| commit | 5d4c0baa65415fddf5d9f9d8cde56053349e6adc | [log] [tgz] |
|---|---|---|
| author | Benoit Jacob <jacob.benoit.1@gmail.com> | Tue Jan 30 12:53:13 2024 -0500 |
| committer | GitHub <noreply@github.com> | Tue Jan 30 17:53:13 2024 +0000 |
| tree | a79df839a7353444c34da0e9c4827955d95ccf1e | |
| parent | eaefd3ff964f14e700b1c468aa2b4f20a4401028 [diff] |
Nothing is unreachable (#16261) This comes from debugging the mysterious crashes on CI on @mariecwhite 's https://github.com/openxla/iree/pull/16259. Debugged: https://github.com/openxla/iree/pull/16259#issuecomment-1917417773 The motivation for these unreachable statements was a performance concern that is no longer current on LLVMCPU now that inlining works reliably. The switch variables here are in practice always compile-time constants so the whole switch statement vanishes in post-linking IR optimization passes. The only places where the unreachable still helps are: VMVX and ukernel own micro benchmarks and tests, and the impact is small, so it's not worth keeping. As a bonus, this makes the fallback now automatic when adding a new mmt4d type not handled by existing architecture-sepcific code, so this just makes ukernels easier to extend, as evidenced by the fact that @mariecwhite did run into this. One place had to retain an unreachable to work around a specific compiler bug. As it's specific to one compiler, it doesn't need a portable macro to wrap around the unreachable builtin.
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.