gistfile1.txt

#include <xmmintrin.h>
#include <immintrin.h>
#include <time.h>
#include <stdio.h>
#include <stdlib.h>
#include <malloc.h>

#define N 32*1024*1024*8

void *aligned_alloc(size_t alignment, size_t size);

void __attribute__ ((noinline)) __attribute__((optimize("no-tree-vectorize")))
uno(float * restrict res, float * restrict uno, float * restrict due) {
	for (int i = 0; i < N; i++)
		res[i] = uno[i] + due[i];
}

void __attribute__ ((noinline))
due(float * restrict res, float * restrict uno, float * restrict due) {
        for (int i = 0; i < N; i+=8) {
            __m256 a1 = _mm256_load_ps(uno + i);
            __m256 a2 = _mm256_load_ps(due + i);
            _mm256_store_ps(res + i, _mm256_add_ps(a1, a2));
       	}
}


void main() {

	srand((unsigned)time(NULL));

	float *a = aligned_alloc (32, N*sizeof(float));
        float *b = aligned_alloc (32, N*sizeof(float));
        float *c = aligned_alloc (32, N*sizeof(float));
        float *d = aligned_alloc (32, N*sizeof(float));

	for (int i = 0; i < N; i++) {
	   a[N] = ((float)rand()/(float)(RAND_MAX));
	   b[N] = ((float)rand()/(float)(RAND_MAX));
	   c[N] = ((float)rand()/(float)(RAND_MAX));
           d[N] = ((float)rand()/(float)(RAND_MAX));
	}

	void diff(struct timespec *res, struct timespec *start, struct timespec *end)
	{
		if ((end->tv_nsec-start->tv_nsec)<0) {
			res->tv_sec = end->tv_sec-start->tv_sec-1;
			res->tv_nsec = 1000000000+end->tv_nsec-start->tv_nsec;
		} else {
			res->tv_sec = end->tv_sec-start->tv_sec;
			res->tv_nsec = end->tv_nsec-start->tv_nsec;
		}
	}

	struct timespec ts1, ts2, diffe;
	clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts1);

	uno(a,b,c);

	clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts2);
	diff(&diffe, &ts1, &ts2);
	printf("Non optimized, passed %d sec and %d nanosec\n", diffe.tv_sec, diffe.tv_nsec);

        clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts1);

	due(d,b,c);	

        clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts2);
        diff(&diffe, &ts1, &ts2);
        printf("Optimized, passed %d sec and %d nanosec\n", diffe.tv_sec, diffe.tv_nsec);

	for (int i = 0; i < N; i++)
		if (a[i] != d[i])
			exit(-1);

	puts("Result checked!\n");
}