#include "header.h"
/*
enum defined in "gsl/gsl_cblas.h" identical in "intel/mkl/10.2.5.035/mkl_cblas.h"
enum CBLAS_ORDER {CblasRowMajor=101, CblasColMajor=102}; 
enum CBLAS_TRANSPOSE {CblasNoTrans=111, CblasTrans=112, CblasConjTrans=113}
*/

void DGEMM_Square_BenchMark(int nMin, int nMax, int step, int mult, int div, int rep){
	int i, n;
	double *A, *B, *C;
	double alpha = 1.8, beta=0.9;
	ticks t1, t2;
	
	A = (double *) malloc(sizeof(double)*nMax*nMax);
	B = (double *) malloc(sizeof(double)*nMax*nMax);
	C = (double *) malloc(sizeof(double)*nMax*nMax);
	
	for(n=nMin;n<nMax;n=(mult*(n+step))/div){
	  for(i=0;i<rep;i++){
	       printf("%d ", n);
	      //cblas in gsl or mkl depends on library
	      // No Trans - No Trans
		t1=getticks();
		cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, 
		     n, n, n, 
		     alpha, 
		     A, nMax, 
		     B, nMax,
		     beta,
		     C, nMax);
		t2=getticks();
		printf("%le ", elapsed(t2,t1));
		// Trans - No Trans
		t1=getticks();
		cblas_dgemm(CblasRowMajor, CblasTrans, CblasNoTrans, 
		     n, n, n, 
		     alpha, 
		     A, nMax, 
		     B, nMax,
		     beta,
		     C, nMax);
		t2=getticks();
		printf("%le ", elapsed(t2,t1));
		// Trans - Trans
		t1=getticks();
		cblas_dgemm(CblasRowMajor, CblasTrans, CblasTrans, 
		     n, n, n, 
		     alpha, 
		     A, nMax, 
		     B, nMax,
		     beta,
		     C, nMax);
		t2=getticks();
		printf("%le ", elapsed(t2,t1));
		// NoTrans - Trans
		t1=getticks();
		cblas_dgemm(CblasRowMajor, CblasTrans, CblasTrans, 
		     n, n, n, 
		     alpha, 
		     A, nMax, 
		     B, nMax,
		     beta,
		     C, nMax);
		t2=getticks();
		printf("%le ", elapsed(t2,t1));
		printf("\n");
		}
	  }
	
	free(A); free(B); free(C);
	
	}
	 
	 
	 /*
	 enum CBLAS_ORDER {CblasRowMajor=101, CblasColMajor=102};
enum CBLAS_TRANSPOSE {CblasNoTrans=111, CblasTrans=112, CblasConjTrans=113};
enum CBLAS_UPLO {CblasUpper=121, CblasLower=122};
enum CBLAS_DIAG {CblasNonUnit=131, CblasUnit=132};
enum CBLAS_SIDE {CblasLeft=141, CblasRight=142};

	 */