stiga-huang opened a new pull request #944:
URL: https://github.com/apache/orc/pull/944
<!--
Thanks for sending a pull request! Here are some tips for you:
1. File a JIRA issue first and use it as a prefix of your PR title, e.g.,
`ORC-001: Fix ABC`.
2. Use your PR title to summarize what this PR proposes instead of
describing the problem.
3. Make PR title and description complete because these will be the
permanent commit log.
4. If possible, provide a concise and reproducible example to reproduce
the issue for a faster review.
5. If the PR is unfinished, use GitHub PR Draft feature.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section
is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class
hierarchy will help reviewers.
2. If there is a discussion in the mailing list, please add the link.
-->
This PR optimizes the C++ implementation of orc::RleDecoderV2::nextDirect.
It provides different methods for different bit sizes, which significantly
reduces the instructions and branches. Note that only bit sizes of
4,8,16,24,32,40,48,56,64 are optimized in this PR. bitSize 1 and 2 are skipped
since they usually have short runs which won't benefit from loop unrolling.
Other deprecated bit sizes, e.g. 3,5,6,7 are also skipped.
When the bit size is specified, we can better exhaust the buffer before
calling readByte(). This also improves performance.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
Perf tests show that orc::RleDecoderV2::nextDirect() dominant the time in
reading random integers. There are more details in the JIRA description.
Performance improvement for scanning 1 billion unsigned numbers in different
bit sizes:
bitSize | original time(s) | optimized time(s) | speedup
-- | -- | -- | --
4 | 5.93 | 3.25 | 1.8246153846
8 | 6.24 | 2.62 | 2.3816793893
16 | 7.24 | 2.59 | 2.7953667954
24 | 8.28 | 3.04 | 2.7236842105
32 | 9.45 | 2.9 | 3.2586206897
40 | 10.62 | 3.42 | 3.1052631579
48 | 11.67 | 3.56 | 3.2780898876
56 | 12.86 | 4.21 | 3.054631829
64 | 14.05 | 7.76 | 1.8105670103
Tested on Ubuntu16.04 with a 6 cores CPU (12 virtual cores) and 32GB RAM.
CPU: Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
-->
Add more tests in TestRleDecoder.cc
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
