RFC: PR rtl-optimization/43360: [4.3/4.4/4.5 Regression] possible wrong code bug

[H.J. Lu]

On Tue, Mar 16, 2010 at 12:28 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> On Tue, Mar 16, 2010 at 12:05:41PM -0700, H.J. Lu wrote:
>> > I'd say preventing the insn from being optimized is not desirable, instead
>> > if the REG_EQUAL is not invariant, the note should be dropped.
>> >
>> >        Jakub
>> >
>>
>> Here is the updated patch.  OK for trunk/4.4/4.3?
>
> Is every invariant insn for which find_invariant_insn succeeds actually
> moved?  I very much doubt so.  Therefore, it would be IMHO better to
> run this check in say create_new_invariant, set a new bool field
> in struct invariant (say drop_reg_equal) and only if/when it is actually
> moved drop the invariant if this field is true?
>

I am not familiar with the code. Would you mind taking it over
and run with it?

Thanks.


-- 
H.J.