On Wed, Apr 16, 2014 at 04:13:19PM +0200, Luca Barbato wrote:
> From: Peter Ross <[email protected]>
> 
> Signed-off-by: Peter Ross <[email protected]>
> Signed-off-by: Michael Niedermayer <[email protected]>
> Signed-off-by: Jean-Baptiste Kempf <[email protected]>
> Signed-off-by: Luca Barbato <[email protected]>
> ---
> 
> The tests are missing, Vittorio willing to bake some?
> 
>  Changelog                 |   1 +
>  doc/general.texi          |   4 +
>  libavcodec/Makefile       |   8 +-
>  libavcodec/allcodecs.c    |   4 +
>  libavcodec/avcodec.h      |   4 +
>  libavcodec/codec_desc.c   |  28 +++++++
>  libavcodec/dsd_tablegen.c |  38 ++++++++++
>  libavcodec/dsd_tablegen.h |  95 ++++++++++++++++++++++++
>  libavcodec/dsddec.c       | 185 
> ++++++++++++++++++++++++++++++++++++++++++++++
>  libavcodec/utils.c        |   4 +
>  10 files changed, 370 insertions(+), 1 deletion(-)
>  create mode 100644 libavcodec/dsd_tablegen.c
>  create mode 100644 libavcodec/dsd_tablegen.h
>  create mode 100644 libavcodec/dsddec.c
> 
> diff --git a/Changelog b/Changelog
> index 55216e8..d895b94 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -19,6 +19,7 @@ version <next>:
>  - LucasArts SMUSH demuxer
>  - MP2 encoding via TwoLAME
>  - asettb filter
> +- Direct Stream Digital (DSD) decoder
> 
> 
>  version 10:
> diff --git a/doc/general.texi b/doc/general.texi
> index 2612eb1..4b16075 100644
> --- a/doc/general.texi
> +++ b/doc/general.texi
> @@ -822,6 +822,10 @@ following image formats are supported:
>  @item DPCM Sol               @tab     @tab  X
>  @item DPCM Xan               @tab     @tab  X
>      @tab Used in Origin's Wing Commander IV AVI files.
> +@item DSD (Direct Stream Digitial), least significant bit first  @tab  @tab  
> X
> +@item DSD (Direct Stream Digitial), most significant bit first   @tab  @tab  
> X
> +@item DSD (Direct Stream Digitial), least significant bit first, planar  
> @tab  @tab  X
> +@item DSD (Direct Stream Digitial), most significant bit first, planar   
> @tab  @tab  X

one item should be enough

>  @item DSP Group TrueSpeech   @tab     @tab  X
>  @item DV audio               @tab     @tab  X
>  @item Enhanced AC-3          @tab  X  @tab  X
[...]
> diff --git a/libavcodec/dsd_tablegen.h b/libavcodec/dsd_tablegen.h
> new file mode 100644
> index 0000000..0bebaea
> --- /dev/null
> +++ b/libavcodec/dsd_tablegen.h
> @@ -0,0 +1,95 @@
> +/*
> + * Header file for hardcoded DSD tables
> + * based on BSD licensed dsd2pcm by Sebastian Gesemann
> + * Copyright (c) 2009, 2011 Sebastian Gesemann. All rights reserved.
> + *
> + * 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
> + */
> +
> +#ifndef AVCODEC_DSD_TABLEGEN_H
> +#define AVCODEC_DSD_TABLEGEN_H
> +
> +#include <stdint.h>
> +#include "libavutil/attributes.h"
> +
> +#define HTAPS   48                /** number of FIR constants */
> +#define CTABLES ((HTAPS + 7) / 8) /** number of "8 MACs" lookup tables */
> +
> +#if CONFIG_HARDCODED_TABLES
> +#define dsd_ctables_tableinit()
> +#include "libavcodec/dsd_tables.h"
> +#else
> +#include "libavutil/common.h"
> +
> +/*
> + * Properties of this 96-tap lowpass filter when applied on a signal
> + * with sampling rate of 44100*64 Hz:
> + *
> + * () has a delay of 17 microseconds.

()?

> + *
> + * () flat response up to 48 kHz
> + *
> + * () if you downsample afterwards by a factor of 8, the
> + *    spectrum below 70 kHz is practically alias-free.
> + *
> + * () stopband rejection is about 160 dB
> + *
> + * The coefficient tables ("ctables") take only 6 Kibi Bytes and

