| commit | 7a9994873fa5ee21fcd393193b4d6e65aa1d406d | [log] [tgz] |
|---|---|---|
| author | bjacob <benoitjacob@google.com> | Fri Nov 03 12:08:00 2023 -0400 |
| committer | GitHub <noreply@github.com> | Fri Nov 03 12:08:00 2023 -0400 |
| tree | 8f7d26cc2f3a377a8e23b12133abe4ae94f00c1f | |
| parent | 573f5e91901ce54c594b0e82f894a98f9f9d53f7 [diff] |
CPU features flags improvements (#15387) The trigger was this discussion with @mariecwhite : https://discord.com/channels/689900678990135345/760577505840463893/1169557140344676372 The initial observation was "IREE is 2x slower than TFLite". The reason was that `--iree-llvmcpu-target-cpu=` was being used on AArch64, where it silently does nothing. It's hard to implement it everywhere, and it's hard to do good error-reporting here as we are deep down a call chain without error-reporting mechanics, including some C++ constructors. Along the way, this PR dusts off some code and structures it better (local static helper function better than private class member), enforces consistency between CPU and CPU feature flags when it comes to specifying `host`, adds helpful fatal errors for a few more scenarios that are "the user passed unhandled flags, better inform them now", resolves a TODO, actually implements CPU-to-CPU-features resolution on AArch64 (so Marie's original command line will work there), etc.
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.