portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Remove warnings under NVCC and move parallel_for to thread-loop

Browse Source
This commit is contained in:
Peter Boyle
2019-01-01 15:08:09 +00:00
parent 0e9b591c1c
commit 802404c78c
7 changed files with 73 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Hadrons/Utilities/Contractor.cc
View File

@ -251,18 +251,18 @@ int main(int argc, char* argv[])
// parse parameter file
ContractorPar par;
unsigned int nMat, nCont;
// unsigned int nMat,nCont;
XmlReader reader(parFilename);
read(reader, "global", par.global);
read(reader, "a2aMatrix", par.a2aMatrix);
read(reader, "product", par.product);
nMat = par.a2aMatrix.size();
nCont = par.product.size();
// nMat = par.a2aMatrix.size();
// nCont = par.product.size();
// create diskvectors
std::map<std::string, EigenDiskVector<ComplexD>> a2aMat;
unsigned int cacheSize;
// unsigned int cacheSize;
for (auto &p: par.a2aMatrix)
{
@ -281,7 +281,8 @@ int main(int argc, char* argv[])
for (auto &p: par.a2aMatrix)
{
std::string filename = p.file;
double t, size;
double t;
// double size;
tokenReplace(filename, "traj", traj);
std::cout << "======== Loading '" << filename << "'" << std::endl;
@ -305,7 +306,8 @@ int main(int argc, char* argv[])
std::vector<A2AMatrixTr<ComplexD>> lastTerm(par.global.nt);
A2AMatrix<ComplexD> prod, buf, tmp;
TimerArray tAr;
double fusec, busec, flops, bytes, tusec;
double fusec, busec, flops, bytes;
// double tusec;
Contractor::CorrelatorResult result;
tAr.startTimer("Total");
@ -351,11 +353,11 @@ int main(int argc, char* argv[])
tAr.startTimer("Transpose caching");
lastTerm[t].resize(ref.rows(), ref.cols());
parallel_for (unsigned int j = 0; j < ref.cols(); ++j)
thread_loop( (unsigned int j = 0; j < ref.cols(); ++j),
for (unsigned int i = 0; i < ref.rows(); ++i)
{
lastTerm[t](i, j) = ref(i, j);
}
});
tAr.stopTimer("Transpose caching");
}
bytes = par.global.nt*lastTerm[0].rows()*lastTerm[0].cols()*sizeof(ComplexD);