[FFmpeg-devel] [PATCH 4/7] lavc: move decoder bsf init into decoder-specific code

James Almer jamrial at gmail.com
Wed Mar 10 21:23:24 EET 2021


On 3/10/2021 10:52 AM, Andreas Rheinhardt wrote:
> James Almer:
>> On 3/10/2021 9:03 AM, Anton Khirnov wrote:
>>> ---
>>>    libavcodec/decode.c | 8 +++++++-
>>>    libavcodec/decode.h | 6 ------
>>>    libavcodec/utils.c  | 6 ------
>>>    3 files changed, 7 insertions(+), 13 deletions(-)
>>>
>>> diff --git a/libavcodec/decode.c b/libavcodec/decode.c
>>> index e5a301ec58..d25b15e95a 100644
>>> --- a/libavcodec/decode.c
>>> +++ b/libavcodec/decode.c
>>> @@ -184,7 +184,7 @@ static int extract_packet_props(AVCodecInternal
>>> *avci, const AVPacket *pkt)
>>>        return 0;
>>>    }
>>>    -int ff_decode_bsfs_init(AVCodecContext *avctx)
>>> +static int decode_bsfs_init(AVCodecContext *avctx)
>>>    {
>>>        AVCodecInternal *avci = avctx->internal;
>>>        int ret;
>>> @@ -2010,6 +2010,8 @@ int ff_reget_buffer(AVCodecContext *avctx,
>>> AVFrame *frame, int flags)
>>>      int ff_decode_preinit(AVCodecContext *avctx)
>>>    {
>>> +    int ret = 0;
>>> +
>>>        /* if the decoder init function was already called previously,
>>>         * free the already allocated subtitle_header before overwriting
>>> it */
>>>        av_freep(&avctx->subtitle_header);
>>> @@ -2046,5 +2048,9 @@ FF_ENABLE_DEPRECATION_WARNINGS
>>>            avctx->export_side_data |= AV_CODEC_EXPORT_DATA_MVS;
>>>        }
>>>    +    ret = decode_bsfs_init(avctx);
>>
>> Did you try to run fate with THREADS set to something other than 1? This
>> moves the bsfs initialization after ff_thread_init(), so all the copies
>> of avctx in frame threading scenarios will be missing the bsfs requested
>> by the decoder.
>>
> 
> For frame threaded decoders the codec's decode function is directly run
> with the child AVCodecContext; they don't need the bsfs: The parent's
> bsf has already been run on the packets before the frame threaded code
> is reached.

Alright, should be ok, then.

> 
>> fate-vp9 should be enough to confirm this, since afaik it's the only
>> frame threaded decoder that uses a bsf.
>>
>>> +    if (ret < 0)
>>> +        return ret;
>>> +
>>>        return 0;
>>>    }
>>> diff --git a/libavcodec/decode.h b/libavcodec/decode.h
>>> index a865fe954f..1467b1eb33 100644
>>> --- a/libavcodec/decode.h
>>> +++ b/libavcodec/decode.h
>>> @@ -69,12 +69,6 @@ int ff_decode_get_packet(AVCodecContext *avctx,
>>> AVPacket *pkt);
>>>     */
>>>    int ff_decode_frame_props(AVCodecContext *avctx, AVFrame *frame);
>>>    -/**
>>> - * Called during avcodec_open2() to initialize avctx->internal->bsf.
>>> - * The bsf should be freed with av_bsf_free().
>>> - */
>>> -int ff_decode_bsfs_init(AVCodecContext *avctx);
>>> -
>>>    /**
>>>     * Make sure avctx.hw_frames_ctx is set. If it's not set, the
>>> function will
>>>     * try to allocate it from hw_device_ctx. If that is not possible,
>>> an error
>>> diff --git a/libavcodec/utils.c b/libavcodec/utils.c
>>> index 6d5aeb4eaf..918cb1b4d1 100644
>>> --- a/libavcodec/utils.c
>>> +++ b/libavcodec/utils.c
>>> @@ -730,12 +730,6 @@ int attribute_align_arg
>>> avcodec_open2(AVCodecContext *avctx, const AVCodec *code
>>>                goto free_and_end;
>>>        }
>>>    -    if (av_codec_is_decoder(avctx->codec)) {
>>> -        ret = ff_decode_bsfs_init(avctx);
>>> -        if (ret < 0)
>>> -            goto free_and_end;
>>> -    }
>>> -
>>>        if (HAVE_THREADS
>>>            && !(avci->frame_thread_encoder &&
>>> (avctx->active_thread_type&FF_THREAD_FRAME))) {
>>>            ret = ff_thread_init(avctx);
>>>
>>
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> 



More information about the ffmpeg-devel mailing list