Bug 227798 – display-profiler-output should be able to print disassembly for the FTL

• Component: New Bugs
• Status: RESOLVED FIXED

[Keith Miller]

display-profiler-output should be able to print disassembly for the FTL

[Keith Miller]

Created attachment 433135 [details]
Patch

[Robin Morisset]

Comment on attachment 433135 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=433135&action=review

> Source/JavaScriptCore/ChangeLog:8
> +        Right now running JSC with the bytecode profiler will will not print

typo: will will

> Source/JavaScriptCore/ChangeLog:11
> +        before the DFG graph despite being printed after before and being added

typo: after before

> Source/JavaScriptCore/ChangeLog:12
> +        to the Profiler::Compilation after the graph. Lastly, the disassmebly

typo: disassmebly

> Source/JavaScriptCore/ChangeLog:13
> +        is the same text as Options::dumpDisassebly so it doesn't include execution

typo: Disassebly

[Keith Miller]

Created attachment 433140 [details]
Patch

[Keith Miller]

Created attachment 433141 [details]
Patch

[Keith Miller]

Created attachment 433142 [details]
Patch

[Saam Barati]

Comment on attachment 433142 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=433142&action=review

> Source/JavaScriptCore/ChangeLog:26
> +        94908                 979/4095/94908/0             Air               Move -824(%rax), %rax, b@402
> +        94908                 979/4095/94908/0             asm                   0x3bf95fa085df: mov -0x338(%rax), %rax
> +        94908                 979/4095/94908/0             b3            Void b@403 = Patchpoint($4396662888(b@90):ColdAny, generator = 0x1066c02e0, earlyClobbered = [], lateClobbered = [], usedRegisters = [%rax, %rbx], resultConstraints = WarmAny, WritesLocalState|Reads:Top, D@65)
> +        94908                 979/4095/94908/0             Air               Patch &Patchpoint0, $0x1060fc068, b@403

It's very deceptive to show DFG node execution counts for B3/Air code. I'd recommend just not showing that info for B3/Air, but still showing it for the parent DFG node.

[Saam Barati]

(In reply to Saam Barati from comment #6)
> Comment on attachment 433142 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=433142&action=review
> 
> > Source/JavaScriptCore/ChangeLog:26
> > +        94908                 979/4095/94908/0             Air               Move -824(%rax), %rax, b@402
> > +        94908                 979/4095/94908/0             asm                   0x3bf95fa085df: mov -0x338(%rax), %rax
> > +        94908                 979/4095/94908/0             b3            Void b@403 = Patchpoint($4396662888(b@90):ColdAny, generator = 0x1066c02e0, earlyClobbered = [], lateClobbered = [], usedRegisters = [%rax, %rbx], resultConstraints = WarmAny, WritesLocalState|Reads:Top, D@65)
> > +        94908                 979/4095/94908/0             Air               Patch &Patchpoint0, $0x1060fc068, b@403
> 
> It's very deceptive to show DFG node execution counts for B3/Air code. I'd
> recommend just not showing that info for B3/Air, but still showing it for
> the parent DFG node.

My reasoning is this:

Many DFG nodes lower to a lot of control flow in B3. Better to not incorrectly say all that control flow executes the same number of times.

[Keith Miller]

Created attachment 433147 [details]
Patch

[Keith Miller]

Landed in r279740.