matrix.c

/*
 * gcc -O2 matrix.c -o matrix
 * 
 * for i in {1..30}; do echo 768 | ./matrix; done > matrix.csv
 * 
 * R
 * data = read.csv("matrix.csv", header = FALSE)
 * wilcox.test(data$V1, data$V2, alternative = "two.sided")
 */

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

void
multiply_static(double *a, double *b, double *c, size_t n)
{
	static const size_t SIZE = 768;

	size_t i, j, k;

	(void)n;
	for (i = 0; i < SIZE; i++)
		for (j = 0; j < SIZE; j++)
			for (k = 0; k < SIZE; k++)
				c[i * SIZE + j] = a[i * SIZE + k] * b[k * SIZE + j];
}

void
multiply_dynamic(double *a, double *b, double *c, size_t n)
{
	size_t i, j, k;
	for (i = 0; i < n; i++)
		for (j = 0; j < n; j++)
			for (k = 0; k < n; k++)
				c[i * n + j] = a[i * n + k] * b[k * n + j];
}

void
fill(double *a, size_t n)
{
	size_t i, j;
	for (i = 0; i < n; i++)
		for (j = 0; j < n; j++)
			a[i * n + j] = (double)rand() / RAND_MAX;
}

typedef void (*func_type)(double *a, double *b, double *c, size_t n);

double
test(func_type func, size_t n)
{
	double *a, *b, *c;
	clock_t t0, t1;

	a = calloc(n * n, sizeof(*a));
	b = calloc(n * n, sizeof(*b));
	c = calloc(n * n, sizeof(*c));
	if (!a || !b || !c) {
		perror("malloc");
		exit(EXIT_FAILURE);
	}

	t0 = clock();
	func(a, b, c, n);
	t1 = clock();
	return (double)(t1 - t0) / CLOCKS_PER_SEC;
}

int
main(int argc, char **argv)
{
	size_t n;
	double dt, st;

	scanf("%zu", &n);
	dt = test(multiply_dynamic, n);
	st = test(multiply_static, n);
	printf("%f,%f\n", dt, st);
}