Let ukernel plugin entry points alias the underlying ukernel (#13597)

Plugin entry points have their own signature, taking 3 pointer args.
They then call the actual ukernel entry point, passing it one of those
pointers.
https://github.com/openxla/iree/blob/2e939b0cb331ba07a62285c3268329394a665355/experimental/cpu_ukernel/plugin.c#L32-L37

Initially I chose to let the plugin advertise the plugin entry points
using a custom namespace, `"ukernel."`, to disambiguate from the
underlying entry point, namespaced `iree_uk_`.

But in #13460, @MaheshRavishankar is introducing bitcode ukernels, where
the true `iree_uk_` ukernel function is called directly. In that case,
the ukernel op needs to have that as its function-name attribute, and so
Mahesh has to change that there, but then that means that the resulting
compiled modules can't call plugins anymore. We would have to introduce
a new iree-compile command-line switch to tell whether to use only
bitcode ukernels or only plugin ukernels. There's no fundamental problem
with that, but it's nice if we can avoid the additional switch and have
everything just work.

Apparently, this works. My only concern with this is that now
`iree_uk_mmt4d` means 2 separate things, the actual ukernel and its
exported name in the plugins. I think that's fine: once bitcode lands,
plugins should be a minority case and for users who care about that
minority case, this will be a lot more convenient.
4 files changed
tree: afc5307f6b8dc8f52e1f6662dd19e1b6f9be9ac9
  1. .devcontainer/
  2. .github/
  3. benchmarks/
  4. build_tools/
  5. compiler/
  6. docs/
  7. experimental/
  8. integrations/
  9. lib/
  10. llvm-external-projects/
  11. runtime/
  12. samples/
  13. tests/
  14. third_party/
  15. tools/
  16. .bazel_to_cmake.cfg.py
  17. .bazelignore
  18. .bazelrc
  19. .bazelversion
  20. .clang-format
  21. .dockerignore
  22. .gitignore
  23. .gitmodules
  24. .pylintrc
  25. .style.yapf
  26. .yamllint.yml
  27. AUTHORS
  28. BUILD.bazel
  29. CITATION.cff
  30. CMakeLists.txt
  31. configure_bazel.py
  32. CONTRIBUTING.md
  33. LICENSE
  34. README.md
  35. WORKSPACE
README.md

IREE: Intermediate Representation Execution Environment

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.

CI Status

Project Status

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!

Communication Channels

Related Project Channels

  • MLIR topic within LLVM Discourse: IREE is enabled by and heavily relies on MLIR. IREE sometimes is referred to in certain MLIR discussions. Useful if you are also interested in MLIR evolution.

Architecture Overview

IREE Architecture IREE Architecture

See our website for more information.

Presentations and Talks

  • 2021-06-09: IREE Runtime Design Tech Talk (recording and slides)
  • 2020-08-20: IREE CodeGen: MLIR Open Design Meeting Presentation (recording and slides)
  • 2020-03-18: Interactive HAL IR Walkthrough (recording)
  • 2020-01-31: End-to-end MLIR Workflow in IREE: MLIR Open Design Meeting Presentation (recording and slides)

License

IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.