Status of Dynamic Parallelism Support in CUDANative.jl

gpu

#1

I found the following issue about dynamic parallelism, and I was wondering if there is any date by which CUDANative is expected to support dynamic parallelism in the future, or if there is any update on that.


#2

No changes yet, and I haven’t had the time to work on new features lately.
I also haven’t really needed dynamic parallelism; is it a feature you use often?


#3

I wouldn’t say I use it often, because I am still a Cuda newbie. But from my readings, I understand that it makes implementing GPU accelerated versions of recursive algorithms more natural. My application is kind of a search tree, actually an optimization algorithm. A BFS might be easy to implement by returning control to the CPU everytime a deeper level needs to be searched, leaving the tree management to the CPU and the number crunching to the GPU. I was just considering dynamic parallelism as the more natural alternative for these kind of problems.


#4

I’ll put it somewhat higher in my TODO list :slight_smile:
No promises though, especially with JuliaCon and the summer holidays coming up.


#5

Great. Thanks a lot.