diff mbox

[FFmpeg-devel,v10,1/2] lavc/svt_hevc: add libsvt hevc encoder wrapper

Message ID 1553847702-11114-1-git-send-email-jing.a.sun@intel.com
State Superseded
Headers show

Commit Message

Jing SUN March 29, 2019, 8:21 a.m. UTC
Signed-off-by: Zhengxu Huang <zhengxu.huang@intel.com>
Signed-off-by: Hassene Tmar <hassene.tmar@intel.com>
Signed-off-by: Jun Zhao <jun.zhao@intel.com>
Signed-off-by: Jing Sun <jing.a.sun@intel.com>
---
 configure                |   4 +
 libavcodec/Makefile      |   1 +
 libavcodec/allcodecs.c   |   1 +
 libavcodec/libsvt_hevc.c | 500 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 506 insertions(+)
 create mode 100644 libavcodec/libsvt_hevc.c

Comments

Zhong Li March 29, 2019, 10:59 a.m. UTC | #1
> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces@ffmpeg.org] On Behalf

> Of Jing Sun

> Sent: Friday, March 29, 2019 4:22 PM

> To: ffmpeg-devel@ffmpeg.org

> Cc: Sun, Jing A <jing.a.sun@intel.com>; Huang, Zhengxu

> <zhengxu.huang@intel.com>; Jun Zhao <jun.zhao@intel.com>; Tmar,

> Hassene <hassene.tmar@intel.com>

> Subject: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc

> encoder wrapper

> 

> Signed-off-by: Zhengxu Huang <zhengxu.huang@intel.com>

> Signed-off-by: Hassene Tmar <hassene.tmar@intel.com>

> Signed-off-by: Jun Zhao <jun.zhao@intel.com>

> Signed-off-by: Jing Sun <jing.a.sun@intel.com>

> ---

>  configure                |   4 +

>  libavcodec/Makefile      |   1 +

>  libavcodec/allcodecs.c   |   1 +

>  libavcodec/libsvt_hevc.c | 500

> +++++++++++++++++++++++++++++++++++++++++++++++

>  4 files changed, 506 insertions(+)

>  create mode 100644 libavcodec/libsvt_hevc.c

> 

> diff --git a/configure b/configure

> index 938ff10..2aabac4 100755

> --- a/configure

> +++ b/configure

> @@ -264,6 +264,7 @@ External library support:

>    --enable-libspeex        enable Speex de/encoding via libspeex [no]

>    --enable-libsrt          enable Haivision SRT protocol via libsrt [no]

>    --enable-libssh          enable SFTP protocol via libssh [no]

> +  --enable-libsvthevc      enable HEVC encoding via svt [no]

>    --enable-libtensorflow   enable TensorFlow as a DNN module backend

>                             for DNN based filters like sr [no]

>    --enable-libtesseract    enable Tesseract, needed for ocr filter [no]

> @@ -1784,6 +1785,7 @@ EXTERNAL_LIBRARY_LIST="

>      libspeex

>      libsrt

>      libssh

> +    libsvthevc

>      libtensorflow

>      libtesseract

>      libtheora

> @@ -3173,6 +3175,7 @@ libshine_encoder_select="audio_frame_queue"

>  libspeex_decoder_deps="libspeex"

>  libspeex_encoder_deps="libspeex"

>  libspeex_encoder_select="audio_frame_queue"

> +libsvt_hevc_encoder_deps="libsvthevc"

>  libtheora_encoder_deps="libtheora"

>  libtwolame_encoder_deps="libtwolame"

>  libvo_amrwbenc_encoder_deps="libvo_amrwbenc"

> @@ -6209,6 +6212,7 @@ enabled libsoxr           && require libsoxr

> soxr.h soxr_create -lsoxr

>  enabled libssh            && require_pkg_config libssh libssh

> libssh/sftp.h sftp_init

>  enabled libspeex          && require_pkg_config libspeex speex

> speex/speex.h speex_decoder_init

>  enabled libsrt            && require_pkg_config libsrt "srt >= 1.3.0"

> srt/srt.h srt_socket

> +enabled libsvthevc        && require_pkg_config libsvthevc SvtHevcEnc

> EbApi.h EbInitHandle

>  enabled libtensorflow     && require libtensorflow tensorflow/c/c_api.h

> TF_Version -ltensorflow

>  enabled libtesseract      && require_pkg_config libtesseract tesseract

> tesseract/capi.h TessBaseAPICreate

>  enabled libtheora         && require libtheora theora/theoraenc.h

> th_info_init -ltheoraenc -ltheoradec -logg

> diff --git a/libavcodec/Makefile b/libavcodec/Makefile index

> 15c43a8..c93e545 100644

> --- a/libavcodec/Makefile

> +++ b/libavcodec/Makefile

> @@ -987,6 +987,7 @@ OBJS-$(CONFIG_LIBOPUS_ENCODER)

