OpenMP 并行堆内存释放问题

Question

malloc函数声明一个二维动态数组A，使用omp并行对A赋值，然后释放A内存时报错。在不使用并行时，不会报错。谢谢！代码如下：
void main(){

void main(){

    int i,j, Njuncs=1827;
    //创建二维数组A
    double**A = (double **)malloc(sizeof(double *)*(Njuncs + 1)); 
    for (i = 0; i < Njuncs + 1; i++)
        *(A + i) = (double  *)malloc(sizeof(double) * (Njuncs + 1)); 

#pragma omp parallel
    {
#pragma omp for
        for (i = 0; i <= Njuncs; i++)
            for (j = 0; j <= Njuncs; j++)
                A[i][j] = 1;
    }
    for (i = 0; i < Njuncs + 1; i++){
        free(*(A + i));//该步报错
        A[i] = NULL;
    }
    free(A);
}

Answer 1

其实并行多重循环这么写肯定是有问题的，内部循环变量j是共享的，被多个线程同时修改，那么就很有可能超出Njuncs，发生内存越界写入，**切记循环内的变量写入是并行的**。
应该把 j 声明为并行体中私有的，这样结果就正确了，参考如下代码：

#include <cstdlib>

int main(int argc, char* argv[]) {
    int i, j, Njuncs = 1827;
    //创建二维数组A
    double** A = (double**)malloc(sizeof(double*) * (Njuncs + 1));
    for (i = 0; i < Njuncs + 1; i++)
        *(A + i) = (double*)malloc(sizeof(double) * (Njuncs + 1));

#pragma omp parallel
    {
#pragma omp for private(j)
        for (i = 0; i <= Njuncs; i++)
            for (j = 0; j <= Njuncs; j++) {
                A[i][j] = 1;
            }
    }

#pragma omp parallel for
    for (i = 0; i < Njuncs + 1; i++) {
        free(*(A + i));//该步报错
        A[i] = NULL;
    }
    free(A);

    return 0;
}