Google Git
Sign in
android / platform / external / tensorflow / 3ebd0c683ef
commit3ebd0c683efa922c3c3988c51856cfbf5b37c8b7[log] [tgz]
authorDerek Murray <mrry@google.com>Sun Mar 22 13:51:11 2020 -0700
committerTensorFlower Gardener <gardener@tensorflow.org>Sun Mar 22 13:54:51 2020 -0700
tree318b3a2b49a194973bf22bc5a506f92bf734b1a6
parentf7b6793c6611210405d066dde84cc67adca4097c [diff]
[Executor] Restructure `ActivateNodes()` to avoid unnecessary loads and branches.

The basic idea of the optimization is to avoid reading each destination `NodeItem` when propagating values/control signals from the outputs of a node.

To do that, we make the following changes:

1. Avoid building executor structures for the sink node, which is a no-op anyway. This avoids the need to compare the endpoint of every edge to the sink node, and removes the `NodeItem::is_sink` bit.
2. Optimize the common case when all consumers of a particular op are neither merge nor control trigger nodes. We record this fact in the source `NodeItem`, using the bit we saved in (1).
3. Repurpose the `EdgeInput::input_slot` field so that it is an offset directly into the vector of input tensors, rather than a relative offset from the destination `NodeItem::input_start`.
4. Move `NodeItem::pending_id` into a dense vector of `PendingCounts::Handle` values owned by the `ExecutorImpl`. This is a more compact and cache-friendly structure for `ActivateNodes()` to access in the common case.
5. Modify `PendingCounts::adjust_for_activations()` to return its value in a register, instead of using output parameters.

PiperOrigin-RevId: 302321563
Change-Id: If1b7feee54705c403fe27d48552545c00cd98601
4 files changed
tree: 318b3a2b49a194973bf22bc5a506f92bf734b1a6
  1. .github/
  2. tensorflow/
  3. third_party/
  4. tools/
  5. .pylintrc ⇨ tensorflow/tools/ci_build/pylintrc
  6. .bazelrc
  7. .bazelversion
  8. .gitignore
  9. ACKNOWLEDGMENTS
  10. ADOPTERS.md
  11. arm_compiler.BUILD
  12. AUTHORS
  13. BUILD
  14. CODE_OF_CONDUCT.md
  15. CODEOWNERS
  16. configure
  17. configure.cmd
  18. configure.py
  19. CONTRIBUTING.md
  20. ISSUE_TEMPLATE.md
  21. ISSUES.md
  22. LICENSE
  23. models.BUILD
  24. README.md
  25. RELEASE.md
  26. SECURITY.md
  27. WORKSPACE
README.md
Documentation
Documentation

TensorFlow is an end-to-end open source platform for machine learning. It has a comprehensive, flexible ecosystem of tools, libraries, and community resources that lets researchers push the state-of-the-art in ML and developers easily build and deploy ML-powered applications.

TensorFlow was originally developed by researchers and engineers working on the Google Brain team within Google's Machine Intelligence Research organization to conduct machine learning and deep neural networks research. The system is general enough to be applicable in a wide variety of other domains, as well.

TensorFlow provides stable Python and C++ APIs, as well as non-guaranteed backward compatible API for other languages.

Keep up-to-date with release announcements and security updates by subscribing to announce@tensorflow.org. See all the mailing lists.

Install

See the TensorFlow install guide for the pip package, to enable GPU support, use a Docker container, and build from source.

To install the current release, which includes support for CUDA-enabled GPU cards (Ubuntu and Windows):

$ pip install tensorflow

A smaller CPU-only package is also available:

$ pip install tensorflow-cpu

To update TensorFlow to the latest version, add --upgrade flag to the above commands.

Nightly binaries are available for testing using the tf-nightly and tf-nightly-cpu packages on PyPi.

Try your first TensorFlow program

$ python

>>> import tensorflow as tf
>>> tf.add(1, 2).numpy()
3
>>> hello = tf.constant('Hello, TensorFlow!')
>>> hello.numpy()
b'Hello, TensorFlow!'

For more examples, see the TensorFlow tutorials.

Contribution guidelines

If you want to contribute to TensorFlow, be sure to review the contribution guidelines. This project adheres to TensorFlow's code of conduct. By participating, you are expected to uphold this code.

We use GitHub issues for tracking requests and bugs, please see TensorFlow Discuss for general questions and discussion, and please direct specific questions to Stack Overflow.

The TensorFlow project strives to abide by generally accepted best practices in open-source software development:

CII Best Practices Contributor Covenant

Continuous build status

Official Builds

Build TypeStatusArtifacts
Linux CPUStatusPyPI
Linux GPUStatusPyPI
Linux XLAStatusTBA
macOSStatusPyPI
Windows CPUStatusPyPI
Windows GPUStatusPyPI
AndroidStatusDownload
Raspberry Pi 0 and 1Status StatusPy2 Py3
Raspberry Pi 2 and 3Status StatusPy2 Py3

Community Supported Builds

Build TypeStatusArtifacts
Linux AMD ROCm GPU NightlyBuild StatusNightly
Linux AMD ROCm GPU Stable ReleaseBuild StatusRelease 1.15 / 2.x
Linux s390x NightlyBuild StatusNightly
Linux s390x CPU Stable ReleaseBuild StatusRelease
Linux ppc64le CPU NightlyBuild StatusNightly
Linux ppc64le CPU Stable ReleaseBuild StatusRelease 1.15 / 2.x
Linux ppc64le GPU NightlyBuild StatusNightly
Linux ppc64le GPU Stable ReleaseBuild StatusRelease 1.15 / 2.x
Linux CPU with Intel® MKL-DNN NightlyBuild StatusNightly
Linux CPU with Intel® MKL-DNN Stable ReleaseBuild StatusRelease 1.15 / 2.x
Red Hat® Enterprise Linux® 7.6 CPU & GPU
Python 2.7, 3.6		Build Status1.13.1 PyPI

Resources

Learn more about the TensorFlow community and how to contribute.

License

Apache License 2.0