> += libopusenc.o libopus.o     \

>  OBJS-$(CONFIG_LIBSHINE_ENCODER)           += libshine.o

>  OBJS-$(CONFIG_LIBSPEEX_DECODER)           += libspeexdec.o

>  OBJS-$(CONFIG_LIBSPEEX_ENCODER)           += libspeexenc.o

> +OBJS-$(CONFIG_LIBSVT_HEVC_ENCODER)        += libsvt_hevc.o

>  OBJS-$(CONFIG_LIBTHEORA_ENCODER)          += libtheoraenc.o

>  OBJS-$(CONFIG_LIBTWOLAME_ENCODER)         += libtwolame.o

>  OBJS-$(CONFIG_LIBVO_AMRWBENC_ENCODER)     +=

> libvo-amrwbenc.o

> diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index

> b26aeca..e93f66f 100644

> --- a/libavcodec/allcodecs.c

> +++ b/libavcodec/allcodecs.c

> @@ -703,6 +703,7 @@ extern AVCodec ff_librsvg_decoder;  extern

> AVCodec ff_libshine_encoder;  extern AVCodec ff_libspeex_encoder;

> extern AVCodec ff_libspeex_decoder;

> +extern AVCodec ff_libsvt_hevc_encoder;

>  extern AVCodec ff_libtheora_encoder;

>  extern AVCodec ff_libtwolame_encoder;

>  extern AVCodec ff_libvo_amrwbenc_encoder; diff --git

> a/libavcodec/libsvt_hevc.c b/libavcodec/libsvt_hevc.c new file mode 100644

> index 0000000..4ef06a8

> --- /dev/null

> +++ b/libavcodec/libsvt_hevc.c

> @@ -0,0 +1,500 @@

> +/*

> +* Scalable Video Technology for HEVC encoder library plugin

> +*

> +* Copyright (c) 2018 Intel Corporation


It is 2019 now.

> +* 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 this program; if not, write to the Free Software

> +* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

> +02110-1301 USA */

> +

> +#include "EbErrorCodes.h"

> +#include "EbTime.h"

> +#include "EbApi.h"

> +

> +#include "libavutil/common.h"

> +#include "libavutil/frame.h"

> +#include "libavutil/opt.h"

> +

> +#include "internal.h"

> +#include "avcodec.h"

> +

> +typedef enum eos_status {

> +    EOS_NOT_REACHED = 0,

> +    EOS_REACHED,

> +    EOS_TOTRIGGER

> +}EOS_STATUS;

> +

> +typedef struct SvtContext {

> +    AVClass *class;

> +

> +    EB_H265_ENC_CONFIGURATION enc_params;

> +    EB_COMPONENTTYPE *svt_handle;

> +    EB_BUFFERHEADERTYPE in_buf;

> +    EOS_STATUS eos_flag;

> +

> +    // User options.

> +    int vui_info;

> +    int hierarchical_level;

> +    int la_depth;

> +    int enc_mode;

> +    int rc_mode;

> +    int scd;

> +    int tune;

> +    int qp;

> +    int hdr;

> +

> +    int forced_idr;

> +

> +    int aud;

> +

> +    int profile;

> +    int tier;

> +    int level;

> +

> +    int base_layer_switch_mode;

> +} SvtContext;

> +

> +static int error_mapping(EB_ERRORTYPE svt_ret) {

> +    int err;

> +

> +    switch (svt_ret) {

> +    case EB_ErrorInsufficientResources:

> +        err = AVERROR(ENOMEM);

> +        break;

> +

> +    case EB_ErrorUndefined:

> +    case EB_ErrorInvalidComponent:

> +    case EB_ErrorBadParameter:

> +        err = AVERROR(EINVAL);

> +        break;

> +

> +    case EB_ErrorDestroyThreadFailed:

> +    case EB_ErrorSemaphoreUnresponsive:

> +    case EB_ErrorDestroySemaphoreFailed:

> +    case EB_ErrorCreateMutexFailed:

> +    case EB_ErrorMutexUnresponsive:

> +    case EB_ErrorDestroyMutexFailed:

> +        err = AVERROR_EXTERNAL;

> +            break;

> +

> +    case EB_NoErrorEmptyQueue:

> +        err = AVERROR(EAGAIN);

> +

> +    case EB_ErrorNone:

> +        err = 0;

> +        break;

> +

> +    default:

> +        err = AVERROR_UNKNOWN;

> +    }


How about add some error message for each case instead of just return?

> +    return err;

> +}

> +

> +static void free_buffer(SvtContext *svt_enc) {

> +    uint8_t *in_data = svt_enc->in_buf.pBuffer;

> +

> +    av_freep(&in_data);

> +}

> +

> +static int alloc_buffer(EB_H265_ENC_CONFIGURATION *config, SvtContext

> +*svt_enc) {

> +    EB_H265_ENC_INPUT *in_data;

> +

> +    // allocate buffer for in and out

> +    in_data  = av_mallocz(sizeof(*in_data));

> +    if (!in_data)

> +        goto failed;

> +    svt_enc->in_buf.pBuffer  = (unsigned char *)in_data;

> +

> +    svt_enc->in_buf.nSize        = sizeof(svt_enc->in_buf);

> +    svt_enc->in_buf.pAppPrivate  = NULL;

> +

> +    return 0;

> +

> +failed:

> +    free_buffer(svt_enc);

> +    return AVERROR(ENOMEM);

