Tim Armstrong has posted comments on this change.

Change subject: IMPALA-4058: ByteSwap256 assumed memory was 16-byte aligned.
......................................................................


Patch Set 5: Code-Review+1

(1 comment)

I also extended the microbenchmark to test perf when loads and stores were 
actually misaligned (not just using the different instruction). AVX2 is still 
significantly faster than SSE. I'll post a CR with that benchmark change 
separately.

    I0901 13:47:02.025802  4956 bswap-benchmark.cc:122] Machine Info: Intel(R) 
Core(TM) i7-4790 CPU @ 3.60GHz
    ByteSwap benchmark alignment=0:Function  iters/ms   10%ile   50%ile   
90%ile     10%ile     50%ile     90%ile
                                                                             
(relative) (relative) (relative)
    
---------------------------------------------------------------------------------------------------------
                             FastScalar                881 1.08e+03 1.14e+03    
     1X         1X         1X
                                  SSSE3           9.33e+03 1.02e+04 1.03e+04    
  10.6X      9.46X      9.03X
                                   AVX2           3.43e+04 3.78e+04 3.84e+04    
  38.9X      35.2X      33.7X
                                   SIMD           3.27e+04 3.78e+04 3.83e+04    
  37.1X      35.2X      33.6X
    ByteSwap benchmark alignment=1:Function  iters/ms   10%ile   50%ile   
90%ile     10%ile     50%ile     90%ile
                                                                             
(relative) (relative) (relative)
    
---------------------------------------------------------------------------------------------------------
                             FastScalar              1e+03 1.08e+03 1.15e+03    
     1X         1X         1X
                                  SSSE3           8.67e+03 9.01e+03 9.11e+03    
  8.66X      8.31X      7.95X
                                   AVX2           2.62e+04 2.75e+04 2.77e+04    
  26.2X      25.4X      24.2X
                                   SIMD           2.56e+04 2.72e+04 2.75e+04    
  25.6X      25.1X        24X

http://gerrit.cloudera.org:8080/#/c/4205/5/be/src/util/bit-util-test.cc
File be/src/util/bit-util-test.cc:

PS5, Line 116: o
or


-- 
To view, visit http://gerrit.cloudera.org:8080/4205
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I39b2b47bb717d5ac9727512a24fcf8a8a6a8dcc6
Gerrit-PatchSet: 5
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Jim Apple <[email protected]>
Gerrit-Reviewer: Dan Hecht <[email protected]>
Gerrit-Reviewer: Jim Apple <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Youwei Wang <[email protected]>
Gerrit-HasComments: Yes

Reply via email to