GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-16 11:23:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

add a small benchmark to quickly bench/compare SMP support

Browse Source
This commit is contained in:
Gael Guennebaud 2010-02-22 11:09:57 +01:00
parent 1c6bdb4060
commit aaaf855a88
2 changed files with 111 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
bench/bench_gemm.cpp Normal file
View File

@ -0,0 +1,43 @@
// g++-4.4 bench_gemm.cpp -I .. -O2 -DNDEBUG -lrt -fopenmp && OMP_NUM_THREADS=2 ./a.out
// icpc bench_gemm.cpp -I .. -O3 -DNDEBUG -lrt -openmp && OMP_NUM_THREADS=2 ./a.out
#include <Eigen/Core>
#include "../../eigen2/bench/BenchTimer.h"
using namespace std;
using namespace Eigen;
#ifndef SCALAR
#define SCALAR float
#endif
typedef SCALAR Scalar;
typedef Matrix<Scalar,Dynamic,Dynamic> M;
void gemm(const M& a, const M& b, M& c)
{
c.noalias() += a * b;
}
int main(int argc, char ** argv)
{
int rep = 2;
int s = 1024;
int m = s;
int n = s;
int p = s;
M a(m,n); a.setOnes();
M b(n,p); b.setOnes();
M c(m,p); c.setOnes();
BenchTimer t;
BENCH(t, 5, rep, gemm(a,b,c));
std::cerr << "cpu " << t.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/t.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << t.total(CPU_TIMER) << "s)\n";
std::cerr << "real " << t.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/t.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << t.total(REAL_TIMER) << "s)\n";
return 0;
}

68
bench/bench_gemm_blas.cpp Normal file
View File

@ -0,0 +1,68 @@
#include <Eigen/Core>
#include <../eigen2/bench/BenchTimer.h>
extern "C"
{
#include <bench/btl/libs/C_BLAS/blas.h>
#include <cblas.h>
}
using namespace std;
using namespace Eigen;
#ifndef SCALAR
#define SCALAR float
#endif
typedef SCALAR Scalar;
typedef Matrix<Scalar,Dynamic,Dynamic> M;
static float fone = 1;
static float fzero = 0;
static double done = 1;
static double szero = 0;
static char notrans = 'N';
static char trans = 'T';
static char nonunit = 'N';
static char lower = 'L';
static char right = 'R';
static int intone = 1;
void blas_gemm(const MatrixXf& a, const MatrixXf& b, MatrixXf& c)
{
// cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, c.rows(), c.cols(), a.cols(), 1, a.data(), a.rows(), b.data(), b.rows(), 1, c.data(), c.rows());
int M = c.rows();
int N = c.cols();
int K = a.cols();
int lda = a.rows();
int ldb = b.rows();
int ldc = c.rows();
sgemm_(&notrans,&notrans,&M,&N,&K,&fone,
const_cast<float*>(a.data()),&lda,
const_cast<float*>(b.data()),&ldb,&fzero,
c.data(),&ldc);
}
int main(int argc, char **argv)
{
int rep = 2;
int s = 1024;
int m = s;
int n = s;
int p = s;
M a(m,n); a.setOnes();
M b(n,p); b.setOnes();
M c(m,p); c.setOnes();
BenchTimer t;
BENCH(t, 5, rep, blas_gemm(a,b,c));
std::cerr << "cpu " << t.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/t.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << t.total(CPU_TIMER) << "s)\n";
std::cerr << "real " << t.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/t.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << t.total(REAL_TIMER) << "s)\n";
return 0;
}