> +}

> +

> +static int config_enc_params(EB_H265_ENC_CONFIGURATION *param,

> +                             AVCodecContext *avctx) {

> +    SvtContext *svt_enc = avctx->priv_data;

> +    int ret;

> +

> +    param->sourceWidth = avctx->width;

> +    param->sourceHeight = avctx->height;

> +    param->encoderBitDepth = 8;

> +

> +    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P10LE) {

> +        av_log(avctx, AV_LOG_DEBUG, "Encoder 10 bits depth input\n");

> +

> +        param->encoderBitDepth = 10;

> +    }


Get the bitdepth informat for pix_fmt should be easier to maintain (e.g: 12bit support).
Here is an example: https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/vaapi_encode_h265.c#L295 

> +    param->encoderColorFormat = EB_YUV420;

> +

> +    // Update param from options

> +    param->hierarchicalLevels   = svt_enc->hierarchical_level - 1;

> +    param->encMode              = svt_enc->enc_mode;

> +    param->profile              = svt_enc->profile;

> +    param->tier                 = svt_enc->tier;

> +    param->level                = svt_enc->level;

> +    param->rateControlMode      = svt_enc->rc_mode;

> +    param->sceneChangeDetection = svt_enc->scd;

> +    param->tune                 = svt_enc->tune;

> +    param->baseLayerSwitchMode  =

> svt_enc->base_layer_switch_mode;

> +    param->qp                   = svt_enc->qp;

> +    param->accessUnitDelimiter  = svt_enc->aud;

> +

> +    param->targetBitRate = avctx->bit_rate;

> +    if (avctx->gop_size > 0)

> +        param->intraPeriodLength = avctx->gop_size - 1;

> +

> +    if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {

> +        param->frameRateNumerator = avctx->framerate.num;

> +        param->frameRateDenominator =

> +            avctx->framerate.den * avctx->ticks_per_frame;

> +    } else {

> +        param->frameRateNumerator = avctx->time_base.den;

> +        param->frameRateDenominator =

> +            avctx->time_base.num * avctx->ticks_per_frame;

> +    }

> +

> +    if (param->rateControlMode) {

> +        param->maxQpAllowed = avctx->qmax;

> +        param->minQpAllowed = avctx->qmin;

> +    }

> +

> +    param->intraRefreshType =

> +        !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) + 1;

> +

> +    // is it repeat headers for MP4 or Annex-b

> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)

> +        param->codeVpsSpsPps = 0;

> +    else

> +        param->codeVpsSpsPps = 1;

> +

> +    param->codeEosNal = 1;

> +

> +    if (svt_enc->hdr) {

> +        av_log(avctx, AV_LOG_WARNING, "Color properties not

> supported

> + yet\n");

> +

> +        svt_enc->vui_info = 1;

> +        param->highDynamicRangeInput = svt_enc->hdr;

> +    }

> +

> +    if (svt_enc->vui_info)

> +        param->videoUsabilityInfo = svt_enc->vui_info;

> +

> +    if (svt_enc->la_depth != -1)


Though the option la_depth is limited to [-1, 256], still would be better to be la_dapth >= 0. 

> +        param->lookAheadDistance = svt_enc->la_depth;

> +

> +    ret = alloc_buffer(param, svt_enc);

> +

> +    return ret;

> +}

> +

> +static void read_in_data(EB_H265_ENC_CONFIGURATION *config,

> +                         const AVFrame *frame,

> +                         EB_BUFFERHEADERTYPE *header_ptr) {

> +    uint8_t is16bit = config->encoderBitDepth > 8;

> +    uint64_t luma_size =

> +        (uint64_t)config->sourceWidth * config->sourceHeight<< is16bit;

> +    EB_H265_ENC_INPUT *in_data = (EB_H265_ENC_INPUT

> +*)header_ptr->pBuffer;

> +

> +    // support yuv420p and yuv420p010


Probably you can give an assert here. Comment for limited format can't avoid any run-time issue.

> +    in_data->luma = frame->data[0];

> +    in_data->cb   = frame->data[1];

> +    in_data->cr   = frame->data[2];

> +

> +    // stride info

> +    in_data->yStride  = frame->linesize[0] >> is16bit;

> +    in_data->cbStride = frame->linesize[1] >> is16bit;

> +    in_data->crStride = frame->linesize[2] >> is16bit;

> +

> +    header_ptr->nFilledLen   += luma_size * 3/2u;

> +}

> +

> +static av_cold int eb_enc_init(AVCodecContext *avctx) {

> +    SvtContext   *svt_enc = avctx->priv_data;

> +    EB_ERRORTYPE svt_ret;

> +

> +    svt_enc->eos_flag = EOS_NOT_REACHED;

> +

> +    svt_ret = EbInitHandle(&svt_enc->svt_handle, svt_enc,

> &svt_enc->enc_params);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error init encoder handle\n");

> +        goto failed;

> +    }

> +

> +    svt_ret = config_enc_params(&svt_enc->enc_params, avctx);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error configure encoder

> parameters\n");

> +        goto failed_init_handle;

> +    }

> +

> +    svt_ret = EbH265EncSetParameter(svt_enc->svt_handle,

> &svt_enc->enc_params);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error setting encoder

