如何正确结合#pragma omp task与simd实现矩形多指标并行计算

我想要编写一个并行程序来计算多个矩形的周长、面积和对角线，计划采用函数级并行的方式实现——将周长、面积和对角线的计算过程并行化处理。目前我在程序中使用#pragma omp for将循环迭代分配给不同线程，并用#pragma omp simd实现计算向量化，但不知道如何正确结合代表非结构化函数级并行的#pragma omp task指令。我清楚当前代码的写法会对任务创建进行向量化，这对SIMD架构来说毫无意义，希望有人能解释如何正确使用#pragma omp for、simd和task指令，以下是我的代码：

#include<stdio.h>
#include<stdlib.h>
#include<omp.h>
#include<math.h>
#include<time.h>

#define NUM_THREADS 4

/**
 *  该程序计算矩形的三个参数：
 *  1. 周长 ==>    P = 2B + 2H
 *  2. 面积      ==>    A = BH
 *  3. 对角线  ==>  D² = B² + H²
 * 
 *  必须使用以下4个指令：parallel, for, simd, task
 */

enum {
    BASE = 0,
    HEIGHT = 1,
};


#pragma omp declare simd
double compute_perimeter(double base, double height){ return (2 * base + 2 * height); }

#pragma omp declare simd
double compute_area(double base, double height){ return (base * height); }

#pragma omp declare simd
double compute_diagonal(double base, double height){ return (sqrt((base * base) + (height * height))); }

double ** generate_rectangles(size_t n){
    double ** rectangles = (double **)malloc(n * sizeof(double *));

    for(size_t i = 0; i < n; ++i){
        rectangles[i] = (double *)malloc(n * sizeof(double));
        for(size_t j = 0; j < n; ++j) rectangles[i][j] = rand() % 100;
    }  

    return rectangles;
}

void destroy_rectangles(double ** rectangles, size_t n){
    for(size_t i = 0; i < n; ++i) free(rectangles[i]);
    free(rectangles);
}

int main(){

    srand(time(NULL));

    size_t n = rand() % 100000;

    double * perimeters = (double *)malloc(n * sizeof(double));
    double * areas = (double *)malloc(n * sizeof(double));
    double * diagonals = (double *)malloc(n * sizeof(double)); 

    double ** rectangles = generate_rectangles(n);


    #pragma omp parallel for simd num_threads(NUM_THREADS) schedule(static)
    for(size_t i = 0; i < n; ++i){
        #pragma omp task
        {perimeters[i] = compute_perimeter(rectangles[i][BASE], rectangles[i][HEIGHT]);}
        #pragma omp task
        {areas[i] = compute_area(rectangles[i][BASE], rectangles[i][HEIGHT]);}
        #pragma omp task
        {diagonals[i] = compute_diagonal(rectangles[i][BASE], rectangles[i][HEIGHT]);}
    }

    for(size_t i = 0; i < n; ++i) printf("矩形 %d:\n 底 = %lf, 高 = %lf, 周长 = %lf, 面积 = %lf, 对角线 = %lf\n\n", i, rectangles[i][BASE], rectangles[i][HEIGHT], perimeters[i], areas[i], diagonals[i]);

    free(perimeters);
    free(areas);
    free(diagonals);
    destroy_rectangles(rectangles, n);

    return 0;
}

内容的提问来源于stack exchange，提问作者Matheus Sousa