barf

> + * should fit into a modern processor's fast cache.
> + */
> +
> +/**
> + * The 2nd half (48 coeffs) of a 96-tap symmetric lowpass filter
> + */
> +static const double htaps[HTAPS] = {
> +     0.09950731974056658,    0.09562845727714668,    0.08819647126516944,
> +     0.07782552527068175,    0.06534876523171299,    0.05172629311427257,
> +     0.0379429484910187,     0.02490921351762261,    0.0133774746265897,
> +     0.003883043418804416,  -0.003284703416210726,  -0.008080250212687497,
> +    -0.01067241812471033,   -0.01139427235000863,   -0.0106813877974587,
> +    -0.009007905078766049,  -0.006828859761015335,  -0.004535184322001496,
> +    -0.002425035959059578,  -0.0006922187080790708,  0.0005700762133516592,
> +     0.001353838005269448,   0.001713709169690937,   0.001742046839472948,
> +     0.001545601648013235,   0.001226696225277855,   0.0008704322683580222,
> +     0.0005381636200535649,  0.000266446345425276,   7.002968738383528e-05,
> +    -5.279407053811266e-05, -0.0001140625650874684, -0.0001304796361231895,
> +    -0.0001189970287491285, -9.396247155265073e-05, -6.577634378272832e-05,
> +    -4.07492895872535e-05,  -2.17407957554587e-05,  -9.163058931391722e-06,
> +    -2.017460145032201e-06,  1.249721855219005e-06,  2.166655190537392e-06,
> +     1.930520892991082e-06,  1.319400334374195e-06,  7.410039764949091e-07,
> +     3.423230509967409e-07,  1.244182214744588e-07,  3.130441005359396e-08
> +};
> +
> +static float ctables[CTABLES][256];
> +
> +static av_cold void dsd_ctables_tableinit(void)
> +{
> +    int t, e, m, k;
> +    double acc;
> +    for (t = 0; t < CTABLES; ++t) {
> +        k = FFMIN(HTAPS - t * 8, 8);

unless one decides to switch to non-multiple-of-8 tap filter k=8 in all cases
(otherwise one can pad filter anyway and see above)

> +        for (e = 0; e < 256; ++e) {
> +            acc = 0.0;
> +            for (m = 0; m < k; ++m)
> +                acc += (((e >> (7 - m)) & 1) * 2 - 1) * htaps[t * 8 + m];
> +            ctables[CTABLES - 1 - t][e] = (float)acc;
> +        }
> +    }
> +}
> +#endif /* CONFIG_HARDCODED_TABLES */
> +
> +#endif /* AVCODEC_DSD_TABLEGEN_H */
> diff --git a/libavcodec/dsddec.c b/libavcodec/dsddec.c
> new file mode 100644
> index 0000000..6876f16
> --- /dev/null
> +++ b/libavcodec/dsddec.c
> @@ -0,0 +1,185 @@
> +/*
> + * Direct Stream Digital (DSD) decoder
> + * based on BSD licensed dsd2pcm by Sebastian Gesemann
> + * Copyright (c) 2009, 2011 Sebastian Gesemann. All rights reserved.
> + * Copyright (c) 2014 Peter Ross
> + *
> + * 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
> + * Direct Stream Digital (DSD) decoder
> + */
> +
> +#include "libavcodec/internal.h"
> +#include "libavcodec/mathops.h"
> +#include "avcodec.h"
> +#include "dsd_tablegen.h"
> +
> +#define FIFOSIZE 16              /** must be a power of two */
> +#define FIFOMASK (FIFOSIZE - 1)  /** bit mask for FIFO offsets */
> +
> +#if FIFOSIZE * 8 < HTAPS * 2
> +#error "FIFOSIZE too small"
> +#endif
> +
> +/*
> + * Per-channel buffer
> + */
> +typedef struct {
> +    unsigned char buf[FIFOSIZE];

uint8_t?

> +    unsigned pos;
> +} DSDContext;
> +
> +static void dsd2pcm_translate(DSDContext* s, size_t samples, int lsbf,
> +                              const uint8_t *src, ptrdiff_t src_stride,
> +                              float *dst, ptrdiff_t dst_stride)
> +{
> +    unsigned int pos, i;
> +    uint8_t *p;
> +    double sum;
> +
> +    pos = s->pos;
> +
> +    while (samples-- > 0) {

while (samples--)

