hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matt McCline (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HIVE-20207) Vectorization: Fix NULL / Wrong Results issues in Filter / Compare
Date Fri, 20 Jul 2018 19:52:00 GMT

     [ https://issues.apache.org/jira/browse/HIVE-20207?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Matt McCline updated HIVE-20207:
--------------------------------
    Description: 
Write new UT tests that use random data and intentional isRepeating batches to checks for
NULL and Wrong Results for vectorized filter and compare.

BUGS:

1) LongColLessLongColumn SIMD optimization do not work for very large integers:
 -7272907770454997143 < 8976171455044006767
 outputVector[i] = (vector1[i] - vector2[i]) >>> 63;
 Produces 0 instead of 1...

Also, add DECIMAL_64 testing. Add missing DECIMAL/DECIMAL_64 IF vectorized expression classes.

  was:
Write new UT tests that use random data and intentional isRepeating batches to checks for
NULL and Wrong Results for vectorized filter and compare.

BUGS:

1) LongColLessLongColumn SIMD optimization do not work for very large integers:
 -7272907770454997143 < 8976171455044006767
 outputVector[i] = (vector1[i] - vector2[i]) >>> 63;
 Produces 0 instead of 1...


> Vectorization: Fix NULL / Wrong Results issues in Filter / Compare
> ------------------------------------------------------------------
>
>                 Key: HIVE-20207
>                 URL: https://issues.apache.org/jira/browse/HIVE-20207
>             Project: Hive
>          Issue Type: Bug
>          Components: Hive
>            Reporter: Matt McCline
>            Assignee: Matt McCline
>            Priority: Critical
>
> Write new UT tests that use random data and intentional isRepeating batches to checks
for NULL and Wrong Results for vectorized filter and compare.
> BUGS:
> 1) LongColLessLongColumn SIMD optimization do not work for very large integers:
>  -7272907770454997143 < 8976171455044006767
>  outputVector[i] = (vector1[i] - vector2[i]) >>> 63;
>  Produces 0 instead of 1...
> Also, add DECIMAL_64 testing. Add missing DECIMAL/DECIMAL_64 IF vectorized expression
classes.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message