[FFmpeg-devel] [PATCH v3 3/6] avformat/s337m: Make available as subdemuxer

[Tomas Härdin]

tis 2019-08-06 klockan 17:08 +0200 skrev Nicolas Gaullier:
> +int avpriv_s337m_probe_stream(void *avc, AVIOContext *pb, AVStream **st, int size)
> +{
> +    if ( size >= S337M_MIN_OFFSET &&
> +        ((*st)->codecpar->codec_id == AV_CODEC_ID_PCM_S16LE || (*st)->codecpar->codec_id == AV_CODEC_ID_PCM_S24LE) &&
> +               (*st)->codecpar->channels == 2) {
> +        uint8_t *buf;
> +        int64_t data_offset;
> +        int pos = 0;
> +
> +        size = FFMIN(size, S337M_MAX_OFFSET);
> +        if (!(buf = av_mallocz(size))) {
> +            return AVERROR(ENOMEM);
> +        }
> +        data_offset = avio_tell(pb);
> +        if (avio_read(pb, buf, size) != size) {
> +            av_freep(&buf);
> +            return AVERROR(EIO);
> +        }
> +        avio_seek(pb, data_offset, SEEK_SET);
> +
> +        while (pos < size - 9 && !buf[pos]) {
> +            pos++;
> +        }
> +        if (pos < size - 9 && pos >= S337M_PROBE_GUARDBAND_MIN_BYTES) {

I think this 9 should be an 11 or 12..

> +            uint64_t state;
> +            int data_type = -1, data_size, offset;
> +
> +            if ((*st)->codecpar->bits_per_coded_sample == 16) {
> +                state = AV_RB32(buf + pos);
> +                if (IS_16LE_MARKER(state)) {
> +                    data_type = AV_RL16(buf + pos + 4);
> +                    data_size = AV_RL16(buf + pos + 6);
> +                }
> +            } else if ((*st)->codecpar->bits_per_coded_sample == 24) {
> +                state = AV_RB48(buf + pos);
> +                if (IS_20LE_MARKER(state) || IS_24LE_MARKER(state)) {
> +                    data_type = AV_RL24(buf + pos + 6);
> +                    data_size = AV_RL24(buf + pos + 9);

.. because of this read

> +                }
> +            }
> +            if (data_type >= 0 && !s337m_get_offset_and_codec(avc, state, data_type, data_size, &offset, &(*st)->codecpar->codec_id)) {
> +                double s337m_phase = pos * 4 / (*st)->codecpar->bits_per_coded_sample / (*st)->codecpar->sample_rate;

This isn't quite what I meant by turning it into an integer test :)
this will likely just be rounded to zero. I meant that things could
likely be rearranged so there's no divisions, so the probing isn't
subject to the vaguaries of float rounding

> +#define DOLBY_E_PHASE_MIN       0.000610
> +#define DOLBY_E_PHASE_MAX       0.001050

Where do these phase values come from? Is there a spec somewhere?

I notice the ratios DOLBY_E_PHASE_MAX/DOLBY_E_PHASE_MIN and
S337M_MAX_OFFSET/S337M_MIN_OFFSET are within 9% of eachother, which
doesn't feel like a coincidence

/Tomas