-4

I just started working with CUDA and tried writing some C code. The checkStr() function is getting an array of strings, the length of that array, quantity of threads and a solution/passphrase. The Threads executing this method then check if a string from that arrray matches the solution/passphrase. Since there might be more strings than threads, one thread might check for multiple strings.

__device__ bool device_strings_equal(const char* a, const char* b) {
    int i = 0;
    while (a[i] == b[i]) {

        if (a[i] == '\0') {
            return true; // matches passkey
        }
        i++;
    }

    return false; // doesnt match 
}

__global__ void checkStr(char** word, int list_length, char* solution, int* results) {
    
    int ThreadID = blockIdx.x * blockDim.x + threadIdx.x;
    int strings_per_thread = (list_length + blockDim.x - 1) / blockDim.x;

    //  define what strings the threads should use
    int start_indx = ThreadID * strings_per_thread;
    int end_indx = start_indx + strings_per_thread;
    if (end_indx > list_length) {
        end_indx = list_length;
    }
    
    for (int i = start_indx; i < end_indx; ++i) {

        results[i] = 0;     //  initializing this value just in case

        if (!device_strings_equal(word[i], solution)) {
            results[i] = 1;
        }
    }

}


int main()
{
    char* list[] = { "abccc\0", "yhdrh\0", "ydhydhyh\0" , "yhfxjtj\0" , "yhrtjksy\0" , "yrgb\0" , "xjjtj\0" , "syjhsrjx\0" , "zsjkjh\0" , "aehhggg\0" };
    char** gpu_list;
    int len = 10;
    char* solution = "yhfxjtj\0";
    int* gpu_results;

    cudaDeviceSynchronize();

    //  allocate memory on gpu
    cudaMalloc((void**) &gpu_list, len * sizeof(char*));
    cudaMalloc((void**) &gpu_results, len * sizeof(int));
    

    // copy list over to gpu
    cudaMemcpy(gpu_list, list, len * sizeof(char*), cudaMemcpyHostToDevice);

    int thread_per_block = 5;
    int blocksPerGrid = (len + thread_per_block - 1) / thread_per_block;

    checkStr << <blocksPerGrid, thread_per_block >> > (gpu_list, len, solution, results);

    //  get result array from cuda methos
    cudaDeviceSynchronize();
    int result_list[10];
    cudaMemcpy(result_list, gpu_results, len * sizeof(int), cudaMemcpyDeviceToHost);

    //  printing results 
    for (int i = 0; i < len; ++i) {
        printf("RESULT: %d\n", result_list[i]);
    }


    cudaFree(gpu_list);
    cudaFree(gpu_results);

    return 0;
}

I expected an output of only 0's and one 1. But instead i only get the output:

RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460
RESULT: -858993460

As far as i know that means that the integers have not been initialized yet. When using the Nsight debugger, the result array does contain the expected values, but a regular debugger or even executing the program leads to the wrong output again. It seems like the ...

cudaMemcpy(result_list, gpu_results, len * sizeof(int), cudaMemcpyDeviceToHost);

... method doesnt work (or isnt used) as intended. I'd really appreciate any help

talonmies
  • 70,661
  • 34
  • 192
  • 269
Timovoncorvin
  • 1
  • 1
  • 2
  • 4
    your kernel is performing illegal activity. Because of that, you are getting garbage. You might wish to learn how to do [proper CUDA error checking](https://stackoverflow.com/questions/14038589/what-is-the-canonical-way-to-check-for-errors-using-the-cuda-runtime-api) and also learn to [run your codes](https://stackoverflow.com/questions/27277365/unspecified-launch-failure-on-memcpy/27278218#27278218) with `compute-sanitizer`. – Robert Crovella Aug 21 '23 at 19:27
  • 2
    This: `cudaMalloc((void**) &gpu_list, len * sizeof(char*));` only allocates **space** for a set of pointers. It does not initialize any of the pointers in that set or provide any device memory allocations for them to point to. The general topic of allocating for double-pointers in CUDA tends to trip people up, so you might want to consider flattening your strings into a single array of `char`. If you really want an array of pointers then you need to [study that](https://stackoverflow.com/questions/45643682/). Additionally, your kernel references a variable `results` which is not in your code. – Robert Crovella Aug 21 '23 at 19:49
  • Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. – Community Aug 22 '23 at 09:15

0 Answers0