From mboxrd@z Thu Jan 1 00:00:00 1970
Return-Path: <ffmpeg-devel-bounces@ffmpeg.org>
Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100])
by master.gitmailbox.com (Postfix) with ESMTPS id 8DE7144B5E
for <ffmpegdev@gitmailbox.com>; Sun, 23 Mar 2025 17:32:10 +0000 (UTC)
Received: from [127.0.1.1] (localhost [127.0.0.1])
by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 97FFB687C7C;
Sun, 23 Mar 2025 19:32:06 +0200 (EET)
Received: from vidala.pars.ee (vidala.pars.ee [116.203.72.101])
by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 3FABE687C01
for <ffmpeg-devel@ffmpeg.org>; Sun, 23 Mar 2025 19:32:00 +0200 (EET)
DKIM-Signature: v=1; a=rsa-sha256; s=202405r; d=lynne.ee; c=relaxed/relaxed;
h=From:To:Subject:Date:Message-ID; t=1742751119; bh=TYdyeWnutBLB5O1iCEuhNda
Sn5VKwWTTrp8hQsZRiXw=; b=lgkKOlj2Qxe/1swNfq/8BaZYN2Gh3OfoScUXeZWRZoBW50yVzp
WxkpmOHjikvNvqLjp0DKyUjDR5+yBQXtMtMuPu7Qqfci9TAzPsi0oYQ0DNzAXBmGAtNldELQeRh
ZoX5afxNBGK/Km6qx+rcJ0FB7CznQhmX6kwHyl9WhHsKMPDryAEc/6xZuengLR8Kxr4gpoBrvJk
0HgNMtzyp5oqE9DzR0j137+6lKfWzYR4Cjj+PnSRT6QI7olz6HAmGhZuCHkHiO05abMiIdg3eGb
R7sAQ54wW7Am8nNI88Q0Un5DEdk6XpRVfUiRsjXiVf1HDap93htJ/EWV+AsU4rjHLDA==;
DKIM-Signature: v=1; a=ed25519-sha256; s=202405e; d=lynne.ee; c=relaxed/relaxed;
h=From:To:Subject:Date:Message-ID; t=1742751119; bh=TYdyeWnutBLB5O1iCEuhNda
Sn5VKwWTTrp8hQsZRiXw=; b=Xp9amNC1Bom8AU+OrEG9lS39HqoU/wlxwOjOzyX8Wo6e6e0Zjf
hCdBGmnL2N3gIgKuweZ4gu88v6egzn4ovkDg==;
Message-ID: <70a141e1-f3f0-45ec-939e-990fe6e80400@lynne.ee>
Date: Sun, 23 Mar 2025 18:31:58 +0100
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
To: ffmpeg-devel@ffmpeg.org
References: <20250322174949.531092-1-eynard.massimo@gmail.com>
Content-Language: en-US
From: Lynne <dev@lynne.ee>
In-Reply-To: <20250322174949.531092-1-eynard.massimo@gmail.com>
Subject: Re: [FFmpeg-devel] [PATCH] avcodec/mlpdec: Add decoding of object
audio data
X-BeenThere: ffmpeg-devel@ffmpeg.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: FFmpeg development discussions and patches <ffmpeg-devel.ffmpeg.org>
List-Unsubscribe: <https://ffmpeg.org/mailman/options/ffmpeg-devel>,
<mailto:ffmpeg-devel-request@ffmpeg.org?subject=unsubscribe>
List-Archive: <https://ffmpeg.org/pipermail/ffmpeg-devel>
List-Post: <mailto:ffmpeg-devel@ffmpeg.org>
List-Help: <mailto:ffmpeg-devel-request@ffmpeg.org?subject=help>
List-Subscribe: <https://ffmpeg.org/mailman/listinfo/ffmpeg-devel>,
<mailto:ffmpeg-devel-request@ffmpeg.org?subject=subscribe>
Reply-To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset="us-ascii"; Format="flowed"
Errors-To: ffmpeg-devel-bounces@ffmpeg.org
Sender: "ffmpeg-devel" <ffmpeg-devel-bounces@ffmpeg.org>
Archived-At: <https://master.gitmailbox.com/ffmpegdev/70a141e1-f3f0-45ec-939e-990fe6e80400@lynne.ee/>
List-Archive: <https://master.gitmailbox.com/ffmpegdev/>
List-Post: <mailto:ffmpegdev@gitmailbox.com>
On 22/03/2025 18:49, Massimo Eynard wrote:
> This patch adds support for decoding the fourth MLP substream
> which contains the 16-channel presentation used for Atmos
> audio objects.
>
> By default only the first three substreams are decoded
> unless the new extract_objects flag is enabled as the resulting
> presentation contains audio object feeds instead of classic
> loudspeaker feeds.
>
> As this introduces interpolation of primitive matrices, precision
> has been increased to 2.18 fixed point. Therefore this requires
> DSP code upgrade which has been done for C and x86 implementations
> but not the ARM implementation.
>
> Adds two FATE tests using existing atmos.thd sample to reflect
> changes.
