During my work on compressed oops I found this problem which is
also reported by one customer who also works with hotspot sources:

gamma -XX:CICompilerCount=8 -XX:+UseCompressedOops -XX:-CheckCompressedOops -Xss4m -Xverify:all -XX:-ShowMessageBoxOnError -XX:+CompileTheWorld -XX:CompileTheWorldStartAt=3962 -Xbootclasspath/p:/net/vmsqe.sfbay/export/backup/testbase/CompileTheWorld/jarfiles/weblogicbig.jar

CompileTheWorld (5243) : utils/dbping
CompileTheWorld (5244) : utils/dbserializec
# To suppress the following error report, specify this argument
# after -XX: or in .hotspotrc:  SuppressErrorAt=/reg_split.cpp:1221
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  Internal Error (/net/jaberwocky/export/home2/work2/hg/6706829/src/share/vm/opto/reg_split.cpp:1221), pid=25949, tid=20
#  Error: assert(def,"must have reaching def")
#
Posted Date : 2008-09-10 17:03:37.0

N/A

Register Allocator does not process a data phi with one unique input correctly.
In PhaseChaitin::Split() when it see all predecessors have the same input
it does not look for existing phi. As result the live range of this phi
is not updated and the new live range is not propagated to successors.
So the new live range is lost on this path.
Posted Date : 2008-09-10 17:03:37.0

http://hg.openjdk.java.net/jdk7/hotspot-comp/hotspot/rev/1c6e3bfb543a
Posted Date : 2008-09-11 01:08:52.0