[FFmpeg-devel] [PATCH v2 00/16] Replace native DCA decoder with libdcadec based one

[Hendrik Leppkes]

On Sat, Jan 30, 2016 at 10:41 PM, James Almer <jamrial at gmail.com> wrote:
> On 1/30/2016 6:15 PM, Hendrik Leppkes wrote:
>> On Sat, Jan 30, 2016 at 7:05 PM, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
>>> On Sat, Jan 30, 2016 at 1:43 AM, Andreas Cadhalpun
>>> <andreas.cadhalpun at googlemail.com> wrote:
>>>> On 25.01.2016 23:47, Hendrik Leppkes wrote:
>>>>> The decoder in itself looks fine to me, short of the regression michael found.
>>>>> If you could look at that, that would be great.
>>>>>
>>>>> I can squash and re-shuffle the commits appropriately for pushing and
>>>>> make sure all intermediate steps still build, so don't worry about
>>>>> that as much.
>>>>> I would love to get this in soon, we can do further improvements and
>>>>> more FATE coverage after.
>>>>
>>>> I agree that it's time to apply these patches. Resending this huge
>>>> patch set for tiny improvements is just not practical.
>>>>
>>>>> There doesn't seem to be much agreement yet of the order of pushing.
>>>>> I would argue that since we're replacing the decoder entirely anyway,
>>>>> a tiny period in between where we don't actually have a dca decoder
>>>>> wouldn't break any bisect flow, since it would probably end there
>>>>> anyway.
>>>>> So considering that, it feels cleaner to me to push the removal first,
>>>>> and then the additions for the new decoder for "prettier" history.
>>>>
>>>> That seems fine to me, but I don't have a strong opinion about this.
>>>>
>>>
>>> I have started to rebase, update and squash it appropriately,
>>> including the patch for Michaels issue I posted earlier.
>>> Once its all ready for pushing, I'll post my GitHub link for a final
>>> review if anyone wants to, and otherwise push it in the next day or
>>> two, so we finally get this done.
>>>
>>
>> Here is the repository, rebased and partially squashed:
>> https://github.com/Nevcairiel/FFmpeg/tree/dca
>>
>> All individual steps build and pass FATE, version bump and Changelog
>> entry in the commit with the new decoder.
>
> It doesn't really need a minor bump. It's not a "new" decoder that wasn't
> available before. Just bump micro because of the avoption changes. This
> change is completely transparent from an API PoV otherwise.

Suppose micro is fine as well.

>
>>
>> - Hendrik
>
> IMO, merge "avcodec/dca: add math helpers and fixed point DCT" into the
> last patch. It's no different than the dsp patch you already squashed.
>

The new synth filter functions depend on those (fixed point synth
filter needs fixed point DCT), so its either squash those as well, or
keep them split.

- Hendrik