Tim Armstrong has posted comments on this change.

Change subject: IMPALA-2680: faster memory copy
......................................................................


Patch Set 10:

I ran an end-to-end benchmark query does a lot of small memcpys (on 
non-codegened paths, some of which will be codegened).

   select distinct * from tpch_parquet.lineitem limit 5

In the baseline version it spends ~4.9% of CPU time in __memcpy_avx_unaligned, 
with this patch it spends ~4.2% in _ZN6impala7MemUtil6memcpyEPvPKvm.avx2 and 
~0.7% in __memcpy_avx_unaligned. So it doesn't seem like there's any gain.

Change in end-to-end query time and operator cpu time was within variance.

I'm not sure if we have any end-to-end workloads that do large memcpys: i 
suspect those would be where we see improvement.

-- 
To view, visit http://gerrit.cloudera.org:8080/1686
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I7f6c046d966883aa66f26d58bee92c427f973e67
Gerrit-PatchSet: 10
Gerrit-Project: Impala
Gerrit-Branch: cdh5-trunk
Gerrit-Owner: Zuo Wang <[email protected]>
Gerrit-Reviewer: Dan Hecht <[email protected]>
Gerrit-Reviewer: Marcel Kornacker <[email protected]>
Gerrit-Reviewer: Mostafa Mokhtar <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Zuo Wang <[email protected]>
Gerrit-HasComments: No

Reply via email to