This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Bug in reload_cse_move2add()
- To: rearnsha at arm dot com
- Subject: Re: Bug in reload_cse_move2add()
- From: Alexandre Oliva <aoliva at redhat dot com>
- Date: 17 Jan 2001 14:01:03 -0200
- Cc: Joern Rennecke <amylaar at redhat dot com>, rearnsha at buzzard dot freeserve dot co dot uk (Richard Earnshaw), gcc-patches at gcc dot gnu dot org, law at redhat dot com
- Organization: GCC Team, Red Hat
- References: <200101171515.PAA27358@cam-mail2.cambridge.arm.com>
On Jan 17, 2001, Richard Earnshaw <rearnsha@arm.com> wrote:
>> On Jan 17, 2001, Joern Rennecke <amylaar@redhat.com> wrote:
>>
>> > You ignore the problem of multi-hard-register stores.
>>
>> Do I? It seems to me that it should work just fine. Not only the
>> luid is used, but also the reg number. So multiple sets in the same
>> insn should work. What am I missing?
> Well the good news is that arm-netbsd has bootstrapped with your latest
> version.
Cool! Ok to install it?
> However, what about the conditional execution issue:
> note_stores just strips the cond_exec part off, so we may falsely set a
> base if we do not handle it explicitly ourselves.
Yep, this is definitely wrong. We should probably pass the cond_exec
down in `x'. But that's a different patch :-)
> I think we could handle it by manually striping that part off and
> passing it as the data operand to move2add_note_stores; then, if
> data was non-null, we would always set the register as unknown if it
> was modified.
I think note_stores is used in more places, and data is to be passed
to fun() exactly as passed to note_stores().
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist *Please* write to mailing lists, not to me