << "Loaded (bin) -----------------" << std::endl;
std::cout << copy2 << std::endl << veccopy2 << std::endl;
}
ioTest<BinaryWriter, BinaryReader>("iotest.bin", obj, "binary (object) ");
ioTest<BinaryWriter, BinaryReader>("iotest.bin", vec, "binary (vector of objects)");
//// text
{
TextWriter TWR("bother.txt");
write(TWR,"discard",copy1 );
write(TWR,"discard",veccopy1 );
}
{
TextReader TRD("bother.txt");
read (TRD,"discard",copy3 );
read (TRD,"discard",veccopy3 );
std::cout << "Loaded (txt) -----------------" << std::endl;
std::cout << copy3 << std::endl << veccopy3 << std::endl;
}
ioTest<TextWriter, TextReader>("iotest.dat", obj, "text (object) ");
ioTest<TextWriter, TextReader>("iotest.dat", vec, "text (vector of objects)");
//// HDF5
#ifdef HAVE_HDF5
ioTest<Hdf5Writer, Hdf5Reader>("iotest.h5", obj, "HDF5 (object) ");
ioTest<Hdf5Writer, Hdf5Reader>("iotest.h5", vec, "HDF5 (vector of objects)");
#endif
std::vector<int> iv = strToVec<int>("1 2 2 4");
std::vector<std::string> sv = strToVec<std::string>("bli bla blu");
std::cout << "\n==== vector flattening/reconstruction" << std::endl;
typedef std::vector<std::vector<std::vector<double>>> vec3d;
for (auto &e: iv)
vec3d dv, buf;
double d = 0.;
dv.resize(4);
for (auto &v1: dv)
{
std::cout << e << " ";
}
std::cout << std::endl;
for (auto &e: sv)
{
std::cout << e << " ";
v1.resize(3);
for (auto &v2: v1)
{
v2.resize(5);
for (auto &x: v2)
{
x = d++;
}
}
}
std::cout << "original 3D vector:" << std::endl;
std::cout << dv << std::endl;
Flatten<vec3d> flatdv(dv);
std::cout << "\ndimensions:" << std::endl;
std::cout << flatdv.getDim() << std::endl;
std::cout << "\nflattened vector:" << std::endl;
std::cout << flatdv.getFlatVector() << std::endl;
Reconstruct<vec3d> rec(flatdv.getFlatVector(), flatdv.getDim());
std::cout << "\nreconstructed vector:" << std::endl;
std::cout << flatdv.getVector() << std::endl;
std::cout << std::endl;
@ -211,4 +234,6 @@ int main(int argc,char **argv)
read(RD,"name", name);
}
}

0
tests/debug/Test_cayley_even_odd_vec.cc → tests/Test_cayley_even_odd_vec.cc
View File

