[FFmpeg-devel] [PATCH] avcodec/binkaudio: add support for >2 channels dct codec

Peter Ross pross at xvid.org
Sun Mar 20 06:37:21 EET 2022


On Fri, Mar 18, 2022 at 04:21:44PM +0100, Paul B Mahol wrote:
> On 3/18/22, Andreas Rheinhardt <andreas.rheinhardt at outlook.com> wrote:
> > Paul B Mahol:
> >> As presented in .binka files.
> >>
> >> Signed-off-by: Paul B Mahol <onemda at gmail.com>
> >> ---
> >>  libavcodec/binkaudio.c | 50 +++++++++++++++++++++++++++---------------
> >>  1 file changed, 32 insertions(+), 18 deletions(-)
> >>
> >> diff --git a/libavcodec/binkaudio.c b/libavcodec/binkaudio.c
> >> index b4ff15beeb..54b7e22854 100644
> >> --- a/libavcodec/binkaudio.c
> >> +++ b/libavcodec/binkaudio.c
> >> @@ -51,13 +51,14 @@ typedef struct BinkAudioContext {
> >>      int version_b;          ///< Bink version 'b'
> >>      int first;
> >>      int channels;
> >> +    int ch_offset;
> >>      int frame_len;          ///< transform size (samples)
> >>      int overlap_len;        ///< overlap size (samples)
> >>      int block_size;
> >>      int num_bands;
> >>      float root;
> >>      unsigned int bands[26];
> >> -    float previous[MAX_CHANNELS][BINK_BLOCK_MAX_SIZE / 16];  ///< coeffs
> >> from previous audio block
> >> +    float previous[6][BINK_BLOCK_MAX_SIZE / 16];  ///< coeffs from
> >> previous audio block
> >>      float quant_table[96];
> >>      AVPacket *pkt;
> >>      union {
> >> @@ -74,6 +75,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
> >>      int sample_rate_half;
> >>      int i, ret;
> >>      int frame_len_bits;
> >> +    int max_channels = avctx->codec->id == AV_CODEC_ID_BINKAUDIO_RDFT ?
> >> MAX_CHANNELS : 6;
> >
> > If you allow up to six channels, then MAX_CHANNELS (i.e. two) needs to
> > be renamed.
> >
> >>      int channels = avctx->ch_layout.nb_channels;
> >>
> >>      /* determine frame length */
> >> @@ -85,7 +87,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
> >>          frame_len_bits = 11;
> >>      }
> >>
> >> -    if (channels < 1 || channels > MAX_CHANNELS) {
> >> +    if (channels < 1 || channels > max_channels) {
> >>          av_log(avctx, AV_LOG_ERROR, "invalid number of channels: %d\n",
> >> channels);
> >>          return AVERROR_INVALIDDATA;
> >>      }
> >> @@ -110,7 +112,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
> >>
> >>      s->frame_len     = 1 << frame_len_bits;
> >>      s->overlap_len   = s->frame_len / 16;
> >> -    s->block_size    = (s->frame_len - s->overlap_len) * s->channels;
> >> +    s->block_size    = (s->frame_len - s->overlap_len) *
> >> FFMIN(MAX_CHANNELS, s->channels);
> >>      sample_rate_half = (sample_rate + 1LL) / 2;
> >>      if (avctx->codec->id == AV_CODEC_ID_BINKAUDIO_RDFT)
> >>          s->root = 2.0 / (sqrt(s->frame_len) * 32768.0);
> >> @@ -166,7 +168,8 @@ static const uint8_t rle_length_tab[16] = {
> >>   * @param[out] out Output buffer (must contain s->block_size elements)
> >>   * @return 0 on success, negative error code on failure
> >>   */
> >> -static int decode_block(BinkAudioContext *s, float **out, int use_dct)
> >> +static int decode_block(BinkAudioContext *s, float **out, int use_dct,
> >> +                        int channels, int ch_offset)
> >>  {
> >>      int ch, i, j, k;
> >>      float q, quant[25];
> >> @@ -176,8 +179,8 @@ static int decode_block(BinkAudioContext *s, float
> >> **out, int use_dct)
> >>      if (use_dct)
> >>          skip_bits(gb, 2);
> >>
> >> -    for (ch = 0; ch < s->channels; ch++) {
> >> -        FFTSample *coeffs = out[ch];
> >> +    for (ch = 0; ch < channels; ch++) {
> >> +        FFTSample *coeffs = out[ch + ch_offset];
> >>
> >>          if (s->version_b) {
> >>              if (get_bits_left(gb) < 64)
> >> @@ -252,17 +255,17 @@ static int decode_block(BinkAudioContext *s, float
> >> **out, int use_dct)
> >>              s->trans.rdft.rdft_calc(&s->trans.rdft, coeffs);
> >>      }
> >>
> >> -    for (ch = 0; ch < s->channels; ch++) {
> >> +    for (ch = 0; ch < channels; ch++) {
> >>          int j;
> >> -        int count = s->overlap_len * s->channels;
> >> +        int count = s->overlap_len * channels;
> >>          if (!s->first) {
> >>              j = ch;
> >> -            for (i = 0; i < s->overlap_len; i++, j += s->channels)
> >> -                out[ch][i] = (s->previous[ch][i] * (count - j) +
> >> -                                      out[ch][i] *          j) / count;
> >> +            for (i = 0; i < s->overlap_len; i++, j += channels)
> >> +                out[ch + ch_offset][i] = (s->previous[ch + ch_offset][i]
> >> * (count - j) +
> >> +                                      out[ch + ch_offset][i] *
> >> j) / count;

