[FFmpeg-devel] [PATCH v3 1/2] avcodec: add external enc libvvenc for H266/VVC
Christian
chris10317h5 at gmail.com
Thu May 16 18:22:26 EEST 2024
> On 14. May 2024, at 18:49, Andreas Rheinhardt <andreas.rheinhardt at outlook.com> wrote:
>
> Christian Bartnik:
>> From: Thomas Siedel <thomas.ff at spin-digital.com>
>>
>> Add external encoder VVenC for H266/VVC encoding.
>> Register new encoder libvvenc.
>> Add libvvenc to wrap the vvenc interface.
>> libvvenc implements encoder option: preset,qp,period,subjopt,
>> vvenc-params,levelidc,tier.
>> Enable encoder by adding --enable-libvvenc in configure step.
>>
>> Co-authored-by: Christian Bartnik chris10317h5 at gmail.com
>> Signed-off-by: Christian Bartnik <chris10317h5 at gmail.com>
>> ---
>> configure | 4 +
>> doc/encoders.texi | 65 +++++
>> libavcodec/Makefile | 1 +
>> libavcodec/allcodecs.c | 1 +
>> libavcodec/libvvenc.c | 566 +++++++++++++++++++++++++++++++++++++++++
>> 5 files changed, 637 insertions(+)
>> create mode 100644 libavcodec/libvvenc.c
>>
>> diff --git a/configure b/configure
>> index a909b0689c..5d9a14821b 100755
>> --- a/configure
>> +++ b/configure
>> @@ -293,6 +293,7 @@ External library support:
>> --enable-libvorbis enable Vorbis en/decoding via libvorbis,
>> native implementation exists [no]
>> --enable-libvpx enable VP8 and VP9 de/encoding via libvpx [no]
>> + --enable-libvvenc enable H.266/VVC encoding via vvenc [no]
>> --enable-libwebp enable WebP encoding via libwebp [no]
>> --enable-libx264 enable H.264 encoding via x264 [no]
>> --enable-libx265 enable HEVC encoding via x265 [no]
>> @@ -1966,6 +1967,7 @@ EXTERNAL_LIBRARY_LIST="
>> libvmaf
>> libvorbis
>> libvpx
>> + libvvenc
>> libwebp
>> libxevd
>> libxeve
>> @@ -3558,6 +3560,7 @@ libvpx_vp8_decoder_deps="libvpx"
>> libvpx_vp8_encoder_deps="libvpx"
>> libvpx_vp9_decoder_deps="libvpx"
>> libvpx_vp9_encoder_deps="libvpx"
>> +libvvenc_encoder_deps="libvvenc"
>> libwebp_encoder_deps="libwebp"
>> libwebp_anim_encoder_deps="libwebp"
>> libx262_encoder_deps="libx262"
>> @@ -7025,6 +7028,7 @@ enabled libvpx && {
>> die "libvpx enabled but no supported decoders found"
>> fi
>> }
>> +enabled libvvenc && require_pkg_config libvvenc "libvvenc >= 1.6.1" "vvenc/vvenc.h" vvenc_get_version
>>
>> enabled libwebp && {
>> enabled libwebp_encoder && require_pkg_config libwebp "libwebp >= 0.2.0" webp/encode.h WebPGetEncoderVersion
>> diff --git a/doc/encoders.texi b/doc/encoders.texi
>> index c82f316f94..92aab17c49 100644
>> --- a/doc/encoders.texi
>> +++ b/doc/encoders.texi
>> @@ -2378,6 +2378,71 @@ Indicates frame duration
>> For more information about libvpx see:
>> @url{http://www.webmproject.org/}
>>
>> + at section libvvenc
>> +
>> +VVenC H.266/VVC encoder wrapper.
>> +
>> +This encoder requires the presence of the libvvenc headers and library
>> +during configuration. You need to explicitly configure the build with
>> + at option{--enable-libvvenc}.
>> +
>> +The VVenC project website is at
>> + at url{https://github.com/fraunhoferhhi/vvenc}.
>> +
>> + at subsection Supported Pixel Formats
>> +
>> +VVenC supports only 10-bit color spaces as input. But the internal (encoded)
>> +bit depth can be set to 8-bit or 10-bit at runtime.
>> +
>> + at subsection Options
>> +
>> + at table @option
>> + at item b
>> +Sets target video bitrate.
>> +
>> + at item g
>> +Set the GOP size. Currently support for g=1 (Intra only) or default.
>> +
>> + at item preset
>> +Set the VVenC preset.
>> +
>> + at item levelidc
>> +Set level idc.
>> +
>> + at item tier
>> +Set vvc tier.
>> +
>> + at item qp
>> +Set constant quantization parameter.
>> +
>> + at item subopt @var{boolean}
>> +Set subjective (perceptually motivated) optimization. Default is 1 (on).
>> +
>> + at item bitdepth8 @var{boolean}
>> +Set 8bit coding mode instead of using 10bit. Default is 0 (off).
>> +
>> + at item period
>> +set (intra) refresh period in seconds.
>> +
>> + at item vvenc-params
>> +Set vvenc options using a list of @var{key}=@var{value} couples separated
>> +by ":". See @command{vvencapp --fullhelp} or @command{vvencFFapp --fullhelp} for a list of options.
