This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch to correct x86 FP extend splits
- To: hubicka at horac dot ta dot jcu dot cz
- Subject: Re: Patch to correct x86 FP extend splits
- From: John Wehle <john at feith dot com>
- Date: Thu, 15 Apr 1999 18:42:44 -0400 (EDT)
- Cc: egcs-patches at egcs dot cygnus dot com, zack at rabi dot columbia dot edu, rth at cygnus dot com
>> > But I think we can at least reduce it by allocating only the larger slot
>> > and use it in both cases.
>>
>> It occurred to me at the time. I don't have any real objection.
> OK. Are you interested in preparing patch for this or do you leave it up to me?
I don't plan to address this at the moment so feel free.
>> > second problem is, that there is full pattern with the clobber and second
>> > pattern w/o them doing the actual extension. The full pattern is translated to
>> > second one by split. But combine during canonization pass removes those
>> > clobbers as well, so we probably need to ensure, that second pattern will
>> > not match before reload.
>>
>> My impression is that combine will remove the clobbers if it allows this insn
>> to be combined with another insn to create a new valid insn. Are you saying
>> that combine will remove the clobbers and match the second insn? I don't
> Thats exactly what happened to me. I was also quite surprised by this fact.
> Maybe that was caused by my bad plays with the move patterns, but I don't think so.
> I think it is more safe to put the reload_completted test there.
Feel free to supply a patch for this also.
-- John
-------------------------------------------------------------------------
| Feith Systems | Voice: 1-215-646-8000 | Email: john@feith.com |
| John Wehle | Fax: 1-215-540-5495 | |
-------------------------------------------------------------------------