>
> Signed-off-by: Massimo Eynard <eynard.massimo@gmail.com>
> ---
> libavcodec/arm/mlpdsp_armv5te.S | 2 +-
> libavcodec/arm/mlpdsp_init_arm.c | 3 +-
> libavcodec/mlp.h | 10 +-
> libavcodec/mlp_parse.c | 31 ++-
> libavcodec/mlp_parse.h | 1 +
> libavcodec/mlp_parser.c | 11 +-
> libavcodec/mlpdec.c | 389 +++++++++++++++++++++++++++----
> libavcodec/mlpdsp.c | 50 +++-
> libavcodec/mlpdsp.h | 25 ++
> libavcodec/x86/mlpdsp.asm | 19 +-
> tests/fate/truehd.mak | 10 +
> 11 files changed, 476 insertions(+), 75 deletions(-)
>
> diff --git a/libavcodec/arm/mlpdsp_armv5te.S b/libavcodec/arm/mlpdsp_armv5te.S
> index d31568611c..5210f391ab 100644
> --- a/libavcodec/arm/mlpdsp_armv5te.S
> +++ b/libavcodec/arm/mlpdsp_armv5te.S
> @@ -21,7 +21,7 @@
>
> #include "libavutil/arm/asm.S"
>
> -#define MAX_CHANNELS 8
> +#define MAX_CHANNELS 16
> #define MAX_FIR_ORDER 8
> #define MAX_IIR_ORDER 4
> #define MAX_RATEFACTOR 4
> diff --git a/libavcodec/arm/mlpdsp_init_arm.c b/libavcodec/arm/mlpdsp_init_arm.c
> index 34a5f61e1d..50c3cf5488 100644
> --- a/libavcodec/arm/mlpdsp_init_arm.c
> +++ b/libavcodec/arm/mlpdsp_init_arm.c
> @@ -113,6 +113,7 @@ static int32_t (*mlp_select_pack_output_armv6(uint8_t *ch_assign,
> ch_index = 2;
> break;
> default:
> + // max_matrix_channel > 7 requires &7 lossless buffer channel shift
> return ff_mlp_pack_output;
> }
>
> @@ -139,7 +140,7 @@ av_cold void ff_mlpdsp_init_arm(MLPDSPContext *c)
>
> if (have_armv5te(cpu_flags)) {
> c->mlp_filter_channel = ff_mlp_filter_channel_arm;
> - c->mlp_rematrix_channel = ff_mlp_rematrix_channel_arm;
> + // c->mlp_rematrix_channel = ff_mlp_rematrix_channel_arm; // TODO: update to 2.18 coeff
> }
> if (have_armv6(cpu_flags))
> c->mlp_select_pack_output = mlp_select_pack_output_armv6;
> diff --git a/libavcodec/mlp.h b/libavcodec/mlp.h
> index bec414c680..264ef64cf1 100644
> --- a/libavcodec/mlp.h
> +++ b/libavcodec/mlp.h
> @@ -31,19 +31,19 @@
>
> /** Last possible matrix channel for each codec */
> #define MAX_MATRIX_CHANNEL_MLP 5
> -#define MAX_MATRIX_CHANNEL_TRUEHD 7
> +#define MAX_MATRIX_CHANNEL_TRUEHD 15
> /** Maximum number of channels in a valid stream.
> * MLP : 5.1 + 2 noise channels -> 8 channels
> - * TrueHD: 7.1 -> 8 channels
> + * TrueHD: up to 16-ch pres -> 16 channels
> */
> -#define MAX_CHANNELS 8
> +#define MAX_CHANNELS 16
>
> /** Maximum number of matrices used in decoding; most streams have one matrix
> * per output channel, but some rematrix a channel (usually 0) more than once.
> */
> #define MAX_MATRICES_MLP 6
> -#define MAX_MATRICES_TRUEHD 8
> -#define MAX_MATRICES 8
> +#define MAX_MATRICES_TRUEHD 16
> +#define MAX_MATRICES 16
>
> /** Maximum number of substreams that can be decoded.
> * MLP's limit is 2. TrueHD supports at least up to 3.
> diff --git a/libavcodec/mlp_parse.c b/libavcodec/mlp_parse.c
> index 924c731439..c94da860d0 100644
> --- a/libavcodec/mlp_parse.c
> +++ b/libavcodec/mlp_parse.c
> @@ -85,7 +85,7 @@ static int mlp_get_major_sync_size(const uint8_t * buf, int bufsize)
>
> int ff_mlp_read_major_sync(void *log, MLPHeaderInfo *mh, GetBitContext *gb)
> {
> - int ratebits, channel_arrangement, header_size;
> + int ratebits, channel_arrangement, header_size, extra_ch_length;
> uint16_t checksum;
>
> av_assert1(get_bits_count(gb) == 0);
> @@ -163,7 +163,34 @@ int ff_mlp_read_major_sync(void *log, MLPHeaderInfo *mh, GetBitContext *gb)
> mh->extended_substream_info = get_bits(gb, 2);
> mh->substream_info = get_bits(gb, 8);
>
> - skip_bits_long(gb, (header_size - 18) * 8);
> + extra_ch_length = 0;
> + mh->channels_thd_stream3 = 0;
> +
> + if (mh->stream_type == 0xba) {
> + skip_bits_long(gb, 63);
> +
> + extra_ch_length = 64;
> + if (get_bits1(gb) && (mh->substream_info & 0x80)) {
> + /* 16ch_channel_meaning */
> + int length = (get_bits(gb, 4) + 1) << 1;
> + if (header_size - 26 < length) {
> + av_log(log, AV_LOG_ERROR, "packet too short, "
> + "unable to read 16ch extra meaning in major sync %d %d\n",
> + header_size, length);
> + return -1;
> + }
> +
> + skip_bits_long(gb, 5+6); // dialogue norm/mix level
> + mh->channels_thd_stream3 = get_bits(gb, 5) + 1;
> + if (!get_bits1(gb)) { // dyn_object_only
> + avpriv_request_sample(log, "16ch presentation with a mixture of channels");
> + return AVERROR_PATCHWELCOME;
> + }
> + extra_ch_length += 21;
> + }
> + }
> +
> + skip_bits_long(gb, (header_size - 18) * 8 - extra_ch_length);
>
> return 0;
> }
> diff --git a/libavcodec/mlp_parse.h b/libavcodec/mlp_parse.h
> index 5f1f953cfe..8011566ff7 100644
> --- a/libavcodec/mlp_parse.h
> +++ b/libavcodec/mlp_parse.h
> @@ -47,6 +47,7 @@ typedef struct MLPHeaderInfo
> int channels_mlp; ///< Channel count for MLP streams
> int channels_thd_stream1; ///< Channel count for substream 1 of TrueHD streams ("6-channel presentation")
> int channels_thd_stream2; ///< Channel count for substream 2 of TrueHD streams ("8-channel presentation")
> + int channels_thd_stream3; ///< Channel count for substream 3 of TrueHD streams ("16-channel presentation")
> uint64_t channel_layout_mlp; ///< Channel layout for MLP streams
> uint64_t channel_layout_thd_stream1; ///< Channel layout for substream 1 of TrueHD streams ("6-channel presentation")
> uint64_t channel_layout_thd_stream2; ///< Channel layout for substream 2 of TrueHD streams ("8-channel presentation")
> diff --git a/libavcodec/mlp_parser.c b/libavcodec/mlp_parser.c
> index d391390dd5..f3d54c40e7 100644
> --- a/libavcodec/mlp_parser.c
> +++ b/libavcodec/mlp_parser.c
> @@ -181,10 +181,15 @@ static int mlp_parse(AVCodecParserContext *s,
> av_channel_layout_from_mask(&avctx->ch_layout, mh.channel_layout_mlp);
> } else { /* mh.stream_type == 0xba */
> /* TrueHD stream */
> - if (!mh.channels_thd_stream2) {
> - av_channel_layout_from_mask(&avctx->ch_layout, mh.channel_layout_thd_stream1);
> - } else {
> + if (mh.channels_thd_stream3) {
> + avctx->ch_layout = (AVChannelLayout) {
> + AV_CHANNEL_ORDER_UNSPEC,
> + mh.channels_thd_stream3
> + };
Is the order really unspecified? Surely there's some understanding which
channels map to which position.