>> +
>> +For example, the options might be provided as:
>> +
>> + at example
>> +intraperiod=64:decodingrefreshtype=idr:poc0idr=1:internalbitdepth=8
>> + at end example
>> +
>> +For example the encoding options for 2-pass encoding might be provided with @option{-vvenc-params}:
>> +
>> + at example
>> +ffmpeg -i input -c:v libvvenc -b 1M -vvenc-params passes=2:pass=1:rcstatsfile=stats.json output.mp4
>> +ffmpeg -i input -c:v libvvenc -b 1M -vvenc-params passes=2:pass=2:rcstatsfile=stats.json output.mp4
>> + at end example
>> +
>> + at end table
>> +
>> @section libwebp
>>
>> libwebp WebP Image encoder wrapper
>> diff --git a/libavcodec/Makefile b/libavcodec/Makefile
>> index 2443d2c6fd..5d7349090e 100644
>> --- a/libavcodec/Makefile
>> +++ b/libavcodec/Makefile
>> @@ -1153,6 +1153,7 @@ OBJS-$(CONFIG_LIBVPX_VP8_DECODER) += libvpxdec.o
>> OBJS-$(CONFIG_LIBVPX_VP8_ENCODER) += libvpxenc.o
>> OBJS-$(CONFIG_LIBVPX_VP9_DECODER) += libvpxdec.o
>> OBJS-$(CONFIG_LIBVPX_VP9_ENCODER) += libvpxenc.o
>> +OBJS-$(CONFIG_LIBVVENC_ENCODER) += libvvenc.o
>> OBJS-$(CONFIG_LIBWEBP_ENCODER) += libwebpenc_common.o libwebpenc.o
>> OBJS-$(CONFIG_LIBWEBP_ANIM_ENCODER) += libwebpenc_common.o libwebpenc_animencoder.o
>> OBJS-$(CONFIG_LIBX262_ENCODER) += libx264.o
>> diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
>> index b102a8069e..59d36dbd56 100644
>> --- a/libavcodec/allcodecs.c
>> +++ b/libavcodec/allcodecs.c
>> @@ -800,6 +800,7 @@ extern const FFCodec ff_libvpx_vp8_encoder;
>> extern const FFCodec ff_libvpx_vp8_decoder;
>> extern FFCodec ff_libvpx_vp9_encoder;
>> extern const FFCodec ff_libvpx_vp9_decoder;
>> +extern const FFCodec ff_libvvenc_encoder;
>> /* preferred over libwebp */
>> extern const FFCodec ff_libwebp_anim_encoder;
>> extern const FFCodec ff_libwebp_encoder;
>> diff --git a/libavcodec/libvvenc.c b/libavcodec/libvvenc.c
>> new file mode 100644
>> index 0000000000..78d4f55a2a
>> --- /dev/null
>> +++ b/libavcodec/libvvenc.c
>> @@ -0,0 +1,566 @@
>> +/*
>> + * H.266 encoding using the VVenC library
>> + *
>> + * Copyright (C) 2022, Thomas Siedel
>> + *
>> + * 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
>> + */
>> +
>> +#include "config_components.h"
>
> Seems unneeded.
>
Thanks, I will remove it.
>> +
>> +#include <vvenc/vvenc.h>
>> +#include <vvenc/vvencCfg.h>
>> +#include <vvenc/version.h>
>> +
>> +#include "avcodec.h"
>> +#include "codec_internal.h"
>> +#include "encode.h"
>> +#include "internal.h"
>> +#include "packet_internal.h"
>> +#include "profiles.h"
>> +
>> +#include "libavutil/avutil.h"
>> +#include "libavutil/mem.h"
>> +#include "libavutil/pixdesc.h"
>> +#include "libavutil/opt.h"
>> +#include "libavutil/common.h"
>> +#include "libavutil/imgutils.h"
>> +#include "libavutil/frame.h"
>> +#include "libavutil/log.h"
>> +
>> +typedef struct VVenCOptions {
>> + int preset; // preset 0: faster 4: slower
>> + int qp; // quantization parameter 0-63
>> + int subjectiveOptimization; // perceptually motivated QP adaptation, XPSNR based
>> + int flag8bitCoding; // encode in 8bit instead of 10bit
>> + int intraRefreshSec; // intra period/refresh in seconds
>> + int levelIdc; // vvc level_idc
>> + int tier; // vvc tier
>> + AVDictionary *vvenc_opts;
>> +} VVenCOptions;
>> +
>> +typedef struct VVenCContext {
>> + AVClass *av_class;
>> + VVenCOptions options; // encoder options
>> + vvencEncoder *vvencEnc;
>> + vvencAccessUnit *pAU;
>> + bool encodeDone;
>
> We do not use CamelCase for variable and struct member names.
>
Thanks, I will change the members.
>> +} VVenCContext;
>> +
>> +
>> +static av_cold void ff_vvenc_log_callback(void *ctx, int level,
>> + const char *fmt, va_list args)
>
> Remove the ff_ prefix from static functions (this is the prefix for
> non-static functions).
>
Thanks. I was not aware of this.
>> +{
>> + vvenc_config params;
>> + vvencEncoder *vvencEnc = (vvencEncoder *)ctx;
>> + if (vvencEnc){
>> + vvenc_config_default(¶ms);
>> + vvenc_get_config(vvencEnc, ¶ms);
>> + if ((int)params.m_verbosity >= level)
>> + vfprintf(level == 1 ? stderr : stdout, fmt, args);
>> + }
>> +}
>> +
>> +static void ff_vvenc_set_verbository(vvenc_config* params )
>
> What's the reason for the space before the closing parentheses?
