[Bug c++/12624] [3.4 Regression] [unit-at-a-time] ICE in notice_global_symbol

["wilson at specifixinc dot com" <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=12624



------- Additional Comments From wilson at specifixinc dot com  2003-10-23 21:45 -------
Subject: Re:  [3.4 Regression] [unit-at-a-time] ICE in notice_global_symbol

pinskia at gcc dot gnu dot org wrote:
> From Phil's regression hunter with that testcase: Search converges between 2003-08-18-trunk 
> (#376) and 2003-08-19-trunk (#377).

That would be Jan's patch that created the notice_global_symbol 
function.  We already knew the problem was with that code.  It shouldn't 
be used for global register variables.

I see two obvious ways to fix the problem.  We can check for the 
DECL_REGISTER bit, or we can check that DECL_RTL contains a MEM of a 
SYMBOL_REF.  I'm leaning towards the former.  That is the one I plan to 
test today with an ia64-linux bootstrap.