------- Comment #8 from wilson at tuliptree dot org 2005-10-06 21:21 -------
Subject: Re: [4.1 Regression] ICE in extract_insn while
compiling libgfortran
On Thu, 2005-10-06 at 14:02, sje at cup dot hp dot com wrote:
> ------- Comment #7 from sje at cup dot hp dot com 2005-10-06 21:02 -------
> I haven't done a full build/test but I did verify that changing operand 0 of
> movti_internal from general_operand to destination_operand fixed this failure.
> I was wondering if operand 1 should be changed from general_operand to
> move_operand.
move_operand is general_operand minus some types of operands containing
symbol_refs that aren't valid. Since FP operands will never contain
symbol addresses, in theory, the use of move_operand instead of
general_operand is extra unnecessary overhead.
By the way, I added fr_spill to the list of patterns that I am fixing.
I am convinced I have the complete list now.
I have started a full build and test, but I have two patches to test and
one machine to test them on, so it will take me a while.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24193
