Google Git
Sign in
opensecura / 3p / openxla / iree / 077168f6c8b2bff9eda724f7d2a764c909d8458c
commit077168f6c8b2bff9eda724f7d2a764c909d8458c[log] [tgz]
authorHan-Chung Wang <hanchung@google.com>Thu Nov 17 07:47:34 2022 +0800
committerGitHub <noreply@github.com>Wed Nov 16 15:47:34 2022 -0800
tree5cff35b24105082b10e080652d8084daf9a419ba
parent541d1f3ac0fa43deee35c940871a9db408b4e6b4 [diff]
[LLVMCPU] Enable vectorization for pack ops. (#11178)

It introduces `SplitFullPartialTransferPass` to hook up upstream
VectorTransferFullPartialRewriter pattern. The pattern splits a
vector.transfer operation into an in-bounds (i.e., no out-of-bounds
masking) fastpath and a slowpath. This needs to be applied before
converting to LLVM dialect.

Verified that we're getting 2x-3x performance improvements from Tracy.

For LHS packing, i.e.,

```mlir
lhs_4d = iree_linalg_ext.pack %lhs
  inner_dims_pos = [0, 1]
  inner_tiles = [8, 8]
  into %lhs_4d_init
  : (tensor<256x256xi8> tensor<32x32x8x8xi8>) -> tensor<32x32x8x8xi8>
```

The performance is improved from `32.59` us to `10.01` us.


For RHS packing, i.e.,

```mlir
%rhs_4d = iree_linalg_ext.pack %rhs
  outer_dims_perm = [1, 0]
  inner_dims_pos = [1, 0]
  inner_tiles = [8, 8]
  into %rhs_4d_init
  : (tensor<256x256xi8> tensor<32x32x8x8xi8>) -> tensor<32x32x8x8xi8>
```

The performance is improved from `32.35` us to `16.11` us.

Fixes https://github.com/iree-org/iree/issues/10889
6 files changed
tree: 5cff35b24105082b10e080652d8084daf9a419ba
  1. .github/
  2. benchmarks/
  3. build_tools/
  4. compiler/
  5. docs/
  6. experimental/
  7. integrations/
  8. llvm-external-projects/
  9. runtime/
  10. samples/
  11. tests/
  12. third_party/
  13. tools/
  14. .bazelignore
  15. .bazelrc
  16. .bazelversion
  17. .clang-format
  18. .dockerignore
  19. .gitignore
  20. .gitmodules
  21. .pylintrc
  22. .style.yapf
  23. .yamllint.yml
  24. AUTHORS
  25. BUILD.bazel
  26. CITATION.cff
  27. CMakeLists.txt
  28. configure_bazel.py
  29. CONTRIBUTING.md
  30. LICENSE
  31. README.md
  32. 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.