We still need to keep on saving r6 as it is used by the MacroAssembler, which we use (we get crashes in some situations otherwise). This issue was discovered because stress/regress-174044.js crashes on a raspberry pi 2 when compiled in -O2.
Created attachment 332334 [details] Patch Patch fixing the issue.
Comment on attachment 332334 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=332334&action=review > Source/JavaScriptCore/ChangeLog:8 > + We still need to keep on saving r6 as it is used by the MacroAssembler you mean r8, not r6?
Comment on attachment 332334 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=332334&action=review >> Source/JavaScriptCore/ChangeLog:8 >> + We still need to keep on saving r6 as it is used by the MacroAssembler > > you mean r8, not r6? Sorry, I meant that in addition to saving $r8, as the title suggest, we can't stop saving $r6, as I initially wrongly deducted in a first version of the patch that was creating weird crashes :). My initial wrong deduction was because the initial register used to be $r6 before #182157.
Comment on attachment 332334 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=332334&action=review >>> Source/JavaScriptCore/ChangeLog:8 >>> + We still need to keep on saving r6 as it is used by the MacroAssembler >> >> you mean r8, not r6? > > Sorry, I meant that in addition to saving $r8, as the title suggest, we can't stop saving $r6, as I initially wrongly deducted in a first version of the patch that was creating weird crashes :). My initial wrong deduction was because the initial register used to be $r6 before #182157. Probably worth rewording this to reflect what the patch is actually doing.
Created attachment 333719 [details] Patch New patch with a more explicit ChangeLog message. Sorry it took me a while to get to it.
Comment on attachment 333719 [details] Patch Clearing flags on attachment: 333719 Committed r228436: <https://trac.webkit.org/changeset/228436>
All reviewed patches have been landed. Closing bug.
<rdar://problem/37514115>