Updating MPB-allocated objects < cache line size in RCCE

wcb.c

#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include <unistd.h>
#include "RCCE.h"
#include "RCCE_lib.h"

#define WORKER(i) if (RCCE_ue() == (i))

static int *fool_wcb;

static int MPB_read(int *n)
{
	return (RC_cache_invalidate(), *n);
}

static void MPB_write(int *n, int v)
{
	(RC_cache_invalidate(), *n = v);

        // Fool WCB by writing to a distinct cache line -> flush
        *fool_wcb = 42;
}

static int MPB_add(int *n, int v)
{
	// Make sure we don't read stale data
	int newval = (RC_cache_invalidate(), *n + v);

	// We need this second invalidation or otherwise the write would hit in L1
	(RC_cache_invalidate(), *n = newval);

	// Fool WCB by writing to a distinct cache line -> flush
	*fool_wcb = 42;
	
	return newval;
}

int RCCE_APP(int argc, char *argv[])
{
	int offset, *shared, i;
	t_vcharp p;

	dup2(STDOUT_FILENO, STDERR_FILENO);

	RCCE_init(&argc, &argv);

	p = RCCE_malloc(2 * RCCE_LINE_SIZE);
	offset = (long)p - (long)RCCE_comm_buffer[RCCE_ue()];
	shared = (int *)(RCCE_comm_buffer[0] + offset);
	fool_wcb = (int *)(RCCE_comm_buffer[0] + offset + RCCE_LINE_SIZE);

	WORKER(0) MPB_write(shared, 0);

	RCCE_barrier(&RCCE_COMM_WORLD);

	assert(MPB_read(shared) == 0);

	for (i = 0; i < RCCE_num_ues(); i++) {
		WORKER(i) {
			while (MPB_read(shared) != i) ;
			MPB_add(shared, 1);
		}
	}

	RCCE_barrier(&RCCE_COMM_WORLD);

	assert(MPB_read(shared) == RCCE_num_ues());

	RCCE_finalize();

	return 0;
}