GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

clean a bit the bench_gemm files

Browse Source
This commit is contained in:
Gael Guennebaud 2010-03-05 11:35:43 +01:00
parent 5f172cd01f
commit c442208358
2 changed files with 3 additions and 115 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
bench/bench_gemm.cpp
View File

@ -20,11 +20,6 @@ typedef Matrix<Scalar,Dynamic,Dynamic> M;
extern "C" {
#include <bench/btl/libs/C_BLAS/blas.h>
void sgemm_kernel(int actual_mc, int cols, int actual_kc, float alpha,
float* blockA, float* blockB, float* res, int resStride);
void sgemm_oncopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
void sgemm_itcopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
}
static float fone = 1;
@ -72,7 +67,9 @@ int main(int argc, char ** argv)
int rep = 1; // number of repetitions per try
int tries = 5; // number of tries, we keep the best
int s = 2048;
int s = argc==2 ? std::atoi(argv[1]) : 2048;
std::cout << "Matrix size = " << s << "\n";
int m = s;
int n = s;
int p = s;

109
bench/bench_gemm_blas.cpp
View File

@ -1,109 +0,0 @@
#include <Eigen/Core>
#include <bench/BenchTimer.h>
extern "C"
{
#include <bench/btl/libs/C_BLAS/blas.h>
#include <cblas.h>
void sgemm_kernel(int actual_mc, int cols, int actual_kc, float alpha,
float* blockA, float* blockB, float* res, int resStride);
void sgemm_otcopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
void sgemm_oncopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
void sgemm_itcopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
void sgemm_incopy(int actual_kc, int cols, const float* rhs, int rhsStride, float* blockB);
}
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)
{
int M = c.rows();
int N = c.cols();
int K = a.cols();
int lda = a.rows();
int ldb = b.rows();
int ldc = c.rows();
// c.noalias() += a * b;
sgemm_(&notrans,&notrans,&M,&N,&K,&fone,
const_cast<float*>(a.data()),&lda,
const_cast<float*>(b.data()),&ldb,&fone,
c.data(),&ldc);
}
void blas_gemm(const MatrixXd& a, const MatrixXd& b, MatrixXd& c)
{
int M = c.rows();
int N = c.cols();
int K = a.cols();
int lda = a.rows();
int ldb = b.rows();
int ldc = c.rows();
// c.noalias() += a * b;
dgemm_(&notrans,&notrans,&M,&N,&K,&done,
const_cast<double*>(a.data()),&lda,
const_cast<double*>(b.data()),&ldb,&done,
c.data(),&ldc);
}
int main(int argc, char **argv)
{
int rep = 1;
int s = 2048;
int m = s;
int n = s;
int p = s;
const int N = 1;
M a[N];
M b[N];
M c[N];
for (int k=0; k<N; ++k)
{
a[k].resize(m,p); a[k].setOnes();
b[k].resize(p,n); b[k].setOnes();
c[k].resize(m,n); c[k].setOnes();
}
BenchTimer t;
BENCH(t, 5, rep,
for(int k=0;k<N;++k)
blas_gemm(a[k],b[k],c[k]));
// BENCH(t, 5, rep,
// _Pragma("omp parallel for schedule(static,1)")
// for(int k=0;k<N;++k)
// blas_gemm(a[k],b[k],c[k]));
std::cerr << "cpu " << t.best(CPU_TIMER)/rep << "s \t" << (double(m)*N*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*n*p*rep*2/t.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << t.total(REAL_TIMER) << "s)\n";
return 0;
}