Matrix.cpp

#include <stdlib.h>
#include <stdio.h>
#include <sys/time.h>
#include <random>
static double dtime(void);
extern int main(void);

const int rndmax = 32767;
static double dtime()/*実行時間を計る関数*/
{
    double q;
    struct timeval tnow;

    gettimeofday(&tnow,NULL);
    q = (double)tnow.tv_sec + (double)tnow.tv_usec * 1.0e-6;

    return q;
}

int main(void){
static long memsize[] = {
        16l, 32l, 64l, 128l, 256l, 512l, 1024l,
        2048l, 4096l
};/*lはlong型を表す接尾辞*/
long max_n=memsize[sizeof(memsize)/sizeof(memsize[0])-1];
long **matrix_1;
long *base_matrix_1;
long **matrix_2;
long *base_matrix_2;
long **matrix_3;
long *base_matrix_3;
long h,i,j,k;
long n = max_n;
long m = max_n;
double start_time,end_time,elapsed_time;
if(!(matrix_1 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_2 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_3 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_1=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_2=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_3=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}

for(k=0; k<sizeof(memsize)/sizeof(memsize[0]); k++) {
	long p = memsize[k];
	long q = memsize[k];
	double start_time,end_time,elapsed_time;
	for (i=0;i<p;i++) {
		matrix_1[i] = base_matrix_1 + i * q;
	}
    std::random_device rnd;
	std::mt19937 mtrnd(rnd());
	std::uniform_int_distribution<> rangernd(0, rndmax);
	for (i=0;i<p;i++) {
		for (j=0;j<q;j++) {
			matrix_1[i][j] = rangernd(mtrnd);
		}
	}

	for (i=0;i<p;i++) {
		matrix_2[i] = base_matrix_2 + i * q;
	}
	for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
			matrix_2[i][j] = rangernd(mtrnd);
		}
	}


	for (i=0;i<p;i++) {
		matrix_3[i] = base_matrix_3 + i * q;
	}
    for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
		    matrix_3[i][j] = 0;
		}
	}
	start_time=dtime();
	for (i=0;i<p;i++){
		for (j=0;j<q;j++){
			for (h=0;h<q;h++){
				matrix_3[i][j] += matrix_1[i][h] * matrix_2[h][j];
			}
		}
	}
	end_time=dtime();
	elapsed_time=end_time-start_time;
	printf("%f\n",elapsed_time);
}
free(base_matrix_1);
free(matrix_1);
free(base_matrix_2);
free(matrix_2);
free(base_matrix_3);
free(matrix_3);
exit(EXIT_SUCCESS);
}

Matrix_block16.cpp

#include <stdlib.h>
#include <stdio.h>
#include <sys/time.h>
#include <random>
static double dtime(void);
extern int main(void);

const int rndmax = 32767;

static double dtime()/*実行時間を計る関数*/
{
    double q;
    struct timeval tnow;

    gettimeofday(&tnow,NULL);
    q = (double)tnow.tv_sec + (double)tnow.tv_usec * 1.0e-6;

    return q;
}

int main(void){
static long memsize[] = {
        16l, 32l, 64l, 128l, 256l, 512l, 1024l,
        2048l, 4096l
};/*すべて16で割り切れる*/
long max_n = memsize[sizeof(memsize)/sizeof(memsize[0])-1];
long **matrix_1;
long *base_matrix_1;
long **matrix_2;
long *base_matrix_2;
long **matrix_3;
long *base_matrix_3;
long h,i,j,k;
long n = max_n;
long m = max_n;
long ibl = 16;
long ib,jb,hb;
double start_time,end_time,elapsed_time;
if(!(matrix_1 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_2 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_3 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_1=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_2=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_3=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}

for(k=0; k<sizeof(memsize)/sizeof(memsize[0]); k++) {
	long p = memsize[k];
	long q = memsize[k];
	double start_time,end_time,elapsed_time;
	for (i=0;i<p;i++) {
		matrix_1[i] = base_matrix_1 + i * q;
	}
    std::random_device rnd;
	std::mt19937 mtrnd(rnd());
	std::uniform_int_distribution<> rangernd(0, rndmax);
	for (i=0;i<p;i++) {
		for (j=0;j<q;j++) {
			matrix_1[i][j] = rangernd(mtrnd);
		}
	}

	for (i=0;i<p;i++) {
		matrix_2[i] = base_matrix_2 + i * q;
	}
	for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
			matrix_2[i][j] = rangernd(mtrnd);
		}
	}


