C++ OpenCL只读取/写入缓冲区内存的1/4，有时会崩溃

OpenCL有一个问题，它执行整个命令队列，但它只读取输入的1/4，只写入结果的1/4。 无论迭代多少次，始终为1/4

而且它有时会随机崩溃。在调试过程中，我得不到任何信息，因为没有调试符号，它会崩溃（0x4c4783f6 in？？？，等等）

源代码：

#include <iostream>
#include <cl/cl.h>
#include <cassert>
#include <cstring>

const char *progsrc[] = {
"#pragma OPENCL EXTENSION cl_intel_printf : enable\n\
__kernel void add(__global const int *a, __global const int *b, __global int *out) \
{ \
    int tid = get_global_id(0);\
    out[tid] = tid/*a[tid]+b[tid]*/;\
    printf(\"krnl: %d = %d + %d \\n\", out[tid], a[tid], b[tid]);\
}"};

const int iterations = 20;

#define CLCheck(a) \
do\
{\
    if(a != CL_SUCCESS)\
    {\
        std::cerr << "OpenCL Error(" << a << ") at " << __LINE__ << std::endl;\
        return -1;\
    }\
} while(0)

int main()
{
    cl_int err = CL_SUCCESS;

    int *aH = NULL;
    int *bH = NULL;
    int *outH = NULL;

    cl_uint platnum, devnum;
    cl_device_id dev;
    cl_platform_id plat;
    err = clGetPlatformIDs(0, 0, &platnum);
    CLCheck(err);
    cl_platform_id pfids[platnum];
    err = clGetPlatformIDs(platnum, pfids, &platnum);
    CLCheck(err);

    if(!platnum)
    {
        std::cerr << "No platform found." << std::endl;
        return -1;
    }
    else
        std::cout << platnum << " OpenCL platform(s) found.\n" << std::endl;

    for(unsigned int i = 0; i != platnum; i++)
    {
        char buf[4096];

        err = clGetDeviceIDs(pfids[i], CL_DEVICE_TYPE_ALL, 0, 0, &devnum);
        CLCheck(err);
        cl_device_id devids[devnum];
        err = clGetDeviceIDs(pfids[i], CL_DEVICE_TYPE_ALL, devnum, devids, &devnum);
        CLCheck(err);
        if(!devnum)
        {
            std::cerr << "No device found." << std::endl;
            return -1;
        }
        else
            std::cout << " " << devnum << " OpenCL device(s) found.\n" << std::endl;

        for(unsigned int i2 = 0; i2 != devnum; i2++)
        {
            char buf[1024];
            std::cout << ": \n\tName: " << buf;
            err = clGetDeviceInfo(devids[i2], CL_DEVICE_VENDOR, 1024, buf, NULL);
            CLCheck(err);
            if(!strncmp(buf, "Intel", 5))
            {
                dev = devids[0];
                plat = pfids[i];
                std::cout << "\n\tFound Intel(R) OpenCL device.";
            }
        }
    }
    cl_context_properties ctxprop[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)plat, 0};
    cl_context ctx = clCreateContext(ctxprop, 1, &dev, NULL, NULL, &err);
    CLCheck(err);

    cl_program program = clCreateProgramWithSource(ctx, 1, progsrc, NULL, &err);
    CLCheck(err);
    err = clBuildProgram(program, 1, &dev, "", NULL, NULL);
    if(err != CL_SUCCESS)
    {
        size_t bufsz;
        err = clGetProgramBuildInfo(program, dev, CL_PROGRAM_BUILD_LOG, 0, 0, &bufsz);
        char buf[bufsz];
        err = clGetProgramBuildInfo(program, dev, CL_PROGRAM_BUILD_LOG, bufsz, buf, &bufsz);
        std::cerr << "OpenCL program building failed: " << buf << std::endl;
        return -1;
    }
    err = clUnloadCompiler();
    CLCheck(err);

    aH = new int[iterations];
    bH = new int[iterations];
    outH = new int[iterations];
    memset(outH, 0, iterations*sizeof(int));
    for(int i = 0; i != iterations; i++)
    {
        aH[i] = i;
        bH[i] = i*2;
    }

    cl_mem aCL = clCreateBuffer(ctx, CL_MEM_READ_ONLY, iterations, NULL, &err);
    cl_mem bCL = clCreateBuffer(ctx, CL_MEM_READ_ONLY, iterations, NULL, &err);
    CLCheck(err);
    cl_mem outCL = clCreateBuffer(ctx, CL_MEM_WRITE_ONLY, iterations, NULL, &err);
    CLCheck(err);

    cl_kernel krnl = clCreateKernel(program, "add", &err);
    CLCheck(err);

    err = clSetKernelArg(krnl, 0, sizeof(aCL), &aCL);
    CLCheck(err);
    err = clSetKernelArg(krnl, 1, sizeof(bCL), &bCL);
    CLCheck(err);
    err = clSetKernelArg(krnl, 2, sizeof(outCL), &outCL);
    CLCheck(err);

    cl_command_queue cmdqueue = clCreateCommandQueue(ctx, dev, 0, &err);
    cl_event evt;
    size_t global_work_size[1] = { iterations };
    err = clEnqueueWriteBuffer(cmdqueue, aCL, CL_TRUE, 0, iterations, aH, 0, NULL, NULL);
    err = clEnqueueWriteBuffer(cmdqueue, bCL, CL_TRUE, 0, iterations, bH, 0, NULL, NULL);
    err = clEnqueueNDRangeKernel(cmdqueue, krnl, 1, NULL, global_work_size, NULL, 0, NULL, &evt);
    err = clWaitForEvents(1, &evt);
    err = clEnqueueReadBuffer(cmdqueue, outCL, CL_TRUE, 0, iterations, outH, 0, NULL, &evt);

    for(int i = 0; i != iterations; i++)
    {
        std::cout << outH[i] << std::endl;
    }

    err = clReleaseEvent(evt);
    err = clReleaseCommandQueue(cmdqueue);
    err = clReleaseKernel(krnl);
    err = clReleaseMemObject(outCL);
    err = clReleaseMemObject(bCL);
    err = clReleaseMemObject(aCL);
    err = clReleaseProgram(program);
    err = clReleaseContext(ctx);

    if(aH)
        delete aH;
    if(bH)
        delete bH;
    if(outH)
        delete outH;
    return 0;
}

---

谢谢：）

我不熟悉openCL，但我认为您缺少一些

大小的
，如下所示：

err = clEnqueueWriteBuffer(cmdqueue, aCL, CL_TRUE, 0, iterations, aH, 0, NULL, NULL);

应该是：

err = clEnqueueWriteBuffer(cmdqueue, aCL, CL_TRUE, 0, iterations * sizeof(int), aH, 0, NULL, NULL);

这同样适用于以下类似的代码

编辑：

这里还有一个地方，您可能错过了一些
sizeof（）
s：

哦，非常感谢……我怎么能成为一个有经验的C++程序员（多年）没有看到这一点。在C++中，你不必在空洞的指针上耍花招，这就是为什么；
err = clEnqueueWriteBuffer(cmdqueue, aCL, CL_TRUE, 0, iterations * sizeof(int), aH, 0, NULL, NULL);

cl_mem aCL = clCreateBuffer(ctx, CL_MEM_READ_ONLY, iterations, NULL, &err);
cl_mem bCL = clCreateBuffer(ctx, CL_MEM_READ_ONLY, iterations, NULL, &err);
CLCheck(err);
cl_mem outCL = clCreateBuffer(ctx, CL_MEM_WRITE_ONLY, iterations, NULL, &err);
CLCheck(err);