portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-14 22:07:05 +01:00
Code Releases Activity

Merge branch 'develop' into feature/hadrons

Browse Source
This commit is contained in:
Antonin Portelli
2017-08-24 17:05:45 +01:00
parent 2bcb704af2 5fa386ddc9
commit 21b02760c3
42 changed files with 1906 additions and 512 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
lib/qcd/action/fermion/ImprovedStaggeredFermion5D.cc
View File

@ -230,8 +230,15 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOr
{
Compressor compressor;
int LLs = in._grid->_rdimensions[0];
DhopTotalTime -= usecond();
DhopCommTime -= usecond();
st.HaloExchange(in,compressor);
DhopCommTime += usecond();
DhopComputeTime -= usecond();
// Dhop takes the 4d grid from U, and makes a 5d index for fermion
if (dag == DaggerYes) {
parallel_for (int ss = 0; ss < U._grid->oSites(); ss++) {
@ -244,12 +251,15 @@ void ImprovedStaggeredFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOr
Kernels::DhopSite(st,lo,U,UUU,st.CommBuf(),LLs,sU,in,out);
}
}
DhopComputeTime += usecond();
DhopTotalTime += usecond();
}
template<class Impl>
void ImprovedStaggeredFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int dag)
{
DhopCalls+=1;
conformable(in._grid,FermionRedBlackGrid()); // verifies half grid
conformable(in._grid,out._grid); // drops the cb check
@ -261,6 +271,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopOE(const FermionField &in, FermionFie
template<class Impl>
void ImprovedStaggeredFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
{
DhopCalls+=1;
conformable(in._grid,FermionRedBlackGrid()); // verifies half grid
conformable(in._grid,out._grid); // drops the cb check
@ -272,6 +283,7 @@ void ImprovedStaggeredFermion5D<Impl>::DhopEO(const FermionField &in, FermionFie
template<class Impl>
void ImprovedStaggeredFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
{
DhopCalls+=2;
conformable(in._grid,FermionGrid()); // verifies full grid
conformable(in._grid,out._grid);
@ -280,6 +292,54 @@ void ImprovedStaggeredFermion5D<Impl>::Dhop(const FermionField &in, FermionField
DhopInternal(Stencil,Lebesgue,Umu,UUUmu,in,out,dag);
}
template<class Impl>
void ImprovedStaggeredFermion5D<Impl>::Report(void)
{
std::vector<int> latt = GridDefaultLatt();
RealD volume = Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
RealD NP = _FourDimGrid->_Nprocessors;
RealD NN = _FourDimGrid->NodeCount();
std::cout << GridLogMessage << "#### Dhop calls report " << std::endl;
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D Number of DhopEO Calls : "
<< DhopCalls << std::endl;
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D TotalTime /Calls : "
<< DhopTotalTime / DhopCalls << " us" << std::endl;
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D CommTime /Calls : "
<< DhopCommTime / DhopCalls << " us" << std::endl;
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D ComputeTime/Calls : "
<< DhopComputeTime / DhopCalls << " us" << std::endl;
// Average the compute time
_FourDimGrid->GlobalSum(DhopComputeTime);
DhopComputeTime/=NP;
RealD mflops = 1154*volume*DhopCalls/DhopComputeTime/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;
std::cout << GridLogMessage << "Average mflops/s per call per node : " << mflops/NN << std::endl;
RealD Fullmflops = 1154*volume*DhopCalls/(DhopTotalTime)/2; // 2 for red black counting
std::cout << GridLogMessage << "Average mflops/s per call (full) : " << Fullmflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per rank (full): " << Fullmflops/NP << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per node (full): " << Fullmflops/NN << std::endl;
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D Stencil" <<std::endl; Stencil.Report();
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilEven"<<std::endl; StencilEven.Report();
std::cout << GridLogMessage << "ImprovedStaggeredFermion5D StencilOdd" <<std::endl; StencilOdd.Report();
}
template<class Impl>
void ImprovedStaggeredFermion5D<Impl>::ZeroCounters(void)
{
DhopCalls = 0;
DhopTotalTime = 0;
DhopCommTime = 0;
DhopComputeTime = 0;
Stencil.ZeroCounters();
StencilEven.ZeroCounters();
StencilOdd.ZeroCounters();
}
/////////////////////////////////////////////////////////////////////////
// Implement the general interface. Here we use SAME mass on all slices

10
lib/qcd/action/fermion/ImprovedStaggeredFermion5D.h
View File

@ -55,6 +55,16 @@ namespace QCD {
FermionField _tmp;
FermionField &tmp(void) { return _tmp; }
////////////////////////////////////////
// Performance monitoring
////////////////////////////////////////
void Report(void);
void ZeroCounters(void);
double DhopTotalTime;
double DhopCalls;
double DhopCommTime;
double DhopComputeTime;
///////////////////////////////////////////////////////////////
// Implement the abstract base
///////////////////////////////////////////////////////////////

