[
https://issues.apache.org/jira/browse/ORC-1137?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Quanlong Huang resolved ORC-1137.
---------------------------------
Fix Version/s: 1.8.0
Resolution: Fixed
> [C++] Improve float/double conversion in DoubleColumnReader::next()
> -------------------------------------------------------------------
>
> Key: ORC-1137
> URL: https://issues.apache.org/jira/browse/ORC-1137
> Project: ORC
> Issue Type: Improvement
> Components: C++
> Reporter: Quanlong Huang
> Assignee: Quanlong Huang
> Priority: Major
> Fix For: 1.8.0
>
> Attachments: DoubleColumnReader.png, DoubleColumnReader_readDouble.png
>
>
> DoubleColumnReader::next() takes a portion of time when reading doubles from
> a tpch-lineitem ORC file.
> !DoubleColumnReader.png|width=783,height=203!
> I can see the loop in readDouble is unrolled. But it still have redundant
> checks. We can manually unroll it to save some instructions.
> !DoubleColumnReader_readDouble.png|width=598,height=653!
> Furthermode, in little-endian machines, the layout of the DATA stream of the
> DOUBLE column matches the memory layout of the output array. We can use
> std::memcpy to copy the data directly.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
