Stack memory trace on the GPU

How does CUDA manage stack memory space on the GPU?
How big the stack memory on the GPU?
Is it possible to trace the size variation of stack memory on the GPU?
Which tool, or approach is able to trace stack memory on the GPU?
Thank you

How to get the stack trace on gpu?
Thanks

Any updates? I would like to know how to get a stack trace at any point during the runtime(complete frames from the point in the host to the point in the device)? Or is it simply not possible right now? But how does cuda-gdb get the backtrace?

Thanks