Bug 115215 - Use push and pop for iOS math function thunks
Summary: Use push and pop for iOS math function thunks
Status: RESOLVED FIXED
Alias: None
Product: WebKit
Classification: Unclassified
Component: New Bugs (show other bugs)
Version: 528+ (Nightly build)
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Benjamin Poulain
URL:
Keywords:
Depends on: 115144
Blocks:
  Show dependency tree / graph
 
Reported: 2013-04-25 17:58 PDT by Benjamin Poulain
Modified: 2013-04-29 16:53 PDT (History)
2 users (show)

See Also:

Attachments
Patch (1.84 KB, patch)
2013-04-25 18:00 PDT, Benjamin Poulain 		no flags Details | Formatted Diff | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Benjamin Poulain 2013-04-25 17:58:07 PDT 
Use push and pop for iOS math function thunks
Comment 1 Benjamin Poulain 2013-04-25 18:00:36 PDT 
Created attachment 199770 [details]
Patch
Comment 2 Filip Pizlo 2013-04-27 15:17:00 PDT 
Comment on attachment 199770 [details]
Patch

Did you test that calling printf() or similar from jsRound() for example, and then calling Math.round(), works without crashing?
Comment 3 Benjamin Poulain 2013-04-27 15:20:10 PDT 
(In reply to comment #2)
> (From update of attachment 199770 [details])
> Did you test that calling printf() or similar from jsRound() for example, and then calling Math.round(), works without crashing?

I'll try that on Monday.

I am pretty sure of the alignment on 4 bytes because I already accidentally ran into that when I wrote TransformationMatrix's optimizations. I assumed the stack would be at least aligned to 8 bytes when writing the VFP code, and that happily crashed in the tests.
Comment 4 Filip Pizlo 2013-04-27 15:21:08 PDT 
Comment on attachment 199770 [details]
Patch

Would be good to do that test before landing.  Otherwise, r=me.
Comment 5 Benjamin Poulain 2013-04-29 16:49:50 PDT 
(In reply to comment #4)
> (From update of attachment 199770 [details])
> Would be good to do that test before landing.  Otherwise, r=me.

I just tried on an Apple A5, no problem with printf called from jsRound.
Comment 6 Benjamin Poulain 2013-04-29 16:53:27 PDT 
Comment on attachment 199770 [details]
Patch

Clearing flags on attachment: 199770

Committed r149331: <http://trac.webkit.org/changeset/149331>
Comment 7 Benjamin Poulain 2013-04-29 16:53:29 PDT 
All reviewed patches have been landed.  Closing bug.