mirror of
https://github.com/paboyle/Grid.git
synced 2025-06-11 03:46:55 +01:00
Hadrons: code cleaning
This commit is contained in:
@ -56,8 +56,8 @@ public:
|
||||
GRID_SERIALIZABLE_CLASS_MEMBERS(A2AMesonFieldPar,
|
||||
int, cacheBlock,
|
||||
int, block,
|
||||
std::string, v,
|
||||
std::string, w,
|
||||
std::string, left,
|
||||
std::string, right,
|
||||
std::string, output,
|
||||
std::string, gammas,
|
||||
std::vector<std::string>, mom);
|
||||
@ -140,11 +140,6 @@ public:
|
||||
virtual void setup(void);
|
||||
// execution
|
||||
virtual void execute(void);
|
||||
private:
|
||||
// IO
|
||||
std::string ioname(const unsigned int m, const unsigned int g) const;
|
||||
std::string filename(const unsigned int m, const unsigned int g) const;
|
||||
void saveBlock(const A2AMatrixSet<MF_IO_TYPE> &mf, IoHelper &h);
|
||||
private:
|
||||
bool hasPhase_{false};
|
||||
std::string momphName_;
|
||||
@ -154,7 +149,6 @@ private:
|
||||
};
|
||||
|
||||
MODULE_REGISTER(A2AMesonField, ARG(TA2AMesonField<FIMPL>), MContraction);
|
||||
MODULE_REGISTER(ZA2AMesonField, ARG(TA2AMesonField<ZFIMPL>), MContraction);
|
||||
|
||||
/******************************************************************************
|
||||
* TA2AMesonField implementation *
|
||||
@ -171,7 +165,7 @@ TA2AMesonField<FImpl>::TA2AMesonField(const std::string name)
|
||||
template <typename FImpl>
|
||||
std::vector<std::string> TA2AMesonField<FImpl>::getInput(void)
|
||||
{
|
||||
std::vector<std::string> in = {par().v, par().w};
|
||||
std::vector<std::string> in = {par().left, par().right};
|
||||
|
||||
return in;
|
||||
}
|
||||
@ -227,7 +221,6 @@ void TA2AMesonField<FImpl>::setup(void)
|
||||
}
|
||||
mom_.push_back(p);
|
||||
}
|
||||
|
||||
envCache(std::vector<ComplexField>, momphName_, 1,
|
||||
par().mom.size(), envGetGrid(ComplexField));
|
||||
envTmpLat(ComplexField, "coor");
|
||||
@ -240,19 +233,19 @@ void TA2AMesonField<FImpl>::setup(void)
|
||||
template <typename FImpl>
|
||||
void TA2AMesonField<FImpl>::execute(void)
|
||||
{
|
||||
auto &v = envGet(std::vector<FermionField>, par().v);
|
||||
auto &w = envGet(std::vector<FermionField>, par().w);
|
||||
auto &left = envGet(std::vector<FermionField>, par().left);
|
||||
auto &right = envGet(std::vector<FermionField>, par().right);
|
||||
|
||||
int nt = env().getDim().back();
|
||||
int N_i = w.size();
|
||||
int N_j = v.size();
|
||||
int N_i = left.size();
|
||||
int N_j = right.size();
|
||||
int ngamma = gamma_.size();
|
||||
int nmom = mom_.size();
|
||||
int block = par().block;
|
||||
int cacheBlock = par().cacheBlock;
|
||||
|
||||
LOG(Message) << "Computing all-to-all meson fields" << std::endl;
|
||||
LOG(Message) << "W: '" << par().w << "' V: '" << par().v << "'" << std::endl;
|
||||
LOG(Message) << "Left: '" << par().left << "' Right: '" << par().right << "'" << std::endl;
|
||||
LOG(Message) << "Momenta:" << std::endl;
|
||||
for (auto &p: mom_)
|
||||
{
|
||||
@ -309,7 +302,7 @@ void TA2AMesonField<FImpl>::execute(void)
|
||||
auto filenameFn = [this, &ionameFn](const unsigned int m, const unsigned int g)
|
||||
{
|
||||
return par().output + "." + std::to_string(vm().getTrajectory())
|
||||
+ "/" + ioname(m, g) + ".h5";
|
||||
+ "/" + ionameFn(m, g) + ".h5";
|
||||
};
|
||||
|
||||
auto metadataFn = [this](const unsigned int m, const unsigned int g)
|
||||
@ -328,189 +321,7 @@ void TA2AMesonField<FImpl>::execute(void)
|
||||
Kernel kernel(gamma_, ph, envGetGrid(FermionField));
|
||||
|
||||
envGetTmp(Computation, computation);
|
||||
computation.execute(w, v, kernel, ionameFn, filenameFn, metadataFn);
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
// i,j is first loop over SchurBlock factors reusing 5D matrices
|
||||
// ii,jj is second loop over cacheBlock factors for high perf contractoin
|
||||
// iii,jjj are loops within cacheBlock
|
||||
// Total index is sum of these i+ii+iii etc...
