Bug 20813

Summary:

MASTER: get faster on v8 benchmarks

Product:

WebKit

Reporter:

Maciej Stachowiak <mjs>

Component:

JavaScriptCore

Assignee:

Geoffrey Garen <ggaren>

Status:

RESOLVED DUPLICATE

Severity:

Normal

CC:

barraclough, ddkilzer, emacemac7, ossy, sam, webmaster, zherczeg, zwarich

Priority:

Version:

528+ (Nightly build)

Hardware:

Mac

OS:

OS X 10.5

Bug Depends on:

20812, 20814, 20815, 20816, 20818, 20819, 20820, 20821, 20823, 20827, 20972, 20982, 20983, 20984, 20989, 21019, 21123, 21200, 21294, 21702, 21703, 21803, 21814, 21943

Bug Blocks:

Attachments:

Description	Flags
Shark Profile of Richards benchmark	none
Shark Profile of DeltaBlue benchmark	none
Shark profile of Crypto benchmark	none
Shark profile of Raytrace benchmark	none
Shark Profile of EarleyBoyer benchmark	none
newer profile of v8 RayTrace benchmark	none

Maciej Stachowiak

Reported 2008-09-12 15:12:28 PDT

The v8 benchmarks overfocus on a few specific language issues, but we should get faster on them than we are currently.

Attachments
Shark Profile of Richards benchmark (deleted) 2008-09-12 15:25 PDT, Maciej Stachowiak	no flags	Details
Shark Profile of DeltaBlue benchmark (deleted) 2008-09-12 15:27 PDT, Maciej Stachowiak	no flags	Details
Shark profile of Crypto benchmark (deleted) 2008-09-12 15:29 PDT, Maciej Stachowiak	no flags	Details
Shark profile of Raytrace benchmark (deleted) 2008-09-12 15:38 PDT, Maciej Stachowiak	no flags	Details
Shark Profile of EarleyBoyer benchmark (deleted) 2008-09-12 15:40 PDT, Maciej Stachowiak	no flags	Details
newer profile of v8 RayTrace benchmark (2.43 MB, application/octet-stream) 2008-10-04 17:47 PDT, Maciej Stachowiak	no flags	Details
View All Add attachment proposed patch, testcase, etc.

Maciej Stachowiak

Comment 1 2008-09-12 15:25:07 PDT

Created attachment 23375 [details] Shark Profile of Richards benchmark

Maciej Stachowiak

Comment 2 2008-09-12 15:27:46 PDT

Created attachment 23376 [details] Shark Profile of DeltaBlue benchmark

Maciej Stachowiak

Comment 3 2008-09-12 15:29:44 PDT

Created attachment 23377 [details] Shark profile of Crypto benchmark

Maciej Stachowiak

Comment 4 2008-09-12 15:38:32 PDT

Created attachment 23378 [details] Shark profile of Raytrace benchmark

Maciej Stachowiak

Comment 5 2008-09-12 15:40:10 PDT

Created attachment 23379 [details] Shark Profile of EarleyBoyer benchmark

Maciej Stachowiak

Comment 6 2008-09-12 15:53:15 PDT

One big issue is bug 20812 Our function call overhead is not so great right now (though faster than all non-v8 engines afaik) and could be improved. This affects all the test cases.

Maciej Stachowiak

Comment 7 2008-09-12 16:21:10 PDT

Another big issue is bug 20814 We spend quite a bit of time serving polymorphic inline cache misses, particularly on DeltaBlue.

Cameron Zwarich (cpst)

Comment 8 2008-09-12 16:43:25 PDT

Slow 'arguments' objection creation is bug 20815.

Cameron Zwarich (cpst)

Comment 9 2008-09-12 16:50:31 PDT

Optimization of op_lesseq is bug 20816.

Cameron Zwarich (cpst)

Comment 10 2008-09-12 17:30:41 PDT

Optimization of the 'instanceof' operator is bug 20818.

Cameron Zwarich (cpst)

Comment 11 2008-09-12 17:40:41 PDT

Optimization of JSValue::isObject() is bug 20819.

Cameron Zwarich (cpst)

Comment 12 2008-09-12 18:05:03 PDT

Optimization of op_stricteq is bug 20820.

Maciej Stachowiak

Comment 13 2008-09-12 18:11:14 PDT

bug 20821 covers the expense of adding new properties to objects as part of initialization, which primarily affects EarleyBoyer.

Cameron Zwarich (cpst)

Comment 14 2008-09-12 18:13:34 PDT

Optimization of op_eq_null and op_neq_null is bug 20823.

Cameron Zwarich (cpst)

Comment 15 2008-09-13 02:41:30 PDT

Optimization of the 'typeof' operator is bug 20827.

Cameron Zwarich (cpst)

Comment 16 2008-09-21 20:45:51 PDT

Removing the IndexToNameMap from the Arguments object was bug 20972.

Cameron Zwarich (cpst)

Comment 17 2008-09-21 20:50:23 PDT

Speeding up the 'apply' method of functions is bug 20982.

Cameron Zwarich (cpst)

Comment 18 2008-09-21 21:30:57 PDT

Inline space for extra arguments in ArgumentsData is 20983.

Cameron Zwarich (cpst)

Comment 19 2008-09-21 21:32:45 PDT

(In reply to comment #18) > Inline space for extra arguments in ArgumentsData is 20983. Oops. Inline space for extra arguments in ArgumentsData is bug 20983.

Cameron Zwarich (cpst)

Comment 20 2008-09-21 21:43:02 PDT

Converting masqueradesAsUndefined() from a virtual function to a TypeInfo flag is bug 20984.

Cameron Zwarich (cpst)

Comment 21 2008-09-22 03:56:59 PDT

Efficiently putting values in the 'callee' and 'length' properties in the Arguments constructor is bug 20989.

Cameron Zwarich (cpst)

Comment 22 2008-09-22 22:24:34 PDT

Optimizing FunctionBodyNode::ref()/deref() was bug 21019.

Cameron Zwarich (cpst)

Comment 23 2008-09-28 11:57:19 PDT

Allowing direct access to 'arguments' without using op_resolve is bug 21200.

Cameron Zwarich (cpst)

Comment 24 2008-09-28 11:58:37 PDT

Making 'arguments' not require an activation is bug 21123.

Cameron Zwarich (cpst)

Comment 25 2008-10-01 23:01:47 PDT

Devirtualization of getOwnPropertySlot() is bug 21294.

Maciej Stachowiak

Comment 26 2008-10-04 17:47:51 PDT

Created attachment 24095 [details] newer profile of v8 RayTrace benchmark

Cameron Zwarich (cpst)

Comment 27 2008-10-22 14:07:47 PDT

Fusing op_jfalse with op_eq_null and op_neq_null is bug 21803.

Cameron Zwarich (cpst)

Comment 28 2008-10-22 16:01:58 PDT

Putting properties directly in JS function creation is bug 21814.

Cameron Zwarich (cpst)

Comment 29 2008-10-24 13:32:40 PDT

Creating the JSFunction prototype property lazily is bug 21862.

Cameron Zwarich (cpst)

Comment 30 2008-10-28 22:55:35 PDT

Avoiding needless reads of temporary values in CTI code is bug 21943.

Gavin Barraclough

Comment 31 2012-03-07 00:05:10 PST

This umbrella bug is now only tracking one subtask, bug#20814. Having two bugs open to track one thing isn't really useful, so I'm going to simplify & close this master bug. *** This bug has been marked as a duplicate of bug 20814 ***

Note You need to log in before you can comment on or make changes to this bug.