> Moreover, we prefer to put the * to the parameter, not the type.
>
Thanks, it´s a type. I´ll fix it.
>> +{
>> + params->m_verbosity = VVENC_VERBOSE;
>> + if (av_log_get_level() >= AV_LOG_DEBUG)
>
> av_log_get_level() can be changed at any time by someone else. So better
> just call it once and cache the value for simplicity and consistency.
>
Good point. I´ll change it.
>> + params->m_verbosity = VVENC_DETAILS;
>> + else if (av_log_get_level() >= AV_LOG_VERBOSE)
>> + params->m_verbosity = VVENC_NOTICE; // output per picture info
>> + else if (av_log_get_level() >= AV_LOG_INFO)
>> + params->m_verbosity = VVENC_WARNING; // ffmpeg default ffmpeg loglevel
>> + else
>> + params->m_verbosity = VVENC_SILENT;
>> +}
>> +
>> +static int ff_vvenc_set_pic_format(AVCodecContext *avctx, vvenc_config* params )
>> +{
>> + VVenCContext *s =(VVenCContext *) avctx->priv_data;
>
> This cast is unnecessary in C.
>
Thanks. I´ll remove it.
>> +
>> + params->m_internChromaFormat = VVENC_CHROMA_420;
>> + params->m_inputBitDepth[0] = 10;
>> +
>> + if (avctx->pix_fmt != AV_PIX_FMT_YUV420P10LE){
>> + av_log(avctx, AV_LOG_ERROR,
>> + "unsupported pixel format %s, currently only support for yuv420p10le\n",
>> + av_get_pix_fmt_name(avctx->pix_fmt));
>> + return AVERROR(EINVAL);
>
> 1. This whole block of code is dead: For encoders that set
> AVCodec.pix_fmts, it is checked generically that the pixel format set is
> one of the pixel formats in AVCodec.pix_fmts.
> 2. Why only LE? Is this also true on BE systems?
>
It was just for double check. I change it to AV_PIX_FMT_YUV420P10
and remove the check.
>> + }
>> +
>> + if (s->options.flag8bitCoding) {
>> +#if VVENC_VERSION_MAJOR > 1 || (VVENC_VERSION_MAJOR == 1 && VVENC_VERSION_MINOR > 9) || (VVENC_VERSION_MAJOR == 1 && VVENC_VERSION_MINOR >= 9 && VVENC_VERSION_PATCH >= 1)
>> + params->m_internalBitDepth[0] = 8;
>> +#else
>> + av_log(avctx, AV_LOG_ERROR,
>> + "unsupported 8bit coding mode. 8bit coding needs at least vvenc version >= 1.9.1 "
>> + "(current version %s)\n", vvenc_get_version() );
>> + return AVERROR(EINVAL);
>> +#endif
>> + }
>> + return 0;
>> +}
>> +
>> +static void ff_vvenc_set_color_format(AVCodecContext *avctx, vvenc_config* params )
>> +{
>> + if (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED)
>> + params->m_colourPrimaries = (int) avctx->color_primaries;
>> + if (avctx->colorspace != AVCOL_SPC_UNSPECIFIED)
>> + params->m_matrixCoefficients = (int) avctx->colorspace;
>> + if (avctx->color_trc != AVCOL_TRC_UNSPECIFIED) {
>> + params->m_transferCharacteristics = (int) avctx->color_trc;
>> +
>> + if (avctx->color_trc == AVCOL_TRC_SMPTE2084)
>> + params->m_HdrMode = (avctx->color_primaries == AVCOL_PRI_BT2020) ?
>> + VVENC_HDR_PQ_BT2020 : VVENC_HDR_PQ;
>> + else if (avctx->color_trc == AVCOL_TRC_BT2020_10
>> + || avctx->color_trc == AVCOL_TRC_ARIB_STD_B67)
>> + params->m_HdrMode = (avctx->color_trc == AVCOL_TRC_BT2020_10 ||
>> + avctx->color_primaries == AVCOL_PRI_BT2020 ||
>> + avctx->colorspace == AVCOL_SPC_BT2020_NCL ||
>> + avctx->colorspace == AVCOL_SPC_BT2020_CL) ?
>> + VVENC_HDR_HLG_BT2020 : VVENC_HDR_HLG;
>> + }
>> +
>> + if (params->m_HdrMode == VVENC_HDR_OFF
>> + && (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED
>> + || avctx->colorspace != AVCOL_SPC_UNSPECIFIED)) {
>> + params->m_vuiParametersPresent = 1;
>> + params->m_colourDescriptionPresent = true;
>> + }
>> +}
>> +
>> +static void ff_vvenc_set_framerate(AVCodecContext *avctx, vvenc_config* params )
>> +{
>> + params->m_FrameRate = avctx->time_base.den;
>> + params->m_FrameScale = avctx->time_base.num;
>> +
>> +FF_DISABLE_DEPRECATION_WARNINGS
>> +
>> +#if FF_API_TICKS_PER_FRAME
>> + if (avctx->ticks_per_frame == 1) {
>> +#endif
>> + params->m_TicksPerSecond = -1; // auto mode for ticks per frame = 1
>> +#if FF_API_TICKS_PER_FRAME
>> + } else {
>> + params->m_TicksPerSecond =
>> + ceil((avctx->time_base.den / (double) avctx->time_base.num) *
>> + (double) avctx->ticks_per_frame);
>> + }
>> +#endif
>> +FF_ENABLE_DEPRECATION_WARNINGS
>> +}
>> +
>> +static int ff_vvenc_parse_vvenc_params(AVCodecContext *avctx, vvenc_config* params, char* statsfile )
>> +{
>> + int parse_ret, ret;
>> + VVenCContext *s;
>> + AVDictionaryEntry *en = NULL;
>
> const
>
>> + s =(VVenCContext *) avctx->priv_data;
>
> Can be initialized directly and without the cast.
