LIVE / thrust /CONTRIBUTING.md
Xu Ma
update
1c3c0d9
|
raw
history blame
18.8 kB

Table of Contents

  1. Contributing to Thrust
  2. CMake Options
  3. Development Model

Contributing to Thrust

Thrust uses Github to manage all open-source development, including bug tracking, pull requests, and design discussions. This document details how to get started as a Thrust contributor.

An overview of this process is:

  1. Clone the Thrust repository
  2. Setup a fork of Thrust
  3. Setup your environment
  4. Create a development branch
  5. Local development loop
  6. Push development branch to your fork
  7. Create pull request
  8. Address feedback and update pull request
  9. When your PR is approved...

Clone the Thrust Repository

To get started, clone the main repository to your local computer. Thrust should be cloned recursively to setup the CUB submodule (required for CUDA acceleration).

git clone --recursive https://github.com/thrust/thrust.git
cd thrust

Setup a Fork of Thrust

You'll need a fork of Thrust on Github to create a pull request. To setup your fork:

  1. Create a Github account (if needed)
  2. Go to the Thrust Github page
  3. Click "Fork" and follow any prompts that appear.

Once your fork is created, setup a new remote repo in your local Thrust clone:

git remote add github-fork [email protected]:<GITHUB_USERNAME>/thrust.git

If you need to modify CUB, too, go to the CUB Github page and repeat this process. Create CUB's github-fork remote in the thrust/dependencies/cub submodule.

Setup Your Environment

Git Environment

If you haven't already, this is a good time to tell git who you are. This information is used to fill out authorship information on your git commits.

git config --global user.name "John Doe"
git config --global user.email [email protected]

Configure CMake builds

Thrust uses CMake for its developer build system. To configure, build, and test your checkout of Thrust:

# Create build directory:
mkdir build
cd build

# Configure -- use one of the following:
cmake ..   # Command line interface.
ccmake ..  # ncurses GUI (Linux only)
cmake-gui  # Graphical UI, set source/build directories in the app

# Build:
cmake --build . -j <num jobs>   # invokes make (or ninja, etc)

# Run tests and examples:
ctest

See CMake Options for details on customizing the build.

Create a Development Branch

All work should be done in a development branch (also called a "topic branch") and not directly in the master branch. This makes it easier to manage multiple in-progress patches at once, and provides a descriptive label for your patch as it passes through the review system.

To create a new branch based on the current master:

# Checkout local master branch:
cd /path/to/thrust/sources
git checkout master

# Sync local master branch with github:
git pull

# Create a new branch named `my_descriptive_branch_name` based on master:
git checkout -b my_descriptive_branch_name

# Verify that the branch has been created and is currently checked out:
git branch

Thrust branch names should follow a particular pattern:

  • For new features, name the branch feature/<name>
  • For bugfixes associated with a github issue, use bug/github/<bug-description>-<bug-id>
    • Internal nvidia and gitlab bugs should use nvidia or gitlab in place of github.

If you plan to work on CUB as part of your patch, repeat this process in the thrust/dependencies/cub submodule.

Local Development Loop

Edit, Build, Test, Repeat

Once the topic branch is created, you're all set to start working on Thrust code. Make some changes, then build and test them:

# Implement changes:
cd /path/to/thrust/sources
emacs thrust/some_file.h # or whatever editor you prefer

# Create / update a unit test for your changes:
emacs testing/some_test.cu

# Check that everything builds and tests pass:
cd /path/to/thrust/build/directory
cmake --build . -j <num jobs>
ctest

Creating a Commit

Once you're satisfied with your patch, commit your changes:

Thrust-only Changes

# Manually add changed files and create a commit:
cd /path/to/thrust
git add thrust/some_file.h
git add testing/some_test.cu
git commit

# Or, if possible, use git-gui to review your changes while building your patch:
git gui

Thrust and CUB Changes

# Create CUB patch first:
cd /path/to/thrust/dependencies/cub
# Manually add changed files and create a commit:
git add cub/some_file.cuh
git commit

# Create Thrust patch, including submodule update:
cd /path/to/thrust/
git add dependencies/cub # Updates submodule info
git add thrust/some_file.h
git add testing/some_test.cu
git commit

# Or, if possible, use git-gui to review your changes while building your patch:
cd /path/to/thrust/dependencies/cub
git gui
cd /path/to/thrust
git gui # Include dependencies/cub as part of your commit

Writing a Commit Message

Your commit message will communicate the purpose and rationale behind your patch to other developers, and will be used to populate the initial description of your Github pull request.

