Bug ID: JDK-6810777 C2Compiler crash libjvm.so:PhaseIdealLoop::build_loop_late

Type: Bug
Component: hotspot
Sub-Component: compiler
Affected Version: 5.0u9,5.0u17,5.0u19,6u12

Priority: P2
Status: Closed
Resolution: Won't Fix
OS:
linux_redhat_4.0,solaris_10,windows_2003 linux_redhat_4.0,solaris_10,windows_2003
CPU: x86,sparc

Submitted: 2009-02-26
Updated: 2010-08-06
Resolved: 2009-06-23

The customer reports hotspot errors when the C2compiler tries to compile in 64-bit java on a Windows system. Everything works fine with 32-bit java. The customer provided a testcase.
just Added ATT to this bug as well. They are seeing this with 32bit Solaris 10 SPARC and using 1.5.0_17. Will this be backported to 1.5.0_xx by default?

All their cores are located under /net/cores.central/cores/70807100

EVALUATION As this is demonstrated not to happen in 6u14 and onwards, am closing will not fix (unless or until there is an escalated issue in an ealier release) as while avoided, it is not a true duplicate of other fixes.

23-06-2009

EVALUATION This appears to a general instance of the long standing issue we have with ConvI2L with type information on it going to top when it's cloned to someplace where that isn't true. 22 ConI === 0 [[ 4114 2393 6416 41 6266 6167 4158 6177 243 1805 84 102 102 8705 6065 205 4158 136 136 157 2069 2211 4158 8731 6038 6032 6125 2562 3243 3187 Run#2051 180 180 180 180 180 180 2119 4812 2335 253 227 243 5734 4070 2153 5644 4158 323 6316 1825 5526 4181 6167 2819 458 489 523 8668 1511 1767 4135 4114 6376 6246 6203 2642 673 704 738 8583 1480 1756 2245 1545 5301 773 6354 4711 2681 3378 894 925 959 8411 2536 1722 1835 1584 5113 4495 1071 1102 1136 8670 1334 1691 1842 1556 1599 4935 1258 1289 1323 7465 7490 8497 7576 8325 7662 Run#2052 8671 7748 7809 7834 7895 7920 7981 8006 8067 8841 8840 8839 8838 8837 8836 8835 8735 8733 8732 8727 8714 8708 8843 ]] #int:0 8733 ConvI2L === _ 22 [[ 8716 ]] #long:1..maxint:www !orig=6253 !jvms: Stack::hasIllegalNearOperatorArguments @ bci:66Run#2053 LRParser::parse @ bci:87 This causes a phi to collapse and the associated control flow to go away leaving a malformed graph. I'm not sure what the proper fix is for this. The most straightforward fix would likely have performance impact that would have to be evaluated first. They may be some bandaids that could be applied. I've worked up one which removes the crash but it may be too unsavory. This is no longer reproducible in 1.7 but I think that's because other changes keep it from happening, not because of any true fix. I can reproduce it using jdk7-b38. Any fix would need to go into 1.7 as well.

04-03-2009

WORK AROUND Two methods excluded from compilation: $ cat .hotspot_compiler exclude com/recommind/prediction/parser/LRParser.parse exclude com/recommind/prediction/parser/Stack.hasIllegalNearOperatorArguments $

27-02-2009