RCCE: measuring round-trip latency between two cores

pingpong.c

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <RCCE.h>
#include "timer.h"

#define PRINT(...) { printf(__VA_ARGS__); fflush(stdout); }
#define CORE(i)	if (RCCE_ue() == (i))

#define MSG_SIZE 32
#define ROUNDS 1000

extern double RC_REFCLOCKGHZ;

int RCCE_APP(int argc, char *argv[])
{
	mytimer_t timer;
	char buf[MSG_SIZE];
	int ME = 0, YOU = 1, i;

	dup2(STDOUT_FILENO, STDERR_FILENO);

	RCCE_init(&argc, &argv);

	timer_new(&timer, RC_REFCLOCKGHZ);

	RCCE_barrier(&RCCE_COMM_WORLD);

	CORE(ME) {
		timer_start(&timer);	
		for (i = 0; i < ROUNDS; i++) {
			RCCE_send(buf, MSG_SIZE, YOU);
			RCCE_recv(buf, MSG_SIZE, YOU);
		}
		timer_end(&timer);
	}

	CORE(YOU) {
		for (i = 0; i < ROUNDS; i++) {
			RCCE_recv(buf, MSG_SIZE, ME);
			RCCE_send(buf, MSG_SIZE, ME);
		}
	} 

	CORE(ME) PRINT("Round-trip latency: %.2lf\n", timer_elapsed(&timer, timer_us) / ROUNDS);

	RCCE_finalize();

	return 0;
}