[
https://issues.apache.org/jira/browse/IMPALA-6344?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Armstrong resolved IMPALA-6344.
-----------------------------------
Resolution: Later
> Optimize decimal multiplication
> -------------------------------
>
> Key: IMPALA-6344
> URL: https://issues.apache.org/jira/browse/IMPALA-6344
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Reporter: Taras Bobrovytsky
> Priority: Major
> Labels: decimal, perf
>
> Our current implementation of decimal multiplication can be slow and
> non-optimal due to having branches in our code.
> [~zamsden] suggested to use
> [https://en.wikipedia.org/wiki/Karatsuba_algorithm] multiplication for int128
> * int128 -> int256 multiply. The following example implements this and uses 3
> hardware 64-bit multiplies to get a full 256 bit result. The code is written
> in inline assembly and has no branches.
> http://coliru.stacked-crooked.com/a/25a697389211189f
> We should consider benchmarking this code and using this approach if it turns
> out to be faster.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