> parameters\n");

> +        goto failed_init_handle;

> +    }

> +

> +    svt_ret = EbInitEncoder(svt_enc->svt_handle);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error init encoder\n");

> +        goto failed_init_handle;

> +    }

> +

> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {

> +        EB_BUFFERHEADERTYPE *header_ptr = NULL;

> +

> +        svt_ret = EbH265EncStreamHeader(svt_enc->svt_handle,

> &header_ptr);

> +        if (svt_ret != EB_ErrorNone) {

> +            av_log(avctx, AV_LOG_ERROR, "Error when build stream

> header.\n");

> +            goto failed_init_enc;

> +        }

> +

> +        avctx->extradata_size = header_ptr->nFilledLen;

> +        avctx->extradata = av_mallocz(avctx->extradata_size +

> AV_INPUT_BUFFER_PADDING_SIZE);


The PADDING memory should be memset to zero.

> +        if (!avctx->extradata) {

> +            av_log(avctx, AV_LOG_ERROR,

> +                   "Cannot allocate HEVC header of size %d.\n",

> avctx->extradata_size);

> +            svt_ret = EB_ErrorInsufficientResources;

> +            goto failed_init_enc;

> +        }

> +        memcpy(avctx->extradata, header_ptr->pBuffer,

> avctx->extradata_size);

> +    }

> +

> +    return 0;

> +

> +failed_init_enc:

> +    EbDeinitEncoder(svt_enc->svt_handle);

> +failed_init_handle:

> +    EbDeinitHandle(svt_enc->svt_handle);

> +failed:

> +    free_buffer(svt_enc);

> +    svt_enc->svt_handle = NULL;

> +    svt_enc = NULL;

> +    return error_mapping(svt_ret);

> +}

> +

> +static int eb_send_frame(AVCodecContext *avctx, const AVFrame *frame) {

> +    SvtContext           *svt_enc = avctx->priv_data;

> +    EB_BUFFERHEADERTYPE  *header_ptr = &svt_enc->in_buf;

> +

> +    if (!frame) {

> +        EB_BUFFERHEADERTYPE header_ptr_last;

> +        header_ptr_last.nAllocLen   = 0;

> +        header_ptr_last.nFilledLen  = 0;

> +        header_ptr_last.nTickCount  = 0;

> +        header_ptr_last.pAppPrivate = NULL;

> +        header_ptr_last.pBuffer     = NULL;

> +        header_ptr_last.nFlags      = EB_BUFFERFLAG_EOS;

> +

> +        EbH265EncSendPicture(svt_enc->svt_handle, &header_ptr_last);

> +        svt_enc->eos_flag = EOS_REACHED;

> +        av_log(avctx, AV_LOG_DEBUG, "Finish sending frames!!!\n");

> +        return 0;

> +    }

> +

> +    read_in_data(&svt_enc->enc_params, frame, header_ptr);

> +

> +    header_ptr->nFlags       = 0;

> +    header_ptr->pAppPrivate  = NULL;

> +    header_ptr->pts          = frame->pts;

> +    switch (frame->pict_type) {

> +    case AV_PICTURE_TYPE_I:

> +        header_ptr->sliceType = svt_enc->forced_idr > 0 ?

> EB_IDR_PICTURE : EB_I_PICTURE;

> +        break;

> +    case AV_PICTURE_TYPE_P:

> +        header_ptr->sliceType = EB_P_PICTURE;

> +        break;

> +    case AV_PICTURE_TYPE_B:

> +        header_ptr->sliceType = EB_B_PICTURE;

> +        break;

> +    default:

> +        header_ptr->sliceType = EB_INVALID_PICTURE;

> +        break;

> +    }

> +    EbH265EncSendPicture(svt_enc->svt_handle, header_ptr);

> +

> +    return 0;

> +}

> +

> +static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) {

> +    SvtContext  *svt_enc = avctx->priv_data;

> +    EB_BUFFERHEADERTYPE   *header_ptr = NULL;

> +    EB_ERRORTYPE          svt_ret;

> +    int ret = 0;

> +

> +    if (EOS_TOTRIGGER == svt_enc->eos_flag) {

> +        pkt = NULL;

> +        return AVERROR_EOF;

> +    }

> +

> +    svt_ret = EbH265GetPacket(svt_enc->svt_handle, &header_ptr,

> svt_enc->eos_flag);

> +    if (svt_ret == EB_NoErrorEmptyQueue)

> +        return AVERROR(EAGAIN);

> +

> +    if ((ret = ff_alloc_packet2(avctx, pkt, header_ptr->nFilledLen, 0)) < 0) {

> +        av_log(avctx, AV_LOG_ERROR, "Failed to allocate output

> packet.\n");

> +        EbH265ReleaseOutBuffer(&header_ptr);

> +        return ret;

> +    }

> +

> +    memcpy(pkt->data, header_ptr->pBuffer, header_ptr->nFilledLen);

> +

> +    pkt->size = header_ptr->nFilledLen;

> +    pkt->pts  = header_ptr->pts;

> +    pkt->dts  = header_ptr->dts;

