cl.c

// compile as:
//     gcc -DCLSIZE=64 -O2 -c cl.c -o cl
// replace 64 with whatever value you want to pad to.
// this bash loop can be helpful:
//     for x in 8 16 32 64 128 256; do gcc -DCLSIZE=${x} -O2 cl.c -o cl; ./cl; done 
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <threads.h>
#include <stdatomic.h>
#include <time.h>

// number of threads to spawn
#define NTT 4

// type to increment
#define atomic_type atomic_llong

//-----------------------------------------
#define PADSIZE (CLSIZE - sizeof(atomic_type))

struct cell {
	atomic_type x;
	char pad[PADSIZE];
};

struct cell arr[NTT];
uint64_t res[NTT];
thrd_t tt[NTT];

int func(void *u) {
	uint64_t uu = (uint64_t)u;
	for(int i = 0; i < 100000000; ++i) {
		arr[uu].x += 1;
	}
	res[uu] = arr[uu].x;
	return 0;
}

int main () {
	struct timespec ta, tb;
	uint64_t na, nb, nd;

	printf("pad to %3u: ", CLSIZE);
	clock_gettime(CLOCK_MONOTONIC, &ta);
	for(uint64_t i = 0; i < NTT; ++i) {
		thrd_create(tt + i, func, (void*)i);
	}
	for(int i = 0; i < NTT; ++i) {
		thrd_join(tt[i], NULL);
	}
	clock_gettime(CLOCK_MONOTONIC, &tb);
	na = ta.tv_sec * 1000000000 + ta.tv_nsec;
	nb = tb.tv_sec * 1000000000 + tb.tv_nsec;
	nd = nb - na;
	printf("%llu ns\n", nd);
//	for(int i = 0; i < NTT; ++i) {
//		printf("#%d: %llu\n", i, res[i]);
//	}
	return 0;
}