在網上找了一些例子，測試了一下都是無法執行的。下面給出一個在msm8953測試通過的可以執行的openc的例項。當然這個例子也是在別人的基礎之上修改的，具體程式碼如下
1.Android.mk
程式碼路徑
：vendor/qcom/proprietary/mm-camera/mm-camera2/opencl/Android.mk

LOCAL_PATH := $(call my-dir)
include $(CLEAR_VARS)
LOCAL_MODULE_TAGS := optional
LOCAL_MODULE    := opencl
LOCAL_SRC_FILES := main.cpp
LOCAL_C_INCLUDES := $(LOCAL_PATH)
LOCAL_C_INCLUDES += $(TARGET_OUT_INTERMEDIATES)/include/adreno/
#./vendor/qcom/proprietary/prebuilt_HY11/target/product/msm8953_64/obj/include/adreno/CL/cl.h
 

#./vendor/qcom/proprietary/prebuilt_HY11/target/product/msm8953_64/system/vendor/lib64/libOpenCL.so
LOCAL_LDFLAGS += -lOpenCL
include $(BUILD_EXECUTABLE)

2.cl_kernel2.cl
程式碼路徑
：vendor/qcom/proprietary/mm-camera/mm-camera2/opencl/cl_kernel2.cl

__kernel void cl_add(__global int *dst, __global int *src1, __global int
 
 *src2)
{
    int index = get_global_id(0);
    dst[index] = src1[index] + src2[index];
}

3.
程式碼路徑
：vendor/qcom/proprietary/mm-camera/mm-camera2/opencl/main.cpp

#include <stdio.h>
#include <stdlib.h>
#include "readyuv.h"
#include "CL/cl.h"
//#include "CL/cl_platform.h"
#include <iostream>
 

using namespace std;
int main(void)
{
    const int  array_size = 1024;
    cl_uint numPlatforms = 0;
    cl_platform_id platform = nullptr;
    cl_context context = nullptr;
    cl_command_queue commandQueue = nullptr;
    cl_program program = nullptr;
    cl_mem input1MemObj = nullptr;
    cl_mem input2MemObj = nullptr;
    cl_mem outputMemObj = nullptr;
    cl_kernel kernel = nullptr;
    //step1. 查詢OpenCL平臺集合
    /*
     * 這個函式一般被呼叫兩次:
     *   第一次呼叫這個函式是獲得可用平臺的數目, 然後為平臺物件分配記憶體空間
     *   第二次呼叫用來獲取平臺物件
     */
    cl_int status = clGetPlatformIDs(0, NULL, &numPlatforms);
    if (status != CL_SUCCESS)
    {
        cout << "Error: Getting platforms!" << endl;
        return -1;
    }
    if (numPlatforms > 0)
    {
        cl_platform_id* platforms = (cl_platform_id*)malloc(numPlatforms* sizeof(cl_platform_id));
        status = clGetPlatformIDs(numPlatforms, platforms, NULL);
        platform = platforms[0];
    }
    else
    {
        puts("Your system does not have any OpenCL platform!");
        return -1;
    }

    cl_uint        numDevices = 0;
    cl_device_id   *devices;
    cl_int errcode_ret ;

    //setp2.獲取一個平臺上的可用裝置清單
    status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices);
    if (numDevices == 0) //no GPU available.        
    {
        cout << "No GPU device available." << endl;
        cout << "Choose CPU as default device." << endl;
        status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, 0, NULL, &numDevices);
        devices = (cl_device_id*)malloc(numDevices * sizeof(cl_device_id));
        status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_CPU, numDevices, devices, NULL);
    }
    else
    {
        devices = (cl_device_id*)malloc(numDevices * sizeof(cl_device_id));
        status = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, numDevices, devices, NULL);
        cout << "The number of devices: " << numDevices << endl;
    }
    //step3.建立一個OpenCL上下文
    /*
    *上下文是 OpenCL runtime 用來管理像命令佇列、記憶體佇列、程式物件和核心物件，並且在上下文中指定的一個或多個裝置上執行核心函式。
    */
    context = clCreateContext(NULL, 1, devices, NULL, NULL, &errcode_ret);
    if(errcode_ret == CL_SUCCESS)
        cout << "craet OpenCL runtime sucess." << endl;
    else
        return -1;
    //step4. 建立命令佇列
    commandQueue = clCreateCommandQueue(context, devices[0], 0, &errcode_ret);
    if(errcode_ret == CL_SUCCESS)
        cout << "craet CommandQueue sucess." << endl;
    else
        return -1;

    char *kernelCodeBuffer = nullptr;
    char *Buffer = nullptr;
    const char *aSource = nullptr;
    size_t kernelLength = 0;
    size_t global_work_size[1] = { array_size };

    // Read the kernel code to the buffer
    FILE *fp = fopen("cl_kernel2.cl", "rb");
    if (fp == nullptr)
    {
        puts("The kernel file not found!");
        goto RELEASE_RESOURCES;
    }
    fseek(fp, 0, SEEK_END);
    kernelLength = ftell(fp);
    cout << "kernelLength : " << kernelLength << endl;
    fseek(fp, 0, SEEK_SET);
    kernelCodeBuffer = (char*)malloc(kernelLength + 1);
    Buffer = (char*)malloc(kernelLength + 1);
    fread(kernelCodeBuffer, 1, kernelLength, fp);
    kernelCodeBuffer[kernelLength] = '\0';
    fclose(fp);
    aSource = kernelCodeBuffer;
    strcpy(Buffer,aSource);
    cout << "kernelCodeBuffer : " << Buffer  << endl;

