| commit | b32e07574551dab08202922f05ef2c078d564da8 | [log] [tgz] |
|---|---|---|
| author | Stella Laurenzo <stellaraccident@gmail.com> | Mon Jul 31 09:12:28 2023 -0700 |
| committer | GitHub <noreply@github.com> | Mon Jul 31 09:12:28 2023 -0700 |
| tree | f928e41a487fc7e17d7fecef6b3d7e5d60c560ac | |
| parent | cf5d348e78eaa893589d8f8553ddc967e38fa2cf [diff] |
Isolate LLVMTarget in the llvm-cpu backend. (#14516) Isolate LLVMTarget in the llvm-cpu backend. This carefully pry's apart the LLVMTargetOptions from the LLVMTarget, making sure that the latter is hermetic for anything affecting the generated code. Further replumbs the target backend so that it doesn't reach into default state and works off of the configuration IR exclusively. Also updates the consteval JIT to use this new capability for targeting the host, even if the overall compilation is cross-compiling. Flags changed: * Removes `iree-consteval-jit-use-vmvx` * Adds `iree-consteval-jit-target-backend`: Currently still defaults to VMVX but with this change it should be possible to reliably flip it to llvm-cpu. I tested this by adding a similar print function to the options in a base branch and exhaustively compared the results of of the target information before and after the change for full builds. There were some tricky things in the original since it was initializing flags off of defaults which would then be re-calculated and such. I've smoothed all of this out so that at least the paths are straight through once the initial configuration is set.
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.