8
lib/qcd/action/scalar/ScalarImpl.h
View File

@ -93,6 +93,8 @@ class ScalarImplTypes {
class ScalarAdjMatrixImplTypes {
public:
typedef S Simd;
typedef QCD::SU<N> Group;
template <typename vtype>
using iImplField = iScalar<iScalar<iMatrix<vtype, N>>>;
template <typename vtype>
@ -108,7 +110,7 @@ class ScalarImplTypes {
typedef Field PropagatorField;
static inline void generate_momenta(Field& P, GridParallelRNG& pRNG) {
QCD::SU<N>::GaussianFundamentalLieAlgebraMatrix(pRNG, P);
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, P);
}
static inline Field projectForce(Field& P) {return P;}
@ -122,11 +124,11 @@ class ScalarImplTypes {
}
static inline void HotConfiguration(GridParallelRNG &pRNG, Field &U) {
QCD::SU<N>::LieRandomize(pRNG, U);
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, U);
}
static inline void TepidConfiguration(GridParallelRNG &pRNG, Field &U) {
QCD::SU<N>::LieRandomize(pRNG, U, 0.01);
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, U, 0.01);
}
static inline void ColdConfiguration(GridParallelRNG &pRNG, Field &U) {

6
lib/qcd/action/scalar/ScalarInteractionAction.h
View File

@ -81,7 +81,7 @@ namespace Grid {
phiStencil.HaloExchange(p, compressor);
Field action(p._grid), pshift(p._grid), phisquared(p._grid);
phisquared = p*p;
action = (2.0*Ndim + mass_square)*phisquared + lambda*phisquared*phisquared;
action = (2.0*Ndim + mass_square)*phisquared - lambda/24.*phisquared*phisquared;
for (int mu = 0; mu < Ndim; mu++) {
// pshift = Cshift(p, mu, +1); // not efficient, implement with stencils
parallel_for (int i = 0; i < p._grid->oSites(); i++) {
@ -98,7 +98,7 @@ namespace Grid {
permute(temp2, *temp, permute_type);
action._odata[i] -= temp2*(*t_p) + (*t_p)*temp2;
} else {
action._odata[i] -= *temp*(*t_p) + (*t_p)*(*temp);
action._odata[i] -= (*temp)*(*t_p) + (*t_p)*(*temp);
}
} else {
action._odata[i] -= phiStencil.CommBuf()[SE->_offset]*(*t_p) + (*t_p)*phiStencil.CommBuf()[SE->_offset];
@ -113,7 +113,7 @@ namespace Grid {
virtual void deriv(const Field &p, Field &force) {
assert(p._grid->Nd() == Ndim);
force = (2.0*Ndim + mass_square)*p + 2.0*lambda*p*p*p;
force = (2.0*Ndim + mass_square)*p - lambda/12.*p*p*p;
// move this outside
static Stencil phiStencil(p._grid, npoint, 0, directions, displacements);
phiStencil.HaloExchange(p, compressor);

2
lib/qcd/hmc/HMC.h
View File

@ -76,7 +76,7 @@ struct HMCparameters: Serializable {
template < class ReaderClass >
void initialize(Reader<ReaderClass> &TheReader){
std::cout << "Reading HMC\n";
std::cout << GridLogMessage << "Reading HMC\n";
read(TheReader, "HMC", *this);
}