>
Thanks. I´ll remove it.
>> + ret = 0;
>> +
>> + while ((en = av_dict_get(s->options.vvenc_opts, "", en,
>> + AV_DICT_IGNORE_SUFFIX))) {
>
> av_dict_iterate()
I was using libx264,libx265,libaomenc as reference and aligned to
the existing code.
I will change it to av_dict_iterate()
>
>> + av_log(avctx, AV_LOG_DEBUG, "vvenc_set_param: '%s:%s'\n", en->key,
>> + en->value);
>> + parse_ret = vvenc_set_param(params, en->key, en->value);
>> + switch (parse_ret) {
>> + case VVENC_PARAM_BAD_NAME:
>> + av_log(avctx, AV_LOG_ERROR, "Unknown vvenc option: %s.\n",
>> + en->key);
>> + ret = AVERROR(EINVAL);
>> + break;
>> + case VVENC_PARAM_BAD_VALUE:
>> + av_log(avctx, AV_LOG_ERROR,
>> + "Invalid vvenc value for %s: %s.\n", en->key, en->value);
>> + ret = AVERROR(EINVAL);
>> + break;
>> + default:
>> + break;
>> + }
>> +
>> + if (memcmp(en->key, "rcstatsfile", 11) == 0 ||
>> + memcmp(en->key, "RCStatsFile", 11) == 0) {
>
> This presumes that en->key is at least 11 byte long (and it would accept
> keys only prefixed by rcstatsfile); this need not be true at all. How
> about using av_strcasecmp()?
Good point. I´ll remove the memcmp.
>
>> + strncpy(statsfile, en->value, strlen(statsfile));
>> + statsfile[strlen(statsfile)] = '\0';
>
> Did you test this at all?
> 1. Using strlen(statsfile) means that the filename of the statsfile is
> bounded by strlen("vvenc-rcstats.json"); more precisely, every iteration
> of this code block can grow strlen() of the buffer by one and there is
> no real bounds check, so this could lead to a stack buffer overflow
> (when using a dictionary that contains rcstatsfile entries multiple times).
> 2. We have av_strlcpy() for this.
> 3. But a better approach is to just not copy the string at all, avoiding
> any length restriction on the statsfile: Use const char **statsfile
> instead of the char *statsfile parameter; in encode_init below you just
> initialize statsfile via 'const char *statsfile = "vvenc-rcstats.json";'.
>
> The typical way to pass the statistics of a first pass to an encoder is
> via an allocated buffer, not via a filename for the encoder to read
> from/write to (see AVCodecContext.stats_out, stats_in). Is this possible
> with libvvenc?
>
The 2pass approach was quite some of a hack by using the vvenc-
params.
Thanks for the review and hints to improve it.
I will change the code and align to x264 behavior by using the
'passlogfile'/'stats' options and use 2pass flags AV_CODEC_FLAG_PASS1 |
AV_CODEC_FLAG_PASS2 instead of using vvenc-params.
I also will use the default ffmpeg 2pass logfile 'ffmpeg2pass-0.log' as
default as libx264 is using it.
Currently there is no way in vvenc to use an allocated buffer for
write/read statistics.
>> + }
>> + }
>> + return ret;
>> +}
>> +
>> +static int ff_vvenc_set_rc_mode(AVCodecContext *avctx, vvenc_config* params)
>> +{
>> + if (params->m_RCPass != -1 && params->m_RCNumPasses == 1)
>> + params->m_RCNumPasses = 2; /* enable 2pass mode */
>
> We actually have AV_CODEC_FLAG_PASS1 and AV_CODEC_FLAG_PASS2 that you
> completely ignore.
>
I´ll change the 2pass behavior as mentioned above.
>> +
>> + if(avctx->rc_max_rate) {
>> +#if VVENC_VERSION_MAJOR > 1 || (VVENC_VERSION_MAJOR == 1 && VVENC_VERSION_MINOR > 8)
>
> Maybe you should map all of the major, minor and patch versions to one
> VVENC_VERSION?
>
Thanks I will use a integer version for better readability.
>> + params->m_RCMaxBitrate = avctx->rc_max_rate;
>> +#endif
>> +
>> +#if VVENC_VERSION_MAJOR == 1 && VVENC_VERSION_MINOR < 11
>> + /* rc_max_rate without a bit_rate enables capped CQF mode.