> + } else if (mh.channels_thd_stream2) {
> av_channel_layout_from_mask(&avctx->ch_layout, mh.channel_layout_thd_stream2);
> + } else {
> + av_channel_layout_from_mask(&avctx->ch_layout, mh.channel_layout_thd_stream1);
> }
> }
>
> diff --git a/libavcodec/mlpdec.c b/libavcodec/mlpdec.c
> index e85dac36a7..14fb953265 100644
> --- a/libavcodec/mlpdec.c
> +++ b/libavcodec/mlpdec.c
> @@ -61,8 +61,8 @@ typedef struct SubStream {
>
> //@{
> /** restart header data */
> - /// The type of noise to be used in the rematrix stage.
> - uint16_t noise_type;
> + /// The type of substream given by the restart header sync word.
> + uint16_t substream_type;
>
> /// The index of the first channel coded in this substream.
> uint8_t min_channel;
> @@ -88,6 +88,13 @@ typedef struct SubStream {
> /// The current seed value for the pseudorandom noise generator(s).
> uint32_t noisegen_seed;
>
> + /// Maximum output_shift value.
> + int8_t max_shift;
> + /// Maximum size of coded audio samples LSBs part.
> + int8_t max_lsbs;
> + /// Maximum bit-depth of output audio samples.
> + int8_t max_bits;
> +
> /// Set if the substream contains extra info to check the size of VLC blocks.
> uint8_t data_check_present;
>
> @@ -104,11 +111,23 @@ typedef struct SubStream {
> /// matrix output channel
> uint8_t matrix_out_ch[MAX_MATRICES];
>
> - /// Whether the LSBs of the matrix output are encoded in the bitstream.
> + /// Size of the LSBs of the matrix output encoded in the bitstream.
> uint8_t lsb_bypass[MAX_MATRICES];
> - /// Matrix coefficients, stored as 2.14 fixed point.
> + /// Matrix coefficients fractional part size in bits.
> + uint8_t matrix_coeff_frac_bits[MAX_MATRICES];
> + /// Matrix coefficients shift amount.
> + int8_t matrix_coeff_shift[MAX_MATRICES];
> + /// Matrix coefficients presence mask.
> + uint16_t matrix_coeff_mask[MAX_MATRICES];
> + /// Matrix coefficients, stored as 2.18 fixed point.
> DECLARE_ALIGNED(32, int32_t, matrix_coeff)[MAX_MATRICES][MAX_CHANNELS];
> - /// Left shift to apply to noise values in 0x31eb substreams.
> + /// Delta matrix coefficients size in bits for 0x31ec substreams.
> + uint8_t delta_matrix_coeff_bits[MAX_MATRICES];
> + /// Delta matrix coefficients precision.
> + uint8_t delta_matrix_coeff_prec[MAX_MATRICES];
> + /// Delta matrix coefficients, stored as 2.18 fixed point.
> + DECLARE_ALIGNED(32, int32_t, delta_matrix_coeff)[MAX_MATRICES][MAX_CHANNELS];
> + /// Left shift to apply to noise values in 0x31eb and 0x31ec substreams.
> uint8_t matrix_noise_shift[MAX_MATRICES];
> //@}
>
> @@ -133,6 +152,8 @@ typedef struct MLPDecodeContext {
> AVCodecContext *avctx;
>
> AVChannelLayout downmix_layout;
> + /// Set to enable decoding of non-loudspeaker feed (objects) audio channels
> + int extract_objects;
>
> /// Current access unit being read has a major sync.
> int is_major_sync_unit;
> @@ -267,14 +288,14 @@ static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
>
> for (mat = 0; mat < s->num_primitive_matrices; mat++)
> if (s->lsb_bypass[mat])
> - m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
> + m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits(gbp, s->lsb_bypass[mat]);
>
> for (channel = s->min_channel; channel <= s->max_channel; channel++) {
> ChannelParams *cp = &s->channel_params[channel];
> int codebook = cp->codebook;
> int quant_step_size = s->quant_step_size[channel];
> int lsb_bits = cp->huff_lsbs - quant_step_size;
> - int result = 0;
> + int32_t result = 0;
>
> if (codebook > 0)
> result = get_vlc2(gbp, huff_vlc[codebook-1].table,
> @@ -410,8 +431,12 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
> m->avctx->profile = AV_PROFILE_TRUEHD_ATMOS;
> }
>
> - /* limit to decoding 3 substreams, as the 4th is used by Dolby Atmos for non-audio data */
> - m->max_decoded_substream = FFMIN(m->num_substreams - 1, 2);
> + /* Limit to decoding the first 3 substreams (or allow the 4th for objects) */
> + m->max_decoded_substream = FFMIN(m->num_substreams - 1,
> + m->extract_objects ? 3 : 2);
> +
> + av_log(m->avctx, AV_LOG_DEBUG, "decoding up to substream %" PRIu8 "\n",
> + m->max_decoded_substream);
>
> m->avctx->sample_rate = mh.group1_samplerate;
> m->avctx->frame_size = mh.access_unit_size;
> @@ -531,23 +556,22 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
> int sync_word, tmp;
> uint8_t checksum;
> uint8_t lossless_check;
> + uint8_t max_bit_depth;
> int start_count = get_bits_count(gbp);
> - int min_channel, max_channel, max_matrix_channel, noise_type;
> + int min_channel, max_channel, max_matrix_channel;
> const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
> ? MAX_MATRIX_CHANNEL_MLP
> : MAX_MATRIX_CHANNEL_TRUEHD;
>
> - sync_word = get_bits(gbp, 13);
> + sync_word = get_bits(gbp, 14);
>
> - if (sync_word != 0x31ea >> 1) {
> + if (sync_word < 0x31ea || 0x31ec < sync_word) {
> av_log(m->avctx, AV_LOG_ERROR,
> "restart header sync incorrect (got 0x%04x)\n", sync_word);
> return AVERROR_INVALIDDATA;
> }
>
> - noise_type = get_bits1(gbp);
> -
> - if (m->avctx->codec_id == AV_CODEC_ID_MLP && noise_type) {
> + if (m->avctx->codec_id == AV_CODEC_ID_MLP && 0x31ea != sync_word) {
> av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
> return AVERROR_INVALIDDATA;
> }
> @@ -567,7 +591,7 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
>
> /* This should happen for TrueHD streams with >6 channels and MLP's noise
> * type. It is not yet known if this is allowed. */
> - if (max_matrix_channel > MAX_MATRIX_CHANNEL_MLP && !noise_type) {
