Vfrdtyky

I'm trying to produce a mean filter of sliding window 3 x 3. I'm inputting a 16 x 16 array (arrays16.txt --> int matrix) and trying to allocate 16 x 16 threads for each block (1 block right now). Using cuda-memcheck i'm receiving a number of sync/async errors and I have been going around and around. Is there something obviously wrong? I understand it is some sort of segfault and probably an issue with my pointers but I can't seem to get past it.Using cuda-memcheck with the -lineinfo flag it seems to be happening in the loop in the kernel.

Here is my code, thank you:

#include <stdio.h>

#include <time.h>

#include <cuda.h>

#include <cuda_runtime.h>

#include <stdlib.h>



#define MAXR 16

#define MAXC 16



__global__ void imagefilter(float ** intermediates_d, int ** result_d) {



    int idx =  threadIdx.x;

    int idy = threadIdx.y;



    int x,y;

    //result_d[2][2]= 5;

    //if ((idx < 15) && (idy < 15)) {

    result_d[x][y] = result_d[idx][idy];

for(x=1; x < MAXR; x++) {

    for(y=1; y < MAXC; y++) {

            result_d[y][x] = (int) (((float) (intermediates_d[idx - 1][idy- 1]

            + intermediates_d[idx - 1][idy]

            + intermediates_d[idx - 1][idy + 1]

            + intermediates_d[idx][idy - 1] + intermediates_d[idx][idy]

            + intermediates_d[idx][idy + 1]

            + intermediates_d[idx + 1][idy - 1]

            + intermediates_d[idx + 1][idy]

            + intermediates_d[idx + 1][idy + 1]) / 9.0F));

            //  result_d[2][2]= 5;

            result_d[idy][idx]= result_d[y][x];

        }

    }

}







    __syncthreads();







int main(void)



{

    int i, j;

    //double cpu_time_used;

    float intermediates[MAXR][MAXC]; // taking input matrix and converting it to floating

    int matrix[MAXR][MAXC]; // This is the input matrix from file

    int result[MAXR][MAXC]={{0}}; //This is where we want to write the mean values. For now set to zeros

    float ** intermediates_d;

    //int **matrix_d;



    int ** result_d;

    int datasize_f = MAXR * MAXC * sizeof(float);

    int datasize_i = MAXR * MAXC * sizeof(int);

    //Allocate memory on the host.



    cudaMalloc((void**) &intermediates_d, datasize_f);

    //cudaMalloc((void**) &matrix_d, datasize);

    cudaMalloc((void**) &result_d, datasize_i);



    FILE *fp;

    fp = fopen("arrays16.txt", "r"); // reads in matrix

    //clock_t start =clock();

    for (i = 0; i < MAXR; i++) // this loop takes the information from .txt file and puts it into arr1 matrix

            {

        for (j = 0; j < MAXC; j++)

        {

            fscanf(fp, "%dt", &matrix[i][j]);

        }

    }



    printf("*****INPUT MATRIX*****n");

    for (i = 0; i < MAXR; i++)

    {

        printf("n");

        for (j = 0; j < MAXC; j++) {

            printf("%d ", matrix[i][j]);

        }

    }

    printf("nn");



    //This is where we convert the input matrix into floating point in intermediate matrix

    for (int y = 0; y < MAXR; y++) {



        for (int x = 0; x < MAXC; x++) {

            intermediates[y][x] = (float) matrix[y][x];

        }

    }

    printf("*******INTERMEDIATE MATRIX*******n");

    for (i = 0; i < 16; i++) {

        printf("n"); // prints out the results array to .txt file

        for (j = 0; j < 16; j++) {

            printf("%.1f ", intermediates[i][j]);

        }

    }

    printf("nn");

    // copying the data from the host array to the device array



    //cudaMemcpy(matrix_d, matrix, datasize,



    //cudaMemcpyHostToDevice);



    cudaMemcpy(intermediates_d, intermediates, datasize_f,cudaMemcpyHostToDevice);

    cudaMemcpy(result_d, result, datasize_i, cudaMemcpyHostToDevice);



    // how many blocks we will allocate

    dim3 blocks(1, 1);



    //how many threads per block we will allocate

    dim3 threadsPerBlock(16, 16);



    //Launch Kernel

    imagefilter<<<blocks, threadsPerBlock,MAXR*MAXC*sizeof(float)>>>(intermediates_d,result_d);



    //Copy back Results Matrix.

    cudaMemcpy(result, result_d, datasize_i, cudaMemcpyDeviceToHost);



    cudaError_t errSync = cudaGetLastError();

    cudaError_t errAsync = cudaDeviceSynchronize();



    if (errSync != cudaSuccess)

        printf("Sync kernel error: %sn", cudaGetErrorString(errSync));

    if (errAsync != cudaSuccess)

        printf("Async kernel error: %sn", cudaGetErrorString(errAsync));



    FILE *file;

    file = fopen("results.txt", "w+"); // writes  matrix to file

    printf("*******RESULTS MATRIX******nn");

    for (i = 1; i < MAXR - 1; i++) { // prints out the results array to .txt file

        for (j = 1; j < MAXC - 1; j++) {

            printf("%d ", result[i][j]);

            fprintf(file, "%d ",result[i][j]);

        }

        printf("n");



        fprintf(file, "n");

    }



    fclose(file);

}