From 3ac2b96a2f131e8162d39f0976cfb31b1a853237 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Fri, 26 Feb 2010 12:32:00 +0100
Subject: implement a smarter parallelization strategy for gemm avoiding
 multiple paking of the same data

---
 bench/bench_gemm_blas.cpp | 83 -----------------------------------------------
 1 file changed, 83 deletions(-)
 delete mode 100644 bench/bench_gemm_blas.cpp

(limited to 'bench/bench_gemm_blas.cpp')

diff --git a/bench/bench_gemm_blas.cpp b/bench/bench_gemm_blas.cpp
deleted file mode 100644
index babf1ec2c..000000000
--- a/bench/bench_gemm_blas.cpp
+++ /dev/null
@@ -1,83 +0,0 @@
-
-#include <Eigen/Core>
-#include <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)
-{
-  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,&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();
-
-  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;
-  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;
-}
-
-- 
cgit v1.2.3