Bug 148559 – Add noDFG() to jsc to prevent DFG compilation of a specified function

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Mark Lam]

Just like noInline() allows us to prevent inlining of a function, noOptimize() allows us to prevent optimization of a function.

[Mark Lam]

Created attachment 260120 [details]
the fix.

[Geoffrey Garen]

Comment on attachment 260120 [details]
the fix.

View in context: https://bugs.webkit.org/attachment.cgi?id=260120&action=review

> Source/JavaScriptCore/jsc.cpp:635
> +        addFunction(vm, "noOptimize", functionNeverOptimizeFunction, 1);

Let's call this noDFG and functionNoDFG. "noOptimize" implies no baseline JIT, which is not the case.

[Saam Barati]

Comment on attachment 260120 [details]
the fix.

r=me

[Saam Barati]

(In reply to comment #2)
> Comment on attachment 260120 [details]
> the fix.
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=260120&action=review
> 
> > Source/JavaScriptCore/jsc.cpp:635
> > +        addFunction(vm, "noOptimize", functionNeverOptimizeFunction, 1);
> 
> Let's call this noDFG and functionNoDFG. "noOptimize" implies no baseline
> JIT, which is not the case.

I also like the name "dontDFGCompile"

[Mark Lam]

(In reply to comment #2)
> Comment on attachment 260120 [details]
> the fix.
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=260120&action=review
> 
> > Source/JavaScriptCore/jsc.cpp:635
> > +        addFunction(vm, "noOptimize", functionNeverOptimizeFunction, 1);
> 
> Let's call this noDFG and functionNoDFG. "noOptimize" implies no baseline
> JIT, which is not the case.

I think fpizlo has always used the term “optimize” (in our code base) to mean DFG and higher, and precludes baseline.  For example, in JITCode.h, he distinguished between isOptimizingJIT() and isBaselineCode().  And he suggested the term “noOptimize” when he asked me to implement this.

But sure, I’ll change it to noDFG().

[Mark Lam]

(In reply to comment #5)
> But sure, I’ll change it to noDFG().

I just realized that you meant for me to only change the function in jsc.cpp and not elsewhere in the patch.  Will do.

[Mark Lam]

Thanks for the reviews.  I opted to go with noDFG() instead of dontDFGCompile() because noDFG() is shorter and adequately conveys the purpose.

Landed in r189086: <http://trac.webkit.org/r189086>.