commit | 71028b229c845b52d0da164b18ea216660c88d5e | [log] [tgz] |
---|---|---|
author | Kunwar Grover <groverkss@gmail.com> | Wed Jun 26 02:51:09 2024 +0100 |
committer | GitHub <noreply@github.com> | Wed Jun 26 01:51:09 2024 +0000 |
tree | c8f29d482907b08f126204bdc76da86611089e0c | |
parent | 36aec8a79cc616d126fe50f213d8b93ef298d6ba [diff] |
[Codegen] Teach Subset hoisting about insertion ops with loop invariant tensors (#17713) This patch teaches subset hoisting to hoist extract/insert pairs operating on equivalent subsets when the destination of the inserted tensor is loop invariant. This condition happens when a tensor is being updated every iteration with a value, but only the final iteration update is what counts. This case occurs frequently when trying to vectorized a linalg.generic , representing an elementwise operation which is not destination passing style. Vectorization creates a write to an empty tensor, but extracts from the loop iteration, which makes subset hoisting fail to reason about it (subset hoisting only works if the same tensor is being updated)
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.
Community meeting recordings: IREE YouTube channel
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.