> +    if ((header_ptr->sliceType == EB_IDR_PICTURE) ||

> +        (header_ptr->sliceType == EB_I_PICTURE))

> +        pkt->flags |= AV_PKT_FLAG_KEY;

> +    if (header_ptr->sliceType == EB_NON_REF_PICTURE)

> +        pkt->flags |= AV_PKT_FLAG_DISPOSABLE;

> +

> +    EbH265ReleaseOutBuffer(&header_ptr);

> +

> +    if (EB_BUFFERFLAG_EOS == header_ptr->nFlags)

> +        svt_enc->eos_flag = EOS_TOTRIGGER;

> +

> +    return 0;

> +}

> +

> +static av_cold int eb_enc_close(AVCodecContext *avctx) {

> +    SvtContext *svt_enc = avctx->priv_data;

> +

> +    if (svt_enc) {

> +        if (svt_enc->svt_handle) {

> +            EbDeinitEncoder(svt_enc->svt_handle);

> +            EbDeinitHandle(svt_enc->svt_handle);

> +            svt_enc->svt_handle = NULL;

> +        }

> +

> +        free_buffer(svt_enc);

> +        svt_enc = NULL;

> +    }

> +

> +    return 0;

> +}

> +

> +#define OFFSET(x) offsetof(SvtContext, x) #define VE

> +AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM static

> const

> +AVOption options[] = {

> +    { "vui", "Enable Video Usability Info", OFFSET(vui_info),

> +      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },

> +

> +    { "aud", "Include Access Unit Delimiter", OFFSET(aud),

> +      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },

> +

> +    { "hielevel", "Hierarchical prediction levels setting",

> OFFSET(hierarchical_level),

> +      AV_OPT_TYPE_INT, { .i64 = 4 }, 1, 4, VE , "hielevel"},

> +        { "flat",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },

> INT_MIN, INT_MAX, VE, "hielevel" },

> +        { "2level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 },  INT_MIN,

> INT_MAX, VE, "hielevel" },

> +        { "3level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 },  INT_MIN,

> INT_MAX, VE, "hielevel" },

> +        { "4level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 4 },

> INT_MIN,

> + INT_MAX, VE, "hielevel" },

> +

> +    { "la_depth", "Look ahead distance [0, 256]", OFFSET(la_depth),

> +      AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 256, VE },

> +

> +    { "preset", "Encoding preset [0, 12] (e,g, for subjective quality tuning

> mode and >=4k resolution), [0, 10] (for >= 1080p resolution), [0, 9] (for all

> resolution and modes)",

> +      OFFSET(enc_mode), AV_OPT_TYPE_INT, { .i64 = 9 }, 0, 12, VE },

> +

> +    { "profile", "Profile setting, Main Still Picture Profile not supported",

> OFFSET(profile),

> +      AV_OPT_TYPE_INT, { .i64 = FF_PROFILE_HEVC_MAIN_10 },


I am quite surprised why you set main10 as default profile. 
You added a debug message for 10bit input, but not added for 8bit case, doesn't mean 8bit case is more common? 

> + FF_PROFILE_HEVC_MAIN, FF_PROFILE_HEVC_MAIN_10, VE, "profile"},

> +

> +    { "tier", "Set tier (general_tier_flag)", OFFSET(tier),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE, "tier" },

> +        { "main", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, VE,

> "tier" },

> +        { "high", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, VE,

> + "tier" },

> +

> +    { "level", "Set level (level_idc)", OFFSET(level),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 0xff, VE, "level" },

> +

> +    { "rc", "Bit rate control mode", OFFSET(rc_mode),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE , "rc"},

> +        { "cqp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN,

> INT_MAX, VE, "rc" },

> +        { "vbr", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN,

> + INT_MAX, VE, "rc" },


Hm. VBR mode can be support but I haven't seen any parameter to set max_rate/vbv buffer. 
Does this mean the bitrate fluctuation is totally out of control?

> +    { "qp", "QP value for intra frames", OFFSET(qp),

> +      AV_OPT_TYPE_INT, { .i64 = 32 }, 0, 51, VE },

> +

> +    { "sc_detection", "Scene change detection", OFFSET(scd),

> +      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },

> +

> +    { "tune", "Quality tuning mode", OFFSET(tune), AV_OPT_TYPE_INT,

> { .i64 = 1 }, 0, 2, VE, "tune" },

> +        { "sq", "Visually optimized mode", 0,

> +          AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN, INT_MAX, VE,

> "tune" },

> +        { "oq",  "PSNR / SSIM optimized mode",  0,

> +          AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, INT_MAX, VE,

> "tune" },

> +        { "vmaf", "VMAF optimized mode", 0,

> +          AV_OPT_TYPE_CONST, { .i64 = 2 },  INT_MIN, INT_MAX, VE,

> + "tune" },

> +

> +    { "bl_mode", "Random Access Prediction Structure type setting",

> OFFSET(base_layer_switch_mode),

> +      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },

> +

> +    { "forced-idr", "If forcing keyframes, force them as IDR frames.",

> OFFSET(forced_idr),

> +      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, -1, 1, VE },

> +

> +    { "hdr", "High dynamic range input", OFFSET(hdr),

> +      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, 0, 1, VE },

> +

> +    {NULL},

> +};

