[Webkit-unassigned] [Bug 74498] ensure consistent heuristics

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=74498


Filip Pizlo <fpizlo at apple.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #119203|review?                     |review-
               Flag|                            |




--- Comment #4 from Filip Pizlo <fpizlo at apple.com>  2011-12-21 14:30:32 PST ---
(From update of attachment 119203)
View in context: https://bugs.webkit.org/attachment.cgi?id=119203&action=review

> Source/JavaScriptCore/runtime/Options.cpp:55
> +    executionCounterValueForOptimizeNextInvocation = 0; // Not configurable.

That's not right. We want to be able to change this.

> Source/JavaScriptCore/runtime/Options.cpp:56
> +    CLAMP(executionCounterValueForOptimizeSoon, std::numeric_limits<int32_t>::min(), executionCounterValueForOptimizeNextInvocation - 1);

That's probably a bit too strict.  Could drop the - 1.

> Source/JavaScriptCore/runtime/Options.cpp:65
> +    CLAMP(desiredProfileFullnessRate, 0, desiredProfileLivenessRate);

Why?  These two nunbers are unrelated.

> Source/JavaScriptCore/runtime/Options.cpp:66
> +    CLAMP(numberOfGCMarkers, 1, 8);

Not sure about this. On Mac the default is to top out at 4 markers *by default*, but we want to enable more markers. What's wrong with having, say, 24 markers?  My machine has 24 logical CPUs.

> Source/JavaScriptCore/runtime/Options.cpp:82
> +    ASSERT((static_cast<int64_t>(executionCounterValueForOptimizeAfterLongWarmUp) << reoptimizationRetryCounterMax) >= static_cast<int64_t>(std::numeric_limits<int32_t>::min()));

I think that CheckedArithmetic.h might be better than this stuff.

> Source/JavaScriptCore/runtime/Options.h:68
> +    m(unsigned, reoptimizationRetryCounterMax,         30) \

Why set this to 30 and then recompute it anyway?

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