> + if (max_matrix_channel > MAX_MATRIX_CHANNEL_MLP && 0x31ea == sync_word) {
> avpriv_request_sample(m->avctx,
> "%d channels (more than the "
> "maximum supported by the decoder)",
> @@ -582,7 +606,7 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
> s->max_channel = max_channel;
> s->coded_channels = ((1LL << (max_channel - min_channel + 1)) - 1) << min_channel;
> s->max_matrix_channel = max_matrix_channel;
> - s->noise_type = noise_type;
> + s->substream_type = sync_word;
>
> if (mlp_channel_layout_subset(&m->downmix_layout, s->mask) &&
> m->max_decoded_substream > substr) {
> @@ -595,8 +619,28 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
>
> s->noise_shift = get_bits(gbp, 4);
> s->noisegen_seed = get_bits(gbp, 23);
> + s->max_shift = get_bits(gbp, 4);
> + s->max_lsbs = get_bits(gbp, 5);
> + s->max_bits = get_bits(gbp, 5);
> +
> + max_bit_depth = (0x31ec == sync_word) ? 31 : 24;
> + if (max_bit_depth < s->max_lsbs) {
> + av_log(m->avctx, AV_LOG_ERROR,
> + "Max LSB size %" PRIu8 " for substream %u exceeds "
> + "%" PRIu8 " bits.\n",
> + s->max_lsbs, substr, max_bit_depth);
> + return AVERROR_INVALIDDATA;
> + }
>
> - skip_bits(gbp, 19);
> + if (max_bit_depth < s->max_bits) {
> + av_log(m->avctx, AV_LOG_ERROR,
> + "Max output bit-depth %" PRIu8 " for substream %u exceeds "
> + "%" PRIu8 " bits.\n",
> + s->max_bits, substr, max_bit_depth);
> + return AVERROR_INVALIDDATA;
> + }
> +
> + skip_bits(gbp, 5);
>
> s->data_check_present = get_bits1(gbp);
> lossless_check = get_bits(gbp, 8);
> @@ -615,7 +659,7 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
>
> for (ch = 0; ch <= s->max_matrix_channel; ch++) {
> int ch_assign = get_bits(gbp, 6);
> - if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
> + if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && s->mask) {
> AVChannelLayout l;
> enum AVChannel channel = thd_channel_layout_extract_channel(s->mask, ch_assign);
>
> @@ -656,12 +700,19 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
> cp->huff_offset = 0;
> cp->sign_huff_offset = -(1 << 23);
> cp->codebook = 0;
> - cp->huff_lsbs = 24;
> + cp->huff_lsbs = (3 == substr) ? 31 : 24;
> }
>
> if (substr == m->max_decoded_substream) {
> av_channel_layout_uninit(&m->avctx->ch_layout);
> - av_channel_layout_from_mask(&m->avctx->ch_layout, s->mask);
> + if (substr < 3) /* Loudspeaker feed channels */
> + av_channel_layout_from_mask(&m->avctx->ch_layout, s->mask);
> + else /* Object channels */
> + m->avctx->ch_layout = (AVChannelLayout) {
> + AV_CHANNEL_ORDER_UNSPEC,
> + s->max_channel+1
> + };
> +
> m->pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
> s->output_shift,
> s->max_matrix_channel,
> @@ -760,17 +811,43 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
> return 0;
> }
>
> -/** Read parameters for primitive matrices. */
> +/** Get the maximum number of primitive matrices allowed. */
>
> -static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
> +static int get_max_nb_primitive_matrices(MLPDecodeContext *m, unsigned int substr)
> +{
> + switch (substr) {
> + case 0: // substream 0 (up to 2 matrix channels)
> + return 2;
> + case 1: // substream 1
> + if (m->substream_info & 0x8) // 6-ch pres carried
> + return 6;
> + if (m->substream_info & 0x20) // 8-ch pres carried
> + return 8;
> + break;
> + case 2: // substream 2
> + if (m->substream_info & 0x40) // 8-ch pres carried
> + return 8;
> + break;
> + case 3: // substream 3
> + if (m->substream_info & 80) // 16-ch pres carried
> + return 16;
> + break;
> + }
> +
> + return MAX_MATRICES_TRUEHD;
> +}
> +
> +/** Read parameters for primitive matrices (0x31ea and 0x31eb substreams). */
> +
> +static int read_31ea_31eb_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
> {
> SubStream *s = &m->substream[substr];
> unsigned int mat, ch;
> const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
> ? MAX_MATRICES_MLP
> - : MAX_MATRICES_TRUEHD;
> + : get_max_nb_primitive_matrices(m, substr);
>
> - if (m->matrix_changed++ > 1) {
> + if (++m->matrix_changed > 1) {
> av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
> return AVERROR_INVALIDDATA;
> }
> @@ -779,8 +856,9 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
>
> if (s->num_primitive_matrices > max_primitive_matrices) {
> av_log(m->avctx, AV_LOG_ERROR,
> - "Number of primitive matrices cannot be greater than %d.\n",
> - max_primitive_matrices);
> + "Number of primitive matrices cannot be greater than %d "
> + "for substream %u of type 0x%04x.\n",
> + max_primitive_matrices, substr, s->substream_type);
> goto error;
> }
>
> @@ -803,7 +881,7 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
> }
>
> max_chan = s->max_matrix_channel;
> - if (!s->noise_type)
> + if (0x31ea == s->substream_type)
> max_chan+=2;
>
> for (ch = 0; ch <= max_chan; ch++) {
> @@ -811,10 +889,10 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
> if (get_bits1(gbp))
> coeff_val = get_sbits(gbp, frac_bits + 2);
>
> - s->matrix_coeff[mat][ch] = coeff_val * (1 << (14 - frac_bits));
> + s->matrix_coeff[mat][ch] = coeff_val * (1 << ((14 + 4) - frac_bits));
> }
>
> - if (s->noise_type)
> + if (0x31eb == s->substream_type)
> s->matrix_noise_shift[mat] = get_bits(gbp, 4);
> else
> s->matrix_noise_shift[mat] = 0;
> @@ -828,6 +906,124 @@ error:
> return AVERROR_INVALIDDATA;
> }
>
> +/** Read parameters for primitive matrices (0x31ec substreams). */
> +
> +static int read_31ec_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
> +{
