[FFmpeg-devel] [RFC] Direct Stream Transfer (DST) decoder
Peter Ross
pross at xvid.org
Tue Sep 30 15:42:20 CEST 2014
Signed-off-by: Peter Ross <pross at xvid.org>
---
DST (described in 14496 Part 3 Subpart 10) is the lossless compression standard
for DSD samples. My implementation is ~45% faster than the reference decoder.
Even so, single threaded decoding is still very demanding and there does not
appear to much scope for SIMD optimisation.
The decoder outputs to AV_SAMPLE_FMT_DSD, which is not in FFmpeg yet.
(I guess it could be converted to output PCM, though that kind of defeats
the purpose having 1-bit audio.)
Changelog | 1 +
libavcodec/Makefile | 1 +
libavcodec/allcodecs.c | 1 +
libavcodec/avcodec.h | 1 +
libavcodec/codec_desc.c | 7 +
libavcodec/dst.h | 28 ++++
libavcodec/dstdec.c | 346 ++++++++++++++++++++++++++++++++++++++++++++++++
7 files changed, 385 insertions(+)
create mode 100644 libavcodec/dst.h
create mode 100644 libavcodec/dstdec.c
diff --git a/Changelog b/Changelog
index 8964774..9d8e2b3 100644
--- a/Changelog
+++ b/Changelog
@@ -6,6 +6,7 @@ version <next>:
- SUP/PGS subtitle demuxer
- DoP (DSD-over-PCM)
- Wideband Single-bit Data (WSD) demuxer
+- Direct Stream Transfer (DST) decoder
version 2.4:
- Icecast protocol
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 08f724e..325ebe1 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -204,6 +204,7 @@ OBJS-$(CONFIG_DSD_MSBF_PLANAR_DECODER) += dsddec.o
OBJS-$(CONFIG_DSD_MSBF_PLANAR_ENCODER) += dsddec.o
OBJS-$(CONFIG_DSICINAUDIO_DECODER) += dsicinaudio.o
OBJS-$(CONFIG_DSICINVIDEO_DECODER) += dsicinvideo.o
+OBJS-$(CONFIG_DST_DECODER) += dstdec.o
OBJS-$(CONFIG_DVBSUB_DECODER) += dvbsubdec.o
OBJS-$(CONFIG_DVBSUB_ENCODER) += dvbsub.o
OBJS-$(CONFIG_DVDSUB_DECODER) += dvdsubdec.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index c988bd1..a1b699d 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -343,6 +343,7 @@ void avcodec_register_all(void)
REGISTER_ENCDEC (DSD_MSBF, dsd_msbf);
REGISTER_ENCDEC (DSD_LSBF_PLANAR, dsd_lsbf_planar);
REGISTER_ENCDEC (DSD_MSBF_PLANAR, dsd_msbf_planar);
+ REGISTER_DECODER(DST, dst);
REGISTER_DECODER(DSICINAUDIO, dsicinaudio);
REGISTER_ENCDEC (EAC3, eac3);
REGISTER_DECODER(EVRC, evrc);
diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index ccffbd0..2d176d6 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -498,6 +498,7 @@ enum AVCodecID {
AV_CODEC_ID_DSD_LSBF_PLANAR = MKBETAG('D','S','D','1'),
AV_CODEC_ID_DSD_MSBF_PLANAR = MKBETAG('D','S','D','8'),
AV_CODEC_ID_DOP_S24LE = MKBETAG('D','O','P','L'),
+ AV_CODEC_ID_DST = MKBETAG('D','S','T',' '),
/* subtitle codecs */
AV_CODEC_ID_FIRST_SUBTITLE = 0x17000, ///< A dummy ID pointing at the start of subtitle codecs.
