请完成代码中的my_lock和my_unlock两个函数，并在代码其他部分加入适当的代码，使这个程序有稳定的输出： my_lock和my_unlock两个函数的实现需要采用两种方式，一种方式是直接使用互斥锁。这种方式实现的程序命名为count_mutex.c。 另一种方式是采用自旋锁。Pthread库中没有提供自旋锁，需要你自己实现。实现自旋锁的硬件支持是test_and_set指令，以下代码可以看做是test_and_set指令，实际上它就是一个函数，调用test_and_set指令：

下面是使用互斥锁实现的count_mutex.c代码：

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>

#define NUM_THREADS 4
#define COUNT_TARGET 10000000

int count = 0;
pthread_mutex_t count_mutex;

void *increment_count(void *thread_id) {
    long tid = (long) thread_id;
    int i;
    for (i = 0; i < COUNT_TARGET; i++) {
        pthread_mutex_lock(&count_mutex);
        count++;
        pthread_mutex_unlock(&count_mutex);
    }
    printf("Thread %ld finished. Count=%d\n", tid, count);
    pthread_exit(NULL);
}

int main() {
    pthread_t threads[NUM_THREADS];
    int rc;
    long t;
    pthread_mutex_init(&count_mutex, NULL);
    for (t = 0; t < NUM_THREADS; t++) {
        rc = pthread_create(&threads[t], NULL, increment_count, (void *) t);
        if (rc) {
            printf("ERROR: return code from pthread_create() is %d\n", rc);
            exit(-1);
        }
    }
    for (t = 0; t < NUM_THREADS; t++) {
        pthread_join(threads[t], NULL);
    }
    printf("Final count=%d\n", count);
    pthread_mutex_destroy(&count_mutex);
    pthread_exit(NULL);
}

下面是使用自旋锁实现的代码：

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>

#define NUM_THREADS 4
#define COUNT_TARGET 10000000

int count = 0;
int lock = 0;

int test_and_set(int *lock) {
    int old_value;
    asm volatile("lock; xchgl %0, %1"
                 : "=r"(old_value), "=m"(*lock)
                 : "0"(1), "m"(*lock)
                 : "memory");
    return old_value;
}

void my_lock(int *lock) {
    while (test_and_set(lock) == 1) {
        // 自旋
    }
}

void my_unlock(int *lock) {
    *lock = 0;
}

void *increment_count(void *thread_id) {
    long tid = (long) thread_id;
    int i;
    for (i = 0; i < COUNT_TARGET; i++) {
        my_lock(&lock);
        count++;
        my_unlock(&lock);
    }
    printf("Thread %ld finished. Count=%d\n", tid, count);
    pthread_exit(NULL);
}

int main() {
    pthread_t threads[NUM_THREADS];
    int rc;
    long t;
    for (t = 0; t < NUM_THREADS; t++) {
        rc = pthread_create(&threads[t], NULL, increment_count, (void *) t);
        if (rc) {
            printf("ERROR: return code from pthread_create() is %d\n", rc);
            exit(-1);
        }
    }
    for (t = 0; t < NUM_THREADS; t++) {
        pthread_join(threads[t], NULL);
    }
    printf("Final count=%d\n", count);
    pthread_exit(NULL);
}