1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 15:55:37 +00:00
Grid/Hadrons/Utilities/ContractorBenchmark.cc

463 lines
13 KiB
C++
Raw Normal View History

2019-02-05 18:55:24 +00:00
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: Hadrons/Utilities/ContractorBenchmark.cc
Copyright (C) 2015-2019
Author: Antonin Portelli <antonin.portelli@me.com>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
2018-10-23 17:08:16 +01:00
#include <Hadrons/Global.hpp>
#include <Hadrons/A2AMatrix.hpp>
2018-10-23 21:00:54 +01:00
#ifdef USE_MKL
#include "mkl.h"
#include "mkl_cblas.h"
#endif
2018-10-23 17:08:16 +01:00
using namespace Grid;
using namespace Hadrons;
2019-07-18 14:47:02 +01:00
const int RowMajor = Eigen::RowMajor;
const int ColMajor = Eigen::ColMajor;
2018-10-23 17:08:16 +01:00
#ifdef GRID_COMMS_MPI3
#define GET_RANK(rank, nMpi) \
MPI_Comm_size(MPI_COMM_WORLD, &(nMpi));\
MPI_Comm_rank(MPI_COMM_WORLD, &(rank));\
assert(rank<nMpi)
#define BARRIER() MPI_Barrier(MPI_COMM_WORLD)
#define INIT() MPI_Init(NULL, NULL)
#define FINALIZE() MPI_Finalize()
#else
#define GET_RANK(rank, nMpi) (nMpi) = 1; (rank) = 0 ; assert(rank<nMpi)
#define BARRIER()
#define INIT()
#define FINALIZE()
#endif
2018-10-23 21:00:54 +01:00
template <typename Function, typename MatLeft, typename MatRight>
inline void trBenchmark(const std::string name, const MatLeft &left,
const MatRight &right, const ComplexD ref, Function fn)
2018-10-23 17:08:16 +01:00
{
double t, flops, bytes, n = left[0].rows()*left[0].cols();
unsigned int nMat = left.size();
int nMpi, rank;
2018-10-23 17:08:16 +01:00
ComplexD buf;
t = 0.;
GET_RANK(rank, nMpi);
t = -usecond();
BARRIER();
for (unsigned int i = rank*nMat/nMpi; i < (rank+1)*nMat/nMpi; ++i)
2018-10-23 17:08:16 +01:00
{
fn(buf, left[i], right[i]);
2018-10-23 17:08:16 +01:00
}
BARRIER();
t += usecond();
2018-10-23 17:08:16 +01:00
flops = nMat*(6.*n + 2.*(n - 1.));
bytes = nMat*(2.*n*sizeof(ComplexD));
if (rank == 0)
{
std::cout << std::setw(34) << name << ": diff= "
<< std::setw(12) << abs(buf-ref)
<< std::setw(10) << t/1.0e6 << " sec "
<< std::setw(10) << flops/t/1.0e3 << " GFlop/s "
<< std::setw(10) << bytes/t*1.0e6/1024/1024/1024 << " GB/s "
<< std::endl;
}
2018-10-23 21:00:54 +01:00
::sleep(1);
}
template <typename Function, typename MatV, typename Mat>
inline void mulBenchmark(const std::string name, const MatV &left,
const MatV &right, const Mat &ref, Function fn)
2018-10-23 21:00:54 +01:00
{
double t, flops, bytes;
double nr = left[0].rows(), nc = left[0].cols();
// double n = nr*nc;
unsigned int nMat = left.size();
int nMpi, rank;
Mat buf(left[0].rows(), left[0].rows());
2018-10-23 21:00:54 +01:00
t = 0.;
GET_RANK(rank, nMpi);
t = -usecond();
BARRIER();
for (unsigned int i = rank*nMat/nMpi; i < (rank+1)*nMat/nMpi; ++i)
2018-10-23 21:00:54 +01:00
{
fn(buf, left[i], right[i]);
2018-10-23 21:00:54 +01:00
}
BARRIER();
t += usecond();
flops = nMat*(nr*nr*(6.*nc + 2.*(nc - 1.)));
bytes = nMat*(2*nc*nr*sizeof(ComplexD));
2018-10-23 21:00:54 +01:00
if (rank == 0)
{
