status of project

[rohany]

Hi,

I was looking at using the MLIR stencil dialect defined in this project to interface with some other MLIR code. I was wondering if the pieces here are extractable in that way, or if this code is expected to build / work correctly against the current mainline MLIR.

Thanks!

[PapyChacal]

Hi @rohany,

I don't know about the extractability, but this dialect's implementation is out of sync and does not build against current mainline MLIR. However, we ported parts of it in xDSL! We successfully used it to run stencils on CUDA or OpenMP, through MLIR. A few disclaimers:

• We still have a subset of the stencil dialect available only, and it's not yet stable or thorougly tested. We plan to work on that over the coming weeks though!
• xDSL is maintained and tested against a fixed commit of MLIR, as both projects evolve rapidly. At the time of writing, we work with the commit 89996621d, dated from January. Though we try to regularly move to the MLIR mainstream, to keep in sync.

I invite you to come chat on xDSL's Zulip, it's our main communication platform.

[gysit]

Hi @rohany,

The project is indeed unmaintained. I assume it would require quite some effort to make it work with a recent version of MLIR and I do not really have the bandwidth to support such an effort.

I would thus recommend to have a look at the xDSL version mentioned by @PapyChacal. Excited to hear the stencil dialect has a second life!

[rohany]

Thanks for the pointer at @PapyChacal! I took a look at xDSL (looks cool!), but I'm looking for a C++ MLIR dialect rather than a Python library that interacts with MLIR through file IO (this could be a wrong take on xDSL, please let me know!). I might give compiling / porting this library to mainline MLIR a shot, maybe it won't be too bad.