> +

> +static const AVClass class = {

> +    .class_name = "libsvt_hevc",

> +    .item_name  = av_default_item_name,

> +    .option     = options,

> +    .version    = LIBAVUTIL_VERSION_INT,

> +};

> +

> +static const AVCodecDefault eb_enc_defaults[] = {

> +    { "b",         "7M"    },

> +    { "flags",     "+cgop" },

> +    { "qmin",      "10"    },

> +    { "qmax",      "48"    },

> +    { "g",         "-2"    },

> +    { NULL },

> +};

> +

> +AVCodec ff_libsvt_hevc_encoder = {

> +    .name           = "libsvt_hevc",

> +    .long_name      = NULL_IF_CONFIG_SMALL("SVT-HEVC(Scalable

> Video Technology for HEVC) encoder"),

> +    .priv_data_size = sizeof(SvtContext),

> +    .type           = AVMEDIA_TYPE_VIDEO,

> +    .id             = AV_CODEC_ID_HEVC,

> +    .init           = eb_enc_init,

> +    .send_frame     = eb_send_frame,

> +    .receive_packet = eb_receive_packet,

> +    .close          = eb_enc_close,

> +    .capabilities   = AV_CODEC_CAP_DELAY |

> AV_CODEC_CAP_AUTO_THREADS,

> +    .pix_fmts       = (const enum

> AVPixelFormat[]){ AV_PIX_FMT_YUV420P,

> +

> AV_PIX_FMT_YUV420P10,

> +

> AV_PIX_FMT_NONE },

> +    .priv_class     = &class,

> +    .defaults       = eb_enc_defaults,

> +    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP,

> +    .wrapper_name   = "libsvt_hevc",

> +};

> --

> 1.8.3.1
Jing SUN April 2, 2019, 6:02 a.m. UTC | #2
-----Original Message-----
From: Li, Zhong 

Sent: Friday, March 29, 2019 6:59 PM
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Cc: Sun, Jing A <jing.a.sun@intel.com>; Huang, Zhengxu <zhengxu.huang@intel.com>; Jun Zhao <jun.zhao@intel.com>; Tmar, Hassene <hassene.tmar@intel.com>
Subject: RE: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc encoder wrapper

> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces@ffmpeg.org] On Behalf 

> Of Jing Sun

> Sent: Friday, March 29, 2019 4:22 PM

> To: ffmpeg-devel@ffmpeg.org

> Cc: Sun, Jing A <jing.a.sun@intel.com>; Huang, Zhengxu 

> <zhengxu.huang@intel.com>; Jun Zhao <jun.zhao@intel.com>; Tmar, 

> Hassene <hassene.tmar@intel.com>

> Subject: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc 

> encoder wrapper

> 

> Signed-off-by: Zhengxu Huang <zhengxu.huang@intel.com>

> Signed-off-by: Hassene Tmar <hassene.tmar@intel.com>

> Signed-off-by: Jun Zhao <jun.zhao@intel.com>

> Signed-off-by: Jing Sun <jing.a.sun@intel.com>

> ---

>  configure                |   4 +

>  libavcodec/Makefile      |   1 +

>  libavcodec/allcodecs.c   |   1 +

>  libavcodec/libsvt_hevc.c | 500

> +++++++++++++++++++++++++++++++++++++++++++++++

>  4 files changed, 506 insertions(+)

>  create mode 100644 libavcodec/libsvt_hevc.c

> 

> diff --git a/configure b/configure

> index 938ff10..2aabac4 100755

> --- a/configure

> +++ b/configure

> @@ -264,6 +264,7 @@ External library support:

>    --enable-libspeex        enable Speex de/encoding via libspeex [no]

>    --enable-libsrt          enable Haivision SRT protocol via libsrt [no]

>    --enable-libssh          enable SFTP protocol via libssh [no]

> +  --enable-libsvthevc      enable HEVC encoding via svt [no]

>    --enable-libtensorflow   enable TensorFlow as a DNN module backend

>                             for DNN based filters like sr [no]

>    --enable-libtesseract    enable Tesseract, needed for ocr filter [no]

> @@ -1784,6 +1785,7 @@ EXTERNAL_LIBRARY_LIST="

>      libspeex

>      libsrt

>      libssh

> +    libsvthevc

>      libtensorflow

>      libtesseract

>      libtheora

> @@ -3173,6 +3175,7 @@ libshine_encoder_select="audio_frame_queue"

>  libspeex_decoder_deps="libspeex"

>  libspeex_encoder_deps="libspeex"

>  libspeex_encoder_select="audio_frame_queue"

> +libsvt_hevc_encoder_deps="libsvthevc"

>  libtheora_encoder_deps="libtheora"

>  libtwolame_encoder_deps="libtwolame"

>  libvo_amrwbenc_encoder_deps="libvo_amrwbenc"

> @@ -6209,6 +6212,7 @@ enabled libsoxr           && require libsoxr

> soxr.h soxr_create -lsoxr

>  enabled libssh            && require_pkg_config libssh libssh

> libssh/sftp.h sftp_init

>  enabled libspeex          && require_pkg_config libspeex speex

> speex/speex.h speex_decoder_init

>  enabled libsrt            && require_pkg_config libsrt "srt >= 1.3.0"

