Re: Re(2): AltiVec optimization in Cocoa?
Re: Re(2): AltiVec optimization in Cocoa?
- Subject: Re: Re(2): AltiVec optimization in Cocoa?
- From: David Remahl <email@hidden>
- Date: Sun, 19 May 2002 22:04:27 +0200
>
Hi David,
>
>
On Sun, 19 May, 2002, David Remahl <email@hidden> wrote:
>
>
> It will work just as well, but you will have to provide alternative
>
> functions for processors without the altivec extensions. There are examples
>
> of how to check for the presence of altivec using the Gestalt manager and
>
> using bsd calls.
>
>
>
> Just enabling -faltivec won't do anything to the generated code. Any
>
> optimized code in the Cocoa frameworks will be used just as well.
>
>
If the -faltivec doesn't do anything to the generated code, ehm, why is
>
this switch added ? :)
To enable the extensions to C that allows you to use the c equivalents of
the altivec instructions. In GCC2.95.x, -faltivec was an extension Apple
made. I believe it has been accepted in the main codebase in 3.1.
>
-I would expect that if you did write your code, so the compiler could
>
recognize that it would be possible to altivec-optimize it, it would; but
>
only if there are benefits in this optimization.
No, i think vectorized optimisation is too complex for today's compilers to
recognize them.
>
No need to optimize one load-modify-and-store operation if it's used only
>
once. ;)
Depends on how lengthy that operation is. If program initialization could be
halved from 10 to 5 seconds, then I think the optimization would be useful
:)
>
-In other words; I would not expect a "controller", which is just telling
>
other objects what to do, to be altivec-optimized at all.
>
(Am I completely misunderstanding this ?)
Nope, you are not understanding.
/ David
_______________________________________________
cocoa-dev mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/cocoa-dev
Do not post admin requests to the list. They will be ignored.