Question

我试图写一个矢量的超级简单的C程序multiply-add＆＃34; axpy＆＃34;整数数据类型的算法。程序输出执行时间以测量机器的性能。矩阵由随机数填充。

int benchmark(void) {
    int N;      /* The matrix size, controlled by user input */
    int r, c;   /* Row and Column number */
    int random; /* Random number to fill the matix */
    int a = rand() % 20; /* Scale number to multiply x matrix */

    printf("Enter the size(N*N) of the matrices(Maximum 1,000,000)\n");
    scanf("%d", &N);

    if (N > 1000000) {
        fprintf(stderr, "Size of matrix is too large!\n");
        return 0;
    }

    /* Initialize and fill the matrix x and y */
    int xMatrix[N][N], yMatrix[N][N], resultMatrix[N][N];

    /* Compute time */
    clock_t t;

    t = clock();

    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            random = rand() % 100;
            xMatrix[r][c] = a * random; /* Multiply matrix x with random value a */
        }
    }
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            int random = rand() % 100;
            yMatrix[r][c] = random;
        }
    }

    /* Add two matrix together */
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            resultMatrix[r][c] = xMatrix[r][c] + yMatrix[r][c];
        }
    }

    t = clock() - t;

    double timeTaken = ((double)t) / CLOCKS_PER_SEC;
    printf("\n -> Total time : %f seconds\n", timeTaken);
    printf("\n -> Vector length : %d", N * N);

}

用户控制矩阵的大小。当N的值小于800时，该程序正常工作。

Answer 1

使用自动存储（在堆栈）上分配的对象的大小太大，您会得到未定义的行为，更具体地说是堆栈溢出。

您应该从堆中分配对象：

    /* Initialize and fill the matix x and y */
    int (*xMatrix)[N] = malloc(N * sizeof(*xMatrix));
    int (*yMatrix)[N] = malloc(N * sizeof(*yMatrix));
    int (*resultMatrix)[N] = malloc(N * sizeof(*resultMatrix));

并验证malloc()返回的指针均不是NULL。

以下是修改后的代码：

int benchmark(void) {
    int N;       /* The matrix size, controlled by user input */
    int r, c;    /* Row and Column number */
    int random;  /* Random number to fill the matix */
    int a = rand() % 20; /* Scale number to multiply x matrix */

    printf("Enter the size(N*N) of the matrices (Maximum 1,000,000)\n");
    if (scanf("%d", &N) != 1) {
        fprintf(stderr, "Input error!\n");
        return 0;
    }

    if (N > 1000000) {
        fprintf(stderr, "Matrix size is too large!\n");
        return 0;
    }

    /* Initialize and fill the matrix x and y */
    int (*xMatrix)[N] = malloc(N * sizeof(*xMatrix));
    int (*yMatrix)[N] = malloc(N * sizeof(*yMatrix));
    int (*resultMatrix)[N] = malloc(N * sizeof(*resultMatrix));

    if (xMatrix == NULL || yMatrix == NULL || resultMatrix == NULL) {
        fprintf(stderr, "Memory allocation failed!\n");
        free(xMatrix);
        free(yMatrix);
        free(resultMatrix);
        return 0;
    }

    /* Compute time */
    clock_t t = clock();

    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            random = rand() % 100;
            xMatrix[r][c] = a * random; /* Multiply matrix x with random value a */
        }
    }
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            random = rand() % 100;
            yMatrix[r][c] = random;
        }
    }

    /* Add two matrix together */
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            resultMatrix[r][c] = xMatrix[r][c] + yMatrix[r][c];
        }
    }

    t = clock() - t;

    double timeTaken = ((double)t) / CLOCKS_PER_SEC;
    printf("\n -> Total time : %f seconds\n", timeTaken);
    printf("\n -> Vector length : %lld", (long long)N * N);

    free(xMatrix);
    free(yMatrix);
    free(resultMatrix);
    return 0;
}

但请注意，您的计算非常简单，大部分时间都可能花在rand()函数上。

Answer 2

您正在尝试动态分配memmory，我建议您使用 stdlib.h 中的 malloc ，如下所示。

另外，请查看以下SO帖子：memory allocation in Stack and Heap和What and where are the stack and heap?

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

int benchmark(void) {
    int N;  /* The matrix size, controlled by user input */
    int r, c; /* Row and Column number */
    int random; /* Random number to fill the matix */
    int a = rand() % 20; /* Scale number to multiply x matrix */

    printf("Enter the size(N*N) of the matrixs(Maximum 1,000,000)\n");
    scanf("%d", &N);

    if(N > 1000000) {
        fprintf(stderr, "Size of matrix is too large!\n");
        return 0;
    }

    /* Initialize and fill the matix x and y */
    int** xMatrix = NULL;
    int** yMatrix = NULL;
    int** resultMatrix = NULL;

    /* Using the heap memory allocation instead of the stack */
    xMatrix = (int **) malloc(N * sizeof(int *));
    yMatrix = (int **) malloc(N * sizeof(int *));
    resultMatrix = (int **) malloc(N * sizeof(int *));
    for (r = 0; r < N; r++) {
        xMatrix[r] = (int *) malloc(N * sizeof(int));
        yMatrix[r] = (int *) malloc(N * sizeof(int));
        resultMatrix[r] = (int *) malloc(N * sizeof(int));
    }

    /* Compute time */
    clock_t t;

    t = clock();

    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            random = rand() % 100;
            xMatrix[r][c] = a * random; /* Multiply matix x with random value a */
        }
    }
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            int random = rand() % 100;
            yMatrix[r][c] = random;
        }
    }

    /* Add two matrix together */
    for (r = 0; r < N; r++) {
        for (c = 0; c < N; c++) {
            resultMatrix[r][c] = xMatrix[r][c] + yMatrix[r][c];
        }
    }

    t = clock() - t;

    double timeTaken = ((double)t)/CLOCKS_PER_SEC;
    printf("\n -> Total time : %f seconds\n", timeTaken);
    printf("\n -> Vector length : %d", N*N);

    /* Always remember to free your allocated memory */
    for (r = 0; r < N; r++) {
        free(xMatrix[r]);
        free(yMatrix[r]);
        free(resultMatrix[r]);
    }
    free(xMatrix);
    free(yMatrix);
    free(resultMatrix);

}

int main() {
    benchmark();
    return 0;
}

两个矩阵大小超过800 * 800时出现分段错误

2 个答案: