Inconsistent behaviour: CUDA 8 on GTX 1080

Accelerated Computing CUDA CUDA Programming and Performance
1

'm experiencing an inconsistent behaviour while running my code using Cuda 8 on a GTX 1080. If I try to run this code on my GTX 1080 (Ubuntu 16.04, Cuda 8)

__global__
void setValue(int *ptr, int index, int val)
{
        ptr[index] = val;
}

void foo(int size)
{
        int *data;
        CudaSafeCall(cudaMallocManaged(&data, size*sizeof(int)));
        memset(data, 0, size*sizeof(int));
        setValue<<<1,1>>>(data, size/2, 5);
        CudaCheckError();
        CudaSafeCall(cudaDeviceSynchronize());
        //useData(data);
        for(int i=0;i<size;i++)
                printf("%u %u\n",i,data[i]);
        CudaSafeCall(cudaFree(data));
}


int main()
{
        foo(5);
}

I get the following output

0 0
 1 0
 2 0
 3 0
 4 0

while if I run it on a machine with a GeForce GT 740 and Cuda 7 I get

0 0
 1 0
 2 5
 3 0
 4 0

I really cannot explain this behavior. I’m checking the CudaErrors but everything is ok even on the faulty code. Any hint about how I can debug this behaviour?

2

My best guess is that the CUDA install on the Ubuntu 16.04 machine is not working correctly.

essentially similar to the comment already on your cross-posting on SO