/*使用二维的grid和block将4*4的矩阵块划分为4个2*2的子阵,即棋盘划分*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <math.h>
#include <cutil.h>
#define SDATA(index) CUT_BANK_CHECKER(sdata,index)
__global__ void testKernel(float* g_idata,float* g_odata,int width,int height)
{
//shared memory没有extern 前缀,静态分配空间,因此[]中必须指定为数组分配的大小
__shared__ float sdata[4];
//计算线程索引
unsigned int bid_in_grid=__mul24(blockIdx.y,gridDim.x)+blockIdx.x;
unsigned int tid_in_block=__mul24(blockIdx.y,gridDim.x)+threadIdx.x;
unsigned int tid_in_grid_x=__mul24(blockDim.x,blockIdx.x)+threadIdx.x;
unsigned int tid_in_grid_y=__mul24(blockDim.y,blockIdx.y)+threadIdx.y;
unsigned int tid_in_grid=__mul24(tid_in_grid_y,width)+tid_in_grid_x;
//从global memory中读入数据。host模拟运行时使用,bank checker检查是否产生bank conflict
SDATA(tid_in_block)=g_idata[tid_in_grid];
__syncthreads();
//进行运算
SDATA(tid_in_block)=(float)bid_in_grid*SDATA(tid_in_block);
__syncthreads();
//将数据写回global memory
g_odata[tid_in_grid]=SDATA(tid_in_block);
}
__host__ void runTest(int argc,char** argv)
{
CUT_DEVICE_INIT(argc,argv);//启动CUDA
//数据大小,这里用每一个线程计算计算一个单精度浮点数
unsigned int mem_size=sizeof(float)*4*4;
//在host端分配内存,注意这里分配的是pinned memory
float* h_idata;
CUDA_SAFE_CALL(cudaMallocHost((void**)&h_idata,mem_size));
//初始化线程中的值
for(unsigned int i=0;i<4;i++)
for(unsigned int j=0;j<4;j++)
h_idata[i*4+j]=2.0f;
//在device端分配显存
float* d_idata;
CUDA_SAFE_CALL(cudaMalloc((void**)&d_idata,mem_size));
//将内存中的值读入显存
CUDA_SAFE_CALL(cudaMemcpy(d_idata,h_idata,mem_size,cudaMemcpyHostToDevice));
//在device端分配显存,用于存储结果
float* d_odata;
CUDA_SAFE_CALL(cudaMalloc((void**)&d_odata,mem_size));
//设置运行参数,即网格的形状和线程块的形状
dim3 grid(2,2,1);
dim3 threads(2,2,1);
//****************运行内核函数,调用调和GPU进行运算
testKernel<<<grid,threads>>>(d_idata,d_odata,4,4);
//检查GPU 是否正常运行
CUT_CHECK_ERROR("Kernel execution failed");
//
float* h_odata;
CUDA_SAFE_CALL(cudaMallocHost((void**)&h_odata,mem_size));
//
CUDA_SAFE_CALL(cudaMemcpy(h_odata,d_odata,mem_size,cudaMemcpyDeviceToHost));
//
for(unsigned int i=0;i<4;i++)
{
for(unsigned int j=0;j<4;j++)
{
printf("%5.0f",h_odata[i*4+j]);
}
printf("\n");
}
//释放存储器
CUDA_SAFE_CALL(cudaFreeHost(h_idata));
CUDA_SAFE_CALL(cudaFreeHost(h_odata));
CUDA_SAFE_CALL(cudaFree(d_idata));
CUDA_SAFE_CALL(cudaFree(d_odata));
}
//主函数
int main(int argc,char** argv)
{
runTest(argc,argv);
CUT_EXIT(argc,argv);// 退出CUDA
}