[FFmpeg-devel] [PATCH] fix Altivec inloop filter

Måns Rullgård mans
Sat Jun 16 23:29:52 CEST 2007


"Guillaume POIRIER" <poirierg at gmail.com> writes:

> Hello folks
>
> Please find below the description what attached patch does to fix
> Altivec-accelerated inloop filter.
> -------------------------
> "From: Graham Booker
> To: Guillaume POIRIER
> Date: Jun 16, 2007 6:46 PM
> Subject: h264 altivec loop_filter_luma
>
> I noticed [9317] and I think I found a potential problem.  In the
> h264_deblock_q1, the result of the deblock is *not* used in future
> deblocks.  I changed the function to an inline function which returns
> this value (it was a #define simply because it used to return the
> value in place, but now that it is no longer in place, there is no
> need for a #define).
>
> I also reworked some of the stuff with the tc0 vector.  It is really a
> signed value, so I treat it as such until after the >=0 check.  Then,
> at that point, after being masked, it can be treated as unsigned."
> -------------------------
>
> I can confirm that this patch allows Alivec code to decode all my
> samples with the same md5sum; now the question is: is this patch OK
> like this, or should h264_deblock_q1 be left as a macro for now, and
> converted to an inline function later, if needed?

The diff would be easier to read if it were left as a macro.

-- 
M?ns Rullg?rd
mans at mansr.com




More information about the ffmpeg-devel mailing list