| commit | f9d7599874087fa30b32446d5459b627a5024558 | [log] [tgz] |
|---|---|---|
| author | Quinn Dawkins <quinn.dawkins@gmail.com> | Fri Nov 10 06:43:36 2023 -0800 |
| committer | GitHub <noreply@github.com> | Fri Nov 10 14:43:36 2023 +0000 |
| tree | 2c7b95444254634f788a228d71d043db38abc9c5 | |
| parent | 86336293a8066b396537fae117d8549460cd85fd [diff] |
[SPIRV] Handle extraction from create_mask during load/store vectorization (#15524) Proper handling of masks requires direct lowerings of the mask, however we can skip the materialization of the mask entirely (and subsequent reliance on unrolling/canonicalization to clean up the masks) by folding `vector.extract(vector.create_mask)` into the appropriate boolean. This works for subgroup reduce (the first dynamic codegen problem we're tackling) because masks and transfers are never unrolled in this pipeline, rather we just distribute them in place to a pre-configured vector size. We handle this during SPIRVVectorizeLoadStore because immediately after scalarizing vector.transfer_read/write, we will introduce the extracts on the mask.
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.