Re: [boost] Re: operator implementation

From: Joel de Guzman (djowel_at_[hidden])
Date: 2002-06-28 13:35:10

Fernando Cacciola wrote:

>----- Original Message -----
>From: "David B. Held" <dheld_at_[hidden]>
>Newsgroups: gmane.comp.lib.boost.devel
>To: <boost_at_[hidden]>
>Sent: Friday, June 28, 2002 2:00 PM
>Subject: [boost] Re: operator implementation
>
>
>>"Joel de Guzman" <djowel_at_[hidden]> wrote in message
>>news:3D1C9097.5040309_at_gmx.co.uk...
>>
>>>[...]
>>>Can we see real world benchmarks? I've read about this along time
>>>ago but unfortunately I haven't seen any benchmarks yet.
>>>
>>Actually, assembly output seems like it would be most appropriate.
>>Shouldn't be too hard to demonstrate whether RVO is occurring or
>>not, either.
>>
>FWIW, BCC55 doesn't apply RVO no matter how you write the code.
>Still, if there is at least a chance that some future compiler release will
>do it, I see no harm in adopting the 'nrvo' idiom you propose.
>Probably, I wouldn't 'change' already written code, but I'll take it into
>consideration from now on.
>

What I am concerned more is the opposite: that future compilers will not
need this and will optimize appropriately with or without the 'nrvo' idiom.
Yet, if it can be shown through benchmarks or assembly code that *current*
compilers do produce better code, it might be worth it to rewrite code.

Regards,
--Joel