1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00

Merge branch 'develop' into feature/CG-reliable-update

This commit is contained in:
Christopher Kelly 2017-08-02 09:48:04 -04:00
commit 7d867a8134
6 changed files with 173 additions and 66 deletions

View File

@ -13,6 +13,10 @@ m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
################ Get git info
#AC_REVISION([m4_esyscmd_s([./scripts/configure.commit])])
################ Set flags
# do not move!
CXXFLAGS="-O3 $CXXFLAGS"
############### Checks for programs
AC_PROG_CXX
AC_PROG_RANLIB
@ -27,7 +31,6 @@ AX_GXX_VERSION
AC_DEFINE_UNQUOTED([GXX_VERSION],["$GXX_VERSION"],
[version of g++ that will compile the code])
CXXFLAGS="-O3 $CXXFLAGS"
############### Checks for typedefs, structures, and compiler characteristics
@ -55,6 +58,10 @@ AC_CHECK_HEADERS(numaif.h)
AC_CHECK_DECLS([ntohll],[], [], [[#include <arpa/inet.h>]])
AC_CHECK_DECLS([be64toh],[], [], [[#include <arpa/inet.h>]])
############## Standard libraries
AC_CHECK_LIB([m],[cos])
AC_CHECK_LIB([stdc++],[abort])
############### GMP and MPFR
AC_ARG_WITH([gmp],
[AS_HELP_STRING([--with-gmp=prefix],