std::cout << std::setw(34) << name << ": diff= "
<< std::setw(12) << (buf-ref).squaredNorm()
<< std::setw(10) << t/1.0e6 << " sec "
<< std::setw(10) << flops/t/1.0e3 << " GFlop/s "
<< std::setw(10) << bytes/t*1.0e6/1024/1024/1024 << " GB/s "
<< std::endl;
}
2018-10-23 17:08:16 +01:00
::sleep(1);
}
#ifdef USE_MKL
template <typename MatLeft, typename MatRight>
static inline void zdotuRow(ComplexD &res, const unsigned int aRow,
const MatLeft &a, const MatRight &b)
2018-10-23 17:08:16 +01:00
{
const ComplexD *aPt, *bPt;
unsigned int aInc, bInc;
2018-10-23 17:08:16 +01:00
2019-07-18 14:47:02 +01:00
if (MatLeft::Options == RowMajor)
2018-10-23 17:08:16 +01:00
{
aPt = a.data() + aRow*a.cols();
aInc = 1;
2018-10-23 17:08:16 +01:00
}
2019-07-18 14:47:02 +01:00
else if (MatLeft::Options == ColMajor)
{
aPt = a.data() + aRow;
aInc = a.rows();
}
2019-07-18 14:47:02 +01:00
if (MatRight::Options == RowMajor)
{
bPt = b.data() + aRow;
bInc = b.cols();
}
2019-07-18 14:47:02 +01:00
else if (MatRight::Options == ColMajor)
{
bPt = b.data() + aRow*b.rows();
bInc = 1;
}
cblas_zdotu_sub(a.cols(), aPt, aInc, bPt, bInc, &res);
}
2018-10-23 17:08:16 +01:00
template <typename MatLeft, typename MatRight>
static inline void zdotuCol(ComplexD &res, const unsigned int aCol,
const MatLeft &a, const MatRight &b)
{
const ComplexD *aPt, *bPt;
unsigned int aInc, bInc;
2018-10-23 21:00:54 +01:00
2019-07-18 14:47:02 +01:00
if (MatLeft::Options == RowMajor)
{
aPt = a.data() + aCol;
aInc = a.cols();
}
2019-07-18 14:47:02 +01:00
else if (MatLeft::Options == ColMajor)
{
aPt = a.data() + aCol*a.rows();
aInc = 1;
}
2019-07-18 14:47:02 +01:00
if (MatRight::Options == RowMajor)
{
bPt = b.data() + aCol*b.cols();
bInc = 1;
}
2019-07-18 14:47:02 +01:00
else if (MatRight::Options == ColMajor)
{
bPt = b.data() + aCol;
bInc = b.rows();
}
cblas_zdotu_sub(a.rows(), aPt, aInc, bPt, bInc, &res);
}
#endif
2018-10-23 17:08:16 +01:00
template <typename MatLeft, typename MatRight>
void fullTrBenchmark(const unsigned int ni, const unsigned int nj, const unsigned int nMat)
{
std::vector<MatLeft> left;
std::vector<MatRight> right;
MatRight buf;
ComplexD ref;
int rank, nMpi;
2018-10-23 17:08:16 +01:00
left.resize(nMat, MatLeft::Random(ni, nj));
right.resize(nMat, MatRight::Random(nj, ni));
GET_RANK(rank, nMpi);
if (rank == 0)
{
std::cout << "==== tr(A*B) benchmarks" << std::endl;
std::cout << "A matrices use ";
2019-07-18 14:47:02 +01:00
if (MatLeft::Options == RowMajor)
{
std::cout << "row-major ordering" << std::endl;
}
2019-07-18 14:47:02 +01:00
else if (MatLeft::Options == ColMajor)
{
std::cout << "col-major ordering" << std::endl;
}
std::cout << "B matrices use ";
2019-07-18 14:47:02 +01:00
if (MatRight::Options == RowMajor)
{
std::cout << "row-major ordering" << std::endl;
}
2019-07-18 14:47:02 +01:00
else if (MatRight::Options == ColMajor)
{
std::cout << "col-major ordering" << std::endl;
}
std::cout << std::endl;
}
BARRIER();
ref = (left.back()*right.back()).trace();
trBenchmark("Hadrons A2AContraction::accTrMul", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
