一个编译优化问题（没解决）

初学多线程，写出了如下程序

#include <assert.h>
#include <stdint.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/time.h>
#include <pthread.h>
#include <semaphore.h>

#define num 1
sem_t p[num];
int finished1[num];
// int t= 0;


void *foo(void *args)
{
    int* a = (int *)args;

    printf("the number is :%d\n", *a);
    sem_wait(&p[*a]);
    printf("--finish: %d\n", finished1[*a]);
    int s = 1;
    for (int i = 1; i <= 1000; i ++)
    s = i * (i + 1) % 43;
        
    finished1[*a] = 1;
    printf("finish: %d\n", finished1[*a]);
    pthread_exit(0);
    
}

int main()
{
    int t= 0;
    for (int i = 0; i < num; i ++)  sem_init(&p[i], 0, 0);
    pthread_t th[1001];
    for (int i = 0; i < 1; i ++)  
    {
        pthread_create(&th[i], NULL, foo, (void *)&t);
    }


    for (int i = 0; i < num; i ++)
    {
        sem_post(&p[i]);
        while(finished1[i] == 0)
        {
        };
        finished1[i] = 0;
        printf("--%d--\n", i);
    }
    return 0;
}

发现如果用O1/O2优化的话,程序不能终止
不优化程序能正常运行

目前太菜了，找不出问题，在此记录…