View File

@ -98,35 +98,39 @@ class BinaryIO {
NerscChecksum(grid,scalardata,nersc_csum);
}
template<class fobj> static inline void NerscChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &nersc_csum)
template <class fobj>
static inline void NerscChecksum(GridBase *grid, std::vector<fobj> &fbuf, uint32_t &nersc_csum)
{
const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t);
const uint64_t size32 = sizeof(fobj) / sizeof(uint32_t);
uint64_t lsites =grid->lSites();
if (fbuf.size()==1) {
lsites=1;
uint64_t lsites = grid->lSites();
if (fbuf.size() == 1)
{
lsites = 1;
}
#pragma omp parallel
{
uint32_t nersc_csum_thr=0;
#pragma omp parallel
{
uint32_t nersc_csum_thr = 0;
#pragma omp for
for(uint64_t local_site=0;local_site<lsites;local_site++){
uint32_t * site_buf = (uint32_t *)&fbuf[local_site];
for(uint64_t j=0;j<size32;j++){
nersc_csum_thr=nersc_csum_thr+site_buf[j];
}
#pragma omp for
for (uint64_t local_site = 0; local_site < lsites; local_site++)
{
uint32_t *site_buf = (uint32_t *)&fbuf[local_site];
for (uint64_t j = 0; j < size32; j++)
{
nersc_csum_thr = nersc_csum_thr + site_buf[j];
}
}
#pragma omp critical
#pragma omp critical
{
nersc_csum += nersc_csum_thr;
nersc_csum += nersc_csum_thr;
}
}
}
template<class fobj> static inline void ScidacChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &scidac_csuma,uint32_t &scidac_csumb)
{
const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t);
@ -266,7 +270,7 @@ class BinaryIO {
grid->Barrier();
GridStopWatch timer;
GridStopWatch bstimer;
nersc_csum=0;
scidac_csuma=0;
scidac_csumb=0;
@ -362,18 +366,22 @@ class BinaryIO {
#else
assert(0);
#endif
} else {
std::cout<< GridLogMessage<< "C++ read I/O "<< file<<" : "
<< iodata.size()*sizeof(fobj)<<" bytes"<<std::endl;
std::ifstream fin;
fin.open(file,std::ios::binary|std::ios::in);
if ( control & BINARYIO_MASTER_APPEND ) {
fin.seekg(-sizeof(fobj),fin.end);
} else {
fin.seekg(offset+myrank*lsites*sizeof(fobj));
}
fin.read((char *)&iodata[0],iodata.size()*sizeof(fobj));assert( fin.fail()==0);
fin.close();
} else {
std::cout << GridLogMessage << "C++ read I/O " << file << " : "
<< iodata.size() * sizeof(fobj) << " bytes" << std::endl;
std::ifstream fin;
fin.open(file, std::ios::binary | std::ios::in);
if (control & BINARYIO_MASTER_APPEND)
{
fin.seekg(-sizeof(fobj), fin.end);
}
else
{
fin.seekg(offset + myrank * lsites * sizeof(fobj));
}
fin.read((char *)&iodata[0], iodata.size() * sizeof(fobj));
assert(fin.fail() == 0);
fin.close();
}
timer.Stop();
@ -416,19 +424,47 @@ class BinaryIO {
assert(0);
#endif
} else {
std::ofstream fout; fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
std::cout<< GridLogMessage<< "C++ write I/O "<< file<<" : "
<< iodata.size()*sizeof(fobj)<<" bytes"<<std::endl;
if ( control & BINARYIO_MASTER_APPEND ) {
std::ofstream fout;
fout.exceptions ( std::fstream::failbit | std::fstream::badbit );
try {
fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
} catch (const std::fstream::failure& exc) {
std::cout << GridLogError << "Error in opening the file " << file << " for output" <<std::endl;
std::cout << GridLogError << "Exception description: " << exc.what() << std::endl;
std::cout << GridLogError << "Probable cause: wrong path, inaccessible location "<< std::endl;
#ifdef USE_MPI_IO
MPI_Abort(MPI_COMM_WORLD,1);
#else
exit(1);
#endif
}
std::cout << GridLogMessage<< "C++ write I/O "<< file<<" : "
<< iodata.size()*sizeof(fobj)<<" bytes"<<std::endl;
if ( control & BINARYIO_MASTER_APPEND ) {
fout.seekp(0,fout.end);
} else {
fout.seekp(offset+myrank*lsites*sizeof(fobj));
}
fout.write((char *)&iodata[0],iodata.size()*sizeof(fobj));assert( fout.fail()==0);
try {
fout.write((char *)&iodata[0],iodata.size()*sizeof(fobj));//assert( fout.fail()==0);
}
catch (const std::fstream::failure& exc) {
std::cout << "Exception in writing file " << file << std::endl;
std::cout << GridLogError << "Exception description: "<< exc.what() << std::endl;
#ifdef USE_MPI_IO
MPI_Abort(MPI_COMM_WORLD,1);
#else
exit(1);
#endif
}
fout.close();
}
timer.Stop();
}
}
timer.Stop();
}
std::cout<<GridLogMessage<<"IOobject: ";
if ( control & BINARYIO_READ) std::cout << " read ";
@ -442,11 +478,14 @@ class BinaryIO {
//////////////////////////////////////////////////////////////////////////////
// Safety check
//////////////////////////////////////////////////////////////////////////////
grid->Barrier();
grid->GlobalSum(nersc_csum);
grid->GlobalXOR(scidac_csuma);
grid->GlobalXOR(scidac_csumb);
grid->Barrier();
// if the data size is 1 we do not want to sum over the MPI ranks
if (iodata.size() != 1){
grid->Barrier();
grid->GlobalSum(nersc_csum);
grid->GlobalXOR(scidac_csuma);
grid->GlobalXOR(scidac_csumb);
grid->Barrier();
}
}
/////////////////////////////////////////////////////////////////////////////
@ -546,9 +585,9 @@ class BinaryIO {
int gsites = grid->gSites();
int lsites = grid->lSites();
uint32_t nersc_csum_tmp;
uint32_t scidac_csuma_tmp;
uint32_t scidac_csumb_tmp;
uint32_t nersc_csum_tmp = 0;
uint32_t scidac_csuma_tmp = 0;
uint32_t scidac_csumb_tmp = 0;
GridStopWatch timer;

View File

@ -84,8 +84,6 @@ class PlaquetteMod: public ObservableModule<PlaquetteLogger<Impl>, NoParameters>
typedef ObservableModule<PlaquetteLogger<Impl>, NoParameters> ObsBase;
using ObsBase::ObsBase; // for constructors
// acquire resource
virtual void initialize(){
this->ObservablePtr.reset(new PlaquetteLogger<Impl>());
@ -94,23 +92,22 @@ class PlaquetteMod: public ObservableModule<PlaquetteLogger<Impl>, NoParameters>
PlaquetteMod(): ObsBase(NoParameters()){}
};
template < class Impl >
class TopologicalChargeMod: public ObservableModule<TopologicalCharge<Impl>, NoParameters>{
typedef ObservableModule<TopologicalCharge<Impl>, NoParameters> ObsBase;
class TopologicalChargeMod: public ObservableModule<TopologicalCharge<Impl>, TopologyObsParameters>{
typedef ObservableModule<TopologicalCharge<Impl>, TopologyObsParameters> ObsBase;
using ObsBase::ObsBase; // for constructors
// acquire resource
virtual void initialize(){
this->ObservablePtr.reset(new TopologicalCharge<Impl>());
this->ObservablePtr.reset(new TopologicalCharge<Impl>(this->Par_));
}
public:
TopologicalChargeMod(): ObsBase(NoParameters()){}
TopologicalChargeMod(TopologyObsParameters Par): ObsBase(Par){}
TopologicalChargeMod(): ObsBase(){}
};
}// QCD temporarily here

