PATCH: PR rtl-optimization/50696: [x32] Unnecessary lea

[H.J. Lu]

On Thu, Oct 13, 2011 at 10:21 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
> On Thu, Oct 13, 2011 at 19:19, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Thu, Oct 13, 2011 at 10:01 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
>>> On 10/13/2011 06:35 PM, Richard Kenner wrote:
>>>>>
>>>>> It never calls make_extraction.  There are several cases handled
>>>>> for AND operation. But
>>>>>
>>>>> (and:DI (plus:DI (subreg:DI (mult:SI (reg/v:SI 85 [ i ])
>>>>>                (const_int 4 [0x4])) 0)
>>>>>        (subreg:DI (reg:SI 106) 0))
>>>>>    (const_int 4294967292 [0xfffffffc]))
>>>>>
>>>>> isn't one of them.
>>>>
>>>> Yes, clearly.  Otherwise it would work!  The correct fix for this problem
>>>> is to make it to do that.  That's where this needs to be fixed: in
>>>> make_compound_operation.
>>>
>>> An and:DI is cheaper than a zero_extend:DI of an and:SI.  So GCC is correct
>>> in not doing this transformation.  I think adding a case to
>>> make_compound_operation that simply undoes the transformation (without
>>> calling make_extraction) is fine if you guard it with if (in_code == MEM).
>>>
>>
>> We first expand zero_extend:DI address to and:DI and then try
>> to restore zero_extend:DI.   Why do we do this transformation
>> to begin with?
>
> Because outside of a MEM it may be beneficial _not_ to restore
> zero_extend:DI in this case (depending on rtx_costs).
>

Why do we do it for MEM then?

-- 
H.J.