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
|
||||
// double flops;
|
||||
// double bytes;
|
||||
// double vol = env().getVolume();
|
||||
// double t_kernel = 0.0;
|
||||
// double nodes = env().getGrid()->NodeCount();
|
||||
// double tot_kernel;
|
||||
|
||||
// envGetTmp(Vector<MF_IO_TYPE>, mfBuf);
|
||||
// envGetTmp(Vector<Complex>, mfCache);
|
||||
|
||||
// double t0 = usecond();
|
||||
// int NBlock_i = N_i/block + (((N_i % block) != 0) ? 1 : 0);
|
||||
// int NBlock_j = N_j/block + (((N_j % block) != 0) ? 1 : 0);
|
||||
|
||||
// for(int i=0;i<N_i;i+=block)
|
||||
// for(int j=0;j<N_j;j+=block)
|
||||
// {
|
||||
// // Get the W and V vectors for this block^2 set of terms
|
||||
// int N_ii = MIN(N_i-i,block);
|
||||
// int N_jj = MIN(N_j-j,block);
|
||||
|
||||
// LOG(Message) << "Meson field block "
|
||||
// << j/block + NBlock_j*i/block + 1
|
||||
// << "/" << NBlock_i*NBlock_j << " [" << i <<" .. "
|
||||
// << i+N_ii-1 << ", " << j <<" .. " << j+N_jj-1 << "]"
|
||||
// << std::endl;
|
||||
|
||||
// A2AMatrixSet<MF_IO_TYPE> mfBlock(mfBuf.data(),nmom,ngamma,nt,N_ii,N_jj);
|
||||
|
||||
// // Series of cache blocked chunks of the contractions within this block
|
||||
// flops = 0.0;
|
||||
// bytes = 0.0;
|
||||
// for(int ii=0;ii<N_ii;ii+=cacheBlock)
|
||||
// for(int jj=0;jj<N_jj;jj+=cacheBlock)
|
||||
// {
|
||||
// int N_iii = MIN(N_ii-ii,cacheBlock);
|
||||
// int N_jjj = MIN(N_jj-jj,cacheBlock);
|
||||
// A2AMatrixSet<Complex> mfCacheBlock(mfCache.data(),nmom,ngamma,nt,N_iii,N_jjj);
|
||||
|
||||
// startTimer("contraction: total");
|
||||
// makeMesonFieldBlock(mfCacheBlock, &w[i+ii], &v[j+jj], gamma_, ph,
|
||||
// env().getNd() - 1, this);
|
||||
// stopTimer("contraction: total");
|
||||
|
||||
// // flops for general N_c & N_s
|
||||
// flops += vol * ( 2 * 8.0 + 6.0 + 8.0*nmom) * N_iii*N_jjj*ngamma;
|
||||
// bytes += vol * (12.0 * sizeof(Complex) ) * N_iii*N_jjj
|
||||
// + vol * ( 2.0 * sizeof(Complex) *nmom ) * N_iii*N_jjj* ngamma;
|
||||
|
||||
// startTimer("cache copy");
|
||||
// parallel_for_nest5(int m =0;m< nmom;m++)
|
||||
// for(int g =0;g< ngamma;g++)
|
||||
// for(int t =0;t< nt;t++)
|
||||
// for(int iii=0;iii< N_iii;iii++)
|
||||
// for(int jjj=0;jjj< N_jjj;jjj++)
|
||||
// {
|
||||
// mfBlock(m,g,t,ii+iii,jj+jjj) = mfCacheBlock(m,g,t,iii,jjj);
|
||||
// }
|
||||
// stopTimer("cache copy");
|
||||
// }
|
||||
|
||||
// // perf
|
||||
// tot_kernel = getDTimer("contraction: colour trace & mom.")