it's more honest here

> +        s->buf[pos] = lsbf ? ff_reverse[*src] : *src;
> +        src += src_stride;
> +
> +        p = s->buf + ((pos - CTABLES) & FIFOMASK);
> +        *p = ff_reverse[*p];
> +
> +        sum = 0.0;
> +        for (i = 0; i < CTABLES; i++) {
> +            const uint8_t a = s->buf[(pos                     - i) & 
> FIFOMASK];
> +            const uint8_t b = s->buf[(pos - (CTABLES * 2 - 1) + i) & 
> FIFOMASK];
> +            sum += ctables[i][a] + ctables[i][b];
> +        }
> +
> +        *dst = (float)sum;
> +        dst += dst_stride;
> +
> +        pos = (pos + 1) & FIFOMASK;
> +    }
> +
> +    s->pos = pos;
> +}
> +
> +static av_cold void dsd_init_static_data(AVCodec *unused)
> +{
> +    dsd_ctables_tableinit();
> +}
> +
> +static av_cold int dsd_decode_init(AVCodecContext *avctx)
> +{
> +    DSDContext * s;
> +    int i;
> +
> +    s = av_malloc(sizeof(DSDContext) * avctx->channels);
> +    if (!s)
> +        return AVERROR(ENOMEM);
> +
> +    for (i = 0; i < avctx->channels; i++) {
> +        s[i].pos = 0;
> +        memset(s[i].buf, 0x69, sizeof(s[i].buf));
> +
> +        /* 0x69 = 01101001
> +         * This pattern "on repeat" makes a low energy 352.8 kHz tone
> +         * and a high energy 1.0584 MHz tone which should be filtered
> +         * out completely by any playback system --> silence
> +         */
> +    }
> +
> +    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
> +    avctx->priv_data  = s;
> +    return 0;
> +}
> +
> +static av_always_inline
> +int dsd_decode_frame_internal(AVCodecContext *avctx, void *data,
> +                              int *got_frame_ptr, AVPacket *avpkt,
> +                              int lsbf, int planar)
> +{
> +    DSDContext * s = avctx->priv_data;
> +    AVFrame *frame = data;
> +    int ret, i;
> +    int src_next;
> +    int src_stride;
> +
> +    frame->nb_samples = avpkt->size / avctx->channels;
> +
> +    if (planar) {
> +        src_next   = frame->nb_samples;
> +        src_stride = 1;
> +    } else {
> +        src_next   = 1;
> +        src_stride = avctx->channels;
> +    }
> +
> +    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
> +        return ret;
> +
> +    for (i = 0; i < avctx->channels; i++) {
> +        float *dst = ((float **)frame->extended_data)[i];
> +        dsd2pcm_translate(&s[i], frame->nb_samples, lsbf,
> +                          avpkt->data + i * src_next, src_stride,
> +                          dst, 1);
> +    }
> +
> +    *got_frame_ptr = 1;
> +    return frame->nb_samples * avctx->channels;
> +}
> +
> +static int dsd_lsbf_decode_frame(AVCodecContext *avctx, void *data,
> +                                 int *got_frame, AVPacket *avpkt)
> +{
> +    return dsd_decode_frame_internal(avctx, data, got_frame, avpkt, 1, 0);
> +}

empty line missing

> +static int dsd_msbf_decode_frame(AVCodecContext *avctx, void *data,
> +                                 int *got_frame, AVPacket *avpkt)
> +{
> +    return dsd_decode_frame_internal(avctx, data, got_frame, avpkt, 0, 0);
> +}
> +
> +static int dsd_lsbf_planar_decode_frame(AVCodecContext *avctx, void *data,
> +                                        int *got_frame, AVPacket *avpkt)
> +{
> +    return dsd_decode_frame_internal(avctx, data, got_frame, avpkt, 1, 1);
> +}
> +static int dsd_msbf_planar_decode_frame(AVCodecContext *avctx, void *data,
> +                                        int *got_frame, AVPacket *avpkt)
> +{
> +    return dsd_decode_frame_internal(avctx, data, got_frame, avpkt, 0, 1);
> +}
> +
> +#define DSD_DECODER(id_, name_, long_name_) \
> +AVCodec ff_##name_##_decoder = { \
> +    .name             = #name_, \
> +    .long_name        = NULL_IF_CONFIG_SMALL(long_name_), \
> +    .type             = AVMEDIA_TYPE_AUDIO, \
> +    .id               = AV_CODEC_ID_##id_, \
> +    .init             = dsd_decode_init, \
> +    .decode           = name_ ## _decode_frame, \
> +    .init_static_data = dsd_init_static_data, \
> +    .sample_fmts      = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP, \
> +                                                       AV_SAMPLE_FMT_NONE }, 
> \
> +};
> +
> +DSD_DECODER(DSD_LSBF, dsd_lsbf, "DSD (Direct Stream Digital), least 
> significant bit first")
> +DSD_DECODER(DSD_MSBF, dsd_msbf, "DSD (Direct Stream Digital), most 
> significant bit first")
> +DSD_DECODER(DSD_MSBF_PLANAR, dsd_msbf_planar, "DSD (Direct Stream Digital), 
> most significant bit first, planar")
> +DSD_DECODER(DSD_LSBF_PLANAR, dsd_lsbf_planar, "DSD (Direct Stream Digital), 
> least significant bit first, planar")

I wonder if it makes sense to contract it to "LSB/MSB first"

> diff --git a/libavcodec/utils.c b/libavcodec/utils.c
> index d9832e2..38a138b 100644
> --- a/libavcodec/utils.c
> +++ b/libavcodec/utils.c
> @@ -1929,6 +1929,10 @@ int av_get_exact_bits_per_sample(enum AVCodecID 
> codec_id)
>      case AV_CODEC_ID_ADPCM_G722:
>      case AV_CODEC_ID_ADPCM_YAMAHA:
>          return 4;
> +    case AV_CODEC_ID_DSD_LSBF:
> +    case AV_CODEC_ID_DSD_MSBF:
> +    case AV_CODEC_ID_DSD_LSBF_PLANAR:
> +    case AV_CODEC_ID_DSD_MSBF_PLANAR:
>      case AV_CODEC_ID_PCM_ALAW:
>      case AV_CODEC_ID_PCM_MULAW:
>      case AV_CODEC_ID_PCM_S8:
> --

in general LGTM
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to