res = 0.;
A2AContraction::accTrMul(res, a, b);
});
trBenchmark("Naive loop rows first", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
2018-10-23 17:08:16 +01:00
{
2018-11-16 20:59:49 +00:00
auto nr = a.rows(), nc = a.cols();
2018-10-23 17:08:16 +01:00
res = 0.;
thread_for(i,nr,
{
2018-11-16 20:59:49 +00:00
ComplexD tmp = 0.;
for (unsigned int j = 0; j < nc; ++j)
{
2018-11-16 20:59:49 +00:00
tmp += a(i, j)*b(j, i);
}
thread_critical
2018-11-16 20:59:49 +00:00
{
res += tmp;
}
});
2018-10-23 17:08:16 +01:00
});
trBenchmark("Naive loop cols first", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
2018-11-16 20:59:49 +00:00
auto nr = a.rows(), nc = a.cols();
2018-10-23 17:08:16 +01:00
res = 0.;
thread_for(j,nc,
{
2018-11-16 20:59:49 +00:00
ComplexD tmp = 0.;
for (unsigned int i = 0; i < nr; ++i)
{
2018-11-16 20:59:49 +00:00
tmp += a(i, j)*b(j, i);
}
thread_critical
2018-11-16 20:59:49 +00:00
{
res += tmp;
}
});
2018-10-23 17:08:16 +01:00
});
trBenchmark("Eigen tr(A*B)", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
2018-10-23 17:08:16 +01:00
{
res = (a*b).trace();
});
trBenchmark("Eigen row-wise dot", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
2018-10-23 17:08:16 +01:00
res = 0.;
thread_for(r,a.rows(),
{
2018-11-16 20:59:49 +00:00
ComplexD tmp;
2018-11-16 20:59:49 +00:00
tmp = a.row(r).conjugate().dot(b.col(r));
thread_critical
{
2018-11-16 20:59:49 +00:00
res += tmp;
}
});
2018-10-23 17:08:16 +01:00
});
trBenchmark("Eigen col-wise dot", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
2018-10-23 17:08:16 +01:00
res = 0.;
thread_for(c,a.cols(),
{
2018-11-16 20:59:49 +00:00
ComplexD tmp;
2018-11-16 20:59:49 +00:00
tmp = a.col(c).conjugate().dot(b.row(c));
thread_critical
{
2018-11-16 20:59:49 +00:00
res += tmp;
}
});
2018-10-23 17:08:16 +01:00
});
trBenchmark("Eigen Hadamard", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
2018-10-23 17:08:16 +01:00
{
res = a.cwiseProduct(b.transpose()).sum();
});
2018-10-23 21:00:54 +01:00
#ifdef USE_MKL
trBenchmark("MKL row-wise zdotu", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
res = 0.;
2019-07-18 21:38:11 +01:00
thread_for(r,a.rows(),
2018-10-23 21:00:54 +01:00
{
ComplexD tmp;
2018-11-16 20:59:49 +00:00
zdotuRow(tmp, r, a, b);
thread_critical
{
2018-11-16 20:59:49 +00:00
res += tmp;
}
});
2018-10-23 21:00:54 +01:00
});
trBenchmark("MKL col-wise zdotu", left, right, ref,
[](ComplexD &res, const MatLeft &a, const MatRight &b)
{
res = 0.;
thread_for(c,a.cols(),
2018-10-23 21:00:54 +01:00
{
ComplexD tmp;
2018-11-16 20:59:49 +00:00
zdotuCol(tmp, c, a, b);
thread_critical
{
2018-11-16 20:59:49 +00:00
res += tmp;
}
});
2018-10-23 21:00:54 +01:00
});
#endif
BARRIER();
if (rank == 0)
{
std::cout << std::endl;
}
}
2018-10-23 21:00:54 +01:00
template <typename Mat>
void fullMulBenchmark(const unsigned int ni, const unsigned int nj, const unsigned int nMat)
{
std::vector<Mat> left, right;
Mat ref;
int rank, nMpi;
2018-10-23 21:00:54 +01:00
left.resize(nMat, Mat::Random(ni, nj));
