Ich implementiere eine Funktion, um die Summe eines Arrays mit Hilfe der Reduktion zu ermitteln. Mein Array hat 32 * 32 Elemente und seine Werte sind 0 ... 1023. Der erwartete Summenwert ist 523776, aber das Ergebnis ist 15872 falsch. Hier ist mein Code:

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

#define w 32
#define h 32
#define N w*h

__global__ void reduce(int *g_idata, int *g_odata);
void fill_array (int *a, int n);

int main( void ) {
    int a[N], b[N]; // copies of a, b, c
    int *dev_a, *dev_b; // device copies of a, b, c
    int size = N * sizeof( int ); // we need space for 512 integers

    // allocate device copies of a, b, c
    cudaMalloc( (void**)&dev_a, size );
    cudaMalloc( (void**)&dev_b, size );

    fill_array( a, N );

    // copy inputs to device
    cudaMemcpy( dev_a, a, size, cudaMemcpyHostToDevice );
    cudaMemcpy( dev_b, b, size, cudaMemcpyHostToDevice );

    dim3 blocksize(16,16);
    dim3 gridsize;

    gridsize.x=(w+blocksize.x-1)/blocksize.x;
    gridsize.y=(h+blocksize.y-1)/blocksize.y;

    reduce<<<gridsize, blocksize>>>(dev_a, dev_b);

    // copy device result back to host copy of c
    cudaMemcpy( b, dev_b, sizeof( int ) , cudaMemcpyDeviceToHost );

    printf("Reduced sum of Array elements = %d \n", b[0]);

    cudaFree( dev_a );
    cudaFree( dev_b );

    return 0;
}

__global__ void reduce(int *g_idata, int *g_odata) {

    __shared__ int sdata[256];

    // each thread loads one element from global to shared mem
    int i = blockIdx.x*blockDim.x + threadIdx.x;

    sdata[threadIdx.x] = g_idata[i];

    __syncthreads();
    // do reduction in shared mem
    for (int s=1; s < blockDim.x; s *=2)
    {
        int index = 2 * s * threadIdx.x;;

        if (index < blockDim.x)
        {
            sdata[index] += sdata[index + s];
        }
        __syncthreads();
    }

    // write result for this block to global mem
    if (threadIdx.x == 0)
        atomicAdd(g_odata,sdata[0]);
}

// CPU function to generate a vector of random integers
void fill_array (int *a, int n)
{
    for (int i = 0; i < n; i++)
        a[i] = i;
}