错误：属性“__global__”在这里不适用

Question

nvcc 无法识别带有 cuda 12.5 的 Windows 11 上的内核 （我可以使用 powershell 编译其他 *.cu 文件）

nvcc -arch=sm_89 .\simplest_kernel.cu

#include <cuda_runtime.h>
#include <iostream>
#include <vector>

__global__ void kernel(uint *A, uint *B, int row) {
  auto x = threadIdx.x / 4;
  auto y = threadIdx.x % 4;
  A[x * row + y] = x;
  B[x * row + y] = y;
}

int main(int argc, char **argv) {
  uint *Xs, *Ys;
  uint *Xs_d, *Ys_d;

  uint SIZE = 4;

  Xs = (uint *)malloc(SIZE * SIZE * sizeof(uint));
  Ys = (uint *)malloc(SIZE * SIZE * sizeof(uint));

  cudaMalloc((void **)&Xs_d, SIZE * SIZE * sizeof(uint));
  cudaMalloc((void **)&Ys_d, SIZE * SIZE * sizeof(uint));

  dim3 grid_size(1, 1, 1);
  dim3 block_size(4 * 4);

  kernel<<<grid_size, block_size>>>(Xs_d, Ys_d, 4);

  cudaMemcpy(Xs, Xs_d, SIZE * SIZE * sizeof(uint), cudaMemcpyDeviceToHost);
  cudaMemcpy(Ys, Ys_d, SIZE * SIZE * sizeof(uint), cudaMemcpyDeviceToHost);

  cudaDeviceSynchronize();

  for (int row = 0; row < SIZE; ++row) {
    for (int col = 0; col < SIZE; ++col) {
      std::cout << "[" << Xs[row * SIZE + col] << "|" << Ys[row * SIZE + col]
                << "] ";
    }
    std::cout << "\n";
  }

  cudaFree(Xs_d);
  cudaFree(Ys_d);
  free(Xs);
  free(Ys);
}

结果

PS D:\samples\api\SGEMM_CUDA> nvcc -arch=sm_89 .\simplest_kernel.cu
simplest_kernel.cu
D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: attribute "__global__" does not apply here
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
             ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: incomplete type "void" is not allowed
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                              ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: identifier "uint" is undefined
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                                     ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: identifier "A" is undefined
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                                           ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: identifier "B" is undefined
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                                                    ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: type name is not allowed
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                                                       ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: expected a ")"
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {
                                                           ^

D:\samples\api\SGEMM_CUDA\simplest_kernel.cu(5): error: expected a ";"
  __declspec(__global__) void kernel(uint *A, uint *B, int row) {

Answer 1

从您的错误消息中我可以看到您正在 Windows 上运行。

仅在 Linux 上定义了

uint

，在 Windows 上则没有定义。

您可以使用

unsigned

轻松解决此问题，或者根据需要定义

uint

。以下修复将使您的代码编译并正确运行。

添加因此根据需要定义 uint （请注意，在现代平台上

unsigned

表示

unsigned int

，又名

uint32_t

）：

#ifndef uint
    typedef unsigned uint;
#endif

完整代码（见下文）现在可以正确编译并运行，输出正确的结果。

#include <cuda_runtime.h>
#include <iostream>
#include <vector>

#ifndef uint
   typedef unsigned uint;
#endif

__global__ void kernel(uint* A, uint* B, int row) {
    auto x = threadIdx.x / 4;
    auto y = threadIdx.x % 4;
    A[x * row + y] = x;
    B[x * row + y] = y;
}

int main(int argc, char** argv) {
    uint* Xs, * Ys;
    uint* Xs_d, * Ys_d;

    uint SIZE = 4;

    Xs = (uint*)malloc(SIZE * SIZE * sizeof(uint));
    Ys = (uint*)malloc(SIZE * SIZE * sizeof(uint));

    cudaMalloc((void**)&Xs_d, SIZE * SIZE * sizeof(uint));
    cudaMalloc((void**)&Ys_d, SIZE * SIZE * sizeof(uint));

    dim3 grid_size(1, 1, 1);
    dim3 block_size(4 * 4);

    kernel << <grid_size, block_size >> > (Xs_d, Ys_d, 4);

    cudaMemcpy(Xs, Xs_d, SIZE * SIZE * sizeof(uint), cudaMemcpyDeviceToHost);
    cudaMemcpy(Ys, Ys_d, SIZE * SIZE * sizeof(uint), cudaMemcpyDeviceToHost);

    cudaDeviceSynchronize();

    for (int row = 0; row < SIZE; ++row) {
        for (int col = 0; col < SIZE; ++col) {
            std::cout << "[" << Xs[row * SIZE + col] << "|" << Ys[row * SIZE + col]
                << "] ";
        }
        std::cout << "\n";
    }

    cudaFree(Xs_d);
    cudaFree(Ys_d);
    free(Xs);
    free(Ys);
}