| commit | 01c4c57d7488dff0da7d117cc19edefe4136a22c | [log] [tgz] |
|---|---|---|
| author | Han-Chung Wang <hanhan0912@gmail.com> | Tue Feb 27 09:35:38 2024 -0800 |
| committer | GitHub <noreply@github.com> | Tue Feb 27 09:35:38 2024 -0800 |
| tree | 503a2a39c7e4bec448361755b335b38713b9cf62 | |
| parent | 6b995b95694fa97cb13765cd556b0d450bf894c2 [diff] |
[CPU] Add a specialized pipeline for LinalgExt::AttentionOp. (#16577) The revision adds a new pipeline for LinalgExt ops. It is an experimental pipeline, and should eventually get merged into MultiTilingPipeline. The new pipeline introduces vector level of tiling to LinalgExt, and vectorization. Some dimension of attention op is not able to tile at this moment, so we set all the tile sizes to 1 which avoids huge vectors. Because the reduction dimension of matmuls is not tiled. Here is selected IR dump: https://gist.githubusercontent.com/hanhanW/db4511da681d4932cb81dd68cc98976f/raw/08c3cc42c9d7fb86b769f60dc712fecb9fb10700/dump.mlir Towards https://github.com/openxla/iree/issues/16421
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.