Bug 127086 – CStack: Fix 32-bit C loop LLINT.

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

[Mark Lam]

Patch coming.

[Mark Lam]

Created attachment 221343 [details]
the patch.

This patch builds and passes run-javascriptcore-tests --debug --32-bit --cloop --no-jsc-stress.  This is the minimum patch that will build and run the 32-bit C loop LLINT.  Since the C loop is based on ARM, this patch also fixes part of the ARMv7 LLINT implementation.  There is still more to do for ARM (e.g. alignment issues) but I didn’t want to overlap with the work in Michael’s patch.  There is still necessarily some minimal overlap with Michael’s patch.  I’ll rebase this patch after Michael’s 32-bit patch is finalized.

[Oliver Hunt]

Comment on attachment 221343 [details]
the patch.

View in context: https://bugs.webkit.org/attachment.cgi?id=221343&action=review

> Source/JavaScriptCore/llint/LowLevelInterpreter32_64.asm:315
> +    addp 8, sp

prior code didn't increment sp prior to the call, should this be in the CLOOP block?

[Mark Lam]

Comment on attachment 221343 [details]
the patch.

View in context: https://bugs.webkit.org/attachment.cgi?id=221343&action=review

>> Source/JavaScriptCore/llint/LowLevelInterpreter32_64.asm:315
>> +    addp 8, sp
> 
> prior code didn't increment sp prior to the call, should this be in the CLOOP block?

No, the sp adjustment (before and after the call) is needed for all 32-bit ports, and Michael’s patch will make that change.  I have to make the same change here or the C loop code won’t work.

[Geoffrey Garen]

Comment on attachment 221343 [details]
the patch.

r=me

[Mark Lam]

Landed in r162248 on the jsCStack branch: <http://trac.webkit.org/r162248>.