CUDA Thrust: Finding the index of the first element in a vector satisfying a predicate (e.g., zero or negative) [Matlab's syntax min(find(x<=0))] - cuda

Question

I am attempting to find the the index of the first zero or negative value of an array using CUDA Thrust. The serial CPU code I am attempting to write using CUDA Thrust is the following:
for (int i = StartIndex; i <= ArrayLimitIndex; i++)
{
if (Array[i] <= 0) { DesiredIndex = i; break; }
}
I am thinking that the easiest way to do this on the GPU will be using the find_if function within the Thrust library.
The array is already on the GPU and I am attempting to search for the index on this array using Thrust as such:
struct less_than_or_eq_zero
{
__host__ __device__
bool operator() (double x)
{
return x <= 0;
}
};
thrust::device_vector<double>::iterator iter;
thrust::device_ptr<double> dev_ptr_Col46 = thrust::device_pointer_cast(dev_Col46);
iter = thrust::find_if(thrust::device, dev_ptr_Col46, dev_ptr_Col46 + size,less_than_or_eq_zero());
Now I would like to use the value of iter as an argument for my next kernel:
newKernel<<<size, 1>>>(*dev_array, iter)
where the newKernel definition is of the form:
__global__ void newKernel(double *dev_array, iter)
{
int x = blockIdx.x;
if(x <= iter)
{
//process data here...
}
}
I know that the code I have here is incorrect and I have a few questions regarding the use of iter. First, iter is a device_vector. Is there any way I can make iter just one value and not a vector? Also, when I have executed the find_if how can I use the value of iter in my next kernel call?
Any help with this be greatly appreciated.
Thanks

Answer 1

I'm summarizing the comments by talonmies and Jared Hoberock above as well as the answer by Sebastian Dressler in a fully compilable and executable example. The code calculates, by CUDA Thrust, the index of the first element of a vector satisfying a predicate (x<=0. in this case), I hope it will be helpful for future readers.
#include <thrust/device_vector.h>
#include <stdio.h>
struct less_than_or_eq_zero
{
__host__ __device__ bool operator() (double x) { return x <= 0.; }
};
int main(void)
{
int N = 6;
thrust::device_vector<float> D(N);
D[0] = 3.;
D[1] = 2.3;
D[2] = -1.3;
D[3] = 0.;
D[4] = 3.;
D[5] = -44.;
thrust::device_vector<float>::iterator iter1 = D.begin();
thrust::device_vector<float>::iterator iter2 = thrust::find_if(D.begin(), D.begin() + N, less_than_or_eq_zero());
int d = thrust::distance(iter1, iter2);
printf("Index = %i\n",d);
getchar();
return 0;
}

Answer 2

As you do not use a device_vector in your kernel but a raw array, you have to pass it an index and not an iterator. You can obtain the index by using thrust::distance to calculate the distance between dev_ptr_Col46 and iter.
You'll also want to read thrust iterators documentation, where distance is documented.

Answer 3

Try this:
thrust::device_ptr<double> val_ptr = thrust::find_if(dev_ptr_Col46, dev_ptr_Col46 + size,less_than_or_eq_zero());
double * val = thrust::raw_pointer_cast(val_ptr);
newKernel<<<size, 1>>>(dev_array, val)
Your kernel will have to have signature
__global__ void newKernel(double * dev_array, double * val)