CUDA memory does not return to host

Question

Okay so I'm trying to learn CUDA for the 'new' FX 570 I bought for $15 ;D now in the code there are NO errors, the the array1_host starts off with it's values correctly, but when I copy the memory from device to host the values remain the same. the same thing happens if I blank out the second kernel call (trying multiple kernels in this project) I'm rather confused so thank you for any help I can achieve :)

#include <cuda_runtime.h>
#include <iostream>

#pragma comment (lib, "cudart")

#define N 5000

__global__ void addArray(float* a, float* b)
{
   a[threadIdx.x] += b[threadIdx.x];
}
__global__ void timesArray(float* a, float* b)
{
   a[threadIdx.x] *= b[threadIdx.x];
}

int main(){
   float array1_host[N];
   float array2_host[N];

   float *array1_device;
   float *array2_device;

   cudaError_t err;

   for(int x = 0; x < N; x++){
       array1_host[x] = (float) x * 2;
       array2_host[x] = (float) x * 6;
   }

   err = cudaMalloc((void**)&array1_device, N*sizeof(float));
   err = cudaMalloc((void**)&array2_device, N*sizeof(float));

   err = cudaMemcpy(array1_device, array1_host, N*sizeof(float),   cudaMemcpyHostToDevice);
   err = cudaMemcpy(array2_device, array2_host, N*sizeof(float), cudaMemcpyHostToDevice);

   dim3 dimBlock( N );
   dim3 dimGrid ( 1 );

   addArray<<<dimGrid, dimBlock>>>(array1_device, array2_device); 
   timesArray<<<dimGrid, dimBlock>>>(array1_device, array2_device);

   err = cudaMemcpy(array1_host, array1_device, N*sizeof(float), cudaMemcpyDeviceToHost);

   cudaFree(array1_device);
   cudaFree(array2_device);

   std::cout << cudaGetErrorString(err) << "\n\n\n\n\n\n";
   std::cout << array1_host;


   cudaDeviceReset();

   system("pause");
   return 0;
}

Answer 1

You have an error, because N is 5000, but there are limits for threds in block - it depends on Compute Capability link to features on wiki .
Try this code:

#define K 200

....

dim3 dimBlock( K );
dim3 dimGrid ( N/K );

To debug your code you can use cudaGetLastError() after each call of kernel or other function to know, where bugs are placed exaple about CUDA errors .