[Bug optimization/9812] [3.3/3.4 regression] [m68k] ICE in extract_insn, at recog.c:2148

["wilson at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9812



------- Additional Comments From wilson at gcc dot gnu dot org  2003-06-27 07:07 -------
I built 3.0.4 and 2.95 m68k-linux cross compilers, and both failed for the
testcase same as 3.3.  This does not appear to be a gcc regression to me.

The problem stems from a patch I added in 1995, which makes
LEGITIMATE_PIC_OPERAND_P reject CONST_DOUBLEs that are in the constant pool.  It
isn't clear if that patch is still useful.  No other target does this.  I am
checking to see if I can reproduce the original problem.  Changing this may be
risky though.

I suspect that this check can be made conditional on reload_completed and/or
reload_in_progress.  I haven't looked into this yet.