Dear all,

I have a array of 100,000 float elements. I want to be able to construct a new array (containing either the elements or just the indices) that fulfill a certain condition (let us say the positive ones).

Is anyone aware of possible approaches to solve this problem on CUDA knowing that the number of size of the results array is not known at the beginning at the procedure and that it has to be fed in parallel with new elements.

I am pretty sure that this problem is common and that it has been studied before (database selection would be an example).

I would appreciate any help.

Thank you in advance.