You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@impala.apache.org by "Daniel Becker (JIRA)" <ji...@apache.org> on 2019/07/05 12:06:00 UTC

[jira] [Created] (IMPALA-8741) Speed up bit unpacking by vectorisation

Daniel Becker created IMPALA-8741:
-------------------------------------

             Summary: Speed up bit unpacking by vectorisation
                 Key: IMPALA-8741
                 URL: https://issues.apache.org/jira/browse/IMPALA-8741
             Project: IMPALA
          Issue Type: Improvement
          Components: Backend
            Reporter: Daniel Becker
            Assignee: Daniel Becker


Using compiler intrinsics for SIMD and bit manipulation instructions (AVX, AVX2 and BMI2), we can speed up bit unpacking by a factor of about 2 to 8 depending on bit width, at most 16.

We need to take care to check that the required instructions are supported by the CPU the impalad is running on and fall back to the scalar implementation if not.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)