> + SubStream *s = &m->substream[substr];
> + unsigned int mat, ch;
> +
> + const int max_primitive_matrices = get_max_nb_primitive_matrices(m, substr);
> +
> + if (++m->matrix_changed > 1) {
> + av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
> + return AVERROR_INVALIDDATA;
> + }
> +
> + /* Seed primitive matrices */
> +
> + if (get_bits1(gbp)) {
> + /* New seed primitive matrices */
> +
> + if (get_bits1(gbp)) {
> + /* New seed matrices parameters */
> + s->num_primitive_matrices = get_bits(gbp, 4) + 1;
> +
> + if (s->num_primitive_matrices > max_primitive_matrices) {
> + av_log(m->avctx, AV_LOG_ERROR,
> + "Number of primitive matrices cannot be greater than %d "
> + "in substream %u of type 0x%04x.\n",
> + max_primitive_matrices, substr, s->substream_type);
> + goto error;
> + }
> +
> + for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> + s->matrix_out_ch [mat] = get_bits(gbp, 4);
> + s->matrix_coeff_frac_bits[mat] = get_bits(gbp, 4);
> + s->matrix_coeff_shift [mat] = ((int) get_bits(gbp, 3)) - 1;
> + s->lsb_bypass [mat] = get_bits(gbp, 2);
> + s->matrix_noise_shift [mat] = get_bits(gbp, 4);
> + s->matrix_coeff_mask [mat] = get_bits(gbp, s->max_matrix_channel + 1);
> +
> + if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
> + av_log(m->avctx, AV_LOG_ERROR,
> + "Invalid channel %d specified as output from matrix.\n",
> + s->matrix_out_ch[mat]);
> + goto error;
> + }
> + if (s->matrix_coeff_frac_bits[mat] > 14) {
> + av_log(m->avctx, AV_LOG_ERROR,
> + "Too many fractional bits specified.\n");
> + goto error;
> + }
> + }
> + }
> +
> + /* Seed matrices coefficients */
> + for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> + const int coeff_shift = s->matrix_coeff_shift[mat]
> + - s->matrix_coeff_frac_bits[mat];
> +
> + memset(s->matrix_coeff[mat], 0, sizeof(s->matrix_coeff[mat]));
> +
> + for (ch = 0; ch <= s->max_matrix_channel; ch++) {
> + int64_t coeff_val;
> +
> + if (!((s->matrix_coeff_mask[mat] >> ch) & 0x1))
> + continue; // skip channel
> +
> + coeff_val = get_sbits(gbp, s->matrix_coeff_frac_bits[mat] + 2);
> + s->matrix_coeff[mat][ch] = coeff_val * (1 << (18 + coeff_shift));
> + }
> + }
> + }
> +
> + if (!get_bits1(gbp)) {
> + /* No primitive matrices interpolation */
> + memset(s->delta_matrix_coeff, 0, sizeof(s->delta_matrix_coeff));
> + }
> + else if (get_bits1(gbp)) {
> + /* New delta primitive matrices */
> +
> + if (get_bits1(gbp)) {
> + /* New delta primitive matrices parameters */
> +
> + for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> + s->delta_matrix_coeff_bits[mat] = get_bits(gbp, 4) + 1;
> + s->delta_matrix_coeff_prec[mat] = get_bits(gbp, 2);
> + }
> + }
> +
> + for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> + const int coeff_shift = s->matrix_coeff_shift[mat]
> + - s->delta_matrix_coeff_prec[mat]
> + - s->matrix_coeff_frac_bits[mat];
> +
> + memset(s->delta_matrix_coeff[mat], 0, sizeof(s->delta_matrix_coeff[mat]));
> +
> + if (s->delta_matrix_coeff_bits[mat] <= 1)
> + continue; // skip matrice
> +
> + for (ch = 0; ch <= s->max_matrix_channel; ch++) {
> + int64_t coeff_val;
> +
> + if (!((s->matrix_coeff_mask[mat] >> ch) & 0x1))
> + continue; // skip channel
> +
> + coeff_val = get_sbits(gbp, s->delta_matrix_coeff_bits[mat]);
> + s->delta_matrix_coeff[mat][ch] = coeff_val * (1 << (18 + coeff_shift));
> + }
> + }
> + }
> +
> + return 0;
> +error:
> + s->num_primitive_matrices = 0;
> + memset(s->matrix_out_ch, 0, sizeof(s->matrix_out_ch));
> +
> + return AVERROR_INVALIDDATA;
> +}
> +
> /** Read channel parameters. */
>
> static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
> @@ -875,8 +1071,10 @@ static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
> cp->codebook = get_bits(gbp, 2);
> cp->huff_lsbs = get_bits(gbp, 5);
>
> - if (cp->codebook > 0 && cp->huff_lsbs > 24) {
> - av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
> + if (cp->codebook > 0 && cp->huff_lsbs > s->max_lsbs) {
> + av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs=%" PRIu8 ", "
> + "exceeds max_lsbs=%" PRIu8 ".\n",
> + cp->huff_lsbs, s->max_lsbs);
> cp->huff_lsbs = 0;
> return AVERROR_INVALIDDATA;
> }
> @@ -910,9 +1108,14 @@ static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
> }
>
> if (s->param_presence_flags & PARAM_MATRIX)
> - if (get_bits1(gbp))
> - if ((ret = read_matrix_params(m, substr, gbp)) < 0)
> + if (get_bits1(gbp)) {
> + if (0x31ec == s->substream_type)
> + ret = read_31ec_matrix_params(m, substr, gbp);
> + else
> + ret = read_31ea_31eb_matrix_params(m, substr, gbp);
> + if (ret < 0)
> return ret;
> + }
>
> if (s->param_presence_flags & PARAM_OUTSHIFT)
> if (get_bits1(gbp)) {
> @@ -922,6 +1125,10 @@ static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
> avpriv_request_sample(m->avctx, "Negative output_shift");
> s->output_shift[ch] = 0;
> }
> + if (s->max_shift < s->output_shift[ch])
> + av_log(m->avctx, AV_LOG_WARNING,
> + "output_shift=%d exceeds max_shift=%d\n",
> + s->output_shift[ch], s->max_shift);
> }
> if (substr == m->max_decoded_substream)
> m->pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
> @@ -1103,6 +1310,56 @@ static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
> s->noisegen_seed = seed;
> }
>
> +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL >= 2
> +
> +/** Check matrices-based channel remapping output for saturation. */
> +
> +static void check_rematrix_output(MLPDecodeContext *m, unsigned int substr)
> +{
> + SubStream *s = &m->substream[substr];
