在NVIDIA GPU 上開發 openCL

在網上查瞭如何在 NVIDIA GPU上來運行openCL ， 結果大部分都是介紹在AMD GPU上開發的。所以在此寫篇簡單例子，介紹在NVIDIA GPU 上開發。

NVIDIA的顯卡上，cuda裝好opencl就自然有了，所以裝cuda=裝opencl. 【CUDA安裝見 CUDA安裝文章】

以下代碼是 功能是實現 向量加： C=A+B ，代碼是參照一本介紹在AMD GPU上開發的openCL教程（具體記不清了），在此我略微有添加改動的等。

系統：redhat Linux

GPU：NVIDIA K40

軟件環境：CUDA-6.0

一：此程序有兩個文件 1.vecadd.c      2 add.cl

vecadd.c

#include<stdio.h>
#include<CL/cl.h>
#include<stdlib.h>
#include<time.h>
#include<string.h>
#include<iostream>
#include<fstream>

using namespace std;

#define NWITEMS 262144  //2^18=262144

//read a text file into a  string
int convertToString(const char *filename, std::string& s)
{
        size_t size;
        char* str;
        std::fstream f(filename,(std::fstream::in | std::fstream::binary));

        if(f.is_open()){
                size_t filesize;
                f.seekg(0,std::fstream::end);
                size=filesize=(size_t)f.tellg();
                f.seekg(0,std::fstream::beg);

                str = new char[size+1];

                if(!str){
                        f.close();
                        return -1;
                }
                f.read(str,filesize);
                f.close();
                str[size]='\0';
                s=str;

                delete[] str;
                return 0;
 }
        printf("Error: Failed to open file %s\n",filename);
        return 1;
}
int main()
{
        float* buff1=NULL;
        float* buff2=NULL;
        float* buff=NULL;

        buff1=(float*)malloc(NWITEMS*sizeof(float));
        buff2=(float*)malloc(NWITEMS*sizeof(float));
        buff=(float*)malloc(NWITEMS*sizeof(float));

        int i;
        srand((unsigned)time(NULL));
        for(i=0;i<NWITEMS;i++){
                buff1[1]=rand()%RAND_MAX;
        }
        srand((unsigned)time(NULL)+1000);
        for(i=0;i<NWITEMS;i++){
                buff2[1]=rand()%RAND_MAX;
        }
        for(i=0;i<NWITEMS;i++){
                buff[i]=buff1[i]+buff2[i];
        }

        /************openCL begain*******/
        cl_int status=0;
        size_t deviceListSize;

        cl_uint numPlatforms;
        cl_platform_id platform=NULL;

        /**********get platform*************/
        status=clGetPlatformIDs(0,NULL,&numPlatforms);//get number of platform
        //printf("status=%d\nnumPlatforms=%d\n",status,numPlatforms);

        status=clGetPlatformIDs(1,&platform,NULL);//choose NV platform


        /**********************get device form NV platform**************/
        cl_uint numdevices;
        cl_device_id device=NULL;

        clGetDeviceIDs(platform,CL_DEVICE_TYPE_GPU,0,NULL,&numdevices);//get device number of NV platform
        printf("numDevice=%d\n",numdevices);

        clGetDeviceIDs(platform,CL_DEVICE_TYPE_GPU,1,&device,NULL);//choose a  NV GPU  device

        cl_context context = clCreateContext(NULL,1,&device,NULL,NULL,NULL); //create Context;

        cl_command_queue queue = clCreateCommandQueue(context,device,CL_QUEUE_PROFILING_ENABLE,NULL);//create commmand queue and enable profiling

        /********************create openCL buffer object, and then copy buff1 to clbuff1, buf2 copy to clbuff2*****************/
        cl_mem clbuf1 = clCreateBuffer(context,CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,NWITEMS*sizeof(cl_float),buff1,NULL);

        cl_mem clbuf2 = clCreateBuffer(context,CL_MEM_READ_ONLY,NWITEMS*sizeof(cl_float),NULL,NULL);
        status = clEnqueueWriteBuffer(queue,clbuf2,1,0,NWITEMS*sizeof(cl_float),buff2,0,0,0);

        cl_mem clbuf = clCreateBuffer(context,CL_MEM_WRITE_ONLY,NWITEMS*sizeof(cl_float),NULL,NULL);

        const char* filename="add.cl";
        std::string sourceStr;
        status = convertToString(filename,sourceStr);
        const char* source = sourceStr.c_str();
        size_t sourceSize[]={ strlen(source) };

        /************create program object***********/
        cl_program program = clCreateProgramWithSource(context,1,&source,sourceSize,NULL);

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

        if(status!=0){
                fprintf(stderr,"clBuild failed:%d\n",status);
                char tbuf[100];
                clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0x10000, tbuf, NULL);
                fprintf(stderr,"\n%s\n",tbuf);
                return -1;
        }
      
        /***********create kernel object************/
        cl_kernel kernel = clCreateKernel(program, "vecadd", NULL);

        //set kernel argument
        cl_int clnum=NWITEMS;
        clSetKernelArg(kernel, 0, sizeof(cl_mem), (void*)&clbuf1);
        clSetKernelArg(kernel, 1, sizeof(cl_mem), (void*)&clbuf2);
        clSetKernelArg(kernel, 2, sizeof(cl_mem), (void*)&clbuf);

        cl_event ev;
        size_t global_work_size=NWITEMS;
        clEnqueueNDRangeKernel(queue,kernel,1,NULL, &global_work_size,NULL,0,NULL,&ev);

        clFinish(queue);

        /**********copy data from device to host memory *****/
        cl_float *ptr;
        ptr=(cl_float *) clEnqueueMapBuffer(queue,clbuf, CL_TRUE, CL_MAP_READ, 0, NWITEMS * sizeof(cl_float),0,NULL, NULL, NULL);

        /***************vertify the result******/
        if(!memcmp(buff,ptr,NWITEMS)){
                printf("vertify passed\n");
        }else{
                printf("vertify failed\n");
        }

        if(buff)
                free(buff);
        if(buff1)
                free(buff1);
        if(buff2)
                free(buff2);

        /*************free OpenCL resource object************/
        clReleaseMemObject(clbuf1);
        clReleaseMemObject(clbuf2);
        clReleaseMemObject(clbuf);
        clReleaseProgram(program);
        clReleaseCommandQueue(queue);
        clReleaseContext(context);

        return 0;
}

add.cl

__kernel void vecadd(__global const float* A, __global const float* B,__global float* C)
{
        int id = get_global_id(0);
        C[id] = A[id] + B[id];
}

二：編譯

 nvcc test.cpp -o test  -lOpenCL  //注意-lOopenCL 的大小寫

三：執行

./vecadd

四：結果

numPlatforms=1  //環境中只有NVIDIA的平臺，所以爲 1
numDevice=3     //共有三塊NVIDIA卡
verify passed