3
tests/hadrons/Makefile.am Normal file
View File

@ -0,0 +1,3 @@
AM_LDFLAGS += -L../../extras/Hadrons
include Make.inc

170
tests/hadrons/Test_hadrons_meson_3pt.cc Normal file
View File

@ -0,0 +1,170 @@
/*******************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: tests/hadrons/Test_hadrons_meson_3pt.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.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
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.
*******************************************************************************/
#include <Grid/Hadrons/Application.hpp>
using namespace Grid;
using namespace Hadrons;
int main(int argc, char *argv[])
{
// initialization //////////////////////////////////////////////////////////
Grid_init(&argc, &argv);
HadronsLogError.Active(GridLogError.isActive());
HadronsLogWarning.Active(GridLogWarning.isActive());
HadronsLogMessage.Active(GridLogMessage.isActive());
HadronsLogIterative.Active(GridLogIterative.isActive());
HadronsLogDebug.Active(GridLogDebug.isActive());
LOG(Message) << "Grid initialized" << std::endl;
// run setup ///////////////////////////////////////////////////////////////
Application application;
std::vector<std::string> flavour = {"l", "s", "c1", "c2", "c3"};
std::vector<double> mass = {.01, .04, .2 , .25 , .3 };
unsigned int nt = GridDefaultLatt()[Tp];
// global parameters
Application::GlobalPar globalPar;
globalPar.trajCounter.start = 1500;
globalPar.trajCounter.end = 1520;
globalPar.trajCounter.step = 20;
globalPar.seed = "1 2 3 4";
globalPar.genetic.maxGen = 1000;
globalPar.genetic.maxCstGen = 200;
globalPar.genetic.popSize = 20;
globalPar.genetic.mutationRate = .1;
application.setPar(globalPar);
// gauge field
application.createModule<MGauge::Unit>("gauge");
for (unsigned int i = 0; i < flavour.size(); ++i)
{
// actions
MAction::DWF::Par actionPar;
actionPar.gauge = "gauge";
actionPar.Ls = 12;
actionPar.M5 = 1.8;
actionPar.mass = mass[i];
application.createModule<MAction::DWF>("DWF_" + flavour[i], actionPar);
// solvers
MSolver::RBPrecCG::Par solverPar;
solverPar.action = "DWF_" + flavour[i];
solverPar.residual = 1.0e-8;
application.createModule<MSolver::RBPrecCG>("CG_" + flavour[i],
solverPar);
}
for (unsigned int t = 0; t < nt; t += 1)
{
std::string srcName;
std::vector<std::string> qName;
std::vector<std::vector<std::string>> seqName;
// Z2 source
MSource::Z2::Par z2Par;
z2Par.tA = t;
z2Par.tB = t;
srcName = "z2_" + std::to_string(t);
application.createModule<MSource::Z2>(srcName, z2Par);
for (unsigned int i = 0; i < flavour.size(); ++i)
{
// sequential sources
MSource::SeqGamma::Par seqPar;
qName.push_back("QZ2_" + flavour[i] + "_" + std::to_string(t));
seqPar.q = qName[i];
seqPar.tA = (t + nt/4) % nt;
seqPar.tB = (t + nt/4) % nt;
seqPar.mom = "1. 0. 0. 0.";
seqName.push_back(std::vector<std::string>(Nd));
for (unsigned int mu = 0; mu < Nd; ++mu)
{
seqPar.gamma = 0x1 << mu;
seqName[i][mu] = "G" + std::to_string(seqPar.gamma)
+ "_" + std::to_string(seqPar.tA) + "-"
+ qName[i];
application.createModule<MSource::SeqGamma>(seqName[i][mu], seqPar);
}
// propagators
Quark::Par quarkPar;
quarkPar.solver = "CG_" + flavour[i];
quarkPar.source = srcName;
application.createModule<Quark>(qName[i], quarkPar);
for (unsigned int mu = 0; mu < Nd; ++mu)
{
quarkPar.source = seqName[i][mu];
seqName[i][mu] = "Q_" + flavour[i] + "-" + seqName[i][mu];
application.createModule<Quark>(seqName[i][mu], quarkPar);
}
}
// contractions
MContraction::Meson::Par mesPar;
for (unsigned int i = 0; i < flavour.size(); ++i)
for (unsigned int j = i; j < flavour.size(); ++j)
{
mesPar.output = "mesons/Z2_" + flavour[i] + flavour[j];
mesPar.q1 = qName[i];
mesPar.q2 = qName[j];
application.createModule<MContraction::Meson>("meson_Z2_"
+ std::to_string(t)
+ "_"
+ flavour[i]
+ flavour[j],
mesPar);
}
for (unsigned int i = 0; i < flavour.size(); ++i)
for (unsigned int j = 0; j < flavour.size(); ++j)
for (unsigned int mu = 0; mu < Nd; ++mu)
{
MContraction::Meson::Par mesPar;
mesPar.output = "3pt/Z2_" + flavour[i] + flavour[j] + "_"
+ std::to_string(mu);
mesPar.q1 = qName[i];
mesPar.q2 = seqName[j][mu];
application.createModule<MContraction::Meson>("3pt_Z2_"
+ std::to_string(t)
+ "_"
+ flavour[i]
+ flavour[j]
+ "_"
+ std::to_string(mu),
mesPar);
}
}
// execution
application.saveParameterFile("meson3pt.xml");
application.run();
// epilogue
LOG(Message) << "Grid is finalizing now" << std::endl;
Grid_finalize();
return EXIT_SUCCESS;
}

