MyDispatchOnce.h

typedef long my_dispatch_once_t;
typedef void (^my_dispatch_block_t)(void);

void my_dispatch_once(my_dispatch_once_t *predicate, my_dispatch_block_t block);

MyDispatchOnce.m

#include "MyDispatchOnce.h"
#import <libkern/OSAtomic.h>

#if defined(__i386__) || defined(__x86_64__)
#define my_hardware_pause() __asm__("pause")
#else
#define my_hardware_pause() __asm__("")
#endif

// this is a shameless copy of dispatch_once()
void my_dispatch_once(my_dispatch_once_t *predicate, my_dispatch_block_t block) {
	volatile long *val = predicate;
	if (OSAtomicCompareAndSwapLong(0l, 1l, val)) {
		block();
		OSMemoryBarrier();
		*val = ~0l;
	} else {
		do {
			// hardware pause
			my_hardware_pause();
		} while (*val != ~0l);
		OSMemoryBarrier();
	}
}