diff --git a/libavcodec/codec_desc.c b/libavcodec/codec_desc.c
index 7c202cc..6d0813e 100644
--- a/libavcodec/codec_desc.c
+++ b/libavcodec/codec_desc.c
@@ -2515,6 +2515,13 @@ static const AVCodecDescriptor codec_descriptors[] = {
.long_name = NULL_IF_CONFIG_SMALL("DoP (DSD-over-PCM); signed 24-bit little-endian"),
.props = AV_CODEC_PROP_LOSSLESS,
},
+ {
+ .id = AV_CODEC_ID_DST,
+ .type = AVMEDIA_TYPE_AUDIO,
+ .name = "dst",
+ .long_name = NULL_IF_CONFIG_SMALL("DST (Direct Stream Transfer)"),
+ .props = AV_CODEC_PROP_LOSSLESS,
+ },
/* subtitle codecs */
{
diff --git a/libavcodec/dst.h b/libavcodec/dst.h
new file mode 100644
index 0000000..83c3f7f
--- /dev/null
+++ b/libavcodec/dst.h
@@ -0,0 +1,28 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * Calculate FS44 ratio
+ */
+#define DSD_FS44(sample_rate) (sample_rate / 44100)
+
+/**
+ * Calculate DST frame size
+ * @return samples per frame (1-bit samples)
+ */
+#define DST_SAMPLES_PER_FRAME(sample_rate) (588 * DSD_FS44(sample_rate))
diff --git a/libavcodec/dstdec.c b/libavcodec/dstdec.c
new file mode 100644
index 0000000..51802ac
--- /dev/null
+++ b/libavcodec/dstdec.c
@@ -0,0 +1,346 @@
+/*
+ * Direct Stream Transfer (DST) decoder
+ * Copyright (c) 2014 Peter Ross <pross at xvid.org>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Direct Stream Transfer (DST) decoder
+ * ISO/IEC 14496-3 Part 3 Subpart 10: Technical description of lossless coding of oversampled audio
+ */
+
+#include "libavcodec/internal.h"
+#include "get_bits.h"
+#include "avcodec.h"
+#include "dst.h"
+#include "golomb.h"
+#include "mathops.h"
+
+#include "libavutil/avassert.h"
+#include "libavutil/intreadwrite.h"
+
+#define DST_MAX_CHANNELS 6
+#define DST_MAX_ELEMENTS (2 * DST_MAX_CHANNELS)
+
+static const int8_t fsets_code_pred_coeff[3][3] = {
+ { -8 },
+ { -16, 8 },
+ { -9, -5, 6 },
+};
+
+static const int8_t probs_code_pred_coeff[3][3] = {
+ { -8 },
+ { -16, 8 },
+ { -24, 24, -8 },
+};
+
+typedef struct {
+ unsigned int elements;
+ unsigned int length[DST_MAX_ELEMENTS];
+ int coeff[DST_MAX_ELEMENTS][128];
+} Table;
+
+static av_cold int decode_init(AVCodecContext *avctx)
+{
+ if (avctx->channels > DST_MAX_CHANNELS) {
+ avpriv_request_sample(avctx, "Channel count %d", avctx->channels);
+ return AVERROR_PATCHWELCOME;
+ }
+
+ avctx->sample_fmt = AV_SAMPLE_FMT_DSD;
+ return 0;
+}
+
+static int read_map(GetBitContext *gb, Table *t, unsigned int map[DST_MAX_CHANNELS], int channels)
+{
+ int ch;
+ t->elements = 1;
+ if (!get_bits1(gb)){
+ map[0] = 0;
+ for (ch = 1; ch < channels; ch++) {
+ int bits = av_log2(t->elements) + 1;
+ map[ch] = get_bits(gb, bits);
+ if (map[ch] == t->elements) {
+ t->elements++;
+ if (t->elements >= DST_MAX_ELEMENTS)
+ return AVERROR_INVALIDDATA;
+ } else if (map[ch] > t->elements) {
+ return AVERROR_INVALIDDATA;
+ }
+ }
+ } else {
+ memset(map, 0, sizeof(*map) * DST_MAX_CHANNELS);
+ }
+ return 0;
+}
+
+static av_always_inline int get_sr_golomb_dst(GetBitContext *gb, unsigned int k)
+{
+#if 0
+ /* 'run_length' upper bound is not specified; we can never be sure it will fit into get_bits cache */
+ int v = get_ur_golomb(gb, k, INT_MAX, 0);
+#else
+ int v = 0;
+ while (!get_bits1(gb))
+ v++;
+ if (k)
+ v = (v << k) | get_bits(gb, k);
+#endif
