Google Git
Sign in
opensecura / 3p / openxla / iree / 245bb9f002582da90d548aeb9dce5c8ae3cdc768
commit245bb9f002582da90d548aeb9dce5c8ae3cdc768[log] [tgz]
authorMaheshRavishankar <1663364+MaheshRavishankar@users.noreply.github.com>Thu Apr 22 11:56:54 2021 -0700
committerGitHub <noreply@github.com>Thu Apr 22 11:56:54 2021 -0700
tree1348665119c618c47b6be0bf2c5df88584646094
parentacd357ff1898411a11c62f2606410b0df4975cfe [diff]
Rework LinalgBufferize pass. (#5539)

The bufferization pass on the linalg on tensors path was an arbitrary
set of methods put in place to connect linalg on tensors to the post
tile+distribute codegeneration on CPU and GPU. This PR cleans up the
logic to better handle the tensor -> memref conversion.
In particular

Adds an analysis step to figure out which tensors can be mapped to
the same memref object. This also provides information of what
tensors are eventually mapped to the output of the dispatch
region. In such cases, the output is used to compute the values in
place reducing the need for copies and intermediate allocations.
Use this analysis step to convert operations from tensors to memrefs.
Use the readonly, readwrite and writeonly attributes to avoid
copies when possible.
All of this uses logic similar to the Tied operands/results approach
that the bufferization in the sandbox uses. Indeed this pass might
eventually subsumed by a bufferization pass in the sandbox and to be
upstreamed. Till that happens, this rewrite gives something that
is similar in spirit.

Fixes #4734
Fixes #5013
2 files changed
tree: 1348665119c618c47b6be0bf2c5df88584646094
  1. .github/
  2. bindings/
  3. build_tools/
  4. colab/
  5. docs/
  6. experimental/
  7. integrations/
  8. iree/
  9. scripts/
  10. third_party/
  11. .bazelignore
  12. .bazelrc
  13. .bazelversion
  14. .clang-format
  15. .gitignore
  16. .gitmodules
  17. .style.yapf
  18. .yamllint.yml
  19. BUILD.bazel
  20. CMakeLists.txt
  21. configure_bazel.py
  22. CONTRIBUTING.md
  23. LICENSE
  24. README.md
  25. SUBMODULE_VERSIONS.txt
  26. WORKSPACE
README.md

IREE: Intermediate Representation Execution Environment

IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler that lowers ML models to a unified IR optimized for real-time mobile/edge inference against heterogeneous hardware accelerators. IREE also provides flexible deployment solutions for the compiled ML models.

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.

Getting Started

Quick Start using Python

Python packages are published on the releases page. See the colab/ directory for examples.

Building from Source

IREE can be built from source using both Bazel and CMake on Windows and Linux. We also have experimental macOS support.

Please see the Getting Started pages on IREE's documentation hub to configure, compile, and run IREE in your favorite development environment!

Documentation and Talks

IREE hosts all its documentation and project status dashboards on GitHub Pages. We are still building up the website; please feel free to create issues for the documentation you'd like to see!

We also have some public talks that explain IREE's concepts and architecture:

  • 2020-03-18: Interactive HAL IR Walkthrough (Ben Vanik and core team) (recording)
  • 2020-01-31: End-to-end MLIR Workflow in IREE (recording and slides)

Architecture and Goals

IREE adopts a holistic approach towards ML model compilation: the IR produced contains both the scheduling logic, required to communicate data dependencies to low-level parallel pipelined hardware/API like Vulkan, and the execution logic, encoding dense computation on the hardware in the form of hardware/API-specific binaries like SPIR-V.

The architecture of IREE is best illustrated by the following picture:

IREE Architecture

Being compilation-based means IREE does not have a traditional runtime that dispatches “ops” to their fat kernel implementations. What IREE provides is a toolbox for different deployment scenarios. It scales from running generated code on a particular API (such as emitting C code calling external DSP kernels), to a HAL (Hardware Abstraction Layer) that allows the same generated code to target multiple APIs (like Vulkan and Direct3D 12), to a full VM allowing runtime model loading for flexible deployment options and heterogeneous execution.

IREE aims to

  • Support advanced models on mobile/edge devices. Dynamic shapes, dynamic flow control, dynamic multi-model dispatch, streaming models, tree-based search algorithms, and other are all good examples of exciting ML evolution. We are trying to build IREE from the ground-up to enable these models and run them efficiently on modern hardware, especially on mobile/edge devices.
  • Demonstrate MLIR‘s ability to develop non-traditional ML compiler backends and runtimes. MLIR enables IREE’s holistic approach of focusing on the math being performed and how that math is scheduled rather than graphs of “ops”.
  • Embrace standard-based ML via Vulkan. The graphics world is shifting towards favoring modern explicit APIs for performance and predictability and Vulkan is emerging as the “compatibility” layer. We would love to allow hardware vendors to be able to make ML efficient on their hardware without the need for bespoke runtimes and special access. We also would love to let developers and users utilize all the hardware available on as many platforms as possible.

Roadmap and Milestones

IREE is in the early stages of development and not yet ready for broad adoption. Check out the long-term design roadmap to get a sense of where we're headed.

We plan on a quarterly basis using OKRs. Review our latest objectives to get a sense of what we're up to in the near term.

We use GitHub Projects to track progress on IREE components and specific efforts. We use GitHub Milestones to track the work associated with plans for each quarter.

Build Status

CI SystemBuild SystemPlatformArchitectureComponentStatus
KokoroBazelLinuxx86Corekokoro_status_bazel_linux_x86_core
KokoroCMake & BazelLinuxx86-swiftshaderIntegrationskokoro_status_cmake-bazel_linux_x86-swiftshader_integrations
KokoroCMake & BazelLinuxx86-turingIntegrationskokoro_status_cmake-bazel_linux_x86-turing_integrations
KokoroCMakeLinuxx86-swiftshaderCore + Bindingskokoro_status_cmake_linux_x86-swiftshader
KokoroCMakeLinuxx86-turingCore + Bindingskokoro_status_cmake_linux_x86-turing
KokoroCMakeAndroidarm64-v8aRuntime (build only)kokoro_status_cmake_android_arm64-v8a
BuildKiteCMakeAndroidarm64-v8aRuntimebuildkite-status-cmake-android-arm

License

IREE is licensed under the terms of the Apache license. See LICENSE for more information.