Ubuntu下使用AMD APP编写OpenCL程序

对于Ubuntu或其近亲（Lubuntu、Kubuntu、Mint等）编写OpenCL程序也不会太难。由于本例用的是AMD APP SDK，因此需要AMD的GPU以及相关驱动。首先，去AMD官网下载GPU驱动——AMD Catalyst。如果你用的是APU并且还有一块独立显卡的话，通过AMD Catalyst Control Center可以选择使用哪个GPU。像我现在用的联想Z475笔记本，搭载了AMD APU A6-3420M以及一块AMD Radeon HD 7400M，但是相比较而言，还是APU自带的6620G的GPU性能更強一些，因此我这边设置的是采用AMD Radeon HD 6620G。

在Linux下，AMD官方的GPU驱动是.run文件，只需使用sudo sh xxx.run即可安装。安装时采用默认安装即可。

然后去developer.amd.com开发者网站下载AMD APP SDK。下载完成之后，将lib里面的动态库文件（xxx.so）取出来，并且把include里的头文件取出来。在你的OpenCL工程中把头文件的路径以及动态加载库都设置好。在你用-l的时候，如果动态库文件后缀名为.so.1，那么得把文件名后缀.1去掉。因为-l只能连接.a、.lib、.so等这些后缀的库文件。

下面我们将举一个最简单的例子，首先看主机端代码：

/*
 ============================================================================
 Name        : OpenCLTest.c
 Author      : Zenny Chen
 Version     :
 Copyright   : Your copyright notice
 Description : Hello World in C, Ansi-style
 ============================================================================
 */

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <CL/cl.h>

static int GetCurrentLocationFilePath(char pDst[512], const char *filename)
{
    if(pDst == NULL || filename == NULL)
        return 0;

    int size = readlink("/proc/self/exe", pDst, 512);
    while(pDst[size - 1] != '/')
        size--;

    strcpy(&pDst[size], filename);
    int retSize = strlen(filename) + size;
    pDst[retSize] = '\0';

    return retSize;
}

int main(void)
{
    /*Step1: Getting platforms and choose an available one.*/
    cl_uint numPlatforms;   //the NO. of platforms
    cl_int  status = clGetPlatformIDs(0, NULL, &numPlatforms);
    if (status != CL_SUCCESS)
    {
        puts("Error: Getting platforms!");
        return 0;
    }

    cl_platform_id platforms[16];

    /*For clarity, choose the first available platform. */
    if(numPlatforms > 0)
    {
        status = clGetPlatformIDs(numPlatforms, platforms, NULL);
        if(status != CL_SUCCESS)
        {
            puts("Failed to get platform IDs");
            return 0;
        }
    }

    /*Step 2:Query the platform and choose the first GPU device if has one.Otherwise use the CPU as device.*/
    cl_uint             numDevices = 0;
    cl_device_id        devices[16];
    clGetDeviceIDs(platforms[0], CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices);
    if(numDevices == 0)    //no GPU available.
    {
        puts("No devices available!");
        return 0;
    }
    else
    {
        printf("The number of available devices is: %u\n", numDevices);
        clGetDeviceIDs(platforms[0], CL_DEVICE_TYPE_GPU, numDevices, devices, NULL);
    }

    /*Step 3: Create context.*/
    cl_context context = clCreateContext(NULL,1, devices,NULL,NULL,NULL);

    /*Step 4: Creating command queue associate with the context.*/
    cl_command_queue commandQueue = clCreateCommandQueue(context, devices[0], 0, NULL);

    /*Step 5: Create program object */
    char filePath[512];
    GetCurrentLocationFilePath(filePath, "test.cl");
    FILE *fp = fopen(filePath, "r");
    if(fp == NULL)
    {
        puts("OpenCL kernel source file open failed!");
        return 0;
    }
    fseek(fp, 0, SEEK_END);
    long fileLength = ftell(fp);
    fseek(fp, 0, SEEK_SET);
    char *source = (char*)malloc(fileLength + 1);
    fread(source, 1, fileLength, fp);
    fclose(fp);
    size_t sourceSize[] = {fileLength};
    cl_program program = clCreateProgramWithSource(context, 1, (const char**)&source, sourceSize, NULL);

    free(source);
    if(program == NULL)
    {
        puts("Failed to create the program!");
        return 0;
    }

    /*Step 6: Build program. */
    status = clBuildProgram(program, 1,devices,NULL,NULL,NULL);
    if(status != CL_SUCCESS)
    {
        puts("Failed to build the program!");
        return 0;
    }

    /*Step 7: Initial input,output for the host and create memory objects for the kernel*/
    int input[128];
    for(int i = 0; i < 128; i++)
        input[i] = i + 1;

    cl_mem inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, sizeof(input), input, NULL);
    cl_mem outputBuffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY , sizeof(input), NULL, NULL);

    /*Step 8: Create kernel object */
    cl_kernel kernel = clCreateKernel(program, "test", NULL);

    /*Step 9: Sets Kernel arguments.*/
    status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &inputBuffer);
    status = clSetKernelArg(kernel, 1, sizeof(cl_mem), &outputBuffer);

    /*Step 10: Running the kernel.*/
    size_t global_work_size[1] = { 128 };
    status = clEnqueueNDRangeKernel(commandQueue, kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);

    /*Step 11: Read the cout put back to host memory.*/
    int output[128];
    status = clEnqueueReadBuffer(commandQueue, outputBuffer, CL_TRUE, 0, sizeof(input), output, 0, NULL, NULL);

    /*Step 12: Clean the resources.*/
    status = clReleaseKernel(kernel);               //Release kernel.
    status = clReleaseProgram(program);             //Release the program object.
    status = clReleaseMemObject(inputBuffer);       //Release mem object.
    status = clReleaseMemObject(outputBuffer);
    status = clReleaseCommandQueue(commandQueue);   //Release  Command queue.
    status = clReleaseContext(context);             //Release context.

    for(int i = 0; i < 128; i++)
    {
        if(output[i] != i + 2)
        {
            printf("Error occurred @%d!", i);
            return 0;
        }
    }

    puts("Pass!");
    return 1;
}

在编译选项中，使用-std=gnu99或-std=gnu11。上述代码为纯C语言，因此即便你没有安装g++也完全没关系。

下面看看内核源代码：

/*
 ============================================================================
 Name        : OpenCLTest.c
 Author      : Zenny Chen
 Version     :
 Copyright   : Your copyright notice
 Description : Simple OpenCL kernel source
 ============================================================================
 */

__kernel void test(__global int* in, __global int* out)
{
    int index = get_global_id(0);
    out[index] = in[index] + 1;
}

将此文件放在可执行文件相同路径下，然后我们就能正常运行了。