| commit | 6233f4f40b4a9223628b8b14c6de75ae1ecb29f3 | [log] [tgz] |
|---|---|---|
| author | Dave Liddell <dave.liddell@amd.com> | Wed May 01 08:16:50 2024 -0600 |
| committer | GitHub <noreply@github.com> | Wed May 01 09:16:50 2024 -0500 |
| tree | 9a10a211dd985a48ffcebb2ea5b84c7f78a65f56 | |
| parent | f54a8619b1ddf0608e1c2ccc5481e2ac1f1a2044 [diff] |
Fixes to passes for custom dispatch to work with bf16 type (#17242) Custom dispatch for bf16 type stopped working because of a couple of problems: 1. `--iree-codegen-expand-strided-metadata` doesn't work with `iree-opt` when given an `iree_codegen.extract_strided_metadata` op due to a lack of a dependency on the `memref` dialect. 2. `--iree-convert-bf16-to-uint16-buffers` doesn't convert types in an `iree_codegen.extract_strided_metadata` op due to a lack of a dependency on the `iree_codegen` dialect. As a result, type casts are introduced between the `iree_codegen.extract_strided_metadata` and its `hal.interface.binding.subspan` predecessor, which confuses `--iree-codegen-expand-strided-metadata`, which fails to do its job of eliminating the `iree_codegen.extract_strided_metadata`. This second problem was described in https://github.com/iree-org/iree/issues/17177. The fix for both problems is to add the missing dialect dependencies. fixes: https://github.com/iree-org/iree/issues/17177
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.