When writing a commit message, the following standard format should be used, since tools in the git ecosystem are designed to parse this correctly:

First line of commit message is a short summary (<80 char)
<Second line left blank>
Detailed description of change begins on third line. This portion can
span multiple lines, try to manually wrap them at something reasonable.

Blank lines can be used to separate multiple paragraphs in the description.

If your patch is associated with another pull request or issue in the main
Thrust repository, you should reference it with a `#` symbol, e.g.
#1023 for issue 1023.

For issues / pull requests in a different github repo, reference them using
the full syntax, e.g. thrust/cub#4 for issue 4 in the thrust/cub repo.

Markdown is recommended for formatting more detailed messages, as these will
be nicely rendered on Github, etc.

Push Development Branch to your Fork

Once you've committed your changes to a local development branch, it's time to push them to your fork:

cd /path/to/thrust/checkout
git checkout my_descriptive_branch_name # if not already checked out
git push --set-upstream github-fork my_descriptive_branch_name

--set-upstream github-fork tells git that future pushes/pulls on this branch should target your github-fork remote by default.

If have CUB changes to commit as part of your patch, repeat this process in the thrust/dependencies/cub submodule.

Create Pull Request

To create a pull request for your freshly pushed branch, open your github fork in a browser by going to https://www.github.com/<GITHUB_USERNAME>/thrust. A prompt may automatically appear asking you to create a pull request if you've recently pushed a branch.

If there's no prompt, go to "Code" > "Branches" and click the appropriate "New pull request" button for your branch.

If you would like a specific developer to review your patch, feel free to request them as a reviewer at this time.

The Thrust team will review your patch, test it on NVIDIA's internal CI, and provide feedback.

If have CUB changes to commit as part of your patch, repeat this process with your CUB branch and fork.

Address Feedback and Update Pull Request

If the reviewers request changes to your patch, use the following process to update the pull request:

# Make changes:
cd /path/to/thrust/sources
git checkout my_descriptive_branch_name
emacs thrust/some_file.h
emacs testing/some_test.cu

# Build + test
cd /path/to/thrust/build/directory
cmake --build . -j <num jobs>
ctest

# Amend commit:
cd /path/to/thrust/sources
git add thrust/some_file.h
git add testing/some_test.cu
git commit --amend
# Or
git gui # Check the "Amend Last Commit" box

# Update the branch on your fork:
git push -f

At this point, the pull request should show your recent changes.

If have CUB changes to commit as part of your patch, repeat this process in the thrust/dependencies/cub submodule, and be sure to include any CUB submodule updates as part of your commit.

When Your PR is Approved

Once your pull request is approved by the Thrust team, no further action is needed from you. We will handle integrating it since we must coordinate changes to master with NVIDIA's internal perforce repository.

CMake Options

A Thrust build is configured using CMake options. These may be passed to CMake using

cmake -D<option_name>=<value> /path/to/thrust/sources

or configured interactively with the ccmake or cmake-gui interfaces.

Thrust supports two build modes. By default, a single configuration is built that targets a specific host system, device system, and C++ dialect. When THRUST_ENABLE_MULTICONFIG is ON, multiple configurations targeting a variety of systems and dialects are generated.

The CMake options are divided into these categories:

  1. Generic CMake Options: Options applicable to all Thrust builds.
  2. Single Config CMake Options Options applicable only when THRUST_ENABLE_MULTICONFIG is disabled.
  3. Multi Config CMake Options Options applicable only when THRUST_ENABLE_MULTICONFIG is enabled.
  4. CUDA Specific CMake Options Options that control CUDA compilation. Only available when one or more configurations targets the CUDA system.
  5. TBB Specific CMake Options Options that control TBB compilation. Only available when one or more configurations targets the TBB system.

