c++ - 使用 clCreateCommandQueueWithProperties 时出现段错误

Question

我正在尝试用 C++ 编写 OpenCL 包装器。昨天我在我的 Windows 10 机器上工作(NVIDIA GTX970 Ti，我相信是最新的 NVIDIA GeForce 驱动程序)并且我的代码运行完美。

今天，我在我的笔记本电脑(Arch Linux、AMD Radeon R7 M265、Mesa 17.3.3)上试用它，我在尝试创建命令队列时遇到段错误。

这是 GDB 回溯:

#0  0x00007f361119db80 in ?? () from /usr/lib/libMesaOpenCL.so.1
#1  0x00007f36125dacb1 in clCreateCommandQueueWithProperties () from /usr/lib/libOpenCL.so.1
#2  0x0000557b2877dfec in OpenCL::createCommandQueue (ctx=..., dev=..., outOfOrderExec=false, profiling=false) at /home/***/OpenCL/Util.cpp:296
#3  0x0000557b2876f0cf in main (argc=1, argv=0x7ffd04fcdac8) at /home/***/main.cpp:27
#4  0x00007f361194cf4a in __libc_start_main () from /usr/lib/libc.so.6
#5  0x0000557b2876ecfa in _start ()

(我删掉了部分路径)这是产生错误的代码:

CommandQueue createCommandQueue(Context ctx, Device dev, bool outOfOrderExec, bool profiling) noexcept
{
    cl_command_queue_properties props [3]= {CL_QUEUE_PROPERTIES, 0, 0};

    if (outOfOrderExec)
    {
        props[1] |= CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE;
    }

    if (profiling)
    {
        props[1] |= CL_QUEUE_PROFILING_ENABLE;
    }

    int error = CL_SUCCESS;

    cl_command_queue queue = clCreateCommandQueueWithProperties(ctx.get(), dev.get(), props, &error);

    if (error != CL_SUCCESS)
    {
        std::cerr << "Error while creating command queue: " << OpenCL::getErrorString(error) << std::endl;
    }

    CommandQueue commQueue = CommandQueue(queue);
    Session::get().registerQueue(commQueue);

    return commQueue;
}

带有 clCreateCommandQueueWithProperties 的行是发生段错误的地方。

Context 是一个cl_context 的包装类，Context::get() 返回原始的cl_context:

class Context
{
    private:
        ...
        cl_context context;

    public:
        ...

        cl_context get() const noexcept;

        ...
};

Device 是 cl_device 的包装器，Device::get() 也返回 cl_device:

class Device
{
    private:
        ...

        cl_device_type type;
        cl_device_id id;

    public:
        ...

        cl_device_id get() const noexcept;

        cl_device_type getType () const noexcept;

        ...
};

这是主要功能:

int main (int argc, char* argv [])
{
    OpenCL::Session::get().init();

    for (const std::string& deviceAddress : OpenCL::Session::get().getAddresses())
    {
        std::cout << "[" << deviceAddress << "]: " << OpenCL::Session::get().getDevice(deviceAddress);
    }

    OpenCL::Context ctx = OpenCL::getContext();

    std::cout << "OpenCL version: " << ctx.getVersionString() << std::endl;

    OpenCL::Kernel kernel = OpenCL::createKernel(OpenCL::createProgram("src/Kernels/Hello.cl", ctx), "SAXPY");
    OpenCL::CommandQueue queue = OpenCL::createCommandQueue(ctx, OpenCL::Session::get().getDevice(ctx.getAssociatedDevices()[0]));

    unsigned int testDataSize = 1 << 13;

    std::vector <float> a = std::vector <float> (testDataSize);
    std::vector <float> b = std::vector <float> (testDataSize);

    for (int  i = 0; i < testDataSize; i++)
    {
        a[i] = static_cast<float>(i);
        b[i] = 0.0;
    }

    OpenCL::Buffer aBuffer = OpenCL::allocateBuffer(ctx, a.data(), sizeof(float), a.size());
    OpenCL::Buffer bBuffer = OpenCL::allocateBuffer(ctx, b.data(), sizeof(float), b.size());

    kernel.setArgument(0, aBuffer);
    kernel.setArgument(1, bBuffer);
    kernel.setArgument(2, 2.0f);

    OpenCL::Event saxpy_event = queue.enqueue(kernel, {testDataSize});
    OpenCL::Event read_event = queue.read(bBuffer, b.data(), bBuffer.size());

    std::cout << "SAXPY kernel took " << saxpy_event.getRunTime() << "ns to complete." << std::endl;
    std::cout << "Read took " << read_event.getRunTime() << "ns to complete." << std::endl;

