[Webkit-unassigned] [Bug 123615] [Win] Enable DFG JIT.

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=123615





--- Comment #34 from peavo at outlook.com  2014-07-07 11:09:41 PST ---
(In reply to comment #32)
> (From update of attachment 234363 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=234363&action=review
>

Thanks for looking into this :)

> This is exciting!  Somebody from the JSC team will have to review it.
> 
> > Source/JavaScriptCore/assembler/MacroAssemblerX86_64.h:162
> > +        // Note: this implementation supports up to 3 parameters.
> 
> Are 3 parameters always going to be enough?
> 

No, there is no guarantee for that.
Currently, there is only one caller (indirectly) of this method (JIT::callOperation(Sprt_JITOperation_EZ operation, int32_t op)),
which uses 2 parameters registers, so there is made room for one more parameter.
On the other hand, we probably don't want to add support for several more parameters than needed for performance reasons.

> > Source/JavaScriptCore/assembler/MacroAssemblerX86_64.h:164
> > +        // Shift the parameters to the right.
> 
> Is there a way to make the parameters be in the right place without shifting them?
> 

Good point, I believe that would require us to add a new method similar to

setupArgumentsWithExecState(TrustedImm32(op));

which skips using the first parameter register (which is needed for the stack pointer to the return value).

Maybe that is the best way to go?

> > Source/WTF/wtf/Platform.h:699
> > +/* Enable the DFG JIT on X86 and X86_64.  Only tested on Mac, GNU/Linux, FreeBSD, and Windows. */
> 
> Couldn't we remove the Only tested... line entirely?

Fixed in latest patch.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.