Generic CMake Options

  • CMAKE_BUILD_TYPE={Release, Debug, RelWithDebInfo, MinSizeRel}
    • Standard CMake build option. Default: RelWithDebInfo
  • THRUST_ENABLE_HEADER_TESTING={ON, OFF}
    • Whether to test compile public headers. Default is ON.
  • THRUST_ENABLE_TESTING={ON, OFF}
    • Whether to build unit tests. Default is ON.
  • THRUST_ENABLE_EXAMPLES={ON, OFF}
    • Whether to build examples. Default is ON.
  • THRUST_ENABLE_MULTICONFIG={ON, OFF}
    • Toggles single-config and multi-config modes. Default is OFF (single config).
  • THRUST_ENABLE_EXAMPLE_FILECHECK={ON, OFF}
    • Enable validation of example outputs using the LLVM FileCheck utility. Default is OFF.

Single Config CMake Options

  • THRUST_HOST_SYSTEM={CPP, TBB, OMP}
    • Selects the host system. Default: CPP
  • THRUST_DEVICE_SYSTEM={CUDA, TBB, OMP, CPP}
    • Selects the device system. Default: CUDA
  • THRUST_CPP_DIALECT={11, 14, 17}
    • Selects the C++ standard dialect to use. Default is 14 (C++14).

Multi Config CMake Options

  • THRUST_MULTICONFIG_ENABLE_DIALECT_CPPXX={ON, OFF}
    • Toggle whether a specific C++ dialect will be targeted.
    • Possible values of XX are {11, 14, 17}.
    • By default, only C++14 is enabled.
  • THRUST_MULTICONFIG_ENABLE_SYSTEM_XXXX={ON, OFF}
    • Toggle whether a specific system will be targeted.
    • Possible values of XXXX are {CPP, CUDA, TBB, OMP}
    • By default, only CPP and CUDA are enabled.
  • THRUST_MULTICONFIG_WORKLOAD={SMALL, MEDIUM, LARGE, FULL}
    • Restricts the host/device combinations that will be targeted.
    • By default, the SMALL workload is used.
    • The full cross product of host x device systems results in 12 configurations, some of which are more important than others. This option can be used to prune some of the less important ones.
    • SMALL: (3 configs) Minimal coverage and validation of each device system against the CPP host.
    • MEDIUM: (6 configs) Cheap extended coverage.
    • LARGE: (8 configs) Expensive extended coverage. Includes all useful build configurations.
    • FULL: (12 configs) The complete cross product of all possible build configurations.
Config Workloads Value Expense Note
CPP/CUDA F L M S Essential Expensive Validates CUDA against CPP
CPP/OMP F L M S Essential Cheap Validates OMP against CPP
CPP/TBB F L M S Essential Cheap Validates TBB against CPP
CPP/CPP F L M Important Cheap Tests CPP as device
OMP/OMP F L M Important Cheap Tests OMP as host
TBB/TBB F L M Important Cheap Tests TBB as host
TBB/CUDA F L Important Expensive Validates TBB/CUDA interop
OMP/CUDA F L Important Expensive Validates OMP/CUDA interop
TBB/OMP F Not useful Cheap Mixes CPU-parallel systems
OMP/TBB F Not useful Cheap Mixes CPU-parallel systems
TBB/CPP F Not Useful Cheap Parallel host, serial device
OMP/CPP F Not Useful Cheap Parallel host, serial device

CUDA Specific CMake Options

  • THRUST_INCLUDE_CUB_CMAKE={ON, OFF}
    • If enabled, the CUB project will be built as part of Thrust. Default is OFF.
    • This adds CUB tests, etc. Useful for working on both CUB and Thrust simultaneously.
    • CUB configurations will be generated for each C++ dialect targeted by the current Thrust build.
  • THRUST_ENABLE_COMPUTE_XX={ON, OFF}
    • Controls the targeted CUDA architecture(s)
    • Multiple options may be selected when using NVCC as the CUDA compiler.
    • Valid values of XX are: {35, 37, 50, 52, 53, 60, 61, 62, 70, 72, 75, 80}
    • Default value depends on THRUST_DISABLE_ARCH_BY_DEFAULT:
  • THRUST_ENABLE_COMPUTE_FUTURE={ON, OFF}
    • If enabled, CUDA objects will target the most recent virtual architecture in addition to the real architectures specified by the THRUST_ENABLE_COMPUTE_XX options.
    • Default value depends on THRUST_DISABLE_ARCH_BY_DEFAULT:
  • THRUST_DISABLE_ARCH_BY_DEFAULT={ON, OFF}
    • When ON, all THRUST_ENABLE_COMPUTE_* options are initially OFF.
    • Default: OFF (meaning all architectures are enabled by default)
  • THRUST_ENABLE_TESTS_WITH_RDC={ON, OFF}
    • Whether to enable Relocatable Device Code when building tests. Default is OFF.
  • THRUST_ENABLE_EXAMPLES_WITH_RDC={ON, OFF}
    • Whether to enable Relocatable Device Code when building examples. Default is OFF.