> srt/srt.h srt_socket

> +enabled libsvthevc        && require_pkg_config libsvthevc SvtHevcEnc

> EbApi.h EbInitHandle

>  enabled libtensorflow     && require libtensorflow tensorflow/c/c_api.h

> TF_Version -ltensorflow

>  enabled libtesseract      && require_pkg_config libtesseract tesseract

> tesseract/capi.h TessBaseAPICreate

>  enabled libtheora         && require libtheora theora/theoraenc.h

> th_info_init -ltheoraenc -ltheoradec -logg diff --git 

> a/libavcodec/Makefile b/libavcodec/Makefile index

> 15c43a8..c93e545 100644

> --- a/libavcodec/Makefile

> +++ b/libavcodec/Makefile

> @@ -987,6 +987,7 @@ OBJS-$(CONFIG_LIBOPUS_ENCODER)

> += libopusenc.o libopus.o     \

>  OBJS-$(CONFIG_LIBSHINE_ENCODER)           += libshine.o

>  OBJS-$(CONFIG_LIBSPEEX_DECODER)           += libspeexdec.o

>  OBJS-$(CONFIG_LIBSPEEX_ENCODER)           += libspeexenc.o

> +OBJS-$(CONFIG_LIBSVT_HEVC_ENCODER)        += libsvt_hevc.o

>  OBJS-$(CONFIG_LIBTHEORA_ENCODER)          += libtheoraenc.o

>  OBJS-$(CONFIG_LIBTWOLAME_ENCODER)         += libtwolame.o

>  OBJS-$(CONFIG_LIBVO_AMRWBENC_ENCODER)     +=

> libvo-amrwbenc.o

> diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index 

> b26aeca..e93f66f 100644

> --- a/libavcodec/allcodecs.c

> +++ b/libavcodec/allcodecs.c

> @@ -703,6 +703,7 @@ extern AVCodec ff_librsvg_decoder;  extern AVCodec 

> ff_libshine_encoder;  extern AVCodec ff_libspeex_encoder; extern 

> AVCodec ff_libspeex_decoder;

> +extern AVCodec ff_libsvt_hevc_encoder;

>  extern AVCodec ff_libtheora_encoder;

>  extern AVCodec ff_libtwolame_encoder;  extern AVCodec 

> ff_libvo_amrwbenc_encoder; diff --git a/libavcodec/libsvt_hevc.c 

> b/libavcodec/libsvt_hevc.c new file mode 100644 index 0000000..4ef06a8

> --- /dev/null

> +++ b/libavcodec/libsvt_hevc.c

> @@ -0,0 +1,500 @@

> +/*

> +* Scalable Video Technology for HEVC encoder library plugin

> +*

> +* Copyright (c) 2018 Intel Corporation


It is 2019 now.

[Jing] To be modified.

> +* 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 this program; if not, write to the Free Software

> +* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

> +02110-1301 USA */

> +

> +#include "EbErrorCodes.h"

> +#include "EbTime.h"

> +#include "EbApi.h"

> +

> +#include "libavutil/common.h"

> +#include "libavutil/frame.h"

> +#include "libavutil/opt.h"

> +

> +#include "internal.h"

> +#include "avcodec.h"

> +

> +typedef enum eos_status {

> +    EOS_NOT_REACHED = 0,

> +    EOS_REACHED,

> +    EOS_TOTRIGGER

> +}EOS_STATUS;

> +

> +typedef struct SvtContext {

> +    AVClass *class;

> +

> +    EB_H265_ENC_CONFIGURATION enc_params;

> +    EB_COMPONENTTYPE *svt_handle;

> +    EB_BUFFERHEADERTYPE in_buf;

> +    EOS_STATUS eos_flag;

> +

> +    // User options.

> +    int vui_info;

> +    int hierarchical_level;

> +    int la_depth;

> +    int enc_mode;

> +    int rc_mode;

> +    int scd;

> +    int tune;

> +    int qp;

> +    int hdr;

> +

> +    int forced_idr;

> +

> +    int aud;

> +

> +    int profile;

> +    int tier;

> +    int level;

> +

> +    int base_layer_switch_mode;

> +} SvtContext;

> +

> +static int error_mapping(EB_ERRORTYPE svt_ret) {

> +    int err;

> +

> +    switch (svt_ret) {

> +    case EB_ErrorInsufficientResources:

> +        err = AVERROR(ENOMEM);

> +        break;

> +

> +    case EB_ErrorUndefined:

> +    case EB_ErrorInvalidComponent:

> +    case EB_ErrorBadParameter:

> +        err = AVERROR(EINVAL);

> +        break;

> +

> +    case EB_ErrorDestroyThreadFailed:

> +    case EB_ErrorSemaphoreUnresponsive:

> +    case EB_ErrorDestroySemaphoreFailed:

> +    case EB_ErrorCreateMutexFailed:

> +    case EB_ErrorMutexUnresponsive:

> +    case EB_ErrorDestroyMutexFailed:

> +        err = AVERROR_EXTERNAL;

> +            break;

> +

> +    case EB_NoErrorEmptyQueue:

> +        err = AVERROR(EAGAIN);

> +

> +    case EB_ErrorNone:

> +        err = 0;

> +        break;

> +

> +    default:

> +        err = AVERROR_UNKNOWN;

> +    }


How about add some error message for each case instead of just return?

