Errors that occur when I start the kernel function in CUDA 5.5

问题

I installed CUDA5.5. Development environment is using Visual Studio 2010 Professional. And I tried to run the source code like the following. However, the red line was drawn to the part of "<<<" for some reason on Visual Studio. It is displayed Error "expression. Required" and. If anyone the same phenomenon is happening, please tell me how to solve.

Development environment--------------------------------------------------------------------------

         OS:Windows7 64bit
         Visual Studio 2010 Professional SP1
         CUDA 5.5

Phenomenon-----------------------------------------------------------------------------------------

↓Underlined red part of the "<<<" of source code you'll see the following. However, the third underline appears only "<". It appears to be: "expression Required. Error" and move the mouse pointer to the location of the red line.

Source code---------------------------------------------------------------------------

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

#define N 256


__global__ void matrix_vector_multi_gpu_1_1(float *A_d, float *B_d, float *C_d){
    int i,j;

    for(j=0;j<N;j++){
        A_d[j]=0.0F;
        for(i=0;i<N;i++){
            A_d[j]=A_d[j]+B_d[j*N+i]*C_d[i];
        }
    }
  }

int main(){
    int i,j;
    float A[N], B[N*N], C[N];
    float *A_d, *B_d, *C_d;

    dim3 blocks(1,1,1);
    dim3 threads(1,1,1);

    for(j=0;j<N;j++){
        for(i=0;i<N;i++){
            B[j*N+i]=((float)j)/256.0;
        }
    }

    for(j=0;j<N;j++){
        C[j]=1.0F;
    }

    cudaMalloc((void**)&A_d, N*sizeof(float));
    cudaMalloc((void**)&B_d, N*N*sizeof(float));
    cudaMalloc((void**)&C_d, N*sizeof(float));

    cudaMemcpy(A_d,A,N*sizeof(float),cudaMemcpyHostToDevice);
    cudaMemcpy(B_d,B,N*N*sizeof(float),cudaMemcpyHostToDevice);
    cudaMemcpy(C_d,C,N*sizeof(float),cudaMemcpyHostToDevice);

    matrix_vector_multi_gpu_1_1<<<blocks,threads>>>(A_d,B_d,C_d);

    cudaMemcpy(A,A_d,N*sizeof(float),cudaMemcpyDeviceToDevice);

    for(j=0;j<N;j++){
        printf("A[ %d ]=%f \n",j,A[j]);
    }
    getchar();
    cudaFree(A_d);
    cudaFree(B_d);
    cudaFree(C_d);
    return 0;
}

The place of occurrence

回答1:

You should at least change from

cudaMemcpy(A,A_d,N*sizeof(float),cudaMemcpyDeviceToDevice);

to

cudaMemcpy(A,A_d,N*sizeof(float),cudaMemcpyDeviceToHost);

A few more suggestions

• Run some CUDA sample code to see if you have setup CUDA correctly.
• Make sure your source code file have the external name .cu

After solving the cudaMemcpyDeviceToDevice issue, I can compile and run your code. And the result is corrected. You code should have no problem that prevent compiling.

回答2:

As I pointed out in my comment above, this is an Intellisense problem. The steps to enable Intellisense support on a Visual Studio 2010 CUDA project are summarized at

Setting Intellisense support for a Visual Studio 2010 CUDA project

来源：https://stackoverflow.com/questions/18954102/errors-that-occur-when-i-start-the-kernel-function-in-cuda-5-5