Re: Octave in Google Summer of Code 2012

[Max Brister]

On Mon, Apr 23, 2012 at 4:44 PM, Luke M <address@hidden> wrote:
>
> Max Brister wrote
>>
>> 2012/4/23 Jordi Gutiérrez Hermoso <jordigh@>:
>>> Good news everyone!
>>>
>>> I'm pleased to announce that Octave will be mentoring three students
>>> this summer under the Google Summer of Code programme.
>>>
>>> Max Dawid (fisheater in #octave) will be primarily mentored by jwe on
>> Wrong last name there Jordi, it should be Max Brister.
>>
>> I'm very excited to get the opportunity to implement JIT compilation
>> in GNU Octave. It is nice to see so many projects accepted for Octave!
>> I wish everyone a good summer, and I hope to see you all at OctCof
>> 2012.
>>
>> Max Brister
>>
>
> These are some great SoC projects, but I have to admit I'm most excited
> about yours.  All I could find was a one sentence abstract, so how about
> some details?

I have attached my proposal (also at
http://jordi.platinum.linux.pl/octave/max-brister-proposal.pdf). I'm
more than happy to discuss project details. I would really like to run
a bit ahead of schedule, so I can get to some of the interesting bits
in section 7.5.

> Trace-based, static?

The idea is to do trace-based JIT, but instead of taking proper timing
information just compile all loops (as we run across them). I *think*
this should work well for most numerical analysis code, as loops in
numerical analysis code tend to run for several thousand iterations.
The implementation should also be easily extendable to be fully
trace-based if deemed necessary. We would just have to add a more
complex metric for determining when to compile code.

This approach is also used by FreeMat.

> LLVM, libjit?

LLVM

> Will you have a public repository for your branch?

My branch will be publicly visible. I'm not sure of the exact details
yet, but I think Jordi was setting something up. I'm sure there will
be something on octave-maintainers list when everything is set up.

> Sorry for all the badgering, just excited and I hope I can help a little
> (and learn a lot).

I don't mind at all, I'm glad that there is interest the project. I
appreciate any input/help as I also have quite a lot to learn.

Max Brister

proposal.tex
Description: TeX document