gemm.cpp
Go to the documentation of this file.
1#define SHARK_USE_SIMD
2#include <shark/LinAlg/BLAS/blas.h>
3#include <shark/Core/Timer.h>
4#include <iostream>
5using namespace shark;
6using namespace std;
7
8template<class AMat, class BMat, class CMat>
9double benchmark(
10 blas::matrix_expression<AMat, blas::cpu_tag> const& A,
11 blas::matrix_expression<BMat, blas::cpu_tag> const& B,
12 blas::matrix_expression<CMat, blas::cpu_tag> & C
13){
14 double minTime = std::numeric_limits<double>::max();
15 for(std::size_t i = 0; i != 10; ++i){
16 Timer time;
17 noalias(C) += prod(A,B);
18 minTime = min(minTime,time.stop());
19 }
20 return (A().size1()*A().size2()*B().size2())/minTime/1024/1024;
21}
22
23int main(int argc, char **argv) {
24 std::size_t size = 100;
25 std::cout<<"Flops"<<std::endl;
26 for(std::size_t iter = 0; iter != 5; ++iter){
27 std::size_t middle = size;
28 blas::matrix<double,blas::row_major> Arow(size,middle);
29 for(std::size_t i = 0; i != size; ++i){
30 for(std::size_t k = 0; k != middle; ++k){
31 Arow(i,k) = 0.1/size*i+0.1/size*k;
32 }
33 }
34
35 blas::matrix<double,blas::row_major> Brow(middle,size);
36 for(std::size_t k = 0; k != middle; ++k){
37 for(std::size_t j = 0; j != size; ++j){
38 Brow(k,j) = 0.1/size*j+0.1/size*k;
39 }
40 }
41 blas::matrix<double,blas::column_major> Acol = Arow;
42 blas::matrix<double,blas::column_major> Bcol = Brow;
43
44 blas::matrix<double,blas::row_major> Crow(size,size,0.0);
45 blas::matrix<double,blas::column_major> Ccol(size,size,0.0);
46 std::cout<<size<<"\t row major results\t"<<benchmark(Arow,Brow,Crow)<<"\t"<< benchmark(Acol,Brow,Crow)
47 <<"\t"<< benchmark(Arow,Bcol,Crow) <<"\t" <<benchmark(Acol,Bcol,Crow) <<std::endl;
48 std::cout<<size<<"\t column major results\t"<<benchmark(Arow,Brow,Ccol)<<"\t"<< benchmark(Acol,Brow,Ccol)
49 <<"\t"<< benchmark(Arow,Bcol,Ccol) <<"\t" <<benchmark(Acol,Bcol,Ccol) <<std::endl;
50 size *=2;
51 }
52}