right.resize(nMat, Mat::Random(nj, ni));
GET_RANK(rank, nMpi);
if (rank == 0)
{
std::cout << "==== A*B benchmarks" << std::endl;
std::cout << "all matrices use ";
2019-07-18 14:47:02 +01:00
if (Mat::Options == RowMajor)
2018-10-23 21:00:54 +01:00
{
std::cout << "row-major ordering" << std::endl;
2018-10-23 21:00:54 +01:00
}
2019-07-18 14:47:02 +01:00
else if (Mat::Options == ColMajor)
2018-10-23 21:00:54 +01:00
{
std::cout << "col-major ordering" << std::endl;
2018-10-23 21:00:54 +01:00
}
std::cout << std::endl;
}
BARRIER();
ref = left.back()*right.back();
mulBenchmark("Hadrons A2AContraction::mul", left, right, ref,
[](Mat &res, const Mat &a, const Mat &b)
{
A2AContraction::mul(res, a, b);
});
mulBenchmark("Eigen A*B", left, right, ref,
[](Mat &res, const Mat &a, const Mat &b)
2018-10-23 21:00:54 +01:00
{
res = a*b;
});
#ifdef USE_MKL
mulBenchmark("MKL A*B", left, right, ref,
[](Mat &res, const Mat &a, const Mat &b)
2018-10-23 21:00:54 +01:00
{
const ComplexD one(1., 0.), zero(0., 0.);
2019-07-18 14:47:02 +01:00
if (Mat::Options == RowMajor)
{
cblas_zgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, a.rows(), b.cols(),
a.cols(), &one, a.data(), a.cols(), b.data(), b.cols(), &zero,
res.data(), res.cols());
}
2019-07-18 14:47:02 +01:00
else if (Mat::Options == ColMajor)
{
cblas_zgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, a.rows(), b.cols(),
a.cols(), &one, a.data(), a.rows(), b.data(), b.rows(), &zero,
res.data(), res.rows());
}
2018-10-23 21:00:54 +01:00
});
#endif
BARRIER();
if (rank == 0)
{
std::cout << std::endl;
}
}
int main(int argc, char *argv[])
{
// parse command line
Eigen::Index ni, nj, nMat;
int nMpi, rank;
if (argc != 4)
{
std::cerr << "usage: " << argv[0] << " <Ni> <Nj> <#matrices>";
std::cerr << std::endl;
return EXIT_FAILURE;
}
ni = std::stoi(argv[1]);
nj = std::stoi(argv[2]);
nMat = std::stoi(argv[3]);
INIT();
GET_RANK(rank, nMpi);
if (rank == 0)
{
std::cout << "\n*** ALL-TO-ALL MATRIX CONTRACTION BENCHMARK ***\n" << std::endl;
std::cout << nMat << " couples of " << ni << "x" << nj << " matrices\n" << std::endl;
std::cout << nMpi << " MPI processes" << std::endl;
#ifdef GRID_OMP
std::cout << omp_get_num_threads() << " threads\n" << std::endl;
#else
std::cout << "Single-threaded\n" << std::endl;
#endif
#ifdef EIGEN_USE_MKL_ALL
std::cout << "Eigen uses the MKL" << std::endl;
#endif
std::cout << "Eigen uses " << Eigen::nbThreads() << " threads" << std::endl;
#ifdef USE_MKL
std::cout << "MKL uses " << mkl_get_max_threads() << " threads" << std::endl;
#endif
std::cout << std::endl;
}
2018-10-23 21:00:54 +01:00
fullTrBenchmark<A2AMatrix<ComplexD>, A2AMatrix<ComplexD>>(ni, nj, nMat);
fullTrBenchmark<A2AMatrix<ComplexD>, A2AMatrixTr<ComplexD>>(ni, nj, nMat);
fullTrBenchmark<A2AMatrixTr<ComplexD>, A2AMatrix<ComplexD>>(ni, nj, nMat);
fullTrBenchmark<A2AMatrixTr<ComplexD>, A2AMatrixTr<ComplexD>>(ni, nj, nMat);
fullMulBenchmark<A2AMatrix<ComplexD>>(ni, nj, nMat);
fullMulBenchmark<A2AMatrixTr<ComplexD>>(ni, nj, nMat);
FINALIZE();
2018-10-23 17:08:16 +01:00
return EXIT_SUCCESS;
}