[FFmpeg-devel] [PATCH v12 5/9] libavcodec/dnxucdec: DNxUncompressed decoder

[Anton Khirnov]

Quoting Martin Schitter (2024-10-21 21:57:18)
> +static int pass_through(AVCodecContext *avctx, AVFrame *frame, const AVPacket *avpkt)
> +{
> +    /* there is no need to copy as the data already match
> +     * a known pixel format */
> +
> +    frame->buf[0] = av_buffer_ref(avpkt->buf);

I said this twice before already - every single format that uses
pass_through() should instead be exported by the demuxer as
AV_CODEC_ID_RAWVIDEO, because that's what it is.

That will also have the following important advantages:
* decoding these formats won't pointlessly waste resources and add
  latency using frame threading, which is useless for them
* your decoder can be marked as AV_CODEC_CAP_DR1


> +static int unpack_y212(AVCodecContext *avctx, AVFrame *frame, const AVPacket *pkt)
> +{
> +    uint8_t *data = &pkt->data[0];
> +    int lw = frame->width;
> +    int msb_bytes = lw * 2;
> +    int line_offset = lw * 3;
> +    int pos_uv = 0, pos_y = 0, pos_in = 0;
> +    for(int y = 0; y < frame->height; y++){
> +        for(int x = 0; x < lw/2; x++){
> +            AV_WL16(&frame->data[1][pos_uv],   get12(data, pos_in++, msb_bytes)); // u
> +            AV_WL16(&frame->data[0][pos_y],    get12(data, pos_in++, msb_bytes)); // y
> +            AV_WL16(&frame->data[2][pos_uv],   get12(data, pos_in++, msb_bytes)); // v
> +            AV_WL16(&frame->data[0][pos_y + 2],get12(data, pos_in++, msb_bytes)); // y
> +            pos_uv += 2;
> +            pos_y += 4;
> +        }
> +        data += line_offset;
> +        pos_in = 0;
> +    }
> +    return pkt->size;
> +}

These functions are now MUCH more readable, and it becomes clear that
you're not accessing frame->linesize, which is immensely suspicious.
Lines of a frame do not need to be contiguous in memory, there can be
arbitrary amounts of paddding between them.

-- 
Anton Khirnov