| commit | 466e812764beb3eeafebba5e5a45324cd40b5341 | [log] [tgz] |
|---|---|---|
| author | Quinn Dawkins <quinn@nod-labs.com> | Wed Oct 25 12:42:26 2023 -0400 |
| committer | GitHub <noreply@github.com> | Wed Oct 25 12:42:26 2023 -0400 |
| tree | 5597cd703e60f4263bf288d8ba774be4d0b43365 | |
| parent | 97a13940bcbb8ab21b8a42044af20cf9dec33c3a [diff] |
[Codegen] Add ability to specify transform dialect libraries (#14788) This adds the ability to specify transform dialect strategies through a symbol pointing to a library call. This is currently available through two flags: `--iree-codegen-use-transform-dialect-strategy=[@<symbol_name>|filename]` `--iree-codegen-transform-dialect-library=filename` The transform library is loaded and cached in the IREE codegen dialect for subsequent invocations within the MaterializeUserConfigs pass. Then, the loaded dialect is immediately used with the symbol name referenced by the transform dialect usage flag. If a filename is specified instead, that is broadcasted to the transform dialect interpreter (intended for microbenchmarking). If the symbol applies successfully, this will send the result through normal IREE codegen. This gives users the option to replace the translation info on the export op with a `<None>` pipeline to send it through <BACKEND>LowerExecutableTarget unperterbed (thereby skipping the initial tile + distribute and bufferization). Additionally this unifies the way the transform dialect testing flags currently duplicated across backends.
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.