[Webkit-unassigned] [Bug 153091] New: DFG's Int52Rep node is adequate to convert doubles to Int52.
bugzilla-daemon at webkit.org
bugzilla-daemon at webkit.org
Wed Jan 13 22:47:48 PST 2016
https://bugs.webkit.org/show_bug.cgi?id=153091
Bug ID: 153091
Summary: DFG's Int52Rep node is adequate to convert doubles to
Int52.
Classification: Unclassified
Product: WebKit
Version: WebKit Local Build
Hardware: Unspecified
OS: Unspecified
Status: NEW
Severity: Normal
Priority: P2
Component: JavaScriptCore
Assignee: webkit-unassigned at lists.webkit.org
Reporter: mark.lam at apple.com
With the patch for https://bugs.webkit.org/show_bug.cgi?id=153019, we can see the following IR for a function AESEncryptCtr#Dp5nqr:
483:<!3:loc24> ArithDiv(DoubleRep:@1009<Double>, DoubleRep:@1010<Double>, Double|MustGen|PureInt|MayHaveNonIntResult|UseAsInt, Int52, NotSet, Exits, bc#598)
484:<!0:-> MovHint(DoubleRep:@483<Double>, MustGen, loc26, W:SideState, ClobbersExit, bc#598, ExitInvalid)
486:< 1:loc28> JSConstant(JS|UseAsInt, Nonboolint32, Int32: 8, bc#603)
487:<!2:loc28> ArithMul(Int32:@473, Int32:@486, Number|MustGen|PureInt|UseAsInt, Int32, Unchecked, Exits, bc#603)
488:<!0:-> MovHint(Untyped:@487, MustGen, loc27, W:SideState, ClobbersExit, bc#603, ExitInvalid)
1012:< 1:loc26> Int52Rep(Check:DoubleRepMachineInt:@483<Double>, Int52|PureInt, Boolint32Nonboolint32Int52, Exits, bc#608)
983:< 1:loc26> ValueToInt32(Int52Rep:@1012<Int52>, Int32|PureInt, Int32, Exits, bc#608)
490:< 2:loc28> BitURShift(KnownInt32:@983, Int32:@487, Int32|UseAsOther, Int32, Exits, bc#608)
1289:<!0:-> Phantom(Check:DoubleRep:@483<Double>, MustGen, bc#608)
491:<!0:-> MovHint(Untyped:@490, MustGen, loc26, W:SideState, ClobbersExit, bc#608, ExitInvalid)
493:< 2:loc28> UInt32ToNumber(Int32:@490, Number|UseAsOther, Int32, CheckOverflow, Exits, bc#612)
AESEncryptCtr#Dp5nqr is OSR exiting at the Int52Rep node (@1012) because it fails to convert the double result of the div (@483) to an Int52. The double value is between 0 and 1. The value would have been used by the BitURShift later. The implementation expects to just truncate the fraction and convert that number to an int 0.
However, the implementation of DFG Int52Rep for converting doubles calls tryConvertToInt52() which will only convert the double to an Int52 if and only if the value it contains is already an Int52 value expressed in double form. It will not truncate the fraction part as we expect.
As a result, the Int52Rep node will trigger an OSR exit (which we don't want).
--
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20160114/92ffcfc0/attachment.html>
More information about the webkit-unassigned
mailing list