optimize hash table construction for UM joins (MCOL-3586)

[MCOL-3587] some low-hanging fruit in UM join construction Created: 2019-11-04  Updated: 2019-11-14  Resolved: 2019-11-14

Status: Closed
Project: MariaDB ColumnStore
Component/s: ExeMgr
Affects Version/s: None
Fix Version/s: 1.4.0

Type: Sub-Task Priority: Major
Reporter: Patrick LeBlanc (Inactive) Assignee: Patrick LeBlanc (Inactive)
Resolution: Done Votes: 0
Labels: None


 Description   

looking into what is necessary for the supertask. Going to use this subtask to track easy optimizations to be made after the main task is done.

  • Unnecessary branching on every row (is it this type, is it that type, etc). TODO: Pass a whole RowGroup into TupleJoiner::insert(), so the branches can be evaluated once for a rowgroup rather than on every row.
  • tuplejoiner could use some cleanup in general. this has gotten more hard to understand than is justified...

 Comments   
Comment by Patrick LeBlanc (Inactive) [ 2019-11-14 ]

Wrote versions of a couple key functions to take whole rowgroups as input. Did not eliminate all usage of the old fcns yet, just did the primary paths. Therefore, cluttered the code a little more by adding and not removing things. The cleanup will have to be another ticket, need to keep moving forward for now.

Generated at Thu Feb 08 02:43:50 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.