search

openxla / stablehlo

Backward compatible ML compute opset inspired by HLO/MHLO
Apache License 2.0
403 stars 110 forks source link

Consider merging RoundOp and RoundNearestEvenOp #35

Open sdasgup3 opened 2 years ago

sdasgup3 commented 2 years ago

Currently we have two versions of the op which differ in rounding mode. The proposal here is to unify them using a attribute to decide the rounding mode.

sdasgup3 commented 2 years ago

This is waiting on setting up public processes for landing such API breaking changes.

burmako commented 2 years ago

At the upcoming OpenXLA meeting (next Tuesday 9/20), we'll talk about the RFC process for StableHLO. Early next week, we'll have a markdown document in this repository talking about how RFCs will be discussed and how decisions about them will be made. Once this process is set up, we will start providing feedback about incoming RFCs within a predictable timeframe, and it will be clearly documented how decisions on these RFCs are being made.

burmako commented 2 years ago

The StableHLO evolution process has been proposed a few days ago (#196), but it'll need some additional work. This RFC will need to wait a little bit longer before we discuss it and make a decision.

burmako commented 2 years ago

Still waiting on finalizing the StableHLO evolution process. We're coordinating with the overall OpenXLA effort about the governance model.

burmako commented 1 year ago

Unassigning this ticket for now, as part of evaluating relative priorities of recently opened spec tickets and tickets that I'm migrating from the MHLO issue tracker today.