/* usual C/C++ includes */
#include <stdio.h>
#include <malloc.h>

/* CUDA includes */
#include <cuda_runtime_api.h>

/* Kernel: set each vector element to the number of the block
   running the corresponding thread */
__global__ void initVector(float *vec, const size_t size) {
    unsigned int idx = blockDim.x*blockIdx.x + threadIdx.x;
    vec[idx] = blockIdx.x;
}

const unsigned int vector_size = 256;
const unsigned int block_size = 32;

int main (int argc, char **argv) {
    int deviceCount = -1; // number of devices
    int dev = 0;

    /* Host and device vector pointers */
    float *hVector, *dVector;

    cudaGetDeviceCount(&deviceCount);

    if (deviceCount == 0) {
        fprintf(stderr, "No CUDA devices found\n");
        return 1;
    }

    cudaError_t error = cudaSetDevice(dev);
    if (error != cudaSuccess) {
        fprintf(stderr, "Error setting device to %d: %s\n",
            dev, cudaGetErrorString(error));
        return 1;
    }

    /* Allocate host vector */
    hVector = (float *)malloc(vector_size*sizeof(float));
    if (hVector == NULL) {
        fprintf(stderr, "Unable to allocate host vector\n");
        return 1;
    }

    /* Allocate device vector */
    error = cudaMalloc(&dVector, vector_size*sizeof(float));
    if (error != cudaSuccess) {
        fprintf(stderr, "Unable to allocate device memory: %s\n",
            cudaGetErrorString(error));
        return 1;
    }

    /* Zero device vector */
    cudaMemset(dVector, 0, vector_size*sizeof(float));

    /* Launch kernel: the first parameter is the number of blocks, the second the block size */
    initVector<<<vector_size/block_size, block_size>>>(dVector, vector_size);

    /* Wait for kernel execution to finish */
    error = cudaThreadSynchronize();
    if (error != cudaSuccess) {
        fprintf(stderr, "initVector failed: %s\n",
            cudaGetErrorString(error));
        return 1;
    }

    /* Copy data from device to host */
    error = cudaMemcpy(hVector, dVector, vector_size*sizeof(float), cudaMemcpyDeviceToHost);
    if (error != cudaSuccess) {
        fprintf(stderr, "cudaMemcpy failed: %s\n",
            cudaGetErrorString(error));
        return 1;
    }

    for (unsigned int i=0; i < vector_size; ++i) {
        printf("%3u: %g", i, hVector[i]);
        if ((i+1) & 0x0f)
            printf(", ");
        else
            printf("\n");
    }

    return 0;

}