On 12/11/18 20:20, Michael Niedermayer wrote:
> On Sun, Nov 11, 2018 at 02:54:41PM +0000, Mark Thompson wrote:
>> Fixes #7519.
>> ---
>> I've had this lying around for a bit with intent to use it to support non-hwaccel hardware decoders (in particular, v4l2m2m), but it actually fixes the force-hwaccel use-case too in a sensible way.
>>
>>
>> doc/ffmpeg.texi | 12 ++++++++++++
>> fftools/ffmpeg.c | 4 ++++
>> fftools/ffmpeg.h | 4 ++++
>> fftools/ffmpeg_opt.c | 15 +++++++++++++++
>> 4 files changed, 35 insertions(+)
>>
>> diff --git a/doc/ffmpeg.texi b/doc/ffmpeg.texi
>> index 3717f22d42..9f9e693898 100644
>> --- a/doc/ffmpeg.texi
>> +++ b/doc/ffmpeg.texi
>> @@ -920,6 +920,18 @@ would be more efficient.
>> When doing stream copy, copy also non-key frames found at the
>> beginning.
>>
>> + at item -decode_format[:@var{stream_specifier}] @var{pixfmt} (@emph{input,per-stream})
>> +Set the output format to be used by the decoder for this stream. If the
>> +decoder does not natively support this format for the input stream then
>> +decoding will fail rather than continuing with a different format.
>> +
>> +In general this should not be set - the decoder will select an output
>> +format based on the input stream parameters and available components, and
>> +that will be automatically converted to whatever the output requires. It
>> +may be useful to force a hardware decoder supporting output in multiple
>> +different memory types to pick the desired one, or to ensure that a
>> +hardware decoder is used when software fallback is also available.
>> +
>> @item -init_hw_device @var{type}[=@var{name}][:@var{device}[, at var{key=value}...]]
>> Initialise a new hardware device of type @var{type} called @var{name}, using the
>> given device parameters.
>> diff --git a/fftools/ffmpeg.c b/fftools/ffmpeg.c
>> index 38c21e944a..73bed55e4d 100644
>> --- a/fftools/ffmpeg.c
>> +++ b/fftools/ffmpeg.c
>> @@ -2800,6 +2800,10 @@ static enum AVPixelFormat get_format(AVCodecContext *s, const enum AVPixelFormat
>> const AVCodecHWConfig *config = NULL;
>> int i;
>>
>> + if (ist->decode_format != AV_PIX_FMT_NONE &&
>> + ist->decode_format != *p)
>> + continue;
>> +
>> if (!(desc->flags & AV_PIX_FMT_FLAG_HWACCEL))
>> break;
>>
>> diff --git a/fftools/ffmpeg.h b/fftools/ffmpeg.h
>> index eb1eaf6363..88e0aa60ea 100644
>> --- a/fftools/ffmpeg.h
>> +++ b/fftools/ffmpeg.h
>> @@ -125,6 +125,8 @@ typedef struct OptionsContext {
>> int nb_ts_scale;
>> SpecifierOpt *dump_attachment;
>> int nb_dump_attachment;
>> + SpecifierOpt *decode_formats;
>> + int nb_decode_formats;
>> SpecifierOpt *hwaccels;
>> int nb_hwaccels;
>> SpecifierOpt *hwaccel_devices;
>
>> @@ -334,6 +336,8 @@ typedef struct InputStream {
>> int top_field_first;
>> int guess_layout_max;
>>
>> + enum AVPixelFormat decode_format;
>> +
>
> I think this should be a list and not a single format
Yes, I agree; that's a very good idea.
New patch to follow.
Thanks,
- Mark