Another hardware mystery

alias.c

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

// STEPS = 4     830ms
// STEPS = 16   4850ms

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

	int STEPS = atoi(argv[1]);
	int stride = 32*1024; // access stride in bytes
	char* arr = malloc(stride * STEPS);
	int iterations = 1000000000 / STEPS;

	clock_t cl = clock();

	char sum = 0;
	for (int i = 0; i < iterations; i++) {
		for (int j = 0; j < STEPS*stride; j += stride) {
			sum += arr[j];

			// this prevents gcc from folding both loops
			// and limits speculative execution
			// (CPU can still prefetch data though)
			arr[j] += sum;
		}
	}

	clock_t end = clock();
	printf("%d %fms\n", STEPS, 1000.0 * (end-cl) / CLOCKS_PER_SEC);
}