[theora-dev] [Fwd: Re: libtheora MMX patch]
andrew at okashistudios.com
Tue Apr 8 21:07:02 PDT 2008
Forwarding an email exchange that I had with Nils Pipenbrinck regarding
the state of the MMX patch for visual studio-style assembly.
I also run with the patches, and everything looks fine as far as I can tell.
Is this enough for a go-ahead to put that stuff into the mainline (if
it's not there already)?
-------- Original Message --------
Subject: Re: libtheora MMX patch
Date: Sun, 06 Apr 2008 15:21:12 +0200
From: Nils Pipenbrinck <n.pipenbrinck at cubic.org>
To: Andrew Chew <andrew at okashistudios.com>
References: <4778152E.5010904 at cubic.org>
<477857B8.1010102 at email.unc.edu> <4778AA4C.9010105 at cubic.org>
<47F8463A.8000007 at okashistudios.com>
I run the original C-code against the mmx optimized version, dumped out
the raw yuv-frames to disk and did a binary compare. It's functional
equivalent down to the bit-level. I used two different test-streams for
this. It's highly unlikely that a bug survives this test.
AFAIK my changes for the gcc inline assembler functions are already in
the trunk (slightly modified - Giles didn't liked the fact that I
unrolled / wrote the loops in asm).
If you want to merge into the win32 baseline let me know I have adopted
the filter and idct mmx functions to match more the gcc style but never
submitted them. Also a little bug crept into the win32 mmx code you
have. I accidently used the same code for idct10 and idct. Functional it
works but I lost a little bit speed that way (barely measurable but
I have a local version where this is fixed as well.
If you're interested I'll dig out the modified codes from my old
harddisk and send it to you.
Andrew Chew schrieb:
> Hi! I tested your assembly patches to libtheora a while back.
> How well have you validated that it works? It works in my environment
> pretty well.
> I'd just like to make sure this gets into mainline libtheora, and so
> far it hasn't (they are waiting for some confirmation that those
> patches are indeed good). I guess they are short on WIN32 testers.
> This was revealed to me at the last libtheora roadmap meeting that we
> just had last week.
> I'd really like to make sure that your stuff gets into their mainline.
More information about the theora-dev