lavc: add a native Opus decoder.

Initial implementation by Andrew D'Addesio <modchipv12@gmail.com> during
GSoC 2012.

Completion by Anton Khirnov <anton@khirnov.net>, sponsored by the
Mozilla Corporation.

Further contributions by:
Christophe Gisquet <christophe.gisquet@gmail.com>
Janne Grunau <janne-libav@jannau.net>
Luca Barbato <lu_zero@gentoo.org>

Changelog

@@ -25,6 +25,7 @@ version <next>:
 - On2 AVC (Audio for Video) decoder
 - support for decoding through DXVA2 in avconv
 - libbs2b-based stereo-to-binaural audio filter
+- native Opus decoder
 
 
 version 10:

configure

@@ -1822,6 +1822,7 @@ nellymoser_decoder_select="mdct sinewin"
 nellymoser_encoder_select="audio_frame_queue mdct sinewin"
 nuv_decoder_select="dsputil lzo"
 on2avc_decoder_select="mdct"
+opus_decoder_deps="avresample"
 png_decoder_deps="zlib"
 png_encoder_deps="zlib"
 png_encoder_select="dsputil"
@@ -4387,6 +4388,8 @@ enabled movie_filter    && prepend avfilter_deps "avformat avcodec"
 enabled resample_filter && prepend avfilter_deps "avresample"
 enabled scale_filter    && prepend avfilter_deps "swscale"
 
+enabled opus_decoder    && prepend avcodec_deps "avresample"
+
 expand_deps(){
     lib_deps=${1}_deps
     eval "deps=\$$lib_deps"

libavcodec/Makefile

@@ -279,6 +279,9 @@ OBJS-$(CONFIG_NELLYMOSER_DECODER)      += nellymoserdec.o nellymoser.o
 OBJS-$(CONFIG_NELLYMOSER_ENCODER)      += nellymoserenc.o nellymoser.o
 OBJS-$(CONFIG_NUV_DECODER)             += nuv.o rtjpeg.o
 OBJS-$(CONFIG_ON2AVC_DECODER)          += on2avc.o on2avcdata.o
+OBJS-$(CONFIG_OPUS_DECODER)            += opusdec.o opus.o opus_celt.o \
+                                          opus_imdct.o opus_silk.o     \
+                                          vorbis_data.o
 OBJS-$(CONFIG_PAF_VIDEO_DECODER)       += paf.o
 OBJS-$(CONFIG_PAF_AUDIO_DECODER)       += paf.o
 OBJS-$(CONFIG_PAM_DECODER)             += pnmdec.o pnm.o
@@ -653,6 +656,7 @@ OBJS-$(CONFIG_MPEGAUDIO_PARSER)        += mpegaudio_parser.o \
                                           mpegaudiodecheader.o mpegaudiodata.o
 OBJS-$(CONFIG_MPEGVIDEO_PARSER)        += mpegvideo_parser.o    \
                                           mpeg12.o mpeg12data.o
+OBJS-$(CONFIG_OPUS_PARSER)             += opus_parser.o opus.o vorbis_data.o
 OBJS-$(CONFIG_PNG_PARSER)              += png_parser.o
 OBJS-$(CONFIG_PNM_PARSER)              += pnm_parser.o pnm.o
 OBJS-$(CONFIG_RV30_PARSER)             += rv34_parser.o

libavcodec/allcodecs.c

@@ -331,6 +331,7 @@ void avcodec_register_all(void)
     REGISTER_DECODER(MPC8,              mpc8);
     REGISTER_ENCDEC (NELLYMOSER,        nellymoser);
     REGISTER_DECODER(ON2AVC,            on2avc);
+    REGISTER_DECODER(OPUS,              opus);
     REGISTER_DECODER(PAF_AUDIO,         paf_audio);
     REGISTER_DECODER(QCELP,             qcelp);
     REGISTER_DECODER(QDM2,              qdm2);
@@ -483,6 +484,7 @@ void avcodec_register_all(void)
     REGISTER_PARSER(MPEG4VIDEO,         mpeg4video);
     REGISTER_PARSER(MPEGAUDIO,          mpegaudio);
     REGISTER_PARSER(MPEGVIDEO,          mpegvideo);
+    REGISTER_PARSER(OPUS,               opus);
     REGISTER_PARSER(PNG,                png);
     REGISTER_PARSER(PNM,                pnm);
     REGISTER_PARSER(RV30,               rv30);

