两个矩阵相加

#include<CL/cl.h>

#include<stdlib.h>

#include<stdio.h>

const char *KernelSource=

"__kernel void add(__global int *A,__global int *B,__global int *C)"

"{"

"    int i=get_global_id(0);"

"    C[i]=A[i]+B[i]; "

"}";

//自定义一个查错函数

void CheckErr(cl_int err, char *information)

{

if(err!=CL_SUCCESS)

{

printf("Error of %s\n",information);

system("pause");

exit(0);

}

}

int main()

{

const int BUFSIZE=20;

       size_t DataSize=sizeof(int)*BUFSIZE;

int MatrixA[BUFSIZE],  

             MatrixB[BUFSIZE], 

             MatrixC[BUFSIZE];

int i;

//为矩阵A随机赋值,并将其输出

for(i=0;i<BUFSIZE;i++)

{

MatrixA[i]=i;

printf("%d ",MatrixA[i]);

}

printf("\n");

//为矩阵B随机赋值，并将其输出

for(i=0;i<BUFSIZE;i++)

{

MatrixB[i]=20-i;

printf("%d ",MatrixB[i]);

}

printf("\n");

printf("CPU Result:\n");

//A与B实现相加存于C中并输出

for(i=0;i<BUFSIZE;i++)

{

MatrixC[i]=MatrixA[i]+MatrixB[i];

printf("%d ",MatrixC[i]);

}

printf("\n");

cl_int err;

//第一步构建OpenCL平台：

cl_platform_id platform;

err=clGetPlatformIDs(1,&platform,NULL);

CheckErr(err,"Platform");

//第二步?查询设备：

cl_device_id device;

err=clGetDeviceIDs(platform,CL_DEVICE_TYPE_GPU,1,&device,NULL);

CheckErr(err,"Device");

//第三步设备上下文：

cl_context context;

context=clCreateContext(NULL,1,&device,NULL,NULL,&err);

CheckErr(err,"Context");

//第四步创建命令队列

cl_command_queue command_queue;

command_queue=clCreateCommandQueue(context,device,CL_QUEUE_PROFILING_ENABLE,&err);

CheckErr(err,"Command_queue");

//第五步创建缓冲

cl_mem MatrixBufferA;

cl_mem MatrixBufferB;

cl_mem MatrixBufferC;

int C[BUFSIZE];

MatrixBufferA=clCreateBuffer(context,CL_MEM_READ_ONLY,DataSize,NULL,&err);

CheckErr(err,"MatrixBufferA");

MatrixBufferB=clCreateBuffer(context,CL_MEM_READ_ONLY,DataSize,NULL,&err);

CheckErr(err,"MatrixBufferB");

MatrixBufferC=clCreateBuffer(context,CL_MEM_WRITE_ONLY,DataSize,NULL,&err);

CheckErr(err,"MatrixBufferC");

//第六步数据拷贝（主机到设备）

err=clEnqueueWriteBuffer(command_queue,MatrixBufferA,CL_FALSE,0,DataSize,MatrixA,0,NULL,NULL);

CheckErr(err,"Copy Data To BufferA");

err=clEnqueueWriteBuffer(command_queue,MatrixBufferB,CL_FALSE,0,DataSize,MatrixB,0,NULL,NULL);

CheckErr(err,"Copy Data To BufferB");

//第7步编译程序

cl_program program;

program=clCreateProgramWithSource(context,1,(const char **)&KernelSource,NULL,&err);

CheckErr(err,"CreateProgramWithSource");

err=clBuildProgram(program,1,&device,NULL,NULL,NULL);

CheckErr(err,"BulidProgram");

//第8步编译内核

cl_kernel kernel;

kernel=clCreateKernel(program,"add",&err);

CheckErr(err,"CreateKernel");

//第9步设置内核参数

err=clSetKernelArg(kernel,0,sizeof(cl_mem),&MatrixBufferA);

CheckErr(err,"SetKernelArg of MatrixBufferA");

err |=clSetKernelArg(kernel,1,sizeof(cl_mem),&MatrixBufferB);

CheckErr(err,"SetKernelArg of MatrixBufferB");

err |=clSetKernelArg(kernel,2,sizeof(cl_mem),&MatrixBufferC);

CheckErr(err,"SetKernelArg of MatrixBufferC");

//第10步配置工作线程

size_t globalWorkSize;

globalWorkSize=BUFSIZE;

//第11步内核执行队列

err=clEnqueueNDRangeKernel(command_queue,kernel,1,NULL,&globalWorkSize,NULL,0,NULL,NULL);

CheckErr(err,"EnqueueNDRangeKernel");

//第12步数据拷贝（设备到主机）

err=clEnqueueReadBuffer(command_queue,MatrixBufferC,CL_TRUE,0,DataSize,C,NULL,NULL,NULL);

printf("\nGPU Result:\n");

for(i=0;i<BUFSIZE;i++)

{

printf("%d ",C[i]);

}

printf("\n");

//第13步释放资源

clReleaseKernel(kernel);

clReleaseProgram(program);

clReleaseCommandQueue(command_queue);

clReleaseMemObject(MatrixBufferA);

clReleaseMemObject(MatrixBufferB);

clReleaseMemObject(MatrixBufferC);--

clReleaseContext(context);

clReleaseDevice(device);

}