Re: [PATCH] rtl-optimization: ppc backend generates unnecessary signed extension.

[Jeff Law <jeffreyalaw@gmail.com>]

On 3/30/23 18:01, Hans-Peter Nilsson wrote:
> On Fri, 24 Mar 2023, Peter Bergner via Gcc-patches wrote:
> 
>> On 3/23/23 6:12 PM, Jeff Law via Gcc-patches wrote:
>>>>>> Is there a reason why REE cannot see that our (reg:QI 4) is a param register
>>>>>> and thus due to our ABI, already correctly sign/zero extended?
>>>>>
>>>>> I don't think REE has ever considered exploiting ABI constraints. Handling
>>>>> that might be a notable improvement on various targets.  It'd be a great
>>>>> place to do some experimentation.
>>>>
>>>> Ok, so sounds like a good follow-on project after this patch is reviewed
>>>> and committed (stage1).  Thanks for your input!
>>>
>>> Agreed.  I suspect that risc-v will benefit from such work as well.
>>> With that in mind, if y'all start poking at this, please loop in Raphael
>>> (on cc) who's expressed an interest in this space.
>>
>> Will do.  I suspect that it'll be best to come up with some generic interface
>> using target hooks like "param regs are sign/zero extended" or "call return
>> values are sign/zero extended", etc. that targets can conditionally opt into
>> depending on their ABI that is in effect.
> 
> Pardon the arm-chair development mode but it sounds like
> re-inventing the TARGET_PROMOTE_* hooks...
> 
> Maybe just hook up TARGET_PROMOTE_FUNCTION_MODE to ree.c (as
> "you" already already define it for "rs6000")?
> 
That's what we touched on up-thread.  Ideally we want to wire up REE to 
utilize the existing mechanisms to specify ABI constraints so that we 
don't have to write all those macros again.

jeff