+ if (v && get_bits1(gb))
+ v = -v;
+ return v;
+}
+
+static void read_uncoded_coeff(GetBitContext *gb, int *dst, unsigned int elements, int coeff_bits, int is_signed, int offset)
+{
+ unsigned int i;
+ for (i = 0; i < elements; i++)
+ dst[i] = (is_signed ? get_sbits(gb, coeff_bits) : get_bits(gb, coeff_bits)) + offset;
+}
+
+static int read_table(GetBitContext *gb, Table *t, const int8_t code_pred_coeff[3][3], int length_bits, int coeff_bits, int is_signed, int offset)
+{
+ unsigned int i, j, k;
+ for (i = 0; i < t->elements; i++) {
+ t->length[i] = get_bits(gb, length_bits) + 1;
+ if (!get_bits1(gb)) {
+ read_uncoded_coeff(gb, t->coeff[i], t->length[i], coeff_bits, is_signed, offset);
+ } else {
+ int method = get_bits(gb, 2), lsb_size;
+ if (method == 3)
+ return AVERROR_INVALIDDATA;
+
+ read_uncoded_coeff(gb, t->coeff[i], method + 1, coeff_bits, is_signed, offset);
+
+ lsb_size = get_bits(gb, 3);
+ for (j = method + 1; j < t->length[i]; j++) {
+ int c, x = 0;
+ for (k = 0; k < method + 1; k++)
+ x += code_pred_coeff[method][k] * t->coeff[i][j - k - 1];
+ c = get_sr_golomb_dst(gb, lsb_size);
+ if (x >= 0)
+ c -= (x + 4) / 8;
+ else
+ c += (-x + 3) / 8;
+ t->coeff[i][j] = c;
+ }
+ }
+ }
+ return 0;
+}
+
+typedef struct {
+ unsigned int a;
+ unsigned int c;
+} Arith;
+
+static void ac_init(Arith * ac, GetBitContext *gb)
+{
+ ac->a = 4095;
+ ac->c = get_bits(gb, 12);
+}
+
+#define AC_GET(ac, re, gb, p, e) \
+{ \
+ unsigned int k = ((ac)->a >> 8) | (((ac)->a >> 7) & 1); \
+ unsigned int q = k * p; \
+ unsigned int a_q = (ac)->a - q; \
+ e = (ac)->c < a_q; \
+ if (e) \
+ (ac)->a = a_q; \
+ else { \
+ (ac)->a = q; \
+ (ac)->c -= a_q; \
+ } \
+ if ((ac)->a < 2048) { \
+ int n = 11 - av_log2((ac)->a); \
+ (ac)->a <<= n; \
+ (ac)->c = ((ac)->c << n) | SHOW_UBITS(re, gb, n); \
+ SKIP_BITS(re, pb, n); \
+ UPDATE_CACHE(re, gb); \
+ } \
+}
+
+static uint8_t prob_dst_x_bit(int c)
+{
+ return (ff_reverse[c & 127] >> 1) + 1;
+}
+
+static void build_filter(int16_t table[DST_MAX_ELEMENTS][16][256], const Table *fsets)
+{
+ int i, j, k, l;
+ for (i = 0; i < fsets->elements; i++) {
+ int length = fsets->length[i];
+ for (j = 0; j < 16; j++) {
+ int total = FFMAX(0, FFMIN(length - j * 8, 8));
+ for (k = 0; k < 256; k++) {
+ int v = 0;
+ for (l = 0; l < total; l++)
+ v += (((k >> l) & 1) * 2 - 1) * fsets->coeff[i][j * 8 + l];
+ table[i][j][k] = v;
+ }
+ }
+ }
+}
+
+static int decode_frame(AVCodecContext *avctx, void *data,
+ int *got_frame_ptr, AVPacket *avpkt)
+{
+ AVFrame *frame = data;
+ unsigned int samples_per_frame = DST_SAMPLES_PER_FRAME(avctx->sample_rate);
+ int ret;
+ unsigned int i, ch, same;
+ GetBitContext gb;
+ Arith ac;
+ Table fsets, probs;
+ unsigned int half_prob[DST_MAX_CHANNELS];
+ unsigned int map_ch_to_felem[DST_MAX_CHANNELS];
+ unsigned int map_ch_to_pelem[DST_MAX_CHANNELS];
+ DECLARE_ALIGNED(16, uint8_t, status)[DST_MAX_CHANNELS][16];
+ DECLARE_ALIGNED(16, int16_t, filter)[DST_MAX_ELEMENTS][16][256];
+
+ frame->nb_samples = samples_per_frame / 8;
+ if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
+ return ret;
+
+ if (!(avpkt->data[0] & 1)) {
+ if (frame->nb_samples > avpkt->size - 1)
+ av_log(avctx, AV_LOG_WARNING, "short frame");
