在CUDA中一起添加字符串？

Question

What I want to do is have both strings processed in the device and called on the host. I get an error telling me that calling host function from global is not allowed. With that in mind, how would I go about re-structuring this code so that it is run through the device?

__global__ void whatever(string *x, string *y){
    *x += *y;}
int main(){
    string x, y, *x_in, *y_in;
    x = "Hello ";
    y = "World!\n";
    cudaMalloc((void**)&x_in, sizeof(int));
    cudaMalloc((void**)&y_in, sizeof(int));
    cudaMemcpy(x_in, &x, sizeof(int), cudaMemcpyHostToDevice);
    cudaMemcpy(y_in, &y, sizeof(int), cudaMemcpyHostToDevice);
    whatever << <1, 1 >> > (x_in, y_in);
    cudaMemcpy(&x, x_in, sizeof(int), cudaMemcpyDeviceToHost);
    cudaMemcpy(&y, y_in, sizeof(int), cudaMemcpyDeviceToHost);
    cudaFree(&x_in);
    cudaFree(&y_in);
    cout << x;
    return 0; }

Answer 1

Answer to your question: you call *x += *y; , ie concatenation and assignment operation for std::string objects (obviously defined only as a host function in STL).

About restructuring - you can pass to device char* buffers (of course, allocated on the device) instead of string objects and work with them.

Some more points about the code:

• You copy only sizeof(int) bytes from/to device, why? I see that your data is more than 4 bytes;
• You allocate only sizeof(int) bytes for each object, why?
• You use << <1, 1> >> syntax instead of <<<1, 1,>>> (I believe that it's only formatting issue after copy-pasting code here);
• And the main question - possibly you describe what do you want to achieve by this piece of code?