Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/43883 Check the result of GCC coverage in OSS is reasonable and ready to ship. The amount of executable lines are not the same between `gcc` and `clang` because of the following reasons: * Lines following are counted in `clang` but not in `gcc`: 1. empty line or line with only “{” or “}” 3. some comments are counted in clang but not in gcc 5. `#define ...` -- not supported by gcc according to official documentation * Besides, a statement that explains to more than one line will be counted as only one executable line in gcc, but several lines in clang ## Advantage of `gcc` coverage 1. Much faster - code coverage tool runtime is onle **4 min** (*ammazzzing!!*) by `gcc`, compared to **3 hours!!** by `clang`, to analyze all the tests' artifacts 2. Use less disk - `Clang`'s artifacts will take as large as 170G, but `GCC` is 980M Besides, also update `README.md`. Test Plan: Compare the result in OSS `clang` and OSS `gcc` with the same command: ``` python oss_coverage.py --run-only atest test_nn.py --interested-folder=aten ``` ---- ## GCC **Summary** > time: 0:15:45 summary percentage: 44.85% **Report and Log** [File Coverage Report](P140825162) [Line Coverage Report](P140825196) [Log](P140825385) ------ ## CLANG **Summary** > time: 0:21:35 summary percentage: 44.08% **Report and Log** [File Coverage Report](P140825845) [Line Coverage Report](P140825923) [Log](P140825950) ---------- # Run all tests ``` # run all tests and get coverage over Pytorch python oss_coverage.py ``` **Summary** > time: 1:27:20. ( time to run tests: 1:23:33) summary percentage: 56.62% **Report and Log** [File Coverage Report](P140837175) [Log](P140837121) Reviewed By: malfet Differential Revision: D23416772 fbshipit-source-id: a6810fa4d8199690f10bd0a4f58a42ab2a22182b
Code Coverage Tool for Pytorch
Overview
This tool is designed for calculating code coverage for Pytorch project. It’s an integrated tool. You can use this tool to run and generate both file-level and line-level report for C++ and Python tests. It will also be the tool we use in CircleCI to generate report for each master commit.
Simple
- Simple command to run:
python oss_coverage.py
- Argument
--clean
will do all the messy clean up things for you
But Powerful
- Choose your own interested folder:
- Default folder will be good enough in most times
- Flexible: you can specify one or more folder(s) that you are interested in
- Run only the test you want:
- By default it will run all the c++ and python tests
- Flexible: you can specify one or more test(s) that you want to run
- Final report:
- File-Level: The coverage percentage for each file you are interested in
- Line-Level: The coverage details for each line in each file you are interested in
- More complex but flexible options:
- Use different stages like --run, --export, --summary to achieve more flexible functionality
How to use
This part will introduce about the arguments you can use when run this tool. The arguments are powerful, giving you full flexibility to do different work.
We have two different compilers, gcc
and clang
, and this tool supports both. But it is recommended to use gcc
because it's much faster and use less disk place. The examples will also be divided to two parts, for gcc
and clang
.
Examples
First step is to set some experimental value if needed:
# pytorch folder, by default all the c++ binaries are in build/bin/
export PYTORCH_FOLDER=...
# set compiler type
export COMPILER_TYPE="GCC" or export COMPILER_TYPE="CLANG"
# make sure llvm-cov is available, by default it is /usr/local/opt/llvm/bin
export LLVM_TOOL_PATH=...
then command will run all the tests in build/bin/
and test/
folder
python oss_coverage.py
Most times you don't want collect coverage for the entire Pytorch folder, use --interested-folder to report coverage only over the folder you want:
python oss_coverage.py --interested-folder=aten
Then, still in most cases, if you only run one or several test(s):
python oss_coverage.py --run-only=atest
python oss_coverage.py --run-only atest basic test_nn.py
For more complex arguments and functionality
To Be Done
Reference
For gcc
- See about how to invoke
gcov
, read Invoking gcov will be helpful
For clang
- If you are not familiar with the procedure of generating code coverage report by using
clang
, read Source-based Code Coverage will be helpful.