	for (i=0;i<p;i++) {
		matrix_3[i] = base_matrix_3 + i * q;
	}
    for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
		    matrix_3[i][j] = 0;
		}
	}
	start_time=dtime();
    for ( ib=0; ib<p; ib+=ibl ) {
        for ( jb=0; jb<p; jb+=ibl ) {
            for ( hb=0; hb<p; hb+=ibl ) {
                for ( i=ib; i<ib+ibl; i++ ) {
                    for ( j=jb; j<jb+ibl; j++ ) {
                        for ( h=hb; h<hb+ibl; h++ ) {
                            matrix_3[i][j] += matrix_1[i][h] * matrix_2[h][j];
                        }
                    }
                }
            }
        }
    }
	end_time=dtime();
	elapsed_time=end_time-start_time;
	printf("%f\n",elapsed_time);
}
free(base_matrix_1);
free(matrix_1);
free(base_matrix_2);
free(matrix_2);
free(base_matrix_3);
free(matrix_3);
exit(EXIT_SUCCESS);

}

Matrix_block64.cpp

#include <stdlib.h>
#include <stdio.h>
#include <sys/time.h>
#include <random>
static double dtime(void);
extern int main(void);

const int rndmax = 32767;

static double dtime()/*実行時間を計る関数*/
{
    double q;
    struct timeval tnow;

    gettimeofday(&tnow,NULL);
    q = (double)tnow.tv_sec + (double)tnow.tv_usec * 1.0e-6;

    return q;
}

int main(void){
static long memsize[] = {
        16l, 32l, 64l, 128l, 256l, 512l, 1024l,
        2048l, 4096l
};/*すべて16で割り切れる*/
long max_n = memsize[sizeof(memsize)/sizeof(memsize[0])-1];
long **matrix_1;
long *base_matrix_1;
long **matrix_2;
long *base_matrix_2;
long **matrix_3;
long *base_matrix_3;
long h,i,j,k;
long n = max_n;
long m = max_n;
long ibl = 64;
long ib,jb,hb;
double start_time,end_time,elapsed_time;
if(!(matrix_1 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_2 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(matrix_3 = (long**)malloc(sizeof(long*)*n))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_1=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_2=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}
if(!(base_matrix_3=(long*)malloc(sizeof(long)*n*m))) {
	puts("memory allocation failed");
	exit(EXIT_FAILURE);
}

for(k=0; k<sizeof(memsize)/sizeof(memsize[0]); k++) {
	long p = memsize[k];
	long q = memsize[k];
	double start_time,end_time,elapsed_time;
	for (i=0;i<p;i++) {
		matrix_1[i] = base_matrix_1 + i * q;
	}
    std::random_device rnd;
	std::mt19937 mtrnd(rnd());
	std::uniform_int_distribution<> rangernd(0, rndmax);
	for (i=0;i<p;i++) {
		for (j=0;j<q;j++) {
			matrix_1[i][j] = rangernd(mtrnd);
		}
	}

	for (i=0;i<p;i++) {
		matrix_2[i] = base_matrix_2 + i * q;
	}
	for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
			matrix_2[i][j] = rangernd(mtrnd);
		}
	}


	for (i=0;i<p;i++) {
		matrix_3[i] = base_matrix_3 + i * q;
	}
    for (i=0;i<p;i++){
		for (j=0;j<q;j++) {
		    matrix_3[i][j] = 0;
		}
	}
	start_time=dtime();
    if(p<64){
        for (i=0;i<p;i++){
    		for (j=0;j<q;j++){
    			for (h=0;h<q;h++){
    				matrix_3[i][j] += matrix_1[i][h] * matrix_2[h][j];
    			}
    		}
    	}
    }else{
        for ( ib=0; ib<p; ib+=ibl ) {
            for ( jb=0; jb<p; jb+=ibl ) {
                for ( hb=0; hb<p; hb+=ibl ) {
                    for ( i=ib; i<ib+ibl; i++ ) {
                        for ( j=jb; j<jb+ibl; j++ ) {
                            for ( h=hb; h<hb+ibl; h++ ) {
                                matrix_3[i][j] += matrix_1[i][h] * matrix_2[h][j];
                            }
                        }
                    }
                }
            }
        }
    }
	end_time=dtime();
	elapsed_time=end_time-start_time;
	printf("%f\n",elapsed_time);
}
free(base_matrix_1);
free(matrix_1);
free(base_matrix_2);
free(matrix_2);
free(base_matrix_3);
free(matrix_3);
exit(EXIT_SUCCESS);

}