[Patch:H8/300] Absolute address mode support for bit_operand predicate
Jeff Law
law@redhat.com
Fri Apr 16 22:56:00 GMT 2010
On 04/16/10 09:25, Jeff Law wrote:
> On 04/16/10 01:19, Kaushik Phatak wrote:
>>> 2010-04-12 Kaushik Phatak<kaushik.phatak@kpitcummins.com>
>>>
>>> * config/h8300/predicate.md (bit_operand): Allow immediate
>>> values that
>>> satisfy 'U' constraint.
>>>
>>>> OK. Please install.
>> Thanks for approval.
>> I do not have check-in rights. Could you apply this to 4.6?
> OK. I installed it for you.
>
>> Also, could you please apply the following approved patch to 4.6,
>>
>> http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00373.html
> I was looking at it yesterday; I'm investigating a codesize regression
> with that patch. I'll post to the list once I've got a handle on what
> the problem is.
OK. The problem is rearranging alternatives in the logical patterns
(and, ior, xor) to prefer the bit variants inhibits the elimination of
tst insns as the bit versions do not set the condition codes and the
condition codes need to be set so that redundant tst insns can be
detected and eliminated.
ISTM that we want to avoid the bit variants when they directly feed a
tst insn. I think the attribute mechanisms are powerful enough to
describe this, but I haven't tried it myself.
For now, I'm going to install the new patterns, but leave the existing
and/ior/xor patterns alone until we can resolve this issue.
Jeff
More information about the Gcc-patches
mailing list