RFA: Revert revision 164552

[Jeff Law]

On 12/01/10 02:32, Bernd Schmidt wrote:
> On 11/03/2010 04:02 PM, Diego Novillo wrote:
>> On Wed, Nov 3, 2010 at 10:02, Rainer Orth<ro@cebitec.uni-bielefeld.de>  wrote:
>>> H.J.,
>>>
>>>> Revision 164552:
>>>>
>>>> http://gcc.gnu.org/ml/gcc-cvs/2010-09/msg00849.html
>>>>
>>>> which fixes:
>>>>
>>>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44374
>>>>
>>>> a mixed optimization bug, but caused many failures:
>>>>
>>>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46257
>>>>
>>>> including bootstrap failures on x86. It was reported more than a month ago
>>>> and nothing has changed.  Should it be reverted for now?
>>> I think it should: it probably also caused PR bootstrap/46018 and
>>> certainly PR rtl-optimization/46114, where Bernd indicated that he would
>>> be away for several weeks.
>> I agree.  Let's revert the patch.  Bernd should be able to figure out
>> a fix after he gets back.
> Now that there are no more vacations and other problems to stop me from
> fixing it, I propose that the patch be reapplied in the form below.
>
> Fixes:
> PR46114: Reported fixed by the patch in PR46238, which is included in
> the patch below.
I'll assume that by including the patch, you think it's the right fix (I 
marked it in my queue, but didn't look at it closely knowing your patch 
had been reverted and that it was going to need to be revisited).

> PR45865: Fixed by testing for NOTE_INSN_EPILOGUE_BEG in two places so as
> to not merge epilogues.
Seems reasonable.

>
> I've also made some changes to preserve an unrelated ifcvt fix from Eric
> that has been applied in the same area in the meantime.
Excellent.


> Bootstrapped and regression tested on i686-linux, all languages. Rainer,
> it would be helpful if you could do a Solaris test.
If there weren't any other changes and the solaris test passes, then 
it's OK with me.

jeff