memprofiletest.c

memprofiletest.c

/*
 * memprofiletest
 *
 * USAGE: ./memprofiletest size_kb
 *
 * eg, ./memprofiletest 10240		# make larger than LLC cache for misses
 */
 
#include <stdio.h>
#include <stdlib.h>

/* pad methods so they can be LLC profiled without PEBS: */
#define PROF_WORKAROUND

void
memreader(char *p, unsigned long long j) {
	char c;
	c = p[j];
#ifdef PROF_WORKAROUND
	asm(
		"mfence\n"
		"mfence\n"
		"mfence\n"
	);
#endif
}

void
cachedreader(char *p, unsigned long long j) {
	char c;
	c = p[0];
#ifdef PROF_WORKAROUND
	asm(
		"mfence\n"
		"mfence\n"
		"mfence\n"
	);
#endif
}

void
spinner() {
	int i, j;
	for (i = 0, j = 0; i < 100; i++) {
		j++;
	}
#ifdef PROF_WORKAROUND
	asm(
		"mfence\n"
		"mfence\n"
		"mfence\n"
	);
#endif
}

int
main(int argc, char *argv[])
{
	unsigned long long size, i, j;
	char *p, c;

	if (argc != 2) {
		printf("USAGE: memstride size_KB\n");
		exit(1);
	}	

	size = atoi(argv[1]) * 1024ULL;

	if ((p = malloc(size)) == NULL) {
		printf("ERROR: malloc failed\n");
		exit(1);
	}

	printf("Populate...\n");
	for (j = 0; j < size; j += 32) {
		p[j] = 'a';
	}

	printf("Stride...\n");
	i = 0;
	for (;;) {
		for (j = 0ULL; j < size; j += 1024) {
			i++;
			if (i > 100000000ULL) { exit(0); }
			memreader(p, j);
			cachedreader(p, j);
			spinner();
		}
	}

	return (0);
}