> + unsigned int mat, sample;
> +
> + for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> + unsigned int dest_ch = s->matrix_out_ch[mat];
> +
> + const uint8_t shift = (3 <= substr && s->min_channel <= dest_ch) ? 31 : 23;
> + const int32_t min_value = -(1u << shift);
> + const int32_t max_value = (1u << shift) - 1;
> +
> + for (sample = 0; sample < s->blockpos; sample++) {
> + if (m->sample_buffer[sample][dest_ch] < min_value)
> + av_log(m->avctx, AV_LOG_WARNING,
> + "rematrix negative saturation substr=%u mat=%u sample=%d "
> + "value=%" PRId32 "\n",
> + substr, mat, sample, m->sample_buffer[sample][dest_ch]);
> + if (m->sample_buffer[sample][dest_ch] > max_value)
> + av_log(m->avctx, AV_LOG_WARNING,
> + "rematrix positive saturation substr=%u mat=%u sample=%d "
> + "value=%" PRId32 "\n",
> + substr, mat, sample, m->sample_buffer[sample][dest_ch]);
> + }
> + }
> +}
> +
> +/** Check output audio bit-depth. */
> +
> +static void check_output_bit_depth(MLPDecodeContext *m, unsigned int substr)
> +{
> + SubStream *s = &m->substream[substr];
> + uint32_t cumul_mask = 0;
> + unsigned int chan, sample;
> +
> + for (chan = 0; chan <= s->max_matrix_channel; chan++)
> + for (sample = 0; sample < s->blockpos; sample++)
> + cumul_mask |= FFABS(m->sample_buffer[sample][chan]);
> +
> + if ((1u << s->max_bits) <= cumul_mask)
> + av_log(m->avctx, AV_LOG_WARNING, "output audio bit-depth exceeds "
> + "expected %u bits.\n",
> + s->max_bits);
> +}
> +#endif
> +
> /** Write the audio data into the output buffer. */
>
> static int output_data(MLPDecodeContext *m, unsigned int substr,
> @@ -1110,8 +1367,7 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
> {
> AVCodecContext *avctx = m->avctx;
> SubStream *s = &m->substream[substr];
> - unsigned int mat;
> - unsigned int maxchan;
> + unsigned int mat, chan, maxchan;
> int ret;
> int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
>
> @@ -1126,7 +1382,7 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
> }
>
> maxchan = s->max_matrix_channel;
> - if (!s->noise_type) {
> + if (0x31ea == s->substream_type) {
> generate_2_noise_channels(m, substr);
> maxchan += 2;
> } else {
> @@ -1137,19 +1393,45 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
> * samples. */
> for (mat = 0; mat < s->num_primitive_matrices; mat++) {
> unsigned int dest_ch = s->matrix_out_ch[mat];
> - m->dsp.mlp_rematrix_channel(&m->sample_buffer[0][0],
> - s->matrix_coeff[mat],
> - &m->bypassed_lsbs[0][mat],
> - m->noise_buffer,
> - s->num_primitive_matrices - mat,
> - dest_ch,
> - s->blockpos,
> - maxchan,
> - s->matrix_noise_shift[mat],
> - m->access_unit_size_pow2,
> - MSB_MASK(s->quant_step_size[dest_ch]));
> +
> + if (substr < 3) {
> + /* Single primitive matrices */
> + m->dsp.mlp_rematrix_channel(&m->sample_buffer[0][0],
> + s->matrix_coeff[mat],
> + &m->bypassed_lsbs[0][mat],
> + m->noise_buffer,
> + s->num_primitive_matrices - mat,
> + dest_ch,
> + s->blockpos,
> + maxchan,
> + s->matrix_noise_shift[mat],
> + m->access_unit_size_pow2,
> + MSB_MASK(s->quant_step_size[dest_ch]));
> + }
> + else {
> + /* Interpolated primitive matrices */
> + m->dsp.mlp_rematrix_interp_channel(&m->sample_buffer[0][0],
> + s->matrix_coeff[mat],
> + s->delta_matrix_coeff[mat],
> + &m->bypassed_lsbs[0][mat],
> + m->noise_buffer,
> + s->num_primitive_matrices - mat,
> + dest_ch,
> + s->blockpos,
> + maxchan,
> + s->matrix_noise_shift[mat],
> + m->access_unit_size_pow2,
> + MSB_MASK(s->quant_step_size[dest_ch]));
> +
> + for (chan = 0; chan <= maxchan; chan++)
> + s->matrix_coeff[mat][chan] += s->delta_matrix_coeff[mat][chan];
> + }
> }
>
> +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL >= 2
> + check_rematrix_output(m, substr);
> +#endif
> +
> /* get output buffer */
> frame->nb_samples = s->blockpos;
> if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
> @@ -1163,6 +1445,10 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
> s->max_matrix_channel,
> is32);
>
> +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL >= 2
> + check_output_bit_depth(m, substr);
> +#endif
> +
> /* Update matrix encoding side data */
> if (s->matrix_encoding != s->prev_matrix_encoding) {
> if ((ret = ff_side_data_update_matrix_encoding(frame, s->matrix_encoding)) < 0)
> @@ -1320,6 +1606,7 @@ static int read_access_unit(AVCodecContext *avctx, AVFrame *frame,
> (avctx->ch_layout.nb_channels == 8 &&
> ((m->substream_info >> 4) & 0x7) != 0x7 &&
> ((m->substream_info >> 4) & 0x7) != 0x6 &&
> + ((m->substream_info >> 4) & 0x7) != 0x4 &&
> ((m->substream_info >> 4) & 0x7) != 0x3)) &&
> substr > 0 && substr < m->max_decoded_substream &&
> (s->min_channel <= m->substream[substr - 1].max_channel)) {
> @@ -1429,8 +1716,10 @@ static void mlp_decode_flush(AVCodecContext *avctx)
> #define OFFSET(x) offsetof(MLPDecodeContext, x)
> #define FLAGS (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
> static const AVOption options[] = {
> - { "downmix", "Request a specific channel layout from the decoder", OFFSET(downmix_layout),
> - AV_OPT_TYPE_CHLAYOUT, {.str = NULL}, .flags = FLAGS },
> + { "downmix", "Request a specific channel layout from the decoder",
> + OFFSET(downmix_layout), AV_OPT_TYPE_CHLAYOUT, {.str = NULL}, .flags = FLAGS },
> + { "extract_objects", "Enable extraction of audio object channels",
> + OFFSET(extract_objects), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = FLAGS },
> { NULL },
IMHO this should be enabled by default. All streams come with
compatibility AC3 mixes too.
