[
https://issues.apache.org/jira/browse/HIVE-16311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15946670#comment-15946670
]
Colin Ma commented on HIVE-16311:
---------------------------------
[~mmccline], I check the TestHiveDecimal and original design for division at
https://cwiki.apache.org/confluence/download/attachments/27362075/Hive_Decimal_Precision_Scale_Support.pdf,
and I think the implementation of OldHiveDecimal is not the same as original
design for division.
In OldHiveDecimal, the result scale is always MAX_SCALE as following code, but
in the original design, the result scale is calculated by the formula *max(6,
s1 + p2 + 1)*
{code}
public HiveDecimalV1 divide(HiveDecimalV1 dec) {
return create(bd.divide(dec.bd, MAX_SCALE, RoundingMode.HALF_UP), true);
}
{code}
For the test case 1/3 in TestHiveDecimal, the result of OldHiveDecimal is
0.33333333333333333333333333333333333333, and the result of original design is
0.333333. I think the TestHiveDecimal also should be updated according to the
original design.
I agree you to write a native or fast version of Decimal division. Actually,
current patch only sets the result scalar according to the original design, it
only benefits the division whose result scalar is less than MAX_SCALE. For the
division between real big Decimal, eg, *Decimal(38, 19)/Decimal(38, 19)*, there
is a little performance improvement. We need a faster way to do the division
like FastHiveDecimalImpl.fastSingleWordDivision, but there is no conflict with
the current patch. What do you think?
Thanks for your comments.
> Improve the performance for FastHiveDecimalImpl.fastDivide
> ----------------------------------------------------------
>
> Key: HIVE-16311
> URL: https://issues.apache.org/jira/browse/HIVE-16311
> Project: Hive
> Issue Type: Improvement
> Affects Versions: 2.2.0
> Reporter: Colin Ma
> Assignee: Colin Ma
> Fix For: 3.0.0
>
> Attachments: HIVE-16311.001.patch
>
>
> FastHiveDecimalImpl.fastDivide is poor performance when evaluate the
> expression as 12345.67/123.45
> There are 2 points can be improved:
> 1. Don't always use HiveDecimal.MAX_SCALE as scale when do the
> BigDecimal.divide.
> 2. Get the precision for BigInteger in a fast way if possible.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
