Google Git
Sign in
opensecura / 3p / openxla / iree / eb0d678894aa6ebb20b056d0f09bad3a16d5d26c
commiteb0d678894aa6ebb20b056d0f09bad3a16d5d26c[log] [tgz]
authorHan-Chung Wang <hanchung@google.com>Fri Apr 08 11:14:31 2022 -0700
committerGitHub <noreply@github.com>Fri Apr 08 11:14:31 2022 -0700
treed2d62053d88b91dd2c4d2dc137dfe5c79bc6febd
parent70949187a7587ef7e0b51d6bd6b06b193364591a [diff]
Improve compilation time and execution time for quantized matmul on ARM (#8815)

The compilation time is cut from 5.5 mins to 1.3 mins.

Perf improvements (sampled from MobileBert Int8):

Before:

```
-----------------------------------------------------------------------------------------------
Benchmark                                                     Time             CPU   Iterations
-----------------------------------------------------------------------------------------------
BM_matmul_i8i8i32_384x128x128/process_time/real_time      0.954 ms        0.933 ms          738
BM_matmul_i8i8i32_384x512x128/process_time/real_time       3.73 ms         3.65 ms          187
BM_matmul_i8i8i32_384x384x32/process_time/real_time        1.12 ms         1.09 ms          628
BM_matmul_i8i8i32_384x128x512/process_time/real_time       3.68 ms         3.60 ms          190
BM_matmul_i8i8i32_384x32x384/process_time/real_time       0.758 ms        0.741 ms          922
BM_matmul_i8i8i32_384x512x384/process_time/real_time       11.7 ms         11.4 ms           60
BM_matmul_i8i8i32_384x2x512/process_time/real_time        0.137 ms        0.133 ms         5106
```

After:

```
-----------------------------------------------------------------------------------------------
Benchmark                                                     Time             CPU   Iterations
-----------------------------------------------------------------------------------------------
BM_matmul_i8i8i32_384x128x128/process_time/real_time      0.728 ms        0.711 ms          968
BM_matmul_i8i8i32_384x512x128/process_time/real_time       2.80 ms         2.74 ms          249
BM_matmul_i8i8i32_384x384x32/process_time/real_time       0.680 ms        0.663 ms         1029
BM_matmul_i8i8i32_384x128x512/process_time/real_time       2.63 ms         2.57 ms          266
BM_matmul_i8i8i32_384x32x384/process_time/real_time       0.521 ms        0.509 ms         1345
BM_matmul_i8i8i32_384x512x384/process_time/real_time       7.99 ms         7.82 ms           88
BM_matmul_i8i8i32_384x2x512/process_time/real_time        0.146 ms        0.142 ms         4789
```

Fixes https://github.com/google/iree/issues/8734
2 files changed
tree: d2d62053d88b91dd2c4d2dc137dfe5c79bc6febd
  1. .github/
  2. benchmarks/
  3. bindings/
  4. build_tools/
  5. colab/
  6. docs/
  7. experimental/
  8. integrations/
  9. iree/
  10. llvm-external-projects/
  11. scripts/
  12. third_party/
  13. .bazelignore
  14. .bazelrc
  15. .bazelversion
  16. .clang-format
  17. .gitignore
  18. .gitmodules
  19. .pylintrc
  20. .style.yapf
  21. .yamllint.yml
  22. AUTHORS
  23. BUILD.bazel
  24. CMakeLists.txt
  25. configure_bazel.py
  26. CONTRIBUTING.md
  27. LICENSE
  28. README.md
  29. 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.

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.

Build Status

CI SystemBuild SystemPlatformArchitectureConfiguration / ComponentStatus
KokoroBazelLinuxx86-64kokoro status bazel/linux/x86-swiftshader/core
KokoroCMake & BazelLinuxx86-64 (swiftshader)Integrationskokoro status cmake-bazel/linux/x86-swiftshader
KokoroCMake & BazelLinuxx86-64 (turing)Integrationskokoro status cmake-bazel/linux/x86-turing
KokoroCMakeLinuxx86-64 (swiftshader)kokoro status cmake/linux/x86-swiftshader
KokoroCMakeLinuxx86-64 (swiftshader)asankokoro status cmake/linux/x86-swiftshader-asan
KokoroCMakeLinuxx86-64 (turing)kokoro status cmake/linux/x86-turing
KokoroCMakeAndroidarm64-v8aRuntime (build only)kokoro status cmake/android/arm64-v8a
KokoroCMakeBare Metalrisc-v-32Runtimekokoro status cmake/baremetal/riscv32
KokoroCMakeLinuxrisc-v-64Runtimekokoro status cmake/linux/riscv64
BuildkiteCMakeAndroidarm64-v8aRuntimebuildkite status iree-android-arm64-v8a
BuildKiteCMakeAndroidarm64-v8aRuntime Benchmarksbuildkite status iree-benchmark
BuildKiteCMakeLinuxx86-64Tracing + Standalone Runtimebuildkite status iree-build-configurations

Architecture Overview

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.