^^^ This line needs to be indented some more, to match the previous line.

> >>          }
> >> -        memcpy(s->previous[ch], &out[ch][s->frame_len - s->overlap_len],
> >> -               s->overlap_len * sizeof(*s->previous[ch]));
> >> +        memcpy(s->previous[ch + ch_offset], &out[ch +
> >> ch_offset][s->frame_len - s->overlap_len],
> >> +               s->overlap_len * sizeof(*s->previous[ch + ch_offset]));
> >>      }
> >>
> >>      s->first = 0;
> >> @@ -293,6 +296,7 @@ static int binkaudio_receive_frame(AVCodecContext
> >> *avctx, AVFrame *frame)
> >>      GetBitContext *gb = &s->gb;
> >>      int ret;
> >>
> >> +again:
> >>      if (!s->pkt->data) {
> >>          ret = ff_decode_get_packet(avctx, s->pkt);
> >>          if (ret < 0)
> >> @@ -313,22 +317,31 @@ static int binkaudio_receive_frame(AVCodecContext
> >> *avctx, AVFrame *frame)
> >>      }
> >>
> >>      /* get output buffer */
> >> -    frame->nb_samples = s->frame_len;
> >> -    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
> >> -        return ret;
> >> +    if (s->ch_offset == 0) {
> >> +        frame->nb_samples = s->frame_len;
> >> +        if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
> >> +            return ret;
> >> +    }
> >>
> >>      if (decode_block(s, (float **)frame->extended_data,
> >> -                     avctx->codec->id == AV_CODEC_ID_BINKAUDIO_DCT)) {
> >> +                     avctx->codec->id == AV_CODEC_ID_BINKAUDIO_DCT,
> >> +                     FFMIN(MAX_CHANNELS, s->channels), s->ch_offset)) {
> >>          av_log(avctx, AV_LOG_ERROR, "Incomplete packet\n");
> >>          return AVERROR_INVALIDDATA;
> >>      }
> >> +    s->ch_offset += MAX_CHANNELS;
> >>      get_bits_align32(gb);
> >>      if (!get_bits_left(gb)) {
> >>          memset(gb, 0, sizeof(*gb));
> >>          av_packet_unref(s->pkt);
> >>      }
> >> +    if (s->ch_offset >= s->channels) {
> >> +        s->ch_offset = 0;
> >> +    } else {
> >> +        goto again;
> >> +    }
> >
> > Is it really intended that the data for one multi-channel frame is
> > divided into several input packets?
> 
> You are missing big picture here, >2 files have channels in different
> packets interleaved.
> Something like in XMA. (And nothing signals how are they interleaved.
> so its worse than in XMA) So it is working fine. I just need another
> look for possible regressions and security implications. Renaming
> MAX_CHANNELS is not useful as that is not property of both codecs.

MAX_CHANNELS (2) *is* a property of both codecs, and should be left alone.

I would prefer the '6' magic number be put into a descriptive macro.

LGTM.

-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20220320/e8a039b3/attachment.sig>


More information about the ffmpeg-devel mailing list