pinto at mit.edu
Fri Nov 20 10:54:17 PST 2009
>> Are there libraries to help with this?
> First of all, since it's a thorny (and unsolved) problem, PyCUDA doesn't try
> to get involved in it. Support it--yes, involved--no. That said, I'm not aware
> of libraries that make autotunig significantly easier. Nicolas mentioned that
> he's eyeing some machine learning techniques like the ones in Milepost gcc.
> Nicolas, care to comment? Aside from that, Cray's "grouped, attributed
> orthogonal search"  sounds useful.
The ideas I have are too immature (and I don't have much time to
explore them, sadly). However, my first step would be to integrate
more information about a given kernel execution (e.g. profiling
counters). I need to see if there is a way to make that fit nicely in
PyCUDA (e.g. regex on the profile log) to help the iterative "machine
learning" approach to auto-tuning find a good answer, quickly.
Hopefully I'll have more to say soon.
Ph.D. Candidate, Brain & Computer Sciences
Massachusetts Institute of Technology, USA
More information about the PyCUDA