--- Comment #6 from steven at gcc dot gnu dot org 2009-06-13 23:45 ---
I would still like to know why these constants can't be lowered earlier.
Exposing the split-up insns may allow CSE to do a better job, etc.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40327
--- Comment #5 from rearnsha at gcc dot gnu dot org 2009-06-13 13:04
---
fixed
--
rearnsha at gcc dot gnu dot org changed:
What|Removed |Added
Status|ASSIGNE
--- Comment #4 from rearnsha at gcc dot gnu dot org 2009-06-13 12:49
---
Subject: Bug 40327
Author: rearnsha
Date: Sat Jun 13 12:49:25 2009
New Revision: 148452
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=148452
Log:
PR target/40327
* arm/constraints.md (Pa,
--- Comment #3 from rearnsha at gcc dot gnu dot org 2009-06-09 22:06
---
Working on a patch
--
rearnsha at gcc dot gnu dot org changed:
What|Removed |Added
Assig
--- Comment #2 from steven at gcc dot gnu dot org 2009-06-03 10:57 ---
FWIW it looks like the define_split at arm.c:5060 (of r147729) is triggered.
Richard E., is there a reason for lowering load-immediates so late in the
pipeline?
--
steven at gcc dot gnu dot org changed:
--- Comment #1 from steven at gcc dot gnu dot org 2009-06-03 10:36 ---
This code comes from the split2 (split insns after reload). So this is target
specific. It surprises me that non-legitimate constants are accepted before
reload on ARM (at least in thumb mode). It may be (and IMHO,