>> + (QP + subj. optimization + max. bitrate) */
>> + if(!avctx->bit_rate) {
>> + av_log( avctx, AV_LOG_ERROR,
>> + "Capped Constant Quality Factor mode (capped CQF) needs at "
>> + "least vvenc version >= 1.11.0 (current version %s)\n",
>> + vvenc_get_version());
>> + return AVERROR(EINVAL);
>> + }
>> +#endif
>> + }
>> + return 0;
>> +}
>> +
>> +static int ff_vvenc_init_extradata(AVCodecContext *avctx, VVenCContext *s)
>> +{
>> + int ret;
>> + if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
>> + ret = vvenc_get_headers(s->vvencEnc, s->pAU);
>> + if (0 != ret) {
>> + av_log(avctx, AV_LOG_ERROR,
>> + "cannot get headers (SPS,PPS) from vvc encoder(vvenc): %s\n",
>> + vvenc_get_last_error(s->vvencEnc));
>> + vvenc_encoder_close(s->vvencEnc);
>> + return AVERROR(EINVAL);
>> + }
>> +
>> + if (s->pAU->payloadUsedSize <= 0) {
>> + vvenc_encoder_close(s->vvencEnc);
>> + return AVERROR_INVALIDDATA;
>> + }
>> +
>> + avctx->extradata_size = s->pAU->payloadUsedSize;
>> + avctx->extradata =
>> + av_mallocz(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
>> + if (!avctx->extradata) {
>> + av_log(avctx, AV_LOG_ERROR,
>> + "Cannot allocate VVC header of size %d.\n",
>> + avctx->extradata_size);
>
> Pointless error message, like so many of your allocation errors.
>
Sorry, but I was using libx265 as reference where all these error
messages are implemented in the same way.
Should I just remove the error messages and return an error?
>> + vvenc_encoder_close(s->vvencEnc);
>> + return AVERROR(ENOMEM);
>> + }
>> +
>> + memcpy(avctx->extradata, s->pAU->payload, avctx->extradata_size);
>> + memset(avctx->extradata + avctx->extradata_size, 0,
>> + AV_INPUT_BUFFER_PADDING_SIZE);
>
> Unnecessary given that you use av_mallocz().
>
Indeed. I´ll remove it.
>> + }
>> + return 0;
>> +}
>> +
>> +static av_cold int ff_vvenc_encode_init(AVCodecContext *avctx)
>> +{
>> + int ret;
>> + int framerate, qp;
>> + VVenCContext *s;
>> + vvenc_config params;
>> + vvencPresetMode preset;
>> + char statsfile[1024] = "vvenc-rcstats.json";
>> +
>> + s = (VVenCContext *) avctx->priv_data;
>> + qp = s->options.qp;
>> + preset = (vvencPresetMode) s->options.preset;
>> +
>> + if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
>> + av_log(avctx, AV_LOG_ERROR,
>> + "ff_vvenc_encode_init::init() interlaced encoding not supported yet\n");
>> + return AVERROR_INVALIDDATA;
>
> This is not invalid data; and there is no need to add the name of the
> function to the error message.
> (And what does "not supported yet" even mean? Does VVC even have
> specific interlaced coding methods? Will this ever be supported?)
>
VVC has interlaces support as HEVC does. There are no special tools
but only signalizing in high-level syntax.
Currently there is no support in vvenc to signalize fields, so I added
an error.
What would be the best way to do it?
Just throw a warning or using an error and return?
I would change it to:
av_log(avctx, AV_LOG_ERROR, "interlaced encoding not supported\n");
return AVERROR(EINVAL);
>> + }
>> +
>> + vvenc_config_default(¶ms);
>> +
>> + framerate = avctx->time_base.den / avctx->time_base.num;
>
> You are aware that the division performed here is an integer division?
> And are you aware that non-cfr content does not really have a framerate?
> And that for cfr content, AVCodecContext.framerate contains the framerate?
>
The interface call 'vvenc_init_default' is expecting an int
framerate and is setting the the ntsc timing depending on the
framerate internally (e.g. 59 is mapped to 60000/1001).
The correct time base is set later in the init call in function
'vvenc_set_framerate'
..
'params->m_FrameRate = avctx->time_base.den;'
'params->m_FrameScale = avctx->time_base.num;'
But indeed I was aware that AVCodecContext.framerate and
AVCodecContext.time_base are handled differently.
I will change it, thanks.
>> + vvenc_init_default(¶ms, avctx->width, avctx->height, framerate,
>> + (qp >= 0) ? 0 : avctx->bit_rate, (qp < 0) ? 32 : qp, preset);
>> +
>> + ff_vvenc_set_verbository(¶ms);
>> +
>> + if (avctx->thread_count > 0)
>> + params.m_numThreads = avctx->thread_count;
>> +
>> + /* GOP settings (IDR/CRA) */
>> + if (avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)
>> + params.m_DecodingRefreshType = VVENC_DRT_IDR;
>> +
>> + if (avctx->gop_size == 1) {
>> + params.m_GOPSize = 1;
>> + params.m_IntraPeriod = 1;
>> + } else {
>> + params.m_IntraPeriodSec = s->options.intraRefreshSec;
>> + }
>> +
>> + params.m_AccessUnitDelimiter = true;
>> + params.m_RCNumPasses = 1;
>> +
>> + params.m_usePerceptQPA = s->options.subjectiveOptimization;
>> + params.m_level = (vvencLevel) s->options.levelIdc;
>> + params.m_levelTier = (vvencTier) s->options.tier;
>> +
>> + ff_vvenc_set_framerate(avctx, ¶ms);
>> +
>> + ret = ff_vvenc_set_pic_format(avctx, ¶ms);
>> + if( ret != 0 )
>> + return ret;
>> +
>> + ff_vvenc_set_color_format(avctx, ¶ms);
>> +
>> + ret = ff_vvenc_parse_vvenc_params(avctx, ¶ms, &statsfile[0]);
>> + if( ret != 0 )
>> + return ret;
>> +
>> +
>> + ret = ff_vvenc_set_rc_mode(avctx, ¶ms);
>> + if( ret != 0 )
>> + return ret;
>> +
>> + s->vvencEnc = vvenc_encoder_create();
>> + if (NULL == s->vvencEnc) {
>
> if (!s->vvencEnc) is the common check here.