> };
>
> diff --git a/libavcodec/mlpdsp.c b/libavcodec/mlpdsp.c
> index cb40160f67..e28006f2c4 100644
> --- a/libavcodec/mlpdsp.c
> +++ b/libavcodec/mlpdsp.c
> @@ -79,11 +79,51 @@ void ff_mlp_rematrix_channel(int32_t *samples,
>
> if (matrix_noise_shift) {
> index &= access_unit_size_pow2 - 1;
> - accum += noise_buffer[index] * (1 << (matrix_noise_shift + 7));
> + accum += noise_buffer[index] * (1 << (matrix_noise_shift + 11));
> index += index2;
> }
>
> - samples[dest_ch] = ((accum >> 14) & mask) + *bypassed_lsbs;
> + samples[dest_ch] = ((accum >> 18) & mask) + *bypassed_lsbs;
> + bypassed_lsbs += MAX_CHANNELS;
> + samples += MAX_CHANNELS;
> + }
> +}
> +
> +void ff_mlp_rematrix_interp_channel(int32_t *samples,
> + const int32_t *seed_coeffs,
> + const int32_t *delta_coeffs,
> + const uint8_t *bypassed_lsbs,
> + const int8_t *noise_buffer,
> + int index,
> + unsigned int dest_ch,
> + uint16_t blockpos,
> + unsigned int maxchan,
> + int matrix_noise_shift,
> + int access_unit_size_pow2,
> + int32_t mask)
> +{
> + unsigned int src_ch, i;
> + int index2 = 2 * index + 1;
> +
> + int32_t delta_inc = (1 << 16) / blockpos;
> +
> + for (i = 0; i < blockpos; i++) {
> + int64_t accum = 0, delta_accum = 0;
> +
> + for (src_ch = 0; src_ch <= maxchan; src_ch++) {
> + accum += (int64_t) samples[src_ch] * seed_coeffs[src_ch];
> + delta_accum += (int64_t) samples[src_ch] * delta_coeffs[src_ch];
> + }
> +
> + accum += ((delta_accum >> 18) * i * delta_inc * (1 << 18)) >> 16;
> +
> + if (matrix_noise_shift) {
> + index &= access_unit_size_pow2 - 1;
> + accum += noise_buffer[index] * (1 << (matrix_noise_shift + 11));
> + index += index2;
> + }
> +
> + samples[dest_ch] = ((accum >> 18) & mask) + *bypassed_lsbs;
> bypassed_lsbs += MAX_CHANNELS;
> samples += MAX_CHANNELS;
> }
> @@ -115,9 +155,10 @@ int32_t ff_mlp_pack_output(int32_t lossless_check_data,
> int mat_ch = ch_assign[out_ch];
> int32_t sample = sample_buffer[i][mat_ch] *
> (1U << output_shift[mat_ch]);
> - lossless_check_data ^= (sample & 0xffffff) << mat_ch;
> + lossless_check_data ^= (sample & 0xffffff) << (mat_ch & 7);
> +
> if (is32)
> - *data_32++ = sample * 256U;
> + *data_32++ = sample * (1 << 8);
> else
> *data_16++ = sample >> 8;
> }
> @@ -129,6 +170,7 @@ av_cold void ff_mlpdsp_init(MLPDSPContext *c)
> {
> c->mlp_filter_channel = mlp_filter_channel;
> c->mlp_rematrix_channel = ff_mlp_rematrix_channel;
> + c->mlp_rematrix_interp_channel = ff_mlp_rematrix_interp_channel;
> c->mlp_select_pack_output = mlp_select_pack_output;
> #if ARCH_ARM
> ff_mlpdsp_init_arm(c);
> diff --git a/libavcodec/mlpdsp.h b/libavcodec/mlpdsp.h
> index 7a9ac228d3..fd29db10a7 100644
> --- a/libavcodec/mlpdsp.h
> +++ b/libavcodec/mlpdsp.h
> @@ -37,6 +37,19 @@ void ff_mlp_rematrix_channel(int32_t *samples,
> int access_unit_size_pow2,
> int32_t mask);
>
> +void ff_mlp_rematrix_interp_channel(int32_t *samples,
> + const int32_t *seed_coeffs,
> + const int32_t *delta_coeffs,
> + const uint8_t *bypassed_lsbs,
> + const int8_t *noise_buffer,
> + int index,
> + unsigned int dest_ch,
> + uint16_t blockpos,
> + unsigned int maxchan,
> + int matrix_noise_shift,
> + int access_unit_size_pow2,
> + int32_t mask);
> +
> int32_t ff_mlp_pack_output(int32_t lossless_check_data,
> uint16_t blockpos,
> int32_t (*sample_buffer)[MAX_CHANNELS],
> @@ -62,6 +75,18 @@ typedef struct MLPDSPContext {
> int matrix_noise_shift,
> int access_unit_size_pow2,
> int32_t mask);
> + void (*mlp_rematrix_interp_channel)(int32_t *samples,
> + const int32_t *seed_coeffs,
> + const int32_t *delta_coeffs,
> + const uint8_t *bypassed_lsbs,
> + const int8_t *noise_buffer,
> + int index,
> + unsigned int dest_ch,
> + uint16_t blockpos,
> + unsigned int maxchan,
> + int matrix_noise_shift,
> + int access_unit_size_pow2,
> + int32_t mask);
> int32_t (*(*mlp_select_pack_output)(uint8_t *ch_assign,
