[MLIR] Speed up compile times

Open

#467 opened on Jan 19, 2024 1 comment 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement

Dominant language: Python
Stars: 234
Forks: 84
Avg merge: 2d 15h
Merged PRs (30d): 66

Description

A strategy discussed in discord:

The easiest way to benefit from parallelism is to use function passes. Then MLIR will process all functions in parallel. This does not work if your pass does module-level changes such as deleting unreachable functions. However, it is perfect if the pass does function local rewrites. 1 Minute compilation time sounds quite slow but I guess it depends a lot on what transformations you are actually running

Is it possible to create the operations for separate functions in parallel? If so, I could create a work queue for threads to go build each function in order to leverage all available cores. Currently I do this on a single thread.

I guess in theory you could first create the functions sequentially and then in parallel fill the function bodies yes.

Filling the functions in parallel works! I'll need to test it with a much larger program to see the benefits though. [...] Generating 100% of the IR takes 2 seconds now. My reachable function analysis brings that down to a fraction of a second during generation now and of course the passes have less to work on so they're faster too.

Contributor guide

Open the contributing guide

First steps

Read the whole issue, then the project's contributing guide.
Comment on the issue to say you are picking it up — it saves two people doing the same work.
Fork the repository and make your change on a branch.
Open a pull request that references the issue number.

Research direction

The issue names no files, tests, or entry points. Start by reviewing the linked Discord discussion about parallel function generation and passes, then identify the relevant compiler entry points and establish a larger-program compile-time comparison; done means measurable speedup without changing compilation results.

Written by the indexing model from the issue text.

Assessment

Domain: compilers, performance
Issue type: Feature
Difficulty: 5/5
Estimated time: Over a week
Activity status: Stale
Clarity: Needs clarification
Newbie friendliness: 20/100