portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 04:37:05 +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);