132
tests/hadrons/Test_hadrons_spectrum.cc Normal file
View File

@ -0,0 +1,132 @@
/*******************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: tests/hadrons/Test_hadrons_spectrum.cc
Copyright (C) 2015
Author: Antonin Portelli <antonin.portelli@me.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
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.
*******************************************************************************/
#include <Grid/Hadrons/Application.hpp>
using namespace Grid;
using namespace Hadrons;
int main(int argc, char *argv[])
{
// initialization //////////////////////////////////////////////////////////
Grid_init(&argc, &argv);
HadronsLogError.Active(GridLogError.isActive());
HadronsLogWarning.Active(GridLogWarning.isActive());
HadronsLogMessage.Active(GridLogMessage.isActive());
HadronsLogIterative.Active(GridLogIterative.isActive());
HadronsLogDebug.Active(GridLogDebug.isActive());
LOG(Message) << "Grid initialized" << std::endl;
// run setup ///////////////////////////////////////////////////////////////
Application application;
std::vector<std::string> flavour = {"l", "s", "c1", "c2", "c3"};
std::vector<double> mass = {.01, .04, .2 , .25 , .3 };
// global parameters
Application::GlobalPar globalPar;
globalPar.trajCounter.start = 1500;
globalPar.trajCounter.end = 1520;
globalPar.trajCounter.step = 20;
globalPar.seed = "1 2 3 4";
application.setPar(globalPar);
// gauge field
application.createModule<MGauge::Unit>("gauge");
// sources
MSource::Z2::Par z2Par;
z2Par.tA = 0;
z2Par.tB = 0;
application.createModule<MSource::Z2>("z2", z2Par);
MSource::Point::Par ptPar;
ptPar.position = "0 0 0 0";
application.createModule<MSource::Point>("pt", ptPar);
for (unsigned int i = 0; i < flavour.size(); ++i)
{
// actions
MAction::DWF::Par actionPar;
actionPar.gauge = "gauge";
actionPar.Ls = 12;
actionPar.M5 = 1.8;
actionPar.mass = mass[i];
application.createModule<MAction::DWF>("DWF_" + flavour[i], actionPar);
// solvers
MSolver::RBPrecCG::Par solverPar;
solverPar.action = "DWF_" + flavour[i];
solverPar.residual = 1.0e-8;
application.createModule<MSolver::RBPrecCG>("CG_" + flavour[i],
solverPar);
// propagators
Quark::Par quarkPar;
quarkPar.solver = "CG_" + flavour[i];
quarkPar.source = "pt";
application.createModule<Quark>("Qpt_" + flavour[i], quarkPar);
quarkPar.source = "z2";
application.createModule<Quark>("QZ2_" + flavour[i], quarkPar);
}
for (unsigned int i = 0; i < flavour.size(); ++i)
for (unsigned int j = i; j < flavour.size(); ++j)
{
MContraction::Meson::Par mesPar;
mesPar.output = "mesons/pt_" + flavour[i] + flavour[j];
mesPar.q1 = "Qpt_" + flavour[i];
mesPar.q2 = "Qpt_" + flavour[j];
application.createModule<MContraction::Meson>("meson_pt_"
+ flavour[i] + flavour[j],
mesPar);
mesPar.output = "mesons/Z2_" + flavour[i] + flavour[j];
mesPar.q1 = "QZ2_" + flavour[i];
mesPar.q2 = "QZ2_" + flavour[j];
application.createModule<MContraction::Meson>("meson_Z2_"
+ flavour[i] + flavour[j],
mesPar);
}
for (unsigned int i = 0; i < flavour.size(); ++i)
for (unsigned int j = i; j < flavour.size(); ++j)
for (unsigned int k = j; k < flavour.size(); ++k)
{
MContraction::Baryon::Par barPar;
barPar.output = "baryons/pt_" + flavour[i] + flavour[j] + flavour[k];
barPar.q1 = "Qpt_" + flavour[i];
barPar.q2 = "Qpt_" + flavour[j];
barPar.q3 = "Qpt_" + flavour[k];
application.createModule<MContraction::Baryon>(
"baryon_pt_" + flavour[i] + flavour[j] + flavour[k], barPar);
}
// execution
application.saveParameterFile("spectrum.xml");
application.run();
// epilogue
LOG(Message) << "Grid is finalizing now" << std::endl;
Grid_finalize();
return EXIT_SUCCESS;
}