Re: renames usage

["Randy Brukardt" <randy@rrsoftware.com>]

"Jeffrey R. Carter" <spam.jrcarter.not@spam.not.acm.org> wrote in message 
news:rss4sk$oqt$1@dont-email.me...
> On 1/3/21 4:19 AM, Randy Brukardt wrote:
>>
>> In this case (a scalar return), the "copy" is a register, and it would be
>> hard (and pointless) to eliminate that.
>
> Thanks for the clarification. I'm not sure I understand what you mean by 
> "the 'copy' is a register". Do you mean
>
> 1. The return value has to be copied into a register
>
> or
>
> 2. The return value is in a register and has to be copied elsewhere
>
> ?

The latter: on every machine on which I'm familar, scalar function results 
are returned in a register. That serves as the "return object", and there's 
not really any way to eliminate that short of inlining the entire call. To 
"rename" the result, one has to copy the register somewhere (unless of 
course the renaming lives a short time with no subprogram calls), and that 
is effectively the same as the constant object declaration.

OTOH, a composite object is typically returned with a reference in a 
register to memory allocated somewhere else (it might be passed in, it might 
be on a "secondary stack", it might be allocated, etc.), and one can rename 
rather than copy that memory.

Of course, if the type requires "build-in-place", then a renames and an 
object are essentially the same again, as no temporary is allowed and the 
memory has to be passed into the function; either for the constant or for 
the renamed temporary.

Which mainly goes to show that nothing is simple... :-)

                                 Randy.