Re: [boost] [Endian] Performance

Subject: Re: [boost] [Endian] Performance
From: Beman Dawes (bdawes_at_[hidden])
Date: 2011-09-13 08:15:33

On Mon, Sep 12, 2011 at 8:43 PM, Adder <adder.thief_at_[hidden]> wrote:
>> Then it's a bug that should be fixed in the compiler.
>> (But then it could also be that your benchmark is imperfect)
>
> Perhaps it is not the compilers who are failing Boost, but the other way around.
>
> For GCC for x86, we can equal and surpass the speed-ups obtained with
> compiler intrinsics even without those intrinsics:
>
>  http://adder.iworks.ro/Boost/RawMemory/#Update-01
>
> Can we not achieve this for other architectures too, in order to
> support programmers ?

Yes; VC++ 2010 (and probably earlier versions) generated better,
faster, code for just plain C++ in my timing tests of some conversion
functions than for intrinsics. And this was for the most important use
case, not just some obscure corner case.

I'm assuming that when folks ask for intrinsics, they really mean high speed.

--Beman