| commit | 6de4a73170e8282220bb1651da92a2e72fa4b7e9 | [log] [tgz] |
|---|---|---|
| author | Oleksandr "Alex" Zinenko <zinenko@google.com> | Mon Jul 03 15:16:57 2023 +0200 |
| committer | GitHub <noreply@github.com> | Mon Jul 03 15:16:57 2023 +0200 |
| tree | 8bca29db081e99bd8f47c8fe7573022f4ddda583 | |
| parent | 566ed4dc140b451f7ec45b0861a73fb97c0c6b4c [diff] |
Make barrier elimination more aggressive (#14293) Contrary to the conceptual description of the barrier elimination technique in the paper by Moses et.al., barrier elimination in IREE is performed by the greedy rewriter. Thus a redundant barrier is erased immediately and the updated IR is considered after for the next barrier. Therefore, there it is sufficient to consider the effects before and after each barrier until hitting the next barrier, rather than pairwise extending those until the end of the parallel region boundary. Specifically, the case that wouldn't be handled correctly with non-eager rewriting, ```mlir store %A barrier // useless because no effects after // nothing barrier // useless because no effects before load %A ``` is handled correctly in IREE that removes one of the barriers and sees the second as required when re-analyzing the eagerly rewrtitten IR. This change lets us eliminate more barriers.
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!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.