>
Thanks.
>> + av_log(avctx, AV_LOG_ERROR, "cannot create vvc encoder (vvenc)\n");
>> + return AVERROR(ENOMEM);
>> + }
>> +
>> + vvenc_set_msg_callback(¶ms, s->vvencEnc, ff_vvenc_log_callback);
>> + ret = vvenc_encoder_open(s->vvencEnc, ¶ms);
>> + if (0 != ret) {
>> + av_log(avctx, AV_LOG_ERROR, "cannot open vvc encoder (vvenc): %s\n",
>> + vvenc_get_last_error(s->vvencEnc));
>> + vvenc_encoder_close(s->vvencEnc);
>> + return AVERROR(EINVAL);
>> + }
>> +
>> + vvenc_get_config(s->vvencEnc, ¶ms); /* get the adapted config */
>> +
>> + av_log(avctx, av_log_get_level(), "vvenc version: %s\n", vvenc_get_version());
>> + av_log(avctx, av_log_get_level(), "%s\n",
>> + vvenc_get_config_as_string(¶ms, params.m_verbosity));
>> +
>> + if (params.m_RCNumPasses == 2) {
>> + ret = vvenc_init_pass(s->vvencEnc, params.m_RCPass - 1, &statsfile[0]);
>> + if (0 != ret) {
>> + av_log(avctx, AV_LOG_ERROR,
>> + "cannot init pass %d for vvc encoder (vvenc): %s\n",
>> + params.m_RCPass, vvenc_get_last_error(s->vvencEnc));
>> + vvenc_encoder_close(s->vvencEnc);
>> + return AVERROR(EINVAL);
>> + }
>> + }
>> +
>> + s->pAU = vvenc_accessUnit_alloc();
>> + if( !s->pAU ){
>> + av_log(avctx, AV_LOG_FATAL, "cannot allocate memory for AU payload\n");
>
> 1. There is a memory leak here, as you don't close the encoder.
> 2. This should be fixed by setting the FF_CODEC_CAP_INIT_CLEANUP
> cap_internal; then you should (in fact, need to) remove all the
> vvenc_encoder_close() calls.
> (This also fixes leaks of pAU (it gets never freed on init errors at all
> and leaks e.g. on ff_vvenc_init_extradata() errors.)
>
Thanks. I will return an error.
I´ll add FF_CODEC_CAP_INIT_CLEANUP and remove all vvenc_encoder_close()
calls.
>> + return AVERROR(ENOMEM);
>> + }
>> + vvenc_accessUnit_alloc_payload(s->pAU, avctx->width * avctx->height);
>> + if( !s->pAU ){
>
> This is a nonsense check: s->pAU is already set and can't be unset by
> the call, so one has to check this somehow else.
>
Thanks for seeing this. It have to be
if (!s->pAU->payload){
I´ll change it.
>> + av_log(avctx, AV_LOG_FATAL, "cannot allocate payload memory of size %d\n",
>
> AV_LOG_ERROR, not AV_LOG_FATAL
>
>> + avctx->width * avctx->height );
>> + return AVERROR(ENOMEM);
>> + }
>> +
>> + ret = ff_vvenc_init_extradata(avctx, s);
>> + if( ret != 0 )
>> + return ret;
>> +
>> + s->encodeDone = false;
>> + return 0;
>> +}
>> +
>> +static av_cold int ff_vvenc_encode_close(AVCodecContext * avctx)
>> +{
>> + VVenCContext *s = (VVenCContext *) avctx->priv_data;
>> + if (s->vvencEnc) {
>> + if (av_log_get_level() >= AV_LOG_VERBOSE)
>> + vvenc_print_summary(s->vvencEnc);
>> +
>> + if (0 != vvenc_encoder_close(s->vvencEnc)) {
>> + av_log(avctx, AV_LOG_ERROR, "cannot close vvenc\n");
>> + return -1;
>
> You should free the access unit even if vvenc_encoder_close() returned
> an error.
Thanks. I´ll move the vvenc_accessUnit_free to the top before
calling vvenc_encoder_close.
>
>> + }
>> + }
>> +
>> + vvenc_accessUnit_free(s->pAU, true);
>
> Can this handle the case in which s->pAU is NULL?
>
Yes, the function vvenc_accessUnit_free will check this internally
as well. However I add a null-check to make it clear.
