prefetch-test.c

/* https://stackoverflow.com/questions/20697215/when-should-we-use-prefetch */

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <sys/timeb.h>
#include <emmintrin.h>

#define MIB  (1024 * 1024)
#define ELEM int
#define SIZE (256 * MIB)  // 64 MiB
#define ELEMS (SIZE / sizeof(ELEM))
#define STEP (64 / sizeof(ELEM))
#define REP 10
#define ITER 10

int main(int argc, char **argv) {

	int do_off = (argc == 1 || strcmp(argv[1], "off") == 0);
	int do_on  = (argc == 1 || strcmp(argv[1],  "on") == 0);


	ELEM *a = aligned_alloc(4096, SIZE);

	for (int i = 0; i < ELEMS; ++i) {
		a[i] = i + 1;
	}

	for (int try = 0; try < ITER; try++) {

		unsigned long long int sum = 0;
		struct timeb start, end;
		unsigned long long delta;

		if (do_off) {
			ftime(&start);
			for (int r = 0; r < REP; r++) {
				for (int i = 0; i < ELEMS; i += STEP) {
					sum += a[i];
				}
			}
			ftime(&end);
			delta = (end.time * 1000 + end.millitm) - (start.time * 1000 + start.millitm);
			printf("Prefetching off: SIZE=%d MiB, sum=%lld, time=%3lld, MiB/s=%lld\n",
					SIZE / MIB, sum, delta, (long long)REP * SIZE * 1000 / (MIB * delta));
		}

		if (do_on) {
			sum = 0;
			ftime(&start);
			for (int r = 0; r < REP; r++) {
				for (int i = 0; i < ELEMS; i += STEP) {
					sum += a[i];
					_mm_prefetch(&a[i + 4096/sizeof(ELEM)], _MM_HINT_T0);
				}
			}

			ftime(&end);
			delta = (end.time * 1000 + end.millitm) - (start.time * 1000 + start.millitm);
			printf("Prefetching  on: SIZE=%d MiB, sum=%lld, time=%3lld, MiB/s=%lld\n",
					SIZE / MIB, sum, delta, (long long)REP * SIZE * 1000 / (MIB * delta));
		}

	}
}