Bug 71873 – Better abstract 'abs' operation through the MacroAssembler.

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Gavin Barraclough]

Currently the x86 specific instruction sequence to perform a double abs is duplicated throughout the JITs / thunk generators.

[Gavin Barraclough]

Created attachment 114190 [details]
Fix

[WebKit Review Bot]

Attachment 114190 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/ChangeLog', u'Source..." exit_code: 1

Source/JavaScriptCore/assembler/X86Assembler.h:1571:  andpd_rr is incorrectly named. Don't use underscores in your identifier names.  [readability/naming] [4]
Source/JavaScriptCore/assembler/X86Assembler.h:1577:  andpd_mr is incorrectly named. Don't use underscores in your identifier names.  [readability/naming] [4]
Source/JavaScriptCore/assembler/X86Assembler.h:1584:  andpd_mr is incorrectly named. Don't use underscores in your identifier names.  [readability/naming] [4]
Total errors found: 3 in 14 files


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

[Gavin Barraclough]

Created attachment 114194 [details]
Fixed typo in changelog.

[Geoffrey Garen]

Comment on attachment 114194 [details]
Fixed typo in changelog.

View in context: https://bugs.webkit.org/attachment.cgi?id=114194&action=review

r=me

> Source/JavaScriptCore/runtime/JSGlobalData.cpp:92
> +const uint64_t MacroAssemblerX86Common::s_maskSignBit = 0x7FFFFFFFFFFFFFFFull;

I think I saw this line of code on Reddit once.

FFFFFFFFFFU!!

[Gavin Barraclough]

Comment on attachment 114194 [details]
Fixed typo in changelog.

committing by hand!

[Gavin Barraclough]

Fixed in r99647.

[Csaba Osztrogonác]

(In reply to comment #6)
> Fixed in r99647.

It broke jsc and layout tests on all bot.

[Csaba Osztrogonác]

(In reply to comment #6)
> Fixed in r99647.

I tried to ping you on #webkit, but I assume you aren't in the office now.

This patch broke almost all jsc tests and layout tests on SL, GTK and Qt, so
I had to roll it out to recover the test coverage: http://trac.webkit.org/changeset/99652 (when nobody was on board to fix it quickly)

[Gavin Barraclough]

(In reply to comment #8)
> (In reply to comment #6)
> > Fixed in r99647.
> 
> I tried to ping you on #webkit, but I assume you aren't in the office now.
> 
> This patch broke almost all jsc tests and layout tests on SL, GTK and Qt, so
> I had to roll it out to recover the test coverage: http://trac.webkit.org/changeset/99652 (when nobody was on board to fix it quickly)

Hey Ossy,

I just saw the rollout, I'm still a little confused – the jsc tests all pass on my machine, I'm just re-running the webkit ones now.  Still, clearly looks like something bad happened on the bots, and understood, we can't leave them broken.

G.

[Gavin Barraclough]

Still not seeing any crashes on my machine.  I'm testing on Lion, rather than Snow Leopard, but not obvious what wound be different.  Hmmmm.

[Tom Schuster]

We have code in SpiderMonkey for absDouble, in think you could use that. We also have double -> int32 conversation.

[Gavin Barraclough]

The important parts of this have been re-landed.