mmul_bm.c

/*
  $ gcc -O3 mmul_bm.c -o mmul_bm
*/
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <time.h>

#define ITERATIONS 1000000

void mmul4x4f (const float * a, const float * b, float * c)
{
    int i,j;
    for (i=0; i<16; i+=4) {
        for (j=0; j<4; ++j) {
            c[i+j] =
                b[i+0] * a[j+0]
              + b[i+1] * a[j+4]
              + b[i+2] * a[j+8]
              + b[i+3] * a[j+12];
        }
    }
}
void mmul4x4d (const double * a, const double * b, double * c)
{
    int i,j;
    for (i=0; i<16; i+=4) {
        for (j=0; j<4; ++j) {
            c[i+j] =
                b[i+0] * a[j+0]
              + b[i+1] * a[j+4]
              + b[i+2] * a[j+8]
              + b[i+3] * a[j+12];
        }
    }
}

void floats_malloc ()
{
    float * m, * n, * p;
    clock_t t1, t2;
    int j;
    m = malloc(16*sizeof(float)); 
    n = malloc(16*sizeof(float)); 
    p = malloc(16*sizeof(float));
    t1 = clock();
    for (j=0; j<ITERATIONS; ++j) {
        mmul4x4f(m, n, p);
    }
    t2 = clock();
    printf("floats_malloc: %f\n", ((double)t2-t1) / CLOCKS_PER_SEC);
}
void floats_memalign ()
{
    float * m, * n, * p;
    clock_t t1, t2;
    int j;
    posix_memalign((void**)&m, 16, 16*sizeof(float)); 
    posix_memalign((void**)&n, 16, 16*sizeof(float)); 
    posix_memalign((void**)&p, 16, 16*sizeof(float)); 
    t1 = clock();
    for (j=0; j<ITERATIONS; ++j) {
        mmul4x4f(m, n, p);
    }
    t2 = clock();
    printf("floats_memalign: %f\n", ((double)t2-t1) / CLOCKS_PER_SEC);
}
void doubles_malloc ()
{
    double * m, * n, * p;
    clock_t t1, t2;
    int j;
    m = malloc(16*sizeof(double)); 
    n = malloc(16*sizeof(double)); 
    p = malloc(16*sizeof(double));
    t1 = clock();
    for (j=0; j<ITERATIONS; ++j) {
        mmul4x4d(m, n, p);
    }
    t2 = clock();
    printf("doubles_malloc: %f\n", ((double)t2-t1) / CLOCKS_PER_SEC);
}
void doubles_memalign ()
{
    double * m, * n, * p;
    clock_t t1, t2;
    int j;
    posix_memalign((void**)&m, 16, 16*sizeof(double)); 
    posix_memalign((void**)&n, 16, 16*sizeof(double)); 
    posix_memalign((void**)&p, 16, 16*sizeof(double)); 
    t1 = clock();
    for (j=0; j<ITERATIONS; ++j) {
        mmul4x4d(m, n, p);
    }
    t2 = clock();
    printf("doubles_memalign: %f\n", ((double)t2-t1) / CLOCKS_PER_SEC);
}

int main(){
  doubles_malloc();
  doubles_memalign();
  floats_malloc();
  floats_memalign();
  return 0;
}