Bug 154674

Summary: Consider using a jump table in element factories
Product: WebKit Reporter: Ryosuke Niwa <rniwa>
Component: DOMAssignee: Nobody <webkit-unassigned>
Status: RESOLVED WONTFIX    
Severity: Normal CC: andersca, benjamin, cdumez, darin, dbates, esprehn+autocc, ews-watchlist, gyuyoung.kim, kangil.han, kling, koivisto, sam
Priority: P2    
Version: WebKit Nightly Build   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
WIP
none
WIP
none
WIP2
none
WIP3
none
WIP4 (fixed build) none

Ryosuke Niwa
Reported 2016-02-25 03:22:34 PST
Created attachment 272199 [details] WIP We've always said that it's silly to be using HashMap for looking up the atomic string name. Here's a super crude prototype of using alphabet jump table & one-level binary search. This appears to be ~0.5% improvement on PerformanceTests/Bindings/create-element.html but this test only creates div so we need a more comprehensive testing.
Attachments
WIP (13.08 KB, patch)
2016-02-25 03:22 PST, Ryosuke Niwa 		no flags
DetailsFormatted DiffDiff
WIP (16.01 KB, patch)
2016-02-25 19:35 PST, Ryosuke Niwa 		no flags
DetailsFormatted DiffDiff
WIP2 (33.90 KB, patch)
2018-03-22 23:55 PDT, Ryosuke Niwa 		no flags
DetailsFormatted DiffDiff
WIP3 (37.62 KB, patch)
2018-03-23 01:26 PDT, Ryosuke Niwa 		no flags
DetailsFormatted DiffDiff
WIP4 (fixed build) (35.32 KB, patch)
2018-03-23 14:57 PDT, Ryosuke Niwa 		no flags
DetailsFormatted DiffDiff
Show Obsolete (4) View All Add attachment proposed patch, testcase, etc.
Ryosuke Niwa
Comment 1 2016-02-25 19:35:14 PST
Created attachment 272281 [details] WIP Here's a new WIP patch with tests. I was observing a consistent 1-2% gain but I'm no longer seeing that now :( I don't know whether my testing was busted. I'm somehow seeing a lot more variance on my test results now.
Ryosuke Niwa
Comment 2 2018-03-22 23:55:41 PDT
Created attachment 336357 [details] WIP2
Ryosuke Niwa
Comment 3 2018-03-23 01:26:12 PDT
Created attachment 336359 [details] WIP3 With this patch, we avoid creating AtomicString when creating built-in HTML elements in the HTML parser & createElement. In the case of createElement, we also avoid lowercasing the string.
EWS Watchlist
Comment 4 2018-03-23 01:28:05 PDT
Attachment 336359 [details] did not pass style-queue: ERROR: Source/WebCore/dom/TagNameLookupTable.h:26: Use #pragma once instead of #ifndef for header guard. [build/header_guard] [5] ERROR: Source/WebCore/dom/TagNameLookupTable.cpp:29: Alphabetical sorting problem. [build/include_order] [4] ERROR: Source/WebCore/html/parser/AtomicHTMLToken.h:30: Alphabetical sorting problem. [build/include_order] [4] Total errors found: 3 in 13 files If any of these errors are false positives, please file a bug against check-webkit-style.
Ryosuke Niwa
Comment 5 2018-03-23 14:57:55 PDT
Created attachment 336429 [details] WIP4 (fixed build)
Ryosuke Niwa
Comment 6 2018-03-23 20:28:20 PDT
Hm... this appears to be an overall regression in Speedometer because there are enough non-standard attributes set. I guess we won't do this.
Note You need to log in before you can comment on or make changes to this bug.