    OpenCL::Session::get().cleanup();

    return 0;
}

(分析将不起作用，因为我已禁用它(认为这是问题的原因)，但是重新启用分析并不能解决问题)。

我正在使用 CLion，所以这是我的调试窗口的屏幕截图:

最后是程序的控制台输出:

/home/***/cmake-build-debug/Main
[gpu0:0]: AMD - AMD OLAND (DRM 2.50.0 / 4.14.15-1-ARCH, LLVM 5.0.1): 6 compute units @ 825MHz
OpenCL version: OpenCL 1.1 Mesa 17.3.3
Signal: SIGSEGV (Segmentation fault)

上下文和设备对象似乎都是在没有任何问题的情况下创建的，所以我真的不知道是什么导致了段错误。

有没有可能是我在 Mesa 驱动程序中发现了错误，还是我遗漏了一些明显的东西？

编辑:This人似乎有类似的问题，不幸的是，他的问题只是一个 C 风格的忘记分配内存问题。

第二次编辑:我可能已经找到了这个问题的可能原因，CMake 正在查找、使用和链接 OpenCL 2.0，而我的 GPU 只支持 OpenCL 1.1。我会调查的。我还没有找到在 Arch Linux 上回滚到 OpenCL 1.1 的方法，但 clinfo 似乎工作正常，blender(依赖于 OpenCL)也是如此，所以我认为这不是问题所在。

这是 clinfo 的输出:

Number of platforms                               1
  Platform Name                                   Clover
  Platform Vendor                                 Mesa
  Platform Version                                OpenCL 1.1 Mesa 17.3.3
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd
  Platform Extensions function suffix             MESA

  Platform Name                                   Clover
Number of devices                                 1
  Device Name                                     AMD OLAND (DRM 2.50.0 / 4.14.15-1-ARCH, LLVM 5.0.1)
  Device Vendor                                   AMD
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 1.1 Mesa 17.3.3
  Driver Version                                  17.3.3
  Device OpenCL C Version                         OpenCL C 1.1 
  Device Type                                     GPU
  Device Available                                Yes
  Device Profile                                  FULL_PROFILE
  Max compute units                               6
  Max clock frequency                             825MHz
  Max work item dimensions                        3
  Max work item sizes                             256x256x256
  Max work group size                             256
  Compiler Available                              Yes
  Preferred work group size multiple              64
  Preferred / native vector sizes                 
    char                                                16 / 16      
    short                                                8 / 8       
    int                                                  4 / 4       
    long                                                 2 / 2       
    half                                                 8 / 8        (cl_khr_fp16)
    float                                                4 / 4       
    double                                               2 / 2        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
  Single-precision Floating-point support         (core)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Address bits                                    64, Little-Endian
  Global memory size                              2147483648 (2GiB)
  Error Correction support                        No
  Max memory allocation                           1503238553 (1.4GiB)
  Unified memory for Host and Device              No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       32768 bits (4096 bytes)
  Global Memory cache type                        None
  Image support                                   No
  Local memory type                               Local
  Local memory size                               32768 (32KiB)
  Max constant buffer size                        1503238553 (1.4GiB)
  Max number of constant args                     16
  Max size of kernel argument                     1024
  Queue properties                                
    Out-of-order execution                        No
    Profiling                                     Yes
  Profiling timer resolution                      0ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
  Device Extensions                               cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_fp64 cl_khr_fp16

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  Clover
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [MESA]
  clCreateContext(NULL, ...) [default]            Success [MESA]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 Clover
    Device Name                                   AMD OLAND (DRM 2.50.0 / 4.14.15-1-ARCH, LLVM 5.0.1)
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 Clover
    Device Name                                   AMD OLAND (DRM 2.50.0 / 4.14.15-1-ARCH, LLVM 5.0.1)
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 Clover
    Device Name                                   AMD OLAND (DRM 2.50.0 / 4.14.15-1-ARCH, LLVM 5.0.1)

ICD loader properties
  ICD loader Name                                 OpenCL ICD Loader
  ICD loader Vendor                               OCL Icd free software
  ICD loader Version                              2.2.12
  ICD loader Profile                              OpenCL 2.2

第三次编辑:我刚刚在我的 NVIDIA 机器上运行了代码，工作没有问题，这是控制台显示的内容:

[gpu0:0]: NVIDIA Corporation - GeForce GTX 970: 13 compute units @ 1253MHz
OpenCL version: OpenCL 1.2 CUDA 9.1.75
SAXPY kernel took 2368149686ns to complete.
Read took 2368158390ns to complete.

我还修复了 2 个问题 Andreas提到

Answer 1

clCreateCommandQueueWithProperties 已添加到 OpenCL 2.0 中。您不应将它用于低于 2.0 版的平台和设备(例如日志中显示的 1.1 和 1.2)。