c - 使用 CUDA 5 的 cudaMemcpyToSymbol 出错

Question

问题

我已经准备了一个使用常量内存的示例 CUDA 代码。我可以在 cuda 4.2 中成功运行它，但是当我使用 CUDA 5 进行编译时，我得到了 "invalid device symbol"。我已在此处附上示例代码。

代码

#include <iostream>
#include <stdio.h>
#include <cuda_runtime.h>
#include <cuda.h>

struct CParameter
{
    int A;  
    float B;
    float C;
    float D;
};

__constant__ CParameter * CONSTANT_PARAMETER;   
#define PARAMETER "CONSTANT_PARAMETER"

bool ERROR_CHECK(cudaError_t Status)
{
    if(Status != cudaSuccess)
    {
        printf(cudaGetErrorString(Status));
        return false;
    }   
    return true;
}

// Kernel that executes on the CUDA device
__global__ void square_array(float *a, int N)
{
  int idx = blockIdx.x * blockDim.x + threadIdx.x;
  if (idx<N)
  {
      a[idx] = CONSTANT_PARAMETER->A * a[idx];
  }
}
////Main Function/////
int main(void)
{
    /////Variable Definition
    const int N = 10;
    size_t size = N * sizeof(float);
    cudaError_t Status = cudaSuccess;

    CParameter * m_dParameter;
    CParameter * m_hParameter;
    float * m_D;
    float * m_H;

    //Memory Allocation Host
    m_hParameter = new CParameter;
    m_H = new float[N];

    //Memory Allocation Device
    cudaMalloc((void **) &m_D, size);
    cudaMalloc((void**)&m_dParameter,sizeof(CParameter));

    ////Data Initialization
    for (int i=0; i<N; i++) 
        m_H[i] = (float)i;

    m_hParameter->A = 5;
    m_hParameter->B = 3;
    m_hParameter->C = 98;
    m_hParameter->D = 100;

    //Memory Copy from Host To Device
    Status = cudaMemcpy(m_D, m_H, size, cudaMemcpyHostToDevice);
    ERROR_CHECK(Status);

    Status = cudaMemcpy(m_dParameter,m_hParameter,sizeof(CParameter),cudaMemcpyHostToDevice);
    ERROR_CHECK(Status);        

    Status = cudaMemcpyToSymbol(PARAMETER, &m_dParameter, sizeof(m_dParameter));
    ERROR_CHECK(Status);

    // Do calculation on device:
    int block_size = 4;

    int n_blocks = N/block_size + (N%block_size == 0 ? 0:1);

    square_array <<<n_blocks, block_size>>>(m_D,N);

    // Retrieve result from device and store it in host array
    cudaMemcpy(m_H, m_D, sizeof(float)*N, cudaMemcpyDeviceToHost);

    // Print results
    for (int i=0; i<N; i++)
        printf("%d %f\n", i, m_H[i]);

    // Cleanup
    free(m_H);
    free(m_hParameter);
    cudaFree(m_dParameter);
    cudaFree(m_D);
    return 0;   
}

我试过 WINDOWS: CUDA 5.0 Production Release 显卡是 GTX 590。
任何帮助将不胜感激。

Answer 1

努力避免成为 "Stringly Typed" ，在 CUDA 4.1 的 CUDA 运行时 API 函数中不推荐使用字符串来引用设备符号，并在 CUDA 5.0 中删除。

CUDA 5 release notes阅读:

** The use of a character string to indicate a device symbol, which was possible with certain API functions, is no longer supported. Instead, the symbol should be used directly.

如果您将代码更改为以下内容，它应该可以工作。

Status = cudaMemcpyToSymbol(CONSTANT_PARAMETER, &m_dParameter, sizeof(m_dParameter));
ERROR_CHECK(Status);