[MCOL-4995] implement basic vectorized filtering for ARM platforms Created: 2022-02-21  Updated: 2022-09-20  Resolved: 2022-08-25

Status: Closed
Project: MariaDB ColumnStore
Component/s: PrimProc
Affects Version/s: 6.2.3
Fix Version/s: 22.08.1

Type: New Feature Priority: Major
Reporter: Roman Assignee: HongdongHe
Resolution: Fixed Votes: 2
Labels: gsoc22

Issue Links:
Relates
relates to MCOL-5207 Update failed MTR test case because o... Closed
Assigned for Review: Roman Roman

 Description   

AS of 6.2.2 Columnstore(MCS) supports vectorization on x86_64 platforms only.
The goal of the project is to implement a low-level filtering for ARM platforms using 128bit ARM NEON extention(SVE is optional).
Low-level filtering algorithm is defined here. The filtering algorithm for x86_64 uses templated class methods that are defined here (SimdProcessor class)

Here are the phases for the project:

  1. get familiar with the code base, compilation and bootstrap process [1]. Fix the bootstrap script if needed.
  2. get familiar with vector extensions(NEON, SVE) for ARM platforms
  3. implement vectorization on ARM

Plz get in touch either here or in MariaDB Zulip chat

1. Here is the automated bootstrap script.



 Comments   
Comment by HongdongHe [ 2022-04-08 ]

Hi,
I'm Hongdong He, computer science student from Beijing Institute of Technology, in China. I would love to contribute to this project. I have read column.cpp and simd_sse.h.How I get started with it?
Thank you.

Comment by Roman [ 2022-04-08 ]

Hey NTH19.
Nice to meet you. Plz reach me in zulip to discuss the topic.

Comment by Roman [ 2022-08-16 ]

Thx to NTH19 the feature has been implemented.

Comment by Roman [ 2022-08-16 ]

You need either AWS instances with their Graviton CPU or ARM-based Apple's MBP.

Generated at Thu Feb 08 02:54:34 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.