libavcodec/opus_imdct.c

+/*
+ * Copyright (c) 2013-2014 Mozilla Corporation
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Celt non-power of 2 iMDCT
+ */
+
+#include <float.h>
+#include <math.h>
+
+#include "libavutil/attributes.h"
+#include "libavutil/common.h"
+
+#include "fft.h"
+#include "opus.h"
+
+// complex c = a * b
+#define CMUL3(cre, cim, are, aim, bre, bim)          \
+do {                                                 \
+    cre = are * bre - aim * bim;                     \
+    cim = are * bim + aim * bre;                     \
+} while (0)
+
+#define CMUL(c, a, b) CMUL3((c).re, (c).im, (a).re, (a).im, (b).re, (b).im)
+
+// complex c = a * b
+//         d = a * conjugate(b)
+#define CMUL2(c, d, a, b)                            \
+do {                                                 \
+    float are = (a).re;                              \
+    float aim = (a).im;                              \
+    float bre = (b).re;                              \
+    float bim = (b).im;                              \
+    float rr  = are * bre;                           \
+    float ri  = are * bim;                           \
+    float ir  = aim * bre;                           \
+    float ii  = aim * bim;                           \
+    (c).re =  rr - ii;                               \
+    (c).im =  ri + ir;                               \
+    (d).re =  rr + ii;                               \
+    (d).im = -ri + ir;                               \
+} while (0)
+
+struct CeltIMDCTContext {
+    int fft_n;
+    int len2;
+    int len4;
+
+    FFTComplex *tmp;
+
+    FFTComplex *twiddle_exptab;
+
+    FFTComplex *exptab[6];
+};
+
+av_cold void ff_celt_imdct_uninit(CeltIMDCTContext **ps)
+{
+    CeltIMDCTContext *s = *ps;
+    int i;
+
+    if (!s)
+        return;
+
+    for (i = 0; i < FF_ARRAY_ELEMS(s->exptab); i++)
+        av_freep(&s->exptab[i]);
+
+    av_freep(&s->twiddle_exptab);
+
+    av_freep(&s->tmp);
+
+    av_freep(ps);
+}
+
+av_cold int ff_celt_imdct_init(CeltIMDCTContext **ps, int N)
+{
+    CeltIMDCTContext *s;
+    int len2 = 15 * (1 << N);
+    int len  = 2 * len2;
+    int i, j;
+
+    if (len2 > CELT_MAX_FRAME_SIZE)
+        return AVERROR(EINVAL);
+
+    s = av_mallocz(sizeof(*s));
+    if (!s)
+        return AVERROR(ENOMEM);
+
+    s->fft_n = N - 1;
+    s->len4 = len2 / 2;
+    s->len2 = len2;
+
+    s->tmp  = av_malloc(len * 2 * sizeof(*s->tmp));
+    if (!s->tmp)
+        goto fail;
+
+    s->twiddle_exptab  = av_malloc(s->len4 * sizeof(*s->twiddle_exptab));
+    if (!s->twiddle_exptab)
+        goto fail;
+
+    for (i = 0; i < s->len4; i++) {
+        s->twiddle_exptab[i].re = cos(2 * M_PI * (i + 0.125 + s->len4) / len);
+        s->twiddle_exptab[i].im = sin(2 * M_PI * (i + 0.125 + s->len4) / len);
+    }
+
+    for (i = 0; i < FF_ARRAY_ELEMS(s->exptab); i++) {
+        int N = 15 * (1 << i);
+        s->exptab[i] = av_malloc(sizeof(*s->exptab[i]) * FFMAX(N, 19));
+        if (!s->exptab[i])
+            goto fail;
+
+        for (j = 0; j < N; j++) {
+            s->exptab[i][j].re = cos(2 * M_PI * j / N);
+            s->exptab[i][j].im = sin(2 * M_PI * j / N);
+        }
+    }
+
+    // wrap around to simplify fft15
+    for (j = 15; j < 19; j++)
+        s->exptab[0][j] = s->exptab[0][j - 15];
+
+    *ps = s;
+
+    return 0;
+fail:
+    ff_celt_imdct_uninit(&s);
+    return AVERROR(ENOMEM);
+}
+
+static void fft5(FFTComplex *out, const FFTComplex *in, int stride)
+{
+    // [0] = exp(2 * i * pi / 5), [1] = exp(2 * i * pi * 2 / 5)
+    static const FFTComplex fact[] = { { 0.30901699437494745,  0.95105651629515353 },
+                                       { -0.80901699437494734, 0.58778525229247325 } };
+
+    FFTComplex z[4][4];
+
+    CMUL2(z[0][0], z[0][3], in[1 * stride], fact[0]);
+    CMUL2(z[0][1], z[0][2], in[1 * stride], fact[1]);
+    CMUL2(z[1][0], z[1][3], in[2 * stride], fact[0]);
