The register allocator allocates two overlapping lifetimes for this bug. This
appears when in BB#009, phi node 592 is allocated to FR8/FR8H, which is used
in BB#013. One of the inputs is 578 SpillNode, which is also allocated to
FR8/8H. In BB#013, we have a reader of 578. So 578 is an input to the phi
node, and is consumed in the loop.

This appears to be a register allocator bug at first look.

The actual assert is
#
# HotSpot Virtual Machine Error, assertion failure
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
#
# Java VM: Java HotSpot(TM) 64-Bit Server VM (robertg_11_54__10_12_2002-debug mixed mode)
#
# assert(!_reg_node[reg_lo] || edge_from_to(_reg_node[reg_lo],def), "before block local scheduling")
#
# Error ID: /.automount/altair/root/export/space10/robertg/sgoldman/build/linux/../../src/share/vm/opto/output.cpp, 1995
#
# Problematic Thread: prio=1 tid=0x60000000000abc78 nid=0x1d4b runnable
#


To reproduce, use the included file rerun.csh

this occurs w/ CompileTheWorld, not running benchmarks.

Note that my version of "output.cpp" has tracing to dump the IR with registers at various places.