TBB Specific CMake Options

  • THRUST_TBB_ROOT=<path to tbb root>
    • When the TBB system is requested, set this to the root of the TBB installation (e.g. the location of lib/, bin/ and include/ for the TBB libraries).

Development Model

The following is a description of the basic development process that Thrust follows. This is a living document that will evolve as our process evolves.

Thrust is distributed in three ways:

  • On GitHub.
  • In the NVIDIA HPC SDK.
  • In the CUDA Toolkit.

Trunk Based Development

Thrust uses trunk based development. There is a single long-lived branch called master. Engineers may create branches for feature development. Such branches always merge into master. There are no release branches. Releases are produced by taking a snapshot of master ("snapping"). After a release has been snapped from master, it will never be changed.

Repositories

As Thrust is developed both on GitHub and internally at NVIDIA, there are three main places where code lives:

  • The Source of Truth, the public Thrust repository, referred to as github later in this document.
  • An internal GitLab repository, referred to as gitlab later in this document.
  • An internal Perforce repository, referred to as perforce later in this document.

Versioning

Thrust has its own versioning system for releases, independent of the versioning scheme of the NVIDIA HPC SDK or the CUDA Toolkit.

Today, Thrust version numbers have a specific semantic meaning. Releases prior to 1.10.0 largely, but not strictly, followed these semantic meanings.

The version number for a Thrust release uses the following format: MMM.mmm.ss-ppp, where:

  • THRUST_VERSION_MAJOR/MMM: Major version, up to 3 decimal digits. It is incremented when changes that are API-backwards-incompatible are made.
  • THRUST_VERSION_MINOR/mmm: Minor version, up to 3 decimal digits. It is incremented when breaking API, ABI, or semantic changes are made.
  • THRUST_VERSION_SUBMINOR/ss: Subminor version, up to 2 decimal digits. It is incremented when notable new features or bug fixes or features that are API-backwards-compatible are made.
  • THRUST_PATCH_NUMBER/ppp: Patch number, up to 3 decimal digits. It is incremented if any change in the repo whatsoever is made and no other version component has been incremented.

The <thrust/version.h> header defines THRUST_* macros for all of the version components mentioned above. Additionally, a THRUST_VERSION macro is defined, which is an integer literal containing all of the version components except for THRUST_PATCH_NUMBER.

Branches and Tags

The following tag names are used in the Thrust project:

  • github/nvhpc-X.Y: the tag that directly corresponds to what has been shipped in the NVIDIA HPC SDK release X.Y.
  • github/cuda-X.Y: the tag that directly corresponds to what has been shipped in the CUDA Toolkit release X.Y.
  • github/A.B.C: the tag that directly corresponds to a Thrust version A.B.C.

The following branch names are used in the Thrust project:

  • github/master: the Source of Truth development branch of Thrust.
  • github/old-master: the old Source of Truth branch, before unification of public and internal repositories.
  • github/feature/<name>: feature branch for a feature under development.
  • github/bug/<bug-system>/<bug-description>-<bug-id>: bug fix branch, where bug-system is github or nvidia.
  • gitlab/master: mirror of github/master.
  • perforce/private: mirrored github/master, plus files necessary for internal NVIDIA testing systems.

On the rare occasion that we cannot do work in the open, for example when developing a change specific to an unreleased product, these branches may exist on gitlab instead of github. By default, everything should be in the open on github unless there is a strong motivation for it to not be open.