Bug 210721 – [JSC] LLInt slow path call should not have third argument

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Yusuke Suzuki]

[JSC] LLInt slow path call should not have third argument

[Yusuke Suzuki]

Created attachment 396908 [details]
Patch

[Yusuke Suzuki]

Created attachment 396909 [details]
Patch

[Mark Lam]

Comment on attachment 396909 [details]
Patch

r=me

[Saam Barati]

Comment on attachment 396909 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=396909&action=review

Wasn’t the goal to speed this up though?

> Source/JavaScriptCore/ChangeLog:8
> +        LLInt callSlowPath does not work with third argument in Windows, CLoop etc. LLInt slow-path should not take third argument,

Why doesn’t it work?

[Mark Lam]

(In reply to Saam Barati from comment #4)
> Comment on attachment 396909 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=396909&action=review
> 
> Wasn’t the goal to speed this up though?
> 
> > Source/JavaScriptCore/ChangeLog:8
> > +        LLInt callSlowPath does not work with third argument in Windows, CLoop etc. LLInt slow-path should not take third argument,
> 
> Why doesn’t it work?

Because the CLoop is able to call arbitrary native functions.  It needs to be taught how call a native function with different arguments.

As for Windows, I suspect it is just a matter teaching it how to call with a 3rd argument due to ABI differences.

[Mark Lam]

(In reply to Mark Lam from comment #5)
> (In reply to Saam Barati from comment #4)
> > Comment on attachment 396909 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=396909&action=review
> > 
> > Wasn’t the goal to speed this up though?
> > 
> > > Source/JavaScriptCore/ChangeLog:8
> > > +        LLInt callSlowPath does not work with third argument in Windows, CLoop etc. LLInt slow-path should not take third argument,
> > 
> > Why doesn’t it work?
> 
> Because the CLoop is able to call arbitrary native functions.

typo: I meant "is NOT able to"

[Yusuke Suzuki]

I think this change does not introduce speed down in LLInt since LLInt is anyway loading metadata in asm side.
For baseline, it would be possible that this causes speed down. I think in this case, the right thing is introducing JIT operation instead of sharing this with LLInt.
I'll create a patch for Baseline after landing this to fix Windows.

[Yusuke Suzuki]

Committed r260344: <https://trac.webkit.org/changeset/260344>

[Radar WebKit Bug Importer]

<rdar://problem/62033808>