| commit | f3b008c6db310f787ad76f151c21a30f72b14794 | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Fri Jan 26 08:35:57 2024 -0800 |
| committer | GitHub <noreply@github.com> | Fri Jan 26 08:35:57 2024 -0800 |
| tree | 0495ec360f55d78491b5f225cbf4feb565a5b8ed | |
| parent | eef24a93ed6235a12828aef6ab94f3ae629f1aab [diff] | |
| parent | 7f88e611a112061d6d6b8bbbad43305db0e039ab [diff] |
Adding util.func/util.call to allow for richer functions. (#16207) This adds util.func/util.call ops that are supersets of func.func/func.call. Indirect calls are not yet implemented as we don't support indirect func.call either. The new ops allow us to enhance the calling convention with tied operands so that we can model in-place function calls on tensors. A new inlining policy attr interface has been added that can be used on both util.func and util.global to control how both are inlined. Today we have the `#util.inline.never` to do what the ad-hoc `noinline` used to but as we refine inlining we can have differing cost models we control both from frontends and from our own generated code (memoized command buffers/etc). The inlining control also supports dialect attrs that can do context-aware inlining control and this will be used in subsequent PRs to avoid inlining functions or globals that are pinned to different devices. In this first PR the ops and interfaces are added but we are still using func.func/func.call exclusively. Future changes will enable support in current func-aware pass (IPO, stream usage refinement, allocation, etc) and convert func->util as part of the input pipeline. Progress on #16187.
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.