Mailing-List: contact dev-help@hive.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@hive.apache.org
Date: Sun, 1 Dec 2013 20:18:35 +0000 (UTC)
From: "Teddy Choi (JIRA)" <jira@apache.org>
To: hive-dev@hadoop.apache.org
Message-ID: <JIRA.12677827.1383761721548.51209.1385929115948@arcas>
In-Reply-To: <JIRA.12677827.1383761721548@arcas>
References: <JIRA.12677827.1383761721548@arcas>
Subject: [jira] [Commented] (HIVE-5761) Implement vectorized support for the
 DATE data type
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable


    [ https://issues.apache.org/jira/browse/HIVE-5761?page=3Dcom.atlassian.=
jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=3D13836=
108#comment-13836108 ]=20

Teddy Choi commented on HIVE-5761:
----------------------------------

I wrote a draft version.

{quote}
DATE shall be implemented within a LongColumnVector. HIVE-4055 represents a=
 DATE value by a number of days since epoch. A vectorized DATE representati=
on will contain this number and its optional cached parse result. A read op=
eration result and a complex date function result, such as date_add and dat=
e_sub, will have an empty cache. During the first simple date function, suc=
h as year, month and day, it will cache its parse result. Then following si=
mple functions will reuse its cache to avoid repeated parses. Its effect on=
 performance will be small, since java.util.Date calculates all fields at o=
nce and caches their results. The first 32-bit set will represent a number =
of days since epoch as a signed integer. Its range is about from BC 2^31/36=
5-1970 to AD 2^31/365+1970. A comparison between vectorized DATE values sho=
uld consider only their first sets. The following 32-bit set will represent=
 its cached parse result; cached state (1 bit; 0 for not cached, 1 for cach=
ed), era (1 bit; 0 for AD, 1 for BC), year (unsigned 21-bit integer), month=
 (unsigned 4-bit integer) and day of month (unsigned 5-bit integer). A valu=
e without a cache will have only zero bits after its first set. A parsed ye=
ar, month and day of month value will start from 1 to represent the exact n=
umber. Its range is from BC 2^21 to AD 2^21, which is shorter than the firs=
t set. If a date is not in the range, its cached state will remain false (0=
). The value 0xFFFFFFFF00000000L shall be reserved for future use to indica=
te data outside the standard range.
{quote}

> Implement vectorized support for the DATE data type
> ---------------------------------------------------
>
>                 Key: HIVE-5761
>                 URL: https://issues.apache.org/jira/browse/HIVE-5761
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Eric Hanson
>            Assignee: Teddy Choi
>
> Add support to allow queries referencing DATE columns and expression resu=
lts to run efficiently in vectorized mode. This should re-use the code for =
the the integer/timestamp types to the extent possible and beneficial. Incl=
ude unit tests and end-to-end tests. Consider re-using or extending existin=
g end-to-end tests for vectorized integer and/or timestamp operations.


--
This message was sent by Atlassian JIRA
(v6.1#6144)