diff options
author | Xueliang Zhong <xueliang.zhong@linaro.org> | 2016-01-26 15:59:00 +0000 |
---|---|---|
committer | Alexandre Rames <alexandre.rames@linaro.org> | 2016-02-01 14:20:40 +0000 |
commit | ba4e60c9450ba122ff0b673d1c1f39e97b108f33 (patch) | |
tree | 1f7904d2e7a4b9f75ac7befab101ece33a55fa49 /README.md | |
parent | dcfe628091365d910a565a0f3824c6caf10f62df (diff) | |
download | art-testing-ba4e60c9450ba122ff0b673d1c1f39e97b108f33.tar.gz |
Guidelines about porting existing benchmark in README.md.
Change-Id: Ifa28745c62da3da5125704bd97f1abc8a1a6385c
Diffstat (limited to 'README.md')
-rw-r--r-- | README.md | 36 |
1 files changed, 35 insertions, 1 deletions
@@ -59,7 +59,7 @@ And similarly on target adb shell "cd /data/local/tmp && dalvikvm -cp /data/local/tmp/bench.apk benchmarks/micro/Base64" -### Comparing the rsults +### Comparing the results The results of `run.py` can be compared using `compare.py`. @@ -103,6 +103,40 @@ benchmarks/ folder. Before contributing, make sure that `test/test.py` passes. + +## How to Port an Existing Benchmark + +Similar to writing a benchmark, above guidelines also applies to porting an +existing benchmark. Besides, developers should also notice: +1. Licenses: + Make sure the benchmark has appropriate license for us to integrate it + freely into our test framework. Apache-v2.0, BSD, MIT licenses are well- + known and preferred. Check with the gatekeepers for other licenses. + The original license header in the ported benchmark MUST be *preserved* and + *unmodified*. + +2. Porting a benchmark should be done in two commits: + (1) Add *untouched* original file *with* its license and copyright header. + (2) Modify the benchmark as necessary. + This allows easily showing (`git diff <first commit> <second commit>`) + what modifications have been made to the original benchmarks. + +3. Keep the original code as it is: + This includes indents, spaces, tabs, etc. Only make changes to original code + when you have to (e.g. fit into our framework), but keep the changes as + minimal as possible. When we have to investigate why we're getting different + results than other projects or developers using the same benchmark, a 'diff' + should show as few changes as possible. If the original code has some coding + style which cannot pass our 'checkstyle' script, use 'CHECKSTYLE.OFF' to + bypass. + +4. Header comment: + When you have modified the code, make sure you comply with the license terms. + Provide a full copy of the license (Apache2, BSD, MIT, etc.) and notices + stating that you changed the files (required by GPLv2 etc) in the header + comment. Also, please put description in the header: where did you find the + benchmark source code and a link to original source. + ### Rules 1. Test method names start with "time" -- Test launcher will find all timeXXX() |