>> +
>> + return 0;
>> +}
>> +
>> +static av_cold int ff_vvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
>> + const AVFrame *frame, int *got_packet)
>> +{
>> + VVenCContext *s = (VVenCContext *) avctx->priv_data;
>> + vvencYUVBuffer *pyuvbuf;
>> + vvencYUVBuffer yuvbuf;
>> + int pict_type;
>> + int ret;
>> +
>> + pyuvbuf = NULL;
>> + if (frame) {
>> + if (avctx->pix_fmt == AV_PIX_FMT_YUV420P10LE) {
>> + vvenc_YUVBuffer_default(&yuvbuf);
>> + yuvbuf.planes[0].ptr = (int16_t *) frame->data[0];
>> + yuvbuf.planes[1].ptr = (int16_t *) frame->data[1];
>> + yuvbuf.planes[2].ptr = (int16_t *) frame->data[2];
>> +
>> + yuvbuf.planes[0].width = frame->width;
>> + yuvbuf.planes[0].height = frame->height;
>> + /* stride is used in 16bitsamples (16bit) in vvenc, ffmpeg uses stride in bytes */
>> + yuvbuf.planes[0].stride = frame->linesize[0] >> 1;
>> +
>> + yuvbuf.planes[1].width = frame->width >> 1;
>> + yuvbuf.planes[1].height = frame->height >> 1;
>> + yuvbuf.planes[1].stride = frame->linesize[1] >> 1;
>> +
>> + yuvbuf.planes[2].width = frame->width >> 1;
>> + yuvbuf.planes[2].height = frame->height >> 1;
>> + yuvbuf.planes[2].stride = frame->linesize[2] >> 1;
>> +
>> + yuvbuf.cts = frame->pts;
>> + yuvbuf.ctsValid = true;
>> + pyuvbuf = &yuvbuf;
>> + } else {
>> + av_log(avctx, AV_LOG_ERROR,
>> + "unsupported input colorspace! input must be yuv420p10le");
>> + return AVERROR(EINVAL);
>
> Checks like these do not belong in a single encoder; they are simply
> allowed to presume that avctx->pix_fmt does not change during an encode
> session and that all frames reaching the decoder actually have this
> pixel format.
>
Thanks, I will remove the check.
>> + }
>> + }
>> +
>> + if (!s->encodeDone) {
>> + ret = vvenc_encode(s->vvencEnc, pyuvbuf, s->pAU, &s->encodeDone);
>> + if (ret != 0) {
>> + av_log(avctx, AV_LOG_ERROR, "error in vvenc::encode - ret:%d\n",
>> + ret);
>> + return AVERROR(EINVAL);
>
> This is not EINVAL. It is AVERROR_EXTERNAL.
>
Thanks. I´ll change it.
>> + }
>> + } else {
>> + *got_packet = 0;
>
> Unecessary: *got_packet is already zero before we enter this function.
>
good point. I´ll remove it.
>> + return 0;
>> + }
>> +
>> + if (s->pAU->payloadUsedSize > 0) {
>> + ret = ff_get_encode_buffer(avctx, pkt, s->pAU->payloadUsedSize, 0);
>> + if (ret < 0) {
>> + av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
>
> Unnecessary, as ff_get_encode_buffer() already emits its own error messages.
>
Good to know. I´ll remove it.
>> + return ret;
>> + }
>> +
>> + memcpy(pkt->data, s->pAU->payload, s->pAU->payloadUsedSize);
>> +
>> + if (s->pAU->ctsValid)
>> + pkt->pts = s->pAU->cts;
>> + if (s->pAU->dtsValid)
>> + pkt->dts = s->pAU->dts;
>> + pkt->flags |= AV_PKT_FLAG_KEY * s->pAU->rap;
>> +
>> + switch (s->pAU->sliceType) {
>> + case VVENC_I_SLICE:
>> + pict_type = AV_PICTURE_TYPE_I;
>> + break;
>> + case VVENC_P_SLICE:
>> + pict_type = AV_PICTURE_TYPE_P;
>> + break;
>> + case VVENC_B_SLICE:
>> + pict_type = AV_PICTURE_TYPE_B;
>> + break;
>> + default:
>> + av_log(avctx, AV_LOG_ERROR, "Unknown picture type encountered.\n");
>> + return AVERROR_EXTERNAL;
>> + }
>> +
>> + ff_side_data_set_encoder_stats(pkt, 0, NULL, 0, pict_type);
>
> Pointless given that you do not really have meaningful statistics to set.
>
vvenc currently only returns the picture type as statistics.
I added this code to call all basic ffmpeg routines during an encoding
step.
However a picture type is also a statistic value.
Shall I remove this code block or leave it to add more statistics in
upcoming versions?
>> +
>> + *got_packet = 1;
>> +
>> + return 0;
>> + } else {
>> + *got_packet = 0;
>> + return 0;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static const enum AVPixelFormat pix_fmts_vvenc[] = {
>> + AV_PIX_FMT_YUV420P10LE,
>> + AV_PIX_FMT_NONE
>> +};
>> +
>> +#define OFFSET(x) offsetof(VVenCContext, x)
>> +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
>> +static const AVOption libvvenc_options[] = {
>> + {"preset", "set encoding preset(0: faster - 4: slower", OFFSET( options.preset), AV_OPT_TYPE_INT, {.i64 = 2} , 0 , 4 , VE, "preset"},
>> + { "faster", "0", 0, AV_OPT_TYPE_CONST, {.i64 = VVENC_FASTER}, INT_MIN, INT_MAX, VE, "preset" },
>> + { "fast", "1", 0, AV_OPT_TYPE_CONST, {.i64 = VVENC_FAST}, INT_MIN, INT_MAX, VE, "preset" },
>> + { "medium", "2", 0, AV_OPT_TYPE_CONST, {.i64 = VVENC_MEDIUM}, INT_MIN, INT_MAX, VE, "preset" },
>> + { "slow", "3", 0, AV_OPT_TYPE_CONST, {.i64 = VVENC_SLOW}, INT_MIN, INT_MAX, VE, "preset" },
>> + { "slower", "4", 0, AV_OPT_TYPE_CONST, {.i64 = VVENC_SLOWER}, INT_MIN, INT_MAX, VE, "preset" },
>> + { "qp" , "set quantization", OFFSET(options.qp), AV_OPT_TYPE_INT, {.i64 = -1}, -1 , 63 ,VE, "qp_mode" },
>
> The "qp_mode" is unnecessary: AVOption.unit exists to link
> AV_OPT_TYPE_CONST options to the actual options.
