[FFmpeg-devel] [PATCH] h264chroma: remove duplicate 9/10 bit functions.

[Ronald S. Bultje]

Hi,

On Sun, Feb 10, 2013 at 7:55 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sun, Feb 10, 2013 at 07:39:11PM -0800, Ronald S. Bultje wrote:
>> Hi,
>>
>> On Sun, Feb 10, 2013 at 7:34 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> > On Sun, Feb 10, 2013 at 07:04:07PM -0800, Ronald S. Bultje wrote:
>> >> Hi,
>> >>
>> >> On Sun, Feb 10, 2013 at 5:22 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> >> > On Sun, Feb 10, 2013 at 04:40:19PM -0800, Ronald S. Bultje wrote:
>> >> >> From: "Ronald S. Bultje" <rsbultje at gmail.com>
>> >> >>
>> >> >> Also use the resulting 16bpp functions for anything >8 and <=16, not just
>> >> >> 9 and 10. This fixes 12 and 14bpp H264 support.
>> >> >
>> >> > How can i test/verify this bugfix ?
>> >> > I tried a few files but notice no difference, they either worked
>> >> > already or still dont work
>> >>
>> >> You see that currently it uses 8bpp chroma for !=9 && !=10, right?
>> >
>> > yes, recent regression introduced by 79dad2a9
>> > the new code should probably also use <=16 instead of <16
>> >
>> > But the SIMD changes look wrong, the code supports 10bit not >10bit
>>
>> Why? Do intermediates overflow 16bit?
>
> yes, for example:
> cglobal %1_h264_chroma_mc8_10, 6,7,8
> does a
> psrlw         m2, 6
> at the end thus would only output upto 10 bits

I suppose it can still be used for 9bits though... Anyway will amend
the x86 portion of the patch.

Ronald