Is OpenScop still interchangeable with isl?

[Ruizhe Zhao]

Hi there,

On the OpenScop website it says isl is interchangeable with the OpenScop format, however, I cannot find anywhere in the isl doc mentioning this.

A bit of the context: I'm working on polyhedral transformations for MLIR, and we've already supported a workflow from MLIR -> OpenScop -> Pluto -> OpenScop -> MLIR (code: https://github.com/kumasento/polymer, paper: https://acohen.gitlabpages.inria.fr/impact/impact2021/papers/IMPACT_2021_paper_1.pdf). And I think it would be a huge advantage if we can support isl and benefit from the schedulers implemented by it.

Obviously we can build isl objects on-the-fly, but I'm just wondering if any of the existing infrastructure, which is based on OpenScop, can be reused. 

A side question: if OpenScop is not supported anyway, would it be recommended to use the JSCoP format from Polly? It seems that JSCoP has better interaction with isl AFAIK.

Any comments are welcomed! Thanks :)

Best regards

Ruizhe 

[Sven Verdoolaege]

On Mon, May 17, 2021 at 03:59:23PM -0700, Ruizhe Zhao wrote:
> Hi there,
>
> On the OpenScop website it says isl is interchangeable with the OpenScop
> format, however, I cannot find anywhere in the isl doc mentioning this.

I have no idea what that is about either. You should ask the OpenScop
developers.
I wanted to ask about that after the presentation of the paper you
mention below, but I can't remember if I did.

I vaguely remember OpenScop has a rather convoluted way of encoding
array accesses. I wouldn't recommend it.

> A bit of the context: I'm working on polyhedral transformations for MLIR,
> and we've already supported a workflow from MLIR -> OpenScop -> Pluto ->
> OpenScop -> MLIR (code: https://github.com/kumasento/polymer, paper:
> https://acohen.gitlabpages.inria.fr/impact/impact2021/papers/IMPACT_2021_paper_1.pdf).
> And I think it would be a huge advantage if we can support isl and benefit
> from the schedulers implemented by it.
>
> Obviously we can build isl objects on-the-fly, but I'm just wondering if
> any of the existing infrastructure, which is based on OpenScop, can be
> reused.
>
> A side question: if OpenScop is not supported anyway, would it be
> recommended to use the JSCoP format from Polly? It seems that JSCoP has
> better interaction with isl AFAIK.

I'm not that familiar with that format either.

skimo

[Ruizhe Zhao]

Thank you so much for your reply Sven :)

Yes, we didn't catch up after the presentation, which is something I really wish I did!

I think I will start with isl from scratch without messing up with the current OpenSCoP logic in my codebase. I will post specific questions later if I meet any.

Best regards

Ruizhe

[Amy Wang]

Hello Ruizhe, I'm curious to know if you've successfully gotten isl to accept OpenScop as an input format?   (e.g. an OpenScop to Schedule Trees converter for isl)  Thank you!

[Amy Wang]

Hi Ruizhe, I'm curious if you've gotten isl to accept OpenScop as an input format?  (e.g. creating an OpenScop to Schedule Trees converter)  Thanks!   -Amy

(I've sent a message earlier but it didn't seem to come through....apology if you accept this message twice)

On Wednesday, May 19, 2021 at 4:39:20 AM UTC-4 Ruizhe Zhao wrote:

[Ruizhe Zhao]

Hey Amy

I’m afraid I didn't spend much time on this, so if no one else has been working on it, I guess the answer is no.

Best regards

Ruizhe

--

---
You received this message because you are subscribed to a topic in the Google Groups "isl Development" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/isl-development/5LjpY2RU7rQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to isl-developme...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/isl-development/d1dc3afe-4043-43f0-829b-43818b12b9e1n%40googlegroups.com.

--

Ruizhe Zhao

https://kumasento.github.io