--- Comment #5 from rearnsha at gcc dot gnu dot org 2009-12-16 11:11
---
Regrename seems to be starting the scan of bb3 (from the reduced testcase) from
somewhere in the middle of the block rather than at the beginning (it's missing
at least 2 insns in the reg chain information:
Basic
--- Comment #6 from bernds_cb1 at t-online dot de 2009-12-16 19:57 ---
I'm having a hard time reproducing this. How do you folks configure your
compilers?
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42372
--- Comment #7 from mikpe at it dot uu dot se 2009-12-16 21:27 ---
I've done a binary search which identified 154688 as the revision which caused
the problem to appear.
To answer Bernd's question, I used an arm cross configured with:
--target=armv5tel-unknown-linux-gnueabi
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Summary|[4.5 regression] Regrename |[4.5 Regression] Regrename
|reuses non-dead
--- Comment #8 from bernds_cb1 at t-online dot de 2009-12-16 23:17 ---
This seems to be a bug in arm.md. Operand 0 in tls_load_dot_plus_eight is
marked in-out, but there's no matching match_dup to show the input. Adding
that to the unspec seems to make the problem go away. Richard
