使用OpenMP并行化实现矩阵乘法

给定两个矩阵A和B，矩阵大小均为1024 * 1024，先使用单线程进行计算，代码如下：

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

#define ORDER 1024
#define AVAL 3.0
#define BVAL 5.0
#define TOL  0.001

int main(int argc, char *argv[])
{
    int Ndim = ORDER, Pdim = ORDER, Mdim = ORDER;
	double *A, *B, *C, tmp;
	double start, end;

   	A = (double *)malloc(Ndim * Pdim * sizeof(double));
    B = (double *)malloc(Pdim * Mdim * sizeof(double));
    C = (double *)malloc(Ndim * Mdim * sizeof(double));

	for (int i = 0; i < Ndim; i++)
		for (int j = 0; j < Pdim; j++)
			*(A + (i * Ndim + j)) = AVAL;

	for (int i = 0; i < Pdim; i++)
		for (int j = 0; j < Mdim; j++)
			*(B + (i * Pdim + j)) = BVAL;

	for (int i = 0; i < Ndim; i++)
		for (int j = 0; j < Mdim; j++)
			*(C + (i * Ndim + j)) = 0.0;

	start = omp_get_wtime();

    for (int i = 0; i < Ndim; i++)
    {
		for (int j = 0; j < Mdim; j++)
        {
			tmp = 0.0;
			for (int k = 0; k < Pdim; k++)
            {
				tmp += *(A + (i * Ndim + k)) * *(B + (k * Pdim + j));
			}
			*(C + (i * Ndim + j)) = tmp;
		}
	}

	end = omp_get_wtime();

	printf("Order %d multiplication in %f seconds \n", ORDER, end - start);

    free(A);
    free(B);
    free(C);
    return 0;
}

执行时间为 63.432267秒

 使用openmp再进行一次计算，代码如下：

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

#define ORDER 1024
#define AVAL 3.0
#define BVAL 5.0
#define TOL  0.001

int main(int argc, char *argv[])
{
    int Ndim = ORDER, Pdim = ORDER, Mdim = ORDER;
	double *A, *B, *C, tmp;
	double start, end;

	A = (double *)malloc(Ndim * Pdim * sizeof(double));
    B = (double *)malloc(Pdim * Mdim * sizeof(double));
    C = (double *)malloc(Ndim * Mdim * sizeof(double));

	for (int i = 0; i < Ndim; i++)
		for (int j = 0; j < Pdim; j++)
			*(A + (i * Ndim + j)) = AVAL;

	for (int i = 0; i < Pdim; i++)
		for (int j = 0; j < Mdim; j++)
			*(B + (i * Pdim + j)) = BVAL;

	for (int i = 0; i < Ndim; i++)
		for (int j = 0; j < Mdim; j++)
			*(C + (i * Ndim + j)) = 0.0;
	
	start = omp_get_wtime();

    #pragma omp parallel for
	for (int i = 0; i < Ndim; i++)
    {
		for (int j = 0; j < Mdim; j++)
        {
            double tmp = 0.0;

			for(int k = 0; k < Pdim; k++)
            {
				tmp += *(A + (i * Ndim + k)) * *(B + (k * Pdim + j));
			}
			*(C + (i * Ndim + j)) = tmp;
		}
	}

	end = omp_get_wtime();

	printf("Order %d multiplication in %f seconds \n", ORDER, end - start);

	free(A);
    free(B);
    free(C);
    return 0;
}

使用 4 线程计算后，执行时间为 19.173928秒