| commit | e876020851afec7b7107a494f27a62c196cdadb9 | [log] [tgz] |
|---|---|---|
| author | Benoit Jacob <jacob.benoit.1@gmail.com> | Fri May 31 16:38:17 2024 -0400 |
| committer | GitHub <noreply@github.com> | Fri May 31 16:38:17 2024 -0400 |
| tree | 5923f22a9d955a02d555b14444b02e05e8689f6a | |
| parent | 62efaee32188a01889822ba5ea547ddc66c23fcc [diff] |
Strided pack/unpack ukernels (#17516) The pack ukernel gains the ability to support inner-strided sources, and the unpack ukernel gains the ability to support inner-strided destinations. Note that the pack and unpack ukernels both use a small temporary buffer to decouple source vs destination layout aspects, limiting the combinatorics. The inner-striding aspect here is local to the "2D" side, orthogonal to the "4D" side which is what we have architecture specific code paths for. The "2D" side was handled by generic loops with generic memcpy/memset primitives. What changes here is that now we need a "memcpy with stride on either src or dst". This is implemented as scalar loops. I tried using AVX-512 gather/scatter instructions, but that turned out 2x slower than a scalar loop on AMD Zen4 where they are micro-coded, and the Intel documentation suggests a very high cost there too. Fixes #17507. Signed-off-by: Benoit Jacob <jacob.benoit.1@gmail.com>
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.
Community meeting recordings: IREE YouTube channel
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.