cuda中矢量加法的分段故障

Question

I was messing with a toy program for cuda .

I declare a float array transfer that to gpu and a number to each element of that float array and transfer it back to the host system and print the array. However this is not working out and it is giving me segmentation fault.

Here's code

#include <iostream>
using namespace std;

__global__ void kern(float *a, float *C){
    for (int i = 0; i < 3; i++) C[i] = a[i] + i;
}

int main(){
    float *A = new float[3];
    for(int i = 0; i < 3; i++){
        A[i] = i;
    }

    float * d;
    float * C;
    cudaMalloc(&C, sizeof(float)*3);
    cudaMalloc(&d, sizeof(float)*3);
    cudaMemcpy(&d, A, sizeof(float)*3, cudaMemcpyHostToDevice);
    kern<<<1, 1>>>(d, C);

    cudaMemcpy(&A, C, sizeof(float)*3, cudaMemcpyDeviceToHost);

    cout << A[2];

}

Also I am not familiar with Malloc most of my experience was with cpp and hence I am more comfortable with new datatype[]; is there a equivalent for Cuda?

Answer 1

Change this to:

cudaMemcpy(&d, A, sizeof(float)*3, cudaMemcpyHostToDevice);
cudaMemcpy(&A, C, sizeof(float)*3, cudaMemcpyDeviceToHost);

To this:

cudaMemcpy(d, A, sizeof(float)*3, cudaMemcpyHostToDevice);
cudaMemcpy(A, C, sizeof(float)*3, cudaMemcpyDeviceToHost);

Also it's always better to store return code by CUDA calls they will give you better idea what going wrong.