Skip to content

Latest commit

 

History

History
143 lines (108 loc) · 6.16 KB

CONTRIBUTING.md

File metadata and controls

143 lines (108 loc) · 6.16 KB

Contributing guidelines

If you have improvements to the oneDNN code, please send us your pull requests! To get started, see the GitHub howto.

You can:

  • Submit your changes directly with a pull request
  • Log a bug or feedback with an issue

See also: Contributor Covenant code of conduct.

Pull request checklist

Before sending your pull requests, make sure that you have followed this list:

Library functionality guidelines

oneDNN focuses on functionality that satisfies all of the following criteria:

  1. Performance: the functionality has material impact on a workload level. In other words, this means that for a new primitive it should be demonstrated that it brings visible performance improvement to some workload.

  2. Generality: the functionality is useful in a wide range of deep learning applications. This implies that when introducing a new primitive, its API needs to be general enough to be integrated into multiple deep learning frameworks that have similar functionality.

  3. Complexity: it is not trivial to implement the functionality directly in a deep learning application.

RFC pull requests

Significant library changes (new primitives, library architecture changes, API modifications, etc) require approval from oneDNN maintainers before opening a pull request with such implementation. For that we use the Request For Comments (RFC) process, which consists of opening, discussing, and accepting (promoting) RFC pull requests.

More information about the process can be found in the dedicated rfcs branch.

Code contribution guidelines

When submitting your contribution, please make sure that it is:

  • Tested: oneDNN uses gtests for lightweight functional testing and benchdnn for functionality that requires both performance and functional testing.

  • Documented: oneDNN uses Doxygen for inline comments in public header files that is used to build reference manual and markdown (also processed by Doxygen) for user guide.

  • Portable: oneDNN supports different operating systems, CPU and GPU architectures, compilers, and run-times. The new code should be compliant with the System Requirements.

All code in oneDNN gets promoted to product branches (master, rls-, and mnt-) only through GitHub pull requests. Requirements for promotion:

  • The request is reviewed and approved by maintainers for all affected components.
  • All discussions in the pull request are resolved.
  • Continuous integration pipeline passed without errors.
  • Promotion to release (rls-) branches can be done only by maintainers (enforced by GitHub)
  • The pull request author is responsible for collecting all the necessary approvals, rebasing of the changes, and resolving the discussions.

To simplify the work of reviewers, make sure that the commits in the pull request adhere to the following requirements:

  • Commit message should be fit into 50 (at most 72) characters and have the imperative mood.
  • Commit message should follow the format: <scope>:[scope: ..] <short description> Scope examples:
    • Top level: build, api, doc, tests, common, cpu, gpu
    • Second level: convolution, pooling, utils, verbose
    • Example commit message:
common: verbose: fix crash when prim_iface_t is empty
  • Commit body should also fit 72 characters. Think of it as a standard e-mail body or a markdown document in terms of styling - write sentences from the very left border keeping capital letters and punctuation in place.
  • oneDNN branches maintain linear history. Rebase the changes on top of target branch before creating a pull request. Rebase again after resolving all the discussions, as well as in case of merge conflicts.
  • Use git add -p and git rebase -i liberally to split unrelated changes into multiple self-contained commits. This is a courtesy to reviewers: smaller commits are easier to comprehend. It also helps with bisecting in the future. Of course judgement needs to be applied whether to split changes or not. For example, split code cleanup and the actual fix into two separate patches.

Coding Standards

Contributions to oneDNN must follow the Coding Standards in order to simplify development and review processes. The general principle is to follow the style of existing/surrounding code.

The Coding Standards are subject to change and contributions to the Coding Standards are welcome.

If you wish to propose changes to the Coding Standards (including clang-tidy checks and clang-format options), please submit the proposal via an RFC pull request. The proposal should contain the following information:

  • Motivation: Why should the proposed standard be introduced and applied?
  • Enforcement: Can the proposed standard be applied via an automated process or other practical means?
  • Example: What does the code base look like with the proposed standard applied?
    • For instance, in case of a clang-tidy check, please open a separate PR with the check applied to the code base alongside the RFC PR.

Unit tests

oneDNN uses gtests for lightweight functional testing and benchdnn for performance and functional testing.

Be sure to extend the existing tests when fixing an issue.

Developing new benchdnn tests can be hard, so it is a good idea to start with gtests first.