Bug 41606 – Refactored string parsing in the lexer

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Zoltan Herczeg]

The current sunspider parse-only results are the following:

=============================================================================

** TOTAL **:                   1.044x as fast    33.1ms +/- 1.9%   31.7ms +/- 1.1%     significant

=============================================================================

  jquery:                      -                  5.1ms +/- 4.4%    5.1ms +/- 4.4%
    1.3.2:                     -                  5.1ms +/- 4.4%    5.1ms +/- 4.4%

  mootools:                    1.080x as fast     5.4ms +/- 6.8%    5.0ms +/- 0.0%     significant
    1.2.2-core-nc:             1.080x as fast     5.4ms +/- 6.8%    5.0ms +/- 0.0%     significant

  prototype:                   -                  6.1ms +/- 3.7%    6.0ms +/- 0.0%
    1.6.0.3:                   -                  6.1ms +/- 3.7%    6.0ms +/- 0.0%

  concat:                      1.058x as fast    16.5ms +/- 2.3%   15.6ms +/- 2.4%     significant
    jquery-mootools-prototype: 1.058x as fast    16.5ms +/- 2.3%   15.6ms +/- 2.4%     significant

But it was not before, actually it was a 5% slowdown. This patch was actually completed a week ago, and I didn't really changed it. I planned to submit a patch with a gcov output and ask for help. (I usually run sunspider 10-15 times before choose the best measurement, to lessen the effect of deviation caused by the OS. So this patch was really slower before, and faster now, albeit I think it is not the effect of the patch, more likely the changed binary structure) Anyway, the gcov output is also attached.

[Zoltan Herczeg]

Created attachment 60514 [details]
patch

[WebKit Review Bot]

Attachment 60514 [details] did not pass style-queue:

Failed to run "['WebKitTools/Scripts/check-webkit-style', '--no-squash']" exit_code: 1
JavaScriptCore/parser/Lexer.cpp:398:  An else if statement should be written as an if statement when the prior "if" concludes with a return, break, continue or goto statement.  [readability/control_flow] [4]
Total errors found: 1 in 3 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Zoltan Herczeg]

Created attachment 60515 [details]
Lexer.cpp.gcov

[Zoltan Herczeg]

The gcov profile is for concat-jquery-mootools-prototype

[Zoltan Herczeg]

Hm, maybe I have an idea what is causing the slowdown. The original code, there was two makeIdentifier call. I did a few changes before, and realized I removed one of them (one was for a fast path when the m_buffer was not needed). I put it back (thought it was a stupid move), and the slowdown appeared again. Later I refactored the code, that even though I keep the fast case, only one makeIdentifier appeared in the code, and the speed was nearly as fast as the patch added to this bug. Maybe the ALWAYS_INLINE makeIdentifier is more computation expensive than I expected... at least on my machine.

[Oliver Hunt]

Comment on attachment 60514 [details]
patch

r=me

[Zoltan Herczeg]

Landed in http://trac.webkit.org/changeset/62628
Closing bug