[tremor] random patch

[Chris Gilbert]

And in perfectly bad form I reply to myself 8)

helps if I average a few runs, doing the initial one is very slow, the
following runs speed along.  Anyway avg times:
asm:  68.087u 5.348s 1:14.55
cver: 69.003u 4.796s 1:14.83

So the patch seems to get cut a small %age off the time, but then I'm
wondering if the 2 blocks can't be merged in some way, IE loop less.

The other strange thing is that I'm not 100% certain the patch is accurate,
I get differences between the cver output and the asm one, gah, must find a
smaller was file to play with

Chris

----- Original Message -----
From: "Chris Gilbert" <chris at dokein.co.uk>
To: <tremor at xiph.org>
Sent: Wednesday, September 11, 2002 12:01 AM
Subject: [tremor] random patch

<p>> I was looking over the tail end of the mdct_backwards, and noted that arm
> asm can be a lot more efficent at it than the compiler produced code (due
to
> being able to increment and decrement pointers, and I can see what it's
> doing 8)
>
> Anyway on a random ogg file it knocked 5 seconds off 298 seconds to
decode,
> so is about a 1.5% improvement, only a small gain, but every bit helps 8)
>
> Certainly someone that knows what the code actually does and why (looks
like
> it's actually reflecting data, to create a wave?)  might want to name the
> functions better.
>
> Oh the warnings on register order are bogus in this case as the same
> registers are used, and in the same order.
>
> Might look at other places to tweak with arm asm.
>
> Is there a proper place to submit patches? or is on here fine?
>
> Cheers
> Chris
>

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.