[FFmpeg-devel] [PATCH] VP8: avoid conditional and division for chroma MV

[Ronald S. Bultje]

Hi,

On Wed, Jun 23, 2010 at 1:40 PM, Stefan Gehrer <stefan.gehrer at gmx.de> wrote:
> On 06/23/2010 06:51 PM, Ronald S. Bultje wrote:
>> On Wed, Jun 23, 2010 at 12:44 PM, Stefan Gehrer<stefan.gehrer at gmx.de>
>> ?wrote:
>>> Are there any recommended samples to test against?
>>
>> http://code.google.com/p/webm/downloads/detail?name=vp8-test-vectors-r1.zip&can=2&q=
>
> Okay, I tested some clips from the test vector and
> the frame CRCs stay the same with the patch.

You can test all of them using a small script (use untested to get ref.md5s):

rm -f test.md5s
for files in 000 001 002 003 004 005 006 007 008 009 \
             010 011 012 013 014 015 016 017; do \
	./ffmpeg -i ~/Desktop/vp8-test-vectors-r1/vp80-00-comprehensive-${files}.ivf \
		-v 0 -y -an -vcodec rawvideo -f md5 - 2>&1 | grep MD5 >> test.md5s
done
diff -u test.md5s ref.md5s && echo "Results identical"

> But when I compile on my machine with gcc 4.4.3 amd -O3
> it seems clever enough to avoid conditional and division
> anyway.
> So now I believe in the correctness of the patch, I am
> just not so sure about the usefulness.

Well, does it make it faster or lead to better assembly?

Ronald