>
Thanks, I will remove it.
>> + { "period" , "set (intra) refresh period in seconds", OFFSET(options.intraRefreshSec), AV_OPT_TYPE_INT, {.i64 = 1}, 1 , INT_MAX ,VE,"irefreshsec" },
>
> We actually have AVCodecContext.gop_size, which is in frames
I know, but 'period' is in seconds for convenience and not in
frames.
In vvenc GOP is a hierarchically group of pictures (size 16 or 32).
The decoding refresh (which you are calling gop) is independent and can
be set by using vvenc-params intraperiod=N, where N=frames
>
>> + { "subjopt", "set subjective (perceptually motivated) optimization", OFFSET(options.subjectiveOptimization), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0 , 1, VE},
>> + { "bitdepth8", "set 8bit coding mode", OFFSET(options.flag8bitCoding), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0 , 1, VE},
>
> Why are these separate options and not simply part of vvenc-params?
>
There are vvenc-params options. It´s just for convenience. It´s
also possible to use
-vvenc-params qpa=1:internalbitdepth=8
Shall I remove these options?
>> + { "vvenc-params", "set the vvenc configuration using a :-separated list of key=value parameters", OFFSET(options.vvenc_opts), AV_OPT_TYPE_DICT, { 0 }, 0, 0, VE },
>> + { "levelidc", "vvc level_idc", OFFSET( options.levelIdc), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 105, VE, "levelidc"},
>> + { "0", "auto", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "1", "1" , 0, AV_OPT_TYPE_CONST, {.i64 = 16}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "2", "2" , 0, AV_OPT_TYPE_CONST, {.i64 = 32}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "2.1", "2.1" , 0, AV_OPT_TYPE_CONST, {.i64 = 35}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "3", "3" , 0, AV_OPT_TYPE_CONST, {.i64 = 48}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "3.1", "3.1" , 0, AV_OPT_TYPE_CONST, {.i64 = 51}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "4", "4" , 0, AV_OPT_TYPE_CONST, {.i64 = 64}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "4.1", "4.1" , 0, AV_OPT_TYPE_CONST, {.i64 = 67}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "5", "5" , 0, AV_OPT_TYPE_CONST, {.i64 = 80}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "5.1", "5.1" , 0, AV_OPT_TYPE_CONST, {.i64 = 83}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "5.2", "5.2" , 0, AV_OPT_TYPE_CONST, {.i64 = 86}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "6", "6" , 0, AV_OPT_TYPE_CONST, {.i64 = 96}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "6.1", "6.1" , 0, AV_OPT_TYPE_CONST, {.i64 = 99}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "6.2", "6.2" , 0, AV_OPT_TYPE_CONST, {.i64 = 102}, INT_MIN, INT_MAX, VE, "levelidc"},
>> + { "6.3", "6.3" , 0, AV_OPT_TYPE_CONST, {.i64 = 105}, INT_MIN, INT_MAX, VE, "levelidc"},
>
> We have a generic level option, so there is no need for this list above.
>
Thanks, I will have a look into that.
>> + { "tier", "set vvc tier", OFFSET( options.tier), AV_OPT_TYPE_INT, {.i64 = 0}, 0 , 1 , VE, "tier"},
>> + { "main", "main", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, VE, "tier"},
>> + { "high", "high", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, VE, "tier"},
>> + {NULL}
>> +};
>> +
>> +static const AVClass class_libvvenc = {
>> + .class_name = "libvvenc-vvc encoder",
>> + .item_name = av_default_item_name,
>> + .option = libvvenc_options,
>> + .version = LIBAVUTIL_VERSION_INT,
>> +};
>> +
>> +static const FFCodecDefault vvenc_defaults[] = {
>> + { "b", "0" },
>> + { "g", "-1" },
>> + { NULL },
>> +};
>> +
>> +FFCodec ff_libvvenc_encoder = {
>
> Missing const
It´s not consistent as some codecs are const other not.
e.g. ff_libaom_av1_encoder, ff_libx265_encoder
I´ll fix it.
>
>> + .p.name = "libvvenc",
>> + CODEC_LONG_NAME("H.266 / VVC Encoder VVenC"),
>> + .p.type = AVMEDIA_TYPE_VIDEO,
>> + .p.id <http://p.id/> = AV_CODEC_ID_VVC,
>> + .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_OTHER_THREADS,
>
> You can add AV_CODEC_CAP_DR1
>
Thanks. I´ll add it.
>> + .p.profiles = NULL_IF_CONFIG_SMALL(ff_vvc_profiles),
>> + .p.priv_class = &class_libvvenc,
>> + .p.wrapper_name = "libvvenc",
>> + .priv_data_size = sizeof(VVenCContext),
>> + .p.pix_fmts = pix_fmts_vvenc,
>> + .init = ff_vvenc_encode_init,
>> + FF_CODEC_ENCODE_CB(ff_vvenc_encode_frame),
>> + .close = ff_vvenc_encode_close,
>> + .defaults = vvenc_defaults,
>> + .caps_internal = FF_CODEC_CAP_AUTO_THREADS,
>> +};
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org <mailto:ffmpeg-devel at ffmpeg.org>
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org <mailto:ffmpeg-devel-request at ffmpeg.org> with subject "unsubscribe".
More information about the ffmpeg-devel
mailing list