[jira] [Commented] (ARROW-6411) [C++][Parquet] DictEncoderImpl::PutIndicesTyped has bad performance on some systems

["Wes McKinney (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-6411?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16920531#comment-16920531 ] 

Wes McKinney commented on ARROW-6411:
-------------------------------------

I'll see if it shows up in a C++ benchmark also, will be easier to run on other systems. 

> [C++][Parquet] DictEncoderImpl<T>::PutIndicesTyped has bad performance on some systems
> --------------------------------------------------------------------------------------
>
>                 Key: ARROW-6411
>                 URL: https://issues.apache.org/jira/browse/ARROW-6411
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: C++
>            Reporter: Wes McKinney
>            Assignee: Wes McKinney
>            Priority: Major
>             Fix For: 0.15.0
>
>
> I was doing some benchmarking and noticed that this function showed up as slow due to {{\_\_memmove_avx_unaligned_erms}}. I'm interested to investigate why this is, but for me it's fixed by changing the {{std::vector::reserve}} call to {{std::vector::resize}} and instead assigning elements into {{buffered_indices_}}. I'll add a Python benchmark that illustrates the problem to see if it shows up on other systems



--
This message was sent by Atlassian Jira
(v8.3.2#803003)