View File

@ -33,9 +33,45 @@ directory
namespace Grid {
namespace QCD {
struct TopologySmearingParameters : Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(TopologySmearingParameters,
int, steps,
float, step_size,
int, meas_interval,
float, maxTau);
TopologySmearingParameters(int s = 0, float ss = 0.0f, int mi = 0, float mT = 0.0f):
steps(s), step_size(ss), meas_interval(mi), maxTau(mT){}
template < class ReaderClass >
TopologySmearingParameters(Reader<ReaderClass>& Reader){
read(Reader, "Smearing", *this);
}
};
struct TopologyObsParameters : Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(TopologyObsParameters,
int, interval,
bool, do_smearing,
TopologySmearingParameters, Smearing);
TopologyObsParameters(int interval = 1, bool smearing = false):
interval(interval), Smearing(smearing){}
template <class ReaderClass >
TopologyObsParameters(Reader<ReaderClass>& Reader){
read(Reader, "TopologyMeasurement", *this);
}
};
// this is only defined for a gauge theory
template <class Impl>
class TopologicalCharge : public HmcObservable<typename Impl::Field> {
TopologyObsParameters Pars;
public:
// here forces the Impl to be of gauge fields
// if not the compiler will complain
@ -44,20 +80,39 @@ class TopologicalCharge : public HmcObservable<typename Impl::Field> {
// necessary for HmcObservable compatibility
typedef typename Impl::Field Field;
TopologicalCharge(int interval = 1, bool do_smearing = false):
Pars(interval, do_smearing){}
TopologicalCharge(TopologyObsParameters P):Pars(P){
std::cout << GridLogDebug << "Creating TopologicalCharge " << std::endl;
}
void TrajectoryComplete(int traj,
Field &U,
GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
Real q = WilsonLoops<Impl>::TopologicalCharge(U);
if (traj%Pars.interval == 0){
// Smearing
Field Usmear = U;
int def_prec = std::cout.precision();
if (Pars.do_smearing){
// using wilson flow by default here
WilsonFlow<PeriodicGimplR> WF(Pars.Smearing.steps, Pars.Smearing.step_size, Pars.Smearing.meas_interval);
WF.smear_adaptive(Usmear, U, Pars.Smearing.maxTau);
Real T0 = WF.energyDensityPlaquette(Usmear);
std::cout << GridLogMessage << std::setprecision(std::numeric_limits<Real>::digits10 + 1)
<< "T0 : [ " << traj << " ] "<< T0 << std::endl;
}
int def_prec = std::cout.precision();
Real q = WilsonLoops<Impl>::TopologicalCharge(Usmear);
std::cout << GridLogMessage
<< std::setprecision(std::numeric_limits<Real>::digits10 + 1)
<< "Topological Charge: [ " << traj << " ] "<< q << std::endl;
std::cout << GridLogMessage
<< std::setprecision(std::numeric_limits<Real>::digits10 + 1)
<< "Topological Charge: [ " << traj << " ] "<< q << std::endl;
std::cout.precision(def_prec);
std::cout.precision(def_prec);
}
}
};

View File

@ -98,7 +98,9 @@ template<class rtype,class vtype,class mtype,int N>
strong_inline void mult(iVector<rtype,N> * __restrict__ ret,
const iVector<vtype,N> * __restrict__ rhs,
const iScalar<mtype> * __restrict__ lhs){
mult(ret,lhs,rhs);
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1],&rhs->_internal[c1],&lhs->_internal);
}
}

View File

@ -66,7 +66,14 @@ int main(int argc, char **argv) {
typedef PlaquetteMod<HMCWrapper::ImplPolicy> PlaqObs;
typedef TopologicalChargeMod<HMCWrapper::ImplPolicy> QObs;
TheHMC.Resources.AddObservable<PlaqObs>();
TheHMC.Resources.AddObservable<QObs>();
TopologyObsParameters TopParams;
TopParams.interval = 5;
TopParams.do_smearing = true;
TopParams.Smearing.steps = 200;
TopParams.Smearing.step_size = 0.01;
TopParams.Smearing.meas_interval = 50;
TopParams.Smearing.maxTau = 2.0;
TheHMC.Resources.AddObservable<QObs>(TopParams);
//////////////////////////////////////////////
/////////////////////////////////////////////////////////////