+ memcpy(frame->data[0], avpkt->data + 1, FFMIN(frame->nb_samples * avctx->channels, avpkt->size - 1));
+ *got_frame_ptr = 1;
+ return avpkt->size;
+ }
+
+ if ((ret = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
+ return ret;
+
+ skip_bits1(&gb);
+
+ /* Segmentation (10.4, 10.5, 10.6) */
+
+ if (!get_bits1(&gb)){
+ avpriv_request_sample(avctx, "Same_Segmentation=0");
+ return AVERROR_PATCHWELCOME;
+ }
+
+ if (!get_bits1(&gb)){
+ avpriv_request_sample(avctx, "Same_Segm_For_All_Channels=0");
+ return AVERROR_PATCHWELCOME;
+ }
+
+ if (!get_bits1(&gb)){
+ avpriv_request_sample(avctx, "End_Of_Channel_Segm=0");
+ return AVERROR_PATCHWELCOME;
+ }
+
+ /* Mapping (10.7, 10.8, 10.9) */
+
+ same = get_bits1(&gb);
+
+ if ((ret = read_map(&gb, &fsets, map_ch_to_felem, avctx->channels)) < 0)
+ return ret;
+
+ if (same) {
+ probs.elements = fsets.elements;
+ memcpy(map_ch_to_pelem, map_ch_to_felem, sizeof(map_ch_to_felem));
+ } else {
+ avpriv_request_sample(avctx, "Same_Mapping=0");
+ if ((ret = read_map(&gb, &probs, map_ch_to_pelem, avctx->channels)) < 0)
+ return ret;
+ }
+
+ /* Half Probability (10.10) */
+
+ for (ch = 0; ch < avctx->channels; ch++)
+ half_prob[ch] = get_bits1(&gb);
+
+ /* Filter Coef Sets (10.12) */
+
+ read_table(&gb, &fsets, fsets_code_pred_coeff, 7, 9, 1, 0);
+
+ /* Probability Tables (10.13) */
+
+ read_table(&gb, &probs, probs_code_pred_coeff, 6, 7, 0, 1);
+
+ /* Arithmetic Coded Data (10.11) */
+
+ memset(frame->data[0], 0, avctx->channels * frame->nb_samples);
+
+ skip_bits1(&gb);
+ ac_init(&ac, &gb);
+
+ build_filter(filter, &fsets);
+ memset(status, 0xAA, sizeof(status));
+
+ {
+ unsigned int dst_x_bit;
+ OPEN_READER(re, &gb);
+ UPDATE_CACHE(re, &gb);
+ AC_GET(&ac, re, &gb, prob_dst_x_bit(fsets.coeff[0][0]), dst_x_bit);
+
+ for (i = 0; i < samples_per_frame; i++) {
+ for (ch = 0; ch < avctx->channels; ch++) {
+ unsigned int felem = map_ch_to_felem[ch], prob, residual, v;
+ uint64_t * s = (uint64_t*)status[ch];
+
+#define F(i) filter[felem][i][status[ch][i]]
+ int16_t predict = F( 0) + F( 1) + F( 2) + F( 3) +
+ F( 4) + F( 5) + F( 6) + F( 7) +
+ F( 8) + F( 9) + F(10) + F(11) +
+ F(12) + F(13) + F(14) + F(15);
+#undef F
+
+ if (!half_prob[ch] || i >= fsets.length[felem]) {
+ unsigned int pelem = map_ch_to_pelem[ch];
+ unsigned int index = FFABS(predict) >> 3;
+ prob = probs.coeff[pelem][FFMIN(index, probs.length[pelem] - 1)];
+ } else {
+ prob = 128;
+ }
+
+ AC_GET(&ac, re, &gb, prob, residual);
+ v = ((predict >> 15) ^ residual) & 1;
+ frame->data[0][ (i >> 3) * avctx->channels + ch] |= v << (7 - (i & 0x7 ));
+
+#if HAVE_BIGENDIAN
+ /* FIXME: not tested */
+ s[0] = (s[0] << 1) | ((s[1] >> 63) & 1);
+ s[1] = (s[1] << 1) | v;
+#else
+ s[1] = (s[1] << 1) | ((s[0] >> 63) & 1);
+ s[0] = (s[0] << 1) | v;
+#endif
+ }
+ }
+ }
+
+ *got_frame_ptr = 1;
+ return avpkt->size;
+}
+
+AVCodec ff_dst_decoder = {
+ .name = "dst",
+ .long_name = NULL_IF_CONFIG_SMALL("Digital Stream Transfer (DST)"),
+ .type = AVMEDIA_TYPE_AUDIO,
+ .id = AV_CODEC_ID_DST,
+ .init = decode_init,
+ .decode = decode_frame,
+ .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_DSD,
+ AV_SAMPLE_FMT_NONE },
+};
--
1.9.1
-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140930/e53edc6c/attachment.asc>
More information about the ffmpeg-devel
mailing list