17
lib/qcd/hmc/HMCResourceManager.h
View File

@ -165,7 +165,7 @@ class HMCResourceManager {
// Grids
//////////////////////////////////////////////////////////////
void AddGrid(std::string s, GridModule& M) {
void AddGrid(const std::string s, GridModule& M) {
// Check for name clashes
auto search = Grids.find(s);
if (search != Grids.end()) {
@ -174,14 +174,24 @@ class HMCResourceManager {
exit(1);
}
Grids[s] = std::move(M);
std::cout << GridLogMessage << "::::::::::::::::::::::::::::::::::::::::" <<std::endl;
std::cout << GridLogMessage << "HMCResourceManager:" << std::endl;
std::cout << GridLogMessage << "Created grid set with name '" << s << "' and decomposition for the full cartesian " << std::endl;
Grids[s].show_full_decomposition();
std::cout << GridLogMessage << "::::::::::::::::::::::::::::::::::::::::" <<std::endl;
}
// Add a named grid set, 4d shortcut
void AddFourDimGrid(std::string s) {
void AddFourDimGrid(const std::string s) {
GridFourDimModule<vComplex> Mod;
AddGrid(s, Mod);
}
// Add a named grid set, 4d shortcut + tweak simd lanes
void AddFourDimGrid(const std::string s, const std::vector<int> simd_decomposition) {
GridFourDimModule<vComplex> Mod(simd_decomposition);
AddGrid(s, Mod);
}
GridCartesian* GetCartesian(std::string s = "") {
@ -253,6 +263,7 @@ class HMCResourceManager {
template<class T, class... Types>
void AddObservable(Types&&... Args){
ObservablesList.push_back(std::unique_ptr<T>(new T(std::forward<Types>(Args)...)));
ObservablesList.back()->print_parameters();
}
std::vector<HmcObservable<typename ImplementationPolicy::Field>* > GetObservables(){
@ -297,4 +308,4 @@ private:
}
}
#endif // HMC_RESOURCE_MANAGER_H
#endif // HMC_RESOURCE_MANAGER_H

92
lib/qcd/hmc/HMC_GridModules.h
View File

@ -33,28 +33,29 @@ directory
namespace Grid {
// Resources
// Modules for grids
// Modules for grids
// Introduce another namespace HMCModules?
class GridModuleParameters: Serializable{
class GridModuleParameters: Serializable{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(GridModuleParameters,
std::string, lattice,
std::string, mpi);
std::vector<int> getLattice(){return strToVec<int>(lattice);}
std::vector<int> getMpi() {return strToVec<int>(mpi);}
std::vector<int> getLattice() const {return strToVec<int>(lattice);}
std::vector<int> getMpi() const {return strToVec<int>(mpi);}
void check(){
if (getLattice().size() != getMpi().size()) {
std::cout << GridLogError
void check() const {
if (getLattice().size() != getMpi().size() ) {
std::cout << GridLogError
<< "Error in GridModuleParameters: lattice and mpi dimensions "
"do not match"
<< std::endl;
exit(1);
}
}
}
template <class ReaderClass>
GridModuleParameters(Reader<ReaderClass>& Reader, std::string n = "LatticeGrid"):name(n) {
@ -75,51 +76,94 @@ private:
// Lower level class
class GridModule {
public:
GridCartesian* get_full() {
GridCartesian* get_full() {
std::cout << GridLogDebug << "Getting cartesian in module"<< std::endl;
return grid_.get(); }
GridRedBlackCartesian* get_rb() {
GridRedBlackCartesian* get_rb() {
std::cout << GridLogDebug << "Getting rb-cartesian in module"<< std::endl;
return rbgrid_.get(); }
void set_full(GridCartesian* grid) { grid_.reset(grid); }
void set_rb(GridRedBlackCartesian* rbgrid) { rbgrid_.reset(rbgrid); }
void show_full_decomposition(){ grid_->show_decomposition(); }
void show_rb_decomposition(){ rbgrid_->show_decomposition(); }
protected:
std::unique_ptr<GridCartesian> grid_;
std::unique_ptr<GridRedBlackCartesian> rbgrid_;
};
////////////////////////////////////
// Classes for the user
////////////////////////////////////
// Note: the space time grid should be out of the QCD namespace
template< class vector_type>
class GridFourDimModule : public GridModule {
public:
GridFourDimModule() {
template <class vector_type>
class GridFourDimModule : public GridModule
{
public:
GridFourDimModule()
{
using namespace QCD;
set_full(SpaceTimeGrid::makeFourDimGrid(
GridDefaultLatt(), GridDefaultSimd(4, vector_type::Nsimd()),
GridDefaultLatt(),
GridDefaultSimd(4, vector_type::Nsimd()),
GridDefaultMpi()));
set_rb(SpaceTimeGrid::makeFourDimRedBlackGrid(grid_.get()));
}
GridFourDimModule(GridModuleParameters Params) {
GridFourDimModule(const std::vector<int> tweak_simd)
{
using namespace QCD;
if (tweak_simd.size() != 4)
{
std::cout << GridLogError
<< "Error in GridFourDimModule: SIMD size different from 4"
<< std::endl;
exit(1);
}
// Checks that the product agrees with the expectation
int simd_sum = 1;
for (auto &n : tweak_simd)
simd_sum *= n;
std::cout << GridLogDebug << "TweakSIMD: " << tweak_simd << " Sum: " << simd_sum << std::endl;
if (simd_sum == vector_type::Nsimd())
{
set_full(SpaceTimeGrid::makeFourDimGrid(
GridDefaultLatt(),
tweak_simd,
GridDefaultMpi()));
set_rb(SpaceTimeGrid::makeFourDimRedBlackGrid(grid_.get()));
}
else
{
std::cout << GridLogError
<< "Error in GridFourDimModule: SIMD lanes must sum to "
<< vector_type::Nsimd()
<< std::endl;
}
}
GridFourDimModule(const GridModuleParameters Params)
{
using namespace QCD;
Params.check();
std::vector<int> lattice_v = Params.getLattice();
std::vector<int> mpi_v = Params.getMpi();
if (lattice_v.size() == 4) {
if (lattice_v.size() == 4)
{
set_full(SpaceTimeGrid::makeFourDimGrid(
lattice_v, GridDefaultSimd(4, vector_type::Nsimd()),
lattice_v,
GridDefaultSimd(4, vector_type::Nsimd()),
mpi_v));
set_rb(SpaceTimeGrid::makeFourDimRedBlackGrid(grid_.get()));
} else {
std::cout << GridLogError
<< "Error in GridFourDimModule: lattice dimension different from 4"
<< std::endl;
}
else
{
std::cout << GridLogError
<< "Error in GridFourDimModule: lattice dimension different from 4"
<< std::endl;
exit(1);
}
}

