From: "James H. Cloos Jr." <[EMAIL PROTECTED]>
Subject: [mp3encoder] Re: Some Amd64 timings
Date: Thu, 26 Feb 2004 18:43:42 -0500

> >>>>> "Alexander" == Alexander Leidinger <[EMAIL PROTECTED]> writes:
> 
> Alexander> As gcc should be able to utilize MMX, SSE, SSE2 and 3dnow
> Alexander> with inline assembly now, it would be nice if someone would
> Alexander> try to use it instead of using a 3rd party assembler.
> 
> Actually, you realy want to the the intrinsics rather than assembly.
> The intrinsics look just like function calls at the C level (making
> them easier to understand), the compiler deals with the minutiae of
> register assignments, function call abi issues, et al so you only
> have the algorithm to deal with, and the compiler's optimizer is not
> blocked.

Do you think current compiler is better than human to optimize the SSE
code written with intrisics ? I don't think so....

> In short, they should be easier to use well than raw assembly,
> whether inlined or external.

Agree. If I need to write a "reasonably" fast program in a hurry from
scratch, I usually use instisic.

But for LAME, I need "extremely" fast program and there's no deadline
and there's GOGO, full-assembler mp3 encoder based on LAME. And in these
days, the intrisic can be used only a newer compilers and they are buggy...

This is why I used NASM instead of intrisics.
Now the situation is changed about availability of intrisics, but the
rest of them are not changed much, IMHO.

Anyway, if someone writes the asm code with intrisics, it will be welcomed
if it runs fine, of course.
-- 
Takehiro TOMINAGA // may the source be with you!
_______________________________________________
mp3encoder mailing list
[EMAIL PROTECTED]
http://minnie.tuhs.org/mailman/listinfo/mp3encoder

Reply via email to