Question

我在Cuda-C应用程序中有这个代码行：

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
#include <thrust/sort.h>
#include <stdio.h>

#include <time.h>

#include <device_functions.h>

int main()
{
    const int size = 32;
    unsigned int * dev_ips_range_end;
    unsigned int * ips_range_end = new unsigned int[size];
    for (int i = 0; i < size; i++)
        ips_range_end[i] = i;



cudaError_t cudaStatus;
    cudaStatus = cudaSetDevice(0);
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "cudaSetDevice failed!  Do you have a CUDA-capable GPU installed?");
        goto Error;
    }
    cudaStatus = cudaMalloc((void**)&dev_ips_range_end, size * sizeof(int));
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "Problem !");
        goto Error;
    }

    // Copy input vectors from host memory to GPU buffers.
    cudaStatus = cudaMemcpy(dev_ips_range_end, ips_range_end, size * sizeof(unsigned char), cudaMemcpyHostToDevice);
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "Problem !");
        goto Error;
    }

    thrust::device_ptr<unsigned int> dev_ips_range_end_ptr(dev_ips_range_end);
    thrust::inclusive_scan(dev_ips_range_end_ptr, dev_ips_range_end_ptr + size, dev_ips_range_end_ptr);
    return 0;

Error:
    cudaFree(dev_ips_range_end);
}

这是我使用的命令和输出：

[测试] $ nvcc -I / usr / local / cuda / include -L / usr / local / cuda / lib kernel.cu -o test.run kernel.cu（27）：错误：控制转移绕过初始化：变量＆＃34; dev_ips_range_end_ptr＆＃34; （42）：这里

kernel.cu（32）：错误：控制转移绕过初始化：变量＆＃34; dev_ips_range_end_ptr＆＃34; （42）：这里

kernel.cu（39）：错误：控制转移绕过初始化：变量＆＃34; dev_ips_range_end_ptr＆＃34; （42）：这里

在＆＃34; /tmp/tmpxft_000022ad_00000000-9_kernel.cpp1.ii"的汇编中检测到3个错误。

相同的代码在Windows上的visual studio中没有任何问题。如何解决这个问题？

Answer 1

有些人可能会告诉你，在C / C ++中使用goto并不是一个好主意。但是为了避免参数，并允许您保持相同的代码结构，您可以将推力设备指针声明在程序的顶部（在任何goto语句之前），然后在准备使用时设置指针值它，像这样：

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
#include <thrust/sort.h>
#include <stdio.h>

#include <time.h>

#include <device_functions.h>

int main()
{
    const int size = 32;
    unsigned int * dev_ips_range_end;
    unsigned int * ips_range_end = new unsigned int[size];
    for (int i = 0; i < size; i++)
        ips_range_end[i] = i;



    thrust::device_ptr<unsigned int> dev_ips_range_end_ptr;
cudaError_t cudaStatus;
    cudaStatus = cudaSetDevice(0);
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "cudaSetDevice failed!  Do you have a CUDA-capable GPU installed?");
        goto Error;
    }
    cudaStatus = cudaMalloc((void**)&dev_ips_range_end, size * sizeof(int));
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "Problem !");
        goto Error;
    }

    // Copy input vectors from host memory to GPU buffers.
    cudaStatus = cudaMemcpy(dev_ips_range_end, ips_range_end, size * sizeof(unsigned char), cudaMemcpyHostToDevice);
    if (cudaStatus != cudaSuccess) {
        fprintf(stderr, "Problem !");
        goto Error;
    }

    dev_ips_range_end_ptr = thrust::device_pointer_cast(dev_ips_range_end);
    thrust::inclusive_scan(dev_ips_range_end_ptr, dev_ips_range_end_ptr + size, dev_ips_range_end_ptr);
    return 0;

Error:
    cudaFree(dev_ips_range_end);
}

CUDA：错误：＆＃34;控制权转移绕过＆＃34;在创建thrust :: device_ptr时

1 个答案: