CUDA--1.计算2*2矩阵并行运算

1.第一次接触CUDA，还有很多不清楚的地方，此程序为计算2*2矩阵并行运算

 

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>
#include "assert.h" 

//cudaError_t addWithCuda(int c[][2],  int a[][2],  int b[][2], unsigned int size);

/*
 1.cpu 与 gpu 一起在运行程序, cpu负责调度, gpu 负责运算, cpu称为HOST , gpu 称为 DEVICE
 2.grid 包含多个 block , block 包含多个 thread
 3.例如"AAAABBBBCCCCDDDDEEEE",ABCDE各代表一个block, 总的为一个Grid, 每个block中有四个thread,
 若此时指向第一个B，则表示第一个block中的第0个thread
 则：
	i = blockDim.x * blockIdx.x + threadIdx.x    
 其中 ：
	blockDim.x 表示block的size行数，当前例子是4  (如果是一维的block的话,即一行有多少个thread)
	blockIdx.x 表示当前运行到的第几个block  ，当前例子是1 (一维grid的话,即该grid中第几个block)
	threadIdx.x 表示当前运行到的第几个thread ，当前例子是0(一维的block的话.即该block中第几个thread)
	所以  i=4*1+0
4.block可以是1，2，3维。根据你用dim3申明时来确定，比如dim3 blockSize(5); 那么就是1维的，其实就是blockSize(5,1,1)，
  后面没写出来的默认为1了。 dim3 blockSize(5,4);就是2维的，类推。

*/

__global__ void addKernel(int a[][2], int b[][2], int c[][2])
{

	printf("我是：threadIdx.x=%d threadIdx.y=%d threadIdx.z=%d || blockIdx.x=%d blockIdx.y=%d blockIdx.z=%d\n"
			, threadIdx.x, threadIdx.y, threadIdx.z
			, blockIdx.x, blockIdx.y, blockIdx.z);
	printf("blockDim.x=%d blockDim.y=%d blockDim.z=%d\n", blockDim.x, blockDim.y, blockDim.z);
	printf("gridDim.x=%d gridDim.y=%d gridDim.z=%d\n", gridDim.x, gridDim.y, gridDim.z);


	c[0][0] = 40; c[0][1] = 40; c[1][0] = 40; c[1][1] = 40;
	
	

	int i = threadIdx.x + blockIdx.x * blockDim.x;
	int j = threadIdx.y + blockIdx.y * blockDim.y;

	if (i < 2 && j < 2) {

		c[i][j] = a[i][j] + b[i][j];

	}
    
}

int main()
{
	//host端申请内存
	int(*a)[2] = new int[2][2];
	int(*b)[2] = new int[2][2];
	int(*c)[2] = new int[2][2];
	

	a[0][0] = 1; a[0][1] = 2; a[1][0] = 3; a[1][1] = 4;
	b[0][0] = 1; b[0][1] = 2; b[1][0] = 3; b[1][1] = 4;
	

	cudaError_t error = cudaSuccess;//断言函数
	//int aqqq = 1 + 2;
	//assert(aqqq == 3);

	int(*device_a)[2], (*device_b)[2], (*device_c)[2];

	//device中申请显存
	error = cudaMalloc((void **)&device_a, sizeof(int) * 4);
	error = cudaMalloc((void **)&device_b, sizeof(int) * 4);
	error = cudaMalloc((void **)&device_c, sizeof(int) * 4);

	//host端内存copy到device
	cudaMemcpy(device_a, a, sizeof(int) * 4, cudaMemcpyHostToDevice);
	cudaMemcpy(device_b, b, sizeof(int) * 4, cudaMemcpyHostToDevice);
	
	//启动kernel函数
	//dim3 threadsPerBlock(1, 1);
	//dim3 numBlocks(2 / threadsPerBlock.x, 2 / threadsPerBlock.y);
	//dim3是NVIDIA的CUDA编程中一种自定义的整型向量类型,dim3类型最终设置的是一个三维向量，三维参数分别为x,y,z;
	dim3 blocks(1,1,1);//一个grid中block数量
	dim3 threads(2,2,1);//一个block中的thread数量
	//addKernel << < blocks, threads >> > (device_a, device_b, device_c);

	//dim3 grid_size(2, 2, 1);//一个grid中blocks的分布和数量
	//dim3 block_size(1, 1, 1);//一个block中的threads的分布和数量
	//dim3 threads_size(1, 1, 1);
	addKernel <<<  blocks, threads >>> (device_a, device_b, device_c);


	cudaMemcpy(c, device_c, sizeof(int) * 4, cudaMemcpyDeviceToHost);

	for (int i = 0; i < 2; i++)
	{
		for (int j = 0; j < 2; j++)
		{
			printf("%10d", c[i][j]);
		}
		printf("\n");
	}

	return 0;
	
}