Mailing-List: contact reviews-help@impala.incubator.apache.org; run by ezmlm
Precedence: bulk
Message-Id: <201708011848.v71ImldY008599@ip-10-146-233-104.ec2.internal>
Date: Tue, 1 Aug 2017 18:48:47 +0000
From: "Tim Armstrong (Code Review)" <gerrit@cloudera.org>
To: Taras Bobrovytsky <tbobrovytsky@cloudera.com>, impala-cr@cloudera.com,
        reviews@impala.incubator.apache.org
CC: Dan Hecht <dhecht@cloudera.com>, Michael Ho <kwho@cloudera.com>,
        Zach Amsden <zamsden@cloudera.com>
Reply-To: tarmstrong@cloudera.com
Subject: =?UTF-8?Q?=5BImpala-ASF-CR=5D_IMPALA-4939=2C_IMPALA-4939=3A_Decimal_V2_multiplication=0A?=
In-Reply-To: <gerrit.1500075819000.I37ad6232d7953bd75c18dc86e665b2b501a1ebe1@gerrit.cloudera.org>
References: <gerrit.1500075819000.I37ad6232d7953bd75c18dc86e665b2b501a1ebe1@gerrit.cloudera.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Content-Disposition: inline
User-Agent: Gerrit/2.12.7
archived-at: Tue, 01 Aug 2017 18:49:01 -0000

Tim Armstrong has posted comments on this change.

Change subject: IMPALA-4939, IMPALA-4939: Decimal V2 multiplication
......................................................................


Patch Set 4:

(4 comments)

http://gerrit.cloudera.org:8080/#/c/7438/4//COMMIT_MSG
Commit Message:

Line 7: IMPALA-4939, IMPALA-4939: Decimal V2 multiplication
The same JIRA is listed twice.


Line 75: DECIMAL_V2 disabled: 2.65s
This is scary and I don't think we should consider merging this until it's fixed (or we have a strong reason why the regression is unavoidable and justified). Do we understand why the int256_t is so much slower?


http://gerrit.cloudera.org:8080/#/c/7438/1/be/src/runtime/decimal-value.inline.h
File be/src/runtime/decimal-value.inline.h:

Line 301:       // The following is frought with apparent difficulty, as there is only 1 bit
> Added a TODO.
It seems most efficient development-wise if we give this a try now while we're working on and benchmarking this code instead of leaving a TODO.

This could be a big win - I know the div here is a big contributor to runtime for some queries, e.g. TPC-H Q1


http://gerrit.cloudera.org:8080/#/c/7438/1/be/src/util/bit-util.h
File be/src/util/bit-util.h:

Line 65:     return value < 0 ? -1 : 1;
> I included the benchmark result in the commit message.
Does it compile to the same assembly (or better assembly) as before? If it's adding a branch, the benchmark may be misleading since the worst case would be a 50/50 mix of positive and negative types, leading to branch mispredicts.

Another option is to keep the existing implementation and specialise the template for int256.


-- 
To view, visit http://gerrit.cloudera.org:8080/7438
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I37ad6232d7953bd75c18dc86e665b2b501a1ebe1
Gerrit-PatchSet: 4
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Taras Bobrovytsky <tbobrovytsky@cloudera.com>
Gerrit-Reviewer: Dan Hecht <dhecht@cloudera.com>
Gerrit-Reviewer: Michael Ho <kwho@cloudera.com>
Gerrit-Reviewer: Taras Bobrovytsky <tbobrovytsky@cloudera.com>
Gerrit-Reviewer: Tim Armstrong <tarmstrong@cloudera.com>
Gerrit-Reviewer: Zach Amsden <zamsden@cloudera.com>
Gerrit-HasComments: Yes