cpu与gpu实现矩阵相乘对比

1、完成矩阵相乘的并行程序的实现
要求：实现2个矩阵（1024*1024）的相乘。数据类型设置为float。
（1） 使用CPU计算;

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <iostream>
int main()
{
    //定义矩阵的长度
    int Ndim = 1024, Mdim = 1024,Pdim=1024;
    int szA = Ndim * Pdim;
    int szB = Pdim * Mdim;
    int szC = Ndim * Mdim;

    float* A;
    float* B;
    float* C;

    A = (float*)malloc(szA * sizeof(float));
    B = (float*)malloc(szB * sizeof(float));
    C = (float*)malloc(szC * sizeof(float));
    int i, j,k;
    float tmp;
    //初始化矩阵，可加学号
    for (i = 0; i < szA; i++)
        A[i] = 8;
    for (i = 0; i < szB; i++)
        B[i] = 3;


    //实现矩阵相乘
    for (i = 0; i < Ndim; i++)
    {
        for(j=0;j<Mdim;j++) {
            tmp = 0.0;
            for (k = 0; k < Pdim; k++)
                tmp += A[i * Pdim + k] * B[k * Mdim + j];
            C[i * Mdim + j] = tmp;
        }

    }
   
    printf("\nArray C:\n");
    for (i = 0; i < Ndim; i++) {
        for (j = 0; j < Mdim; j++)
            printf("%.1f\t", C[i * Mdim + j]);
        printf("\n");
    }
    if (A)
        free(A);
    if (B)
        free(B);
    if (C)
        free(C);
    system("pause");
    return 0;

}

本地cpu上运行结果如下：

（2） 使用GPU全局内存存放输入的矩阵M和N，P矩阵的结果存放于全局内存；

#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <stdio.h>
#include <stdlib.h>
#define index 1024
# define TILE_WIDTH 2

__global__ void  calcSum(float* AA, float* BB,float* CC,int Width)
{
	int Row = blockIdx.y * blockDim.y + threadIdx.y;
	int Col = blockIdx.x * blockDim.x + threadIdx.x;
	if ((Row < Width) && (Col < Width)) {
		float Pvalue = 0;
		// 每一个线程处理输出数据d_P的一个元素
		for (int k = 0; k < Width; ++k)
			Pvalue += AA[Row * Width + k] * BB[k * Width + Col];
		CC[Row * Width + Col] = Pvalue;
	}
}

int main()
{
	cudaError_t cudaStatus = cudaSuccess;
	//初始化cpu矩阵
	int Ndim=0,Mdim=0 , Pdim=0 ,Width=0;
	Ndim = Mdim = Pdim = Width = index;
	int szA = Ndim * Pdim;
	int szB = Pdim * Mdim;
	int szC = Ndim * Mdim;
	float* A,*AA;
	float* B,*BB;
	float* C,*CC;
	A = (float*)malloc(szA * sizeof(float));
	B = (float*)malloc(szB * sizeof(float));
	C = (float*)malloc(szC * sizeof(float));
	int i, j;
	for (i = 0; i < szA; i++)
		A[i] = 8;
	for (i = 0; i < szB; i++)
		B[i] = 3;

	cudaStatus = cudaMalloc((void**)&AA, szA * sizeof(float));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc1 failed!");
	}
	cudaStatus = cudaMalloc((void**)&BB, szB* sizeof(float));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc2 failed!");
	}
	cudaStatus = cudaMalloc((void**)&CC, szC * sizeof(float));
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMalloc3 failed!");
	}

	// Copy input vectors from host memory to GPU buffers.
	cudaStatus = cudaMemcpy(AA, A, szA * sizeof(float), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy1 failed!");
	}
	cudaStatus = cudaMemcpy(BB, B, szB * sizeof(float), cudaMemcpyHostToDevice);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy2 failed!");
	}

	// 线程配置
	// TILE_WIDTH 是一个用“#define”定义的常量
	dim3 dimGrid(Width / TILE_WIDTH, Width / TILE_WIDTH, 1);
	dim3 dimBlock(TILE_WIDTH, TILE_WIDTH, 1);
	calcSum << <dimGrid, dimBlock >> > (AA, BB, CC, Width);

	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "calcSum failed!");
		return 1;
	}
	// cudaDeviceSynchronize waits for the kernel to finish, and returns
	// any errors encountered during the launch.
	cudaStatus = cudaDeviceSynchronize();
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching addKernel!\n", cudaStatus);
	}

	// Copy output vector from GPU buffer to host memory.
	cudaStatus = cudaMemcpy(C, CC, szC * sizeof(float), cudaMemcpyDeviceToHost);
	if (cudaStatus != cudaSuccess) {
		fprintf(stderr, "cudaMemcpy failed!");
	}
	//打印
	printf("\nArray C:\n");
	for (i = 0; i < Ndim; i++) {
		for (j = 0; j < Mdim; j++)
			printf("%.1f\t", C[i * Mdim + j]);
		printf("\n");
	}
	cudaFree(AA);
	cudaFree(BB);
	cudaFree(CC);
	free(A);
	free(B);
	free(C);

	return 0;
}

服务器上gpu矩阵相乘结果：