[vlc-devel] [PATCH] Soft field repeat support for deinterlacer

[Juha Jeronen]

Hi,

On 02/10/11 14:10, Rémi Denis-Courmont wrote:
> Hmm, first it probably needs its own subdirectory in video_filter. We don't
> need to add a new Makefile though.

Sounds fine to me.

>> - filter_sys_t: this definition needs to be seen by all parts of the
>> deinterlacer. Move to header?
> Unavoidable to some extent, yes. Then again, algorithms could have their
> own data allocated in private structures on the fly.

If it's genuinely private data, then I agree. I think this needs an
init/uninit API (maybe just two functions) for clean alloc, init and
release of the data in the private structures (see e.g. ClearIVTC() and
DeallocateIVTC() in patch #3).

For general-use data that is useful to more than one algorithm, maybe
those should go in filter_sys_t.

I'm thinking that maybe e.g. the history mechanism should be moved to
Deinterlace(). Right now, it's only used by Yadif, but both IVTC and
Phosphor duplicate it. Three strikes and refactor...


>> - Simple rendering algorithms with just one function, or maybe two
>> functions at most: make a new file that gathers these.
> Is there even such an algorithm? Typically the simpler algorithm have the
> most variants of SIMD inline assembler.
True.

I was mainly thinking of RenderDiscard() and RenderBob(), but if those
are the only ones, maybe it's not that useful to do so. OTOH, placing
each of those in its own file seems to me overkill :)


 -J