[GSoC 2016] Enabling Polyhedral Optimizations in Julia

[Matthias Reisinger]

Dear Julia contributors,

during the last weeks I extensively explored Polly, a framework for polyhedral optimizations in LLVM. From discussions with the contributors of Polly I learned that there is great interest in making it possible to integrate Polly into Julia. I plan to apply for Google Summer of Code 2016 and therefore I wanted to ask if there would be a mentor available to supervise this topic. I also created a first draft for my according project proposal which can be found here https://docs.google.com/document/d/1s5mmSW965qmOEbHiM3O4XFz-Vd7cy9TxX9RQaTK_SQo/edit?usp=sharing. I also would be grateful for any comments or any kind of advice on the draft.

Thanks in advance.

Best regards,

Matthias

[Tobias Grosser]

Dear all,

I would be happy to co-mentor this project from the Polly/LLVM side, but
it would be great if someone with more experience with Julia could cover
the Julia side of things as well.

Best,
Tobias

[Mike Innes]

Hey Matthias,

Sorry to not be more immediately helpful, but at this stage I recommend going ahead and submitting the proposal through the GSoC page without a mentor. If the project looks like it'd be a good fit, I'm sure we can find you a great mentor later on.

Cheers,
Mike

[Tim Holy]

Don't know what I can promise in terms of concrete help (it will depend in
part on whether I have other GSoC commitments), but it's certainly a topic I
find interesting.

Best,
--Tim

[Shashi Gowda]

The proposal is labeled "incomplete" on the GSoC UI, could you complete it and do the final submission ASAP?

[Matthias Reisinger]

Hello Shashi,

I was working on the draft until the last minute to round it down, but I handed it in on time. Nevertheless, thank you for pointing that out. I am sorry for my late response, however, it seems I don't get notified when somebody posts in my topics. I should check my settings ;)

Best regards,

Matthias

[Douglas Bates]

I have an initial implementation of code to read the binary files in the soon to be registered "FlatBuffers" package.  (Repository is https://github.com/dmbates/FlatBuffers.jl)

Right now the IDL file needs to be translated to Julia code by hand but the correspondence is reasonably transparent.  At the end of next week I will take a look at writing a flatbuffers/src/idl_gen_julia.cpp code unless someone beats me to it.

[Matthias Reisinger]

Unfortunately I am not that familiar with FlatBuffers. I see that FlatBuffer could change the memory access behavior for some code parts, but I don't expect this to pose a problem to a first integration of Polly. Does your change have any further implications that would have to be considered when optimizing Julia's LLVM code via Polly?

Best regards,

Matthias

[Tony Kelman]

I think Doug posted that in the wrong thread.

[Matthias Reisinger]

Thank you, I've already supposed it was an accident :)