+    CMUL2(z[1][1], z[1][2], in[2 * stride], fact[1]);
+    CMUL2(z[2][0], z[2][3], in[3 * stride], fact[0]);
+    CMUL2(z[2][1], z[2][2], in[3 * stride], fact[1]);
+    CMUL2(z[3][0], z[3][3], in[4 * stride], fact[0]);
+    CMUL2(z[3][1], z[3][2], in[4 * stride], fact[1]);
+
+    out[0].re = in[0].re + in[stride].re + in[2 * stride].re + in[3 * stride].re + in[4 * stride].re;
+    out[0].im = in[0].im + in[stride].im + in[2 * stride].im + in[3 * stride].im + in[4 * stride].im;
+
+    out[1].re = in[0].re + z[0][0].re + z[1][1].re + z[2][2].re + z[3][3].re;
+    out[1].im = in[0].im + z[0][0].im + z[1][1].im + z[2][2].im + z[3][3].im;
+
+    out[2].re = in[0].re + z[0][1].re + z[1][3].re + z[2][0].re + z[3][2].re;
+    out[2].im = in[0].im + z[0][1].im + z[1][3].im + z[2][0].im + z[3][2].im;
+
+    out[3].re = in[0].re + z[0][2].re + z[1][0].re + z[2][3].re + z[3][1].re;
+    out[3].im = in[0].im + z[0][2].im + z[1][0].im + z[2][3].im + z[3][1].im;
+
+    out[4].re = in[0].re + z[0][3].re + z[1][2].re + z[2][1].re + z[3][0].re;
+    out[4].im = in[0].im + z[0][3].im + z[1][2].im + z[2][1].im + z[3][0].im;
+}
+
+static void fft15(CeltIMDCTContext *s, FFTComplex *out, const FFTComplex *in, int stride)
+{
+    const FFTComplex *exptab = s->exptab[0];
+    FFTComplex tmp[5];
+    FFTComplex tmp1[5];
+    FFTComplex tmp2[5];
+    int k;
+
+    fft5(tmp,  in,              stride * 3);
+    fft5(tmp1, in +     stride, stride * 3);
+    fft5(tmp2, in + 2 * stride, stride * 3);
+
+    for (k = 0; k < 5; k++) {
+        FFTComplex t1, t2;
+
+        CMUL(t1, tmp1[k], exptab[k]);
+        CMUL(t2, tmp2[k], exptab[2 * k]);
+        out[k].re = tmp[k].re + t1.re + t2.re;
+        out[k].im = tmp[k].im + t1.im + t2.im;
+
+        CMUL(t1, tmp1[k], exptab[k + 5]);
+        CMUL(t2, tmp2[k], exptab[2 * (k + 5)]);
+        out[k + 5].re = tmp[k].re + t1.re + t2.re;
+        out[k + 5].im = tmp[k].im + t1.im + t2.im;
+
+        CMUL(t1, tmp1[k], exptab[k + 10]);
+        CMUL(t2, tmp2[k], exptab[2 * k + 5]);
+        out[k + 10].re = tmp[k].re + t1.re + t2.re;
+        out[k + 10].im = tmp[k].im + t1.im + t2.im;
+    }
+}
+
+/*
+ * FFT of the length 15 * (2^N)
+ */
+static void fft_calc(CeltIMDCTContext *s, FFTComplex *out, const FFTComplex *in, int N, int stride)
+{
+    if (N) {
+        const FFTComplex *exptab = s->exptab[N];
+        const int len2 = 15 * (1 << (N - 1));
+        int k;
+
+        fft_calc(s, out,        in,          N - 1, stride * 2);
+        fft_calc(s, out + len2, in + stride, N - 1, stride * 2);
+
+        for (k = 0; k < len2; k++) {
+            FFTComplex t;
+
+            CMUL(t, out[len2 + k], exptab[k]);
+
+            out[len2 + k].re = out[k].re - t.re;
+            out[len2 + k].im = out[k].im - t.im;
+
+            out[k].re += t.re;
+            out[k].im += t.im;
+        }
+    } else
+        fft15(s, out, in, stride);
+}
+
+void ff_celt_imdct_half(CeltIMDCTContext *s, float *dst, const float *src,
+                        int stride, float scale)
+{
+    FFTComplex *z = (FFTComplex *)dst;
+    const int len8 = s->len4 / 2;
+    const float *in1 = src;
+    const float *in2 = src + (s->len2 - 1) * stride;
+    int i;
+
+    for (i = 0; i < s->len4; i++) {
+        FFTComplex tmp = { *in2, *in1 };
+        CMUL(s->tmp[i], tmp, s->twiddle_exptab[i]);
+        in1 += 2 * stride;
+        in2 -= 2 * stride;
+    }
+
+    fft_calc(s, z, s->tmp, s->fft_n, 1);
+
+    for (i = 0; i < len8; i++) {
+        float r0, i0, r1, i1;
+
+        CMUL3(r0, i1, z[len8 - i - 1].im, z[len8 - i - 1].re,  s->twiddle_exptab[len8 - i - 1].im, s->twiddle_exptab[len8 - i - 1].re);
+        CMUL3(r1, i0, z[len8 + i].im,     z[len8 + i].re,      s->twiddle_exptab[len8 + i].im,     s->twiddle_exptab[len8 + i].re);
+        z[len8 - i - 1].re = scale * r0;
+        z[len8 - i - 1].im = scale * i0;
+        z[len8 + i].re     = scale * r1;
+        z[len8 + i].im     = scale * i1;
+    }
+}

