[x264-devel] [PATCH] AltiVec implementation of hadamard_ac routines

Guillaume POIRIER gpoirier at mplayerhq.hu
Sun Feb 1 22:59:31 CET 2009


Hello Takashi-san (is this the right way to greet in Japanese?) ;-)


On Sun, Feb 1, 2009 at 4:29 AM, Takashi Mochizuki
<mochi at da2.so-net.ne.jp> wrote:
> Guillaume-san,
> I have made a test binary of my QuickTime Component x264Encoder, using
> r1096 with your hadamard_ac altivec patch.
>
> To build with your patch, gcc does warn some printf with %vhd. I have
> cut them all.

That's odd. Maybe I didn't attach the right patch then. This newer
version compiles without additional warnings.


> Tried with 15sec VGA sample, and it runs well. No binary diff found in
> each video stream.

That's odd too since there was a bug in a previous patch version I
had. It definitely looks like I didn't attach the right one :-(

Please use the newer version attached to this mail, since it fixes
alignment issues that previous patch had. It may be faster than the
previous one you tried... In any case, it doesn't suffer from know
bugs so you should consider using it instead.

I'll try to find some time to understand the optimizations in x86's
SIMD routines to use them in my AltiVec implementation of hadamard_ac.


> On my G4 1.67GHz(MacOS X 10.5.6), it shows pretty positive performance
> gain. (+1.8% overall)
>  r1096 + patch : 165.5 sec
>  r1096 : 168.5 sec
>  r1080 : 174.8 sec
> test options(SEI) : cabac=1 ref=3 deblock=1:0:0 analyse=0x3:0x113
> me=umh subme=7 psy_rd=1.0:0.0 mixed_ref=1 me_range=16 chroma_me=1
> trellis=1 8x8dct=1 cqm=0 deadzone=21,11 chroma_qp_offset=-2 threads=2
> nr=0 decimate=0 mbaff=0 bframes=3 b_pyramid=0 b_adapt=2 b_bias=0
> direct=3 wpredb=1 keyint=24 keyint_min=1 scenecut=80(pre) rc=crf
> crf=24.0 qcomp=0.60 qpmin=10 qpmax=51 qpstep=4 ip_ratio=1.40
> pb_ratio=1.30 aq=1:1.00
>
> Other test options set required?

Shouldn't be necessary.


> Takashi Mochizuki
> FYI: my blog topic URL is: mycometg3.blog.so-net.ne.jp/2009-02-01

So you're the person developing x264Encoder? Good job!

Guillaume
-- 
Only a very small fraction of our DNA does anything; the rest is all
comments and ifdefs.

Calvin Trillin  - "Health food makes me sick."
-------------- next part --------------
A non-text attachment was scrubbed...
Name: hadamard_ac.1.diff
Type: application/octet-stream
Size: 7791 bytes
Desc: not available
Url : http://mailman.videolan.org/pipermail/x264-devel/attachments/20090201/2f4af4cb/attachment.obj 


More information about the x264-devel mailing list