Open MLIR Meeting 3/30/2023: Representing Dataflow with MLIR

This Thursday, March 30th (9am California Time, 16:00 UTC ), @jabcross will talk about an IR for static dataflow. Here is the abstract:

Existing dataflow tools are difficult to integrate with the state-of-the-art industrial compilers. We approach this problem by creating an IR for static dataflow using MLIR, and demonstrate it by implementing a novel dead code elimination optimization that relies on both actor and graph information. In this presentation, we’ll focus on the challenges related to the use of MLIR and open the way for feedback and discussion regarding the found solutions.

Zoom Meeting Link is unchanged, the presentation will be recorded and posted here and on our talks page on the website as usual.
You can also subscribe to this Google calendar to keep informed about incoming meetings.

Meeting ID: 851 5109 0498
Passcode: 828404


Here are the slides and the recording for the meeting this morning.

Great presentation and work, @jabcross! I use a graphical dataflow programming language extensively and am looking at using MLIR to create a compiler for that language. Your work looks very relevant.

BTW, I strongly agree with your proposal (at the end of your presentation) for MLIR to support named outputs. Yes, referring outputs by index (since we cannot reference by name) makes code hard to maintain.

Do you have a github repo with any example code from your talk?

Thank you very much!

I do have a repository, but I’m currently in the process of cleaning it up and adding a simple multicore scheduler with the OpenMP dialect.

Once that’s done I’ll be sure to share it here!

1 Like