|
||||
// + getDTimer("contraction: local space sum");
|
||||
// t_kernel = tot_kernel - t_kernel;
|
||||
// LOG(Message) << "Kernel perf " << flops/t_kernel/1.0e3/nodes
|
||||
// << " Gflop/s/node " << std::endl;
|
||||
// LOG(Message) << "Kernel perf " << bytes/t_kernel*1.0e6/1024/1024/1024/nodes
|
||||
// << " GB/s/node " << std::endl;
|
||||
// t_kernel = tot_kernel;
|
||||
|
||||
// // IO
|
||||
// if (!par().output.empty())
|
||||
// {
|
||||
// double blockSize, ioTime;
|
||||
// unsigned int myRank = env().getGrid()->ThisRank(),
|
||||
// nRank = env().getGrid()->RankCount();
|
||||
|
||||
// LOG(Message) << "Writing block to disk" << std::endl;
|
||||
// ioTime = -getDTimer("IO: write block");
|
||||
// startTimer("IO: total");
|
||||
// makeFileDir(filename(0, 0), env().getGrid());
|
||||
// #ifdef MF_PARALLEL_IO
|
||||
// env().getGrid()->Barrier();
|
||||
// // make task list for current node
|
||||
// nodeIo_.clear();
|
||||
// for(int f = myRank; f < nmom*ngamma; f += nRank)
|
||||
// {
|
||||
// IoHelper h;
|
||||
|
||||
// h.i = i;
|
||||
// h.j = j;
|
||||
// h.m = f/ngamma;
|
||||
// h.g = f % ngamma;
|
||||
// h.io = A2AMatrixIo<MF_IO_TYPE>(filename(h.m, h.g),
|
||||
// ioname(h.m, h.g), nt, N_i, N_j);
|
||||
// for (auto pmu: mom_[h.m])
|
||||
// {
|
||||
// h.md.momentum.push_back(pmu);
|
||||
// }
|
||||
// h.md.gamma = gamma_[h.g];
|
||||
// nodeIo_.push_back(h);
|
||||
// }
|
||||
// // parallel IO
|
||||
// for (auto &h: nodeIo_)
|
||||
// {
|
||||
// saveBlock(mfBlock, h);
|
||||
// }
|
||||
// env().getGrid()->Barrier();
|
||||
// #else
|
||||
// // serial IO, for testing purposes only
|
||||
// for(int m = 0; m < nmom; m++)
|
||||
// for(int g = 0; g < ngamma; g++)
|
||||
// {
|
||||
// IoHelper h;
|
||||
|
||||
// h.i = i;
|
||||
// h.j = j;
|
||||
// h.m = m;
|
||||
// h.g = g;
|
||||
// h.io = A2AMatrixIo<MF_IO_TYPE>(filename(h.m, h.g),
|
||||
// ioname(h.m, h.g), nt, N_i, N_j);
|
||||
// for (auto pmu: mom_[h.m])
|
||||
// {
|
||||
// h.md.momentum.push_back(pmu);
|
||||
// }
|
||||
// h.md.gamma = gamma_[h.g];
|
||||
// saveBlock(mfBlock, h);
|
||||
// }
|
||||
// #endif
|
||||
// stopTimer("IO: total");
|
||||
// blockSize = static_cast<double>(nmom*ngamma*nt*N_ii*N_jj*sizeof(MF_IO_TYPE));
|
||||
// ioTime += getDTimer("IO: write block");
|
||||
// LOG(Message) << "HDF5 IO done " << sizeString(blockSize) << " in "
|
||||
// << ioTime << " us ("
|
||||
// << blockSize/ioTime*1.0e6/1024/1024
|
||||
// << " MB/s)" << std::endl;
|
||||
// }
|
||||
// }
|
||||
}
|
||||
|
||||
// IO
|
||||
template <typename FImpl>
|
||||
std::string TA2AMesonField<FImpl>::ioname(unsigned int m, unsigned int g) const
|
||||
{
|
||||
std::stringstream ss;
|
||||
|
||||
ss << gamma_[g] << "_";
|
||||
for (unsigned int mu = 0; mu < mom_[m].size(); ++mu)
|
||||
{
|
||||
ss << mom_[m][mu] << ((mu == mom_[m].size() - 1) ? "" : "_");
|
||||
}
|
||||
|
||||
return ss.str();
|
||||
}
|
||||
|
||||
template <typename FImpl>
|
||||
std::string TA2AMesonField<FImpl>::filename(unsigned int m, unsigned int g) const
|
||||
{
|
||||
return par().output + "." + std::to_string(vm().getTrajectory())
|
||||
+ "/" + ioname(m, g) + ".h5";
|
||||
}
|
||||
|
||||
template <typename FImpl>
|
||||
void TA2AMesonField<FImpl>::saveBlock(const A2AMatrixSet<MF_IO_TYPE> &mf, IoHelper &h)
|
||||
{
|
||||
if ((h.i == 0) and (h.j == 0))
|
||||
{
|
||||
startTimer("IO: file creation");
|
||||
h.io.initFile(h.md, par().block);
|
||||
stopTimer("IO: file creation");
|
||||
}
|
||||
startTimer("IO: write block");
|
||||
h.io.saveBlock(mf, h.m, h.g, h.i, h.j);
|
||||
stopTimer("IO: write block");
|
||||
computation.execute(left, right, kernel, ionameFn, filenameFn, metadataFn);
|
||||
}
|
||||
|
||||
END_MODULE_NAMESPACE
|
||||
|
||||
Reference in New Issue
Block a user