15
lib/qcd/modules/ObservableModules.h
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

69
lib/qcd/observables/topological_charge.h
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);
}
}
};

9
lib/qcd/smearing/WilsonFlow.h
View File

@ -108,7 +108,7 @@ void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, Real
if (maxTau - taus < epsilon){
epsilon = maxTau-taus;
}
std::cout << GridLogMessage << "Integration epsilon : " << epsilon << std::endl;
//std::cout << GridLogMessage << "Integration epsilon : " << epsilon << std::endl;
GaugeField Z(U._grid);
GaugeField Zprime(U._grid);
GaugeField tmp(U._grid), Uprime(U._grid);
@ -138,10 +138,10 @@ void WilsonFlow<Gimpl>::evolve_step_adaptive(typename Gimpl::GaugeField &U, Real
// adjust integration step
taus += epsilon;
std::cout << GridLogMessage << "Adjusting integration step with distance: " << diff << std::endl;
//std::cout << GridLogMessage << "Adjusting integration step with distance: " << diff << std::endl;
epsilon = epsilon*0.95*std::pow(1e-4/diff,1./3.);
std::cout << GridLogMessage << "New epsilon : " << epsilon << std::endl;
//std::cout << GridLogMessage << "New epsilon : " << epsilon << std::endl;
}
@ -166,7 +166,6 @@ void WilsonFlow<Gimpl>::smear(GaugeField& out, const GaugeField& in) const {
out = in;
for (unsigned int step = 1; step <= Nstep; step++) {
auto start = std::chrono::high_resolution_clock::now();
std::cout << GridLogMessage << "Evolution time :"<< tau(step) << std::endl;
evolve_step(out);
auto end = std::chrono::high_resolution_clock::now();
std::chrono::duration<double> diff = end - start;
@ -191,7 +190,7 @@ void WilsonFlow<Gimpl>::smear_adaptive(GaugeField& out, const GaugeField& in, Re
unsigned int step = 0;
do{
step++;
std::cout << GridLogMessage << "Evolution time :"<< taus << std::endl;
//std::cout << GridLogMessage << "Evolution time :"<< taus << std::endl;
evolve_step_adaptive(out, maxTau);
std::cout << GridLogMessage << "[WilsonFlow] Energy density (plaq) : "
<< step << " "

11
lib/qcd/utils/GaugeFix.h
View File

@ -26,12 +26,14 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
/* END LEGAL */
//#include <Grid/Grid.h>
using namespace Grid;
using namespace Grid::QCD;
#ifndef GRID_QCD_GAUGE_FIX_H
#define GRID_QCD_GAUGE_FIX_H
namespace Grid {
namespace QCD {
template <class Gimpl>
class FourierAcceleratedGaugeFixer : public Gimpl {
public:
public:
INHERIT_GIMPL_TYPES(Gimpl);
typedef typename Gimpl::GaugeLinkField GaugeMat;
@ -186,3 +188,6 @@ class FourierAcceleratedGaugeFixer : public Gimpl {
}
};
}
}
#endif

3
lib/qcd/utils/SUn.h
View File

@ -716,8 +716,7 @@ template<typename GaugeField,typename GaugeMat>
for (int a = 0; a < AdjointDimension; a++) {
generator(a, Ta);
auto tmp = - 2.0 * (trace(timesI(Ta) * in)) * scale;// 2.0 for the normalization of the trace in the fundamental rep
pokeColour(h_out, tmp, a);
pokeColour(h_out, - 2.0 * (trace(timesI(Ta) * in)) * scale, a);
}
}