> int8_t *output_shift,
> uint8_t max_matrix_channel,
> diff --git a/libavcodec/x86/mlpdsp.asm b/libavcodec/x86/mlpdsp.asm
> index 3dc641e89e..3b232d4551 100644
> --- a/libavcodec/x86/mlpdsp.asm
> +++ b/libavcodec/x86/mlpdsp.asm
> @@ -61,12 +61,12 @@ SECTION .text
> paddq xm0, xm1
> movq accumq, xm0
> movzx blsbsd, byte [blsbs_ptrq] ; load *bypassed_lsbs
> - sar accumq, 14 ; accum >>= 14
> + sar accumq, 18 ; accum >>= 18
> and accumd, maskd ; accum &= mask
> add accumd, blsbsd ; accum += *bypassed_lsbs
> mov [samplesq + dest_chq], accumd ; samples[dest_ch] = accum
> - add blsbs_ptrq, 8 ; bypassed_lsbs += MAX_CHANNELS;
> - add samplesq, 32 ; samples += MAX_CHANNELS;
> + add blsbs_ptrq, 16 ; bypassed_lsbs += MAX_CHANNELS;
> + add samplesq, 64 ; samples += MAX_CHANNELS;
> cmp blsbs_ptrq, cntq
> %endmacro
>
> @@ -80,12 +80,12 @@ SECTION .text
> SHLX noiseq, mns ; noise_buffer[index] <<= matrix_noise_shift
> add accumq, noiseq ; accum += noise_buffer[index]
> movzx noised, byte [blsbs_ptrq] ; load *bypassed_lsbs (reuse tmp noise register)
> - sar accumq, 14 ; accum >>= 14
> + sar accumq, 18 ; accum >>= 18
> and accumd, maskd ; accum &= mask
> add accumd, noised ; accum += *bypassed_lsbs
> mov [samplesq + dest_chq], accumd ; samples[dest_ch] = accum
> - add blsbs_ptrq, 8 ; bypassed_lsbs += MAX_CHANNELS;
> - add samplesq, 32 ; samples += MAX_CHANNELS;
> + add blsbs_ptrq, 16 ; bypassed_lsbs += MAX_CHANNELS;
> + add samplesq, 64 ; samples += MAX_CHANNELS;
> cmp blsbs_ptrq, cntq
> %endmacro
>
> @@ -106,7 +106,8 @@ cglobal mlp_rematrix_channel, 0, 13, 5, samples, coeffs, blsbs_ptr, blsbs, \
> mov dest_chd, dest_chm ; load dest_chd (not needed on UNIX64)
> %endif
> shl dest_chd, 2
> - lea cntq, [blsbs_ptrq + blockposq*8]
> + lea cntq, [blsbs_ptrq + blockposq*8] ; loop end address (bypassed_lsbs + blockpos * MAX_CHANNELS)
> + lea cntq, [cntq + blockposq*8]
> test mnsd, mnsd ; is matrix_noise_shift != 0?
> jne .shift ; jump if true
> cmp maxchand, 4 ; is maxchan < 4?
> @@ -144,7 +145,7 @@ align 16
> DEFINE_ARGS samples, coeffs, blsbs_ptr, noise_buffer, \
> index, dest_ch, accum, index2, mns, \
> ausp, mask, cnt, noise
> - add mnsd, 7 ; matrix_noise_shift += 7
> + add mnsd, 11 ; matrix_noise_shift += 11
> %else ; sse4
> mov r6, rcx ; move rcx elsewhere so we can use cl for matrix_noise_shift
> %if WIN64
> @@ -156,7 +157,7 @@ align 16
> DEFINE_ARGS samples, coeffs, blsbs_ptr, mns, index, dest_ch, noise_buffer, \
> index2, accum, ausp, mask, cnt, noise
> %endif
> - lea mnsd, [r8 + 7] ; rcx = matrix_noise_shift + 7
> + lea mnsd, [r8 + 11] ; rcx = matrix_noise_shift + 11
> %endif ; cpuflag
> sub auspd, 1 ; access_unit_size_pow2 -= 1
> cmp r7d, 4 ; is maxchan < 4?
If the changes are that few, why not do the aarch64 version too?
> diff --git a/tests/fate/truehd.mak b/tests/fate/truehd.mak
> index b0bc86a965..30c0e9628b 100644
> --- a/tests/fate/truehd.mak
> +++ b/tests/fate/truehd.mak
> @@ -18,5 +18,15 @@ fate-truehd-mono1726: CMD = md5pipe -f truehd -i $(TARGET_SAMPLES)/truehd/ticket
> fate-truehd-mono1726: CMP = oneline
> fate-truehd-mono1726: REF = 9be9551fac418440bb02101bfdb11df9
>
> +FATE_TRUEHD-$(call DEMDEC, TRUEHD, TRUEHD) += fate-truehd-atmos-no-obj
> +fate-truehd-atmos-no-obj: CMD = md5pipe -f truehd -extract_objects 0 -i $(TARGET_SAMPLES)/truehd/atmos.thd -f s32le
> +fate-truehd-atmos-no-obj: CMP = oneline
> +fate-truehd-atmos-no-obj: REF = 53da6ce35c778bcc2182ef2160bf16a2
> +
> +FATE_TRUEHD-$(call DEMDEC, TRUEHD, TRUEHD) += fate-truehd-atmos-obj
> +fate-truehd-atmos-obj: CMD = md5pipe -f truehd -extract_objects 1 -i $(TARGET_SAMPLES)/truehd/atmos.thd -f s32le
> +fate-truehd-atmos-obj: CMP = oneline
> +fate-truehd-atmos-obj: REF = f19f6f8ec8b040050aaa019f016f7ddc
> +
> FATE_SAMPLES_AUDIO += $(FATE_TRUEHD-yes)
> fate-truehd: $(FATE_TRUEHD-yes)
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".