[FFmpeg-devel] [PATCH] vp9: add subpel MC SIMD for 10/12bpp.

[James Almer]

On 9/16/2015 5:48 PM, Ronald S. Bultje wrote:
> Hi,
> 
> On Wed, Sep 16, 2015 at 4:31 PM, James Almer <jamrial at gmail.com> wrote:
> 
>> On 9/16/2015 4:17 PM, Ronald S. Bultje wrote:
>>> ---
>>>  libavcodec/x86/Makefile                     |   5 +-
>>>  libavcodec/x86/vp9dsp_init.c                | 197 +++----------
>>>  libavcodec/x86/vp9dsp_init.h                | 110 +++++++-
>>>  libavcodec/x86/vp9dsp_init_10bpp.c          |  25 ++
>>>  libavcodec/x86/vp9dsp_init_12bpp.c          |  25 ++
>>>  libavcodec/x86/vp9dsp_init_16bpp.c          |   2 +-
>>>  libavcodec/x86/vp9dsp_init_16bpp_template.c |  63 +++++
>>
>> Using a template file seems unnecessarily complex. Why not just call the
>> macros
>> to declare both 10 and 12 bits prototypes inside vp9dsp_init_16bpp.c, then
>> in
>> ff_vp9dsp_init_16bpp_x86() do
>>
>> if (bpp == 10) { init stuff }
>> else if (bpp == 12) { init stuff }
>>
>> Sort of like how h264 and hevc currently do.
>> Afaics all 10/12bit function prototypes are going to be the same so the
>> above
>> approach is probably cleaner (One file instead of four).
> 
> 
> Well, two of these files are only 2 lines, so it's really just 2 files. The
> nice thing is that it's half the code for the bpp-specific stuff.
> 
> If you don't like it I'll remove it, but I personally liked the fact that
> it prevents duplicate source code between the 10/12bpp setup.

Keep it like this, then. Since you're the one writing the stuff of course use
whatever you like the most. It's just cosmetics and I'm fine either way.