libavcodec/opus_parser.c

+/*
+ * Copyright (c) 2013-2014 Mozilla Corporation
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Opus parser
+ *
+ * Determines the duration for each packet.
+ */
+
+#include "avcodec.h"
+#include "opus.h"
+
+typedef struct OpusParseContext {
+    OpusContext ctx;
+    OpusPacket pkt;
+    int extradata_parsed;
+} OpusParseContext;
+
+static int opus_parse(AVCodecParserContext *ctx, AVCodecContext *avctx,
+                      const uint8_t **poutbuf, int *poutbuf_size,
+                      const uint8_t *buf, int buf_size)
+{
+    OpusParseContext *s = ctx->priv_data;
+    int ret;
+
+    if (!buf_size)
+        return 0;
+
+    if (avctx->extradata && !s->extradata_parsed) {
+        ret = ff_opus_parse_extradata(avctx, &s->ctx);
+        if (ret < 0) {
+            av_log(avctx, AV_LOG_ERROR, "Error parsing Ogg extradata.\n");
+            goto fail;
+        }
+        av_freep(&s->ctx.channel_maps);
+        s->extradata_parsed = 1;
+    }
+
+    ret = ff_opus_parse_packet(&s->pkt, buf, buf_size, s->ctx.nb_streams > 1);
+    if (ret < 0) {
+        av_log(avctx, AV_LOG_ERROR, "Error parsing Opus packet header.\n");
+        goto fail;
+    }
+
+    ctx->duration = s->pkt.frame_count * s->pkt.frame_duration;
+
+fail:
+    *poutbuf = buf;
+    *poutbuf_size = buf_size;
+    return buf_size;
+}
+
+AVCodecParser ff_opus_parser = {
+    .codec_ids      = { AV_CODEC_ID_OPUS },
+    .priv_data_size = sizeof(OpusParseContext),
+    .parser_parse   = opus_parse,
+};

