一个关于Memory Reordering的实验

Instruction Reordering有两种，包括Compiler Reordering和Memory Reordering。

Intel官方列出的有关Memory Reordering的情况总共有8种：

Neither Loads Nor Stores Are Reordered with Like Operations

Stores Are Not Reordered With Earlier Loads

Loads May Be Reordered with Earlier Stores to Different Locations

Intra-Processor Forwarding Is Allowed

Stores Are Transitively Visible

Stores Are Seen in a Consistent Order by Other Processors

Locked Instructions Have a Total Order

Loads and Stores Are Not Reordered with Locked Instructions



可以看出，第三点是会发生指令重排的情况。

下面做一个验证第三点的实验，参考《Memory Reordering Caught in the Act》一文，

原文链接：http://preshing.com/20120515/memory-reordering-caught-in-the-act

（注：原文采用的是windows平台，这里采用linux平台）

#define _GNU_SOURCE
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <pthread.h>
#include <semaphore.h>

sem_t beginSema1;
sem_t beginSema2;
sem_t endSema;

int X,Y;
int r1,r2;

void* thread1Func(void* param) {
	while (1) {
		sem_wait(&beginSema1);
		while ( (rand() / (double)RAND_MAX) > 0.2 ) ;
		X=1;
		__asm__ __volatile__("":::"memory");
		r1 = Y;
		sem_post(&endSema);
	}
	return NULL;
}

void* thread2Func(void* param) {
	while (1) {
		sem_wait(&beginSema2);
		while ( (rand() / (double)RAND_MAX) > 0.2 ) ;
		Y=1;
		__asm__ __volatile__("":::"memory");
		r2 = X;
		sem_post(&endSema);
	}
	return NULL;
}

int main() {
	sem_init(&beginSema1,0,0);
	sem_init(&beginSema2,0,0);
	sem_init(&endSema,0,0);

	pthread_t thread1,thread2;

	pthread_create(&thread1,NULL,thread1Func,NULL);
	pthread_create(&thread2,NULL,thread2Func,NULL);

	int detected = 0;
	int iterations = 0;
	for (iterations=1;;iterations++) {
		X=0;
		Y=0;
		sem_post(&beginSema1);
		sem_post(&beginSema2);
		sem_wait(&endSema);
		sem_wait(&endSema);
		if (r1 == 0 && r2 == 0) {
			detected++;
			printf("%d reorders detected after %d iterations\n", detected, iterations);
		}
	}
	return 0;
}

其中，__asm__ __volatile__("":::"memory") 是禁止编译器进行指令重排，保证了store操作和load操作在编译后的先后顺序。

可以发现，输出结果出现了 r1==0&&r2==0 的情况，证明CPU对指令进行了重排。

下面，再将__asm__ __volatile__("":::"memory") 改为 __asm__ __volatile__("mfence":::"memory")，强制使用strong ordering的模式，保证CPU不对该句前后的store和load操作进行重排：

#define _GNU_SOURCE
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <pthread.h>
#include <semaphore.h>

sem_t beginSema1;
sem_t beginSema2;
sem_t endSema;

int X,Y;
int r1,r2;

void* thread1Func(void* param) {
	while (1) {
		sem_wait(&beginSema1);
		while ( (rand() / (double)RAND_MAX) > 0.2 ) ;
		X=1;
		__asm__ __volatile__("mfence":::"memory");
		r1 = Y;
		sem_post(&endSema);
	}
	return NULL;
}

void* thread2Func(void* param) {
	while (1) {
		sem_wait(&beginSema2);
		while ( (rand() / (double)RAND_MAX) > 0.2 ) ;
		Y=1;
		__asm__ __volatile__("mfence":::"memory");
		r2 = X;
		sem_post(&endSema);
	}
	return NULL;
}

int main() {
	sem_init(&beginSema1,0,0);
	sem_init(&beginSema2,0,0);
	sem_init(&endSema,0,0);

	pthread_t thread1,thread2;

	pthread_create(&thread1,NULL,thread1Func,NULL);
	pthread_create(&thread2,NULL,thread2Func,NULL);

	int detected = 0;
	int iterations = 0;
	for (iterations=1;;iterations++) {
		X=0;
		Y=0;
		sem_post(&beginSema1);
		sem_post(&beginSema2);
		sem_wait(&endSema);
		sem_wait(&endSema);
		if (r1 == 0 && r2 == 0) {
			detected++;
			printf("%d reorders detected after %d iterations\n", detected, iterations);
		}
	}
	return 0;
}

可以发现，输出结果中没有了 r1==0&&r2==0 的情况。