Cuda equivlent of alloca

Question

I am wondering if there is a CUDA equivalent of the alloca function.

I need to create arrays of floats which act as the arguments to the mathematical function I am trying to optimize. The issue is that I don't really want to have to know the number of arguments at compile time which is what I am doing now with templates. I could use the new operator in CUDA but I feel that it is slow (maybe I could preallocate it or something). I would use shared memory but it is not big enough.

Answer 1

There is nothing I am aware of which works like alloca for CUDA. The stack frame in the CUDA ABI is statically allocated by the assembler at compile time anyway, so I doubt there would be any scope for dynamically allocating memory on the stack frame