Implement direct lowering of broadcasts from chlo. * This is step 1 of the larger plan to do a one step HLO->IREE conversion as part of the input pipeline. We are choosing to directly legalize broadcasts at a higher level where we have more context to resolve dynamic ambiguities. * Introduced via a pass option that is used for testing the new behavior and will be flipped as part of the overall switch, which is a few PRs down the line. * This uncovered several missing legalizations (now nicely flagged because of enforcing a full conversion of the mhlo/chlo dialects) for shift left/right. * Also, broadcast_select semantics were changed recently and this only covers the old forms (new forms are in the missing_legalizations.mlir) test and will be implemented in a follow-up. * Still iterating on some expanded lit tests but what is there is ready for review.
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.