    //step5. 建立程式物件
    program = clCreateProgramWithSource(context, 1, &aSource, &kernelLength, NULL);

    //step6. 編譯程式物件
    status = clBuildProgram(program, 1, devices, NULL, NULL, NULL);


    // Do initialization
    int i;
    int input1Buffer[array_size];
    int input2Buffer[array_size];
    int outputBuffer[array_size];
    for (i = 0; i < array_size; i++)
        input1Buffer[i] = input2Buffer[i] = i + 1;
    memset(outputBuffer, 0, sizeof(outputBuffer));

    // Create mmory object
    input1MemObj = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, array_size * sizeof(int), input1Buffer, nullptr);
    input2MemObj = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, array_size * sizeof(int), input2Buffer, nullptr);
    outputMemObj = clCreateBuffer(context, CL_MEM_WRITE_ONLY, array_size * sizeof(int), NULL, NULL);

    //step7.建立核心物件
    kernel = clCreateKernel(program, "cl_add", NULL);

    //step8.設定核心引數,想要執行核心，就必須設定核心引數
    status = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&outputMemObj);
    status = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&input1MemObj);
    status = clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&input2MemObj);

    status = clEnqueueNDRangeKernel(commandQueue, kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);
/*
 * clEnqueueNDRangeKernel(
 *
 * cl_command_queue queue,
 *
 * cl_kernel kernel,
 *
 * cl_uint work_dims,
 *  //if you deal with image object, you should probably set work_dims equal 2 or 3. But for buffer objects, you can set whatever dimensionality you think best.
 *  //For a buffer object containing a two-dimensional matrix, you might set work-dims equal 2.
 *
 * const size_t *global_work_offset,
 *  //the global ID offset in each dimension
 *
 * const size_t *global_work_size, 
 *  //the number of work items in each dimension
 * const size_t *local_work_size,
 *  //the number of work_items in a work_group，in each dimension
 * cl_uint num_events,
 *
 * const cl_event *wait_list,
 *
 * cl_event *event)
 */
/*
 * __kernel void cl_add(__global int *dst, __global int *src1, __global int *src2)
 * {
 *     int index = get_global_id(0);//get_global_id獲取執行緒索引，其取值範圍是0 到 global_work_size - 1
 *     dst[index] = src1[index] + src2[index];
 * }        
 */
    clFinish(commandQueue);

    //GPU中執行完成後需要Copy結果到記憶體
    status = clEnqueueReadBuffer(commandQueue, outputMemObj, CL_TRUE, 0, global_work_size[0] * sizeof(int), outputBuffer, 0, NULL, NULL);

    printf("Veryfy the rsults... ");
    for (i = 0; i < array_size; i++)
    {
        if (outputBuffer[i] != (i + 1) * 2)
        {
            puts("Results not correct!");
            break;
        }
    }
    if (i == array_size)
        puts("Correct!");

RELEASE_RESOURCES:

    status = clReleaseKernel(kernel);//*Release kernel.
    status = clReleaseProgram(program);    //Release the program object.
    status = clReleaseMemObject(input1MemObj);//Release mem object.
    status = clReleaseMemObject(input2MemObj);
    status = clReleaseMemObject(outputMemObj);
    status = clReleaseCommandQueue(commandQueue);//Release  Command queue.
    status = clReleaseContext(context);//Release context.
    free(devices);
    getchar();
    return 0;
}
/*
 * clGetPlatformIDs---------------------------獲取平臺ID
 *
 * clGetDeviceIDs-----------------------------獲取裝置ID
 *
 * clCreateContext----------------------------建立上下文
 *
 * clCreateCommandQueue-----------------------建立命令佇列
 *
 * clCreateBuffer-----------------------------建立裝置記憶體
 *
 * clCreateProgramWithSource------------------建立程式
 *
 * clBuildProgram-----------------------------編譯程式
 *
 * clGetProgramBuildInfo----------------------獲取編譯資訊
 *
 * clCreateKernel-----------------------------建立核
 *
 * clSetKernelArg-----------------------------設定核引數
 *
 * clEnqueueNDRangeKernel---------------------執行核
 *
 * clEnqueueReadBuffer------------------------讀取裝置記憶體
 *
 * clReleaseMemObject-------------------------釋放記憶體物件
 *
 * clReleaseKernel----------------------------釋放核
 *
 * clReleaseCommandQueue----------------------釋放命令佇列
 *
 * clReleaseContext---------------------------釋放上下文
 */

執行編譯命令後驗證方法如下
一、把檔案推送到平臺端
1.adb root
2.adb remount
3.adb push out\target\product\msm8953_64\system\bin\opencl data
4.adb push vendor\qcom\proprietary\mm-camera\mm-camera2\opencl\cl_kernel2.cl data
二、執行命令
1 ) chmod 777 data/opencl
2 ) cd data/ (若不執行，會報錯誤如下The kernel file not found!)
3 ) ./opencl

執行完後可得成功log如下

The number of devices: 1
craet OpenCL runtime sucess.
craet CommandQueue sucess.
kernelLength : 162
kernelCodeBuffer : __kernel void cl_add(__global int *dst, __global int *src1, __global int *src2)
{
    int index = get_global_id(0);
    dst[index] = src1[index] + src2[index];
}

Veryfy the rsults... Correct!