libavcodec/version.h

@@ -29,8 +29,8 @@
 #include "libavutil/version.h"
 
 #define LIBAVCODEC_VERSION_MAJOR 55
-#define LIBAVCODEC_VERSION_MINOR 50
-#define LIBAVCODEC_VERSION_MICRO  3
+#define LIBAVCODEC_VERSION_MINOR 51
+#define LIBAVCODEC_VERSION_MICRO  0
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \

tests/Makefile

@@ -93,6 +93,7 @@ include $(SRC_PATH)/tests/fate/microsoft.mak
 include $(SRC_PATH)/tests/fate/monkeysaudio.mak
 include $(SRC_PATH)/tests/fate/mp3.mak
 include $(SRC_PATH)/tests/fate/mpc.mak
+include $(SRC_PATH)/tests/fate/opus.mak
 include $(SRC_PATH)/tests/fate/pcm.mak
 include $(SRC_PATH)/tests/fate/probe.mak
 include $(SRC_PATH)/tests/fate/prores.mak

tests/fate/opus.mak

+# The samples were produced by simply rewrapping the official test vectors from
+# their custom format into Matroska.
+# The reference files were created with our decoder and tested against the
+# libopus output with the official opus_compare tool. We cannot use libopus
+# output as reference directly, because the use of different resamplers would
+# require too high fuzz values, which can hide bugs.
+# Before adding new tests here, always make sure they pass opus_compare.
+
+OPUS_CELT_SAMPLES   = $(addprefix testvector, 01 07 11) tron.6ch.tinypkts
+OPUS_HYBRID_SAMPLES = $(addprefix testvector, 05 06)
+OPUS_SILK_SAMPLES   = $(addprefix testvector, 02 03 04)
+OPUS_SAMPLES        = $(addprefix testvector, 08 09 10 12)
+
+define FATE_OPUS_TEST
+FATE_OPUS     += fate-opus-$(1)
+FATE_OPUS$(2) += fate-opus-$(1)
+fate-opus-$(1): CMD = avconv -i $(TARGET_SAMPLES)/opus/$(1).mka -f f32le -
+fate-opus-$(1): REF = $(TARGET_SAMPLES)/opus/$(1).f32
+endef
+
+$(foreach N,$(OPUS_CELT_SAMPLES),  $(eval $(call FATE_OPUS_TEST,$(N),_CELT)))
+$(foreach N,$(OPUS_HYBRID_SAMPLES),$(eval $(call FATE_OPUS_TEST,$(N),_HYBRID)))
+$(foreach N,$(OPUS_SILK_SAMPLES),  $(eval $(call FATE_OPUS_TEST,$(N),_SILK)))
+$(foreach N,$(OPUS_SAMPLES),       $(eval $(call FATE_OPUS_TEST,$(N),)))
+
+FATE_OPUS := $(sort $(FATE_OPUS))
+
+$(FATE_OPUS): CMP = stddev
+$(FATE_OPUS): CMP_UNIT = f32
+$(FATE_OPUS): FUZZ = 3
+
+$(FATE_OPUS_CELT): CMP = oneoff
+$(FATE_OPUS_CELT): FUZZ = 5
+
+FATE_SAMPLES_AVCONV-$(call DEMDEC, MATROSKA, OPUS) += $(FATE_OPUS)
+fate-opus-celt: $(FATE_OPUS_CELT)
+fate-opus-hybrid: $(FATE_OPUS_HYBRID)
+fate-opus-silk: $(FATE_OPUS_SILK)
+fate-opus: $(FATE_OPUS)