CUDA（warpSize為負數）

Question

我有Nvidia geforce 310M，當我編寫此代碼時

int main()
{
    int dev;
    cudaDeviceProp prop;
    cudaGetDevice(&dev);
    cudaGetDeviceProperties(&prop,dev);
    cout << "Device name " << prop.name << endl;
    cout << "warp size  = " << prop.warpSize << endl;
    cout << "mutiprocesosrs = " << prop.multiProcessorCount << endl;
    return 0;
}

它寫道：

Device name ♀
warp size  = -2144521342
mutiprocesosrs = 0
Press any key to continue . . .

每次我運行此程序時，它都會產生不同的變形大小。 我該如何解決？

（ps我正在使用Visual Studio 2010）

Answer 1

首先檢查函數調用是否成功！

cudaDeviceProp prop;
int dev;

if (cudaGetDevice(&dev) != cudaSuccess)
{
  /* error */
  abort();
}

if (cudaGetDeviceProperties(&prop, dev) != cudaSuccess)
{
  /* error */
  abort();
}

---

我制作了一個小宏，用於調用所有CUDA函數，該宏負責檢查並引發自定義異常：

#define CUDACALL(F, ARGS...) do { e = F(ARGS); if (e != cudaSuccess) throw cudaException(#F, e); } while (false)

有了這個，我只是說，

try
{
   int dev;
   cudaDeviceProp prop;
   CUDACALL(cudaGetDevice, &dev);
   CUDACALL(cudaGetDeviceProperties, &prop, dev);
   // ...
}
catch (cudeException const & e)
{
  std::cerr << "Cuda error: " << e.what() << "\n";
}

異常類的定義如下：

struct cudaException : public std::exception
{
  cudaException(const std::string & str, cudaError_t err) : s(str), e(err) { }

  ~cudaException() throw() { }

  virtual const char * what() const throw()
  {
    const std::string r = s + ": \"" + cudaGetErrorString(e) + "\" (" + (char)(e + '0') + ')';
    return r.c_str();
  }
private:
  std::string s;
  cudaError_t e;
};

Answer 2

使用此代碼，您可以檢查您的cuda設置是否錯誤。 （來自cuda deviceQuery示例）

    if (cudaGetDeviceCount((int*)&_deviceCount) != cudaSuccess) 
    {
        printf("ERROR: CUDA Driver and Runtime version may be mismatched.\n");
        return false;
    }

    // This function call returns 0 if there are no CUDA capable devices.
    if (_deviceCount == 0)
    {
        printf("ERROR: There is no device supporting CUDA!\n");
        return false;
    }

    _deviceProperties = SAVE_ALLOCATE(cudaDeviceProp, _deviceCount * sizeof(cudaDeviceProp));

    for (unsigned int dev = 0; dev < _deviceCount; ++dev) 
    {
        cudaGetDeviceProperties(&_deviceProperties[dev], dev);

        printf("\nDevice %d: \"%s\"\n", dev, _deviceProperties[dev].name);

    #if CUDART_VERSION >= 2020
        // Console log
        cudaDriverGetVersion(&_driverVersion);
        printf("  CUDA Driver Version:                           %d.%d\n",              _driverVersion/1000, _driverVersion%100);
        cudaRuntimeGetVersion(&_runtimeVersion);
        printf("  CUDA Runtime Version:                          %d.%d\n",              _runtimeVersion/1000, _runtimeVersion%100);
    #endif
        printf("  CUDA Capability revision number:               %d.%d\n",              

_deviceProperties[dev].major,_deviceProperties[dev].minor);
}

您會發現是否有問題;）