[Jing] When an error happens, logs are dumped before this func is called, such as:
        ret = EbH265EncStreamHeader(svt_enc->svt_handle, &headerPtr);
        if (ret != EB_ErrorNone) {
            av_log(avctx, AV_LOG_ERROR, "Error when build stream header.\n");
            av_freep(&headerPtr.pBuffer);
            goto failed_init_enc;
        }

> +    return err;

> +}

> +

> +static void free_buffer(SvtContext *svt_enc) {

> +    uint8_t *in_data = svt_enc->in_buf.pBuffer;

> +

> +    av_freep(&in_data);

> +}

> +

> +static int alloc_buffer(EB_H265_ENC_CONFIGURATION *config, SvtContext

> +*svt_enc) {

> +    EB_H265_ENC_INPUT *in_data;

> +

> +    // allocate buffer for in and out

> +    in_data  = av_mallocz(sizeof(*in_data));

> +    if (!in_data)

> +        goto failed;

> +    svt_enc->in_buf.pBuffer  = (unsigned char *)in_data;

> +

> +    svt_enc->in_buf.nSize        = sizeof(svt_enc->in_buf);

> +    svt_enc->in_buf.pAppPrivate  = NULL;

> +

> +    return 0;

> +

> +failed:

> +    free_buffer(svt_enc);

> +    return AVERROR(ENOMEM);

> +}

> +

> +static int config_enc_params(EB_H265_ENC_CONFIGURATION *param,

> +                             AVCodecContext *avctx) {

> +    SvtContext *svt_enc = avctx->priv_data;

> +    int ret;

> +

> +    param->sourceWidth = avctx->width;

> +    param->sourceHeight = avctx->height;

> +    param->encoderBitDepth = 8;

> +

> +    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P10LE) {

> +        av_log(avctx, AV_LOG_DEBUG, "Encoder 10 bits depth input\n");

> +

> +        param->encoderBitDepth = 10;

> +    }


Get the bitdepth informat for pix_fmt should be easier to maintain (e.g: 12bit support).
Here is an example: https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/vaapi_encode_h265.c#L295 

[Jing] It requires an extra step av_pix_fmt_desc_get(). Currently only 420 is supported, and checking avctx->pix_fmt is more straightforward. When 422 and 444 are supported later, I will change to the suggested method.
 
> +    param->encoderColorFormat = EB_YUV420;

> +

> +    // Update param from options

> +    param->hierarchicalLevels   = svt_enc->hierarchical_level - 1;

> +    param->encMode              = svt_enc->enc_mode;

> +    param->profile              = svt_enc->profile;

> +    param->tier                 = svt_enc->tier;

> +    param->level                = svt_enc->level;

> +    param->rateControlMode      = svt_enc->rc_mode;

> +    param->sceneChangeDetection = svt_enc->scd;

> +    param->tune                 = svt_enc->tune;

> +    param->baseLayerSwitchMode  =

> svt_enc->base_layer_switch_mode;

> +    param->qp                   = svt_enc->qp;

> +    param->accessUnitDelimiter  = svt_enc->aud;

> +

> +    param->targetBitRate = avctx->bit_rate;

> +    if (avctx->gop_size > 0)

> +        param->intraPeriodLength = avctx->gop_size - 1;

> +

> +    if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {

> +        param->frameRateNumerator = avctx->framerate.num;

> +        param->frameRateDenominator =

> +            avctx->framerate.den * avctx->ticks_per_frame;

> +    } else {

> +        param->frameRateNumerator = avctx->time_base.den;

> +        param->frameRateDenominator =

> +            avctx->time_base.num * avctx->ticks_per_frame;

> +    }

> +

> +    if (param->rateControlMode) {

> +        param->maxQpAllowed = avctx->qmax;

> +        param->minQpAllowed = avctx->qmin;

> +    }

> +

> +    param->intraRefreshType =

> +        !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) + 1;

> +

> +    // is it repeat headers for MP4 or Annex-b

> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)

> +        param->codeVpsSpsPps = 0;

> +    else

> +        param->codeVpsSpsPps = 1;

> +

> +    param->codeEosNal = 1;

> +

> +    if (svt_enc->hdr) {

> +        av_log(avctx, AV_LOG_WARNING, "Color properties not

> supported

> + yet\n");

> +

> +        svt_enc->vui_info = 1;

> +        param->highDynamicRangeInput = svt_enc->hdr;

> +    }

> +

> +    if (svt_enc->vui_info)

> +        param->videoUsabilityInfo = svt_enc->vui_info;

> +

> +    if (svt_enc->la_depth != -1)


Though the option la_depth is limited to [-1, 256], still would be better to be la_dapth >= 0. 

[Jing] That range is defined by the SVT-HEVC lib, and -1 means no setting and the encoding lib will decide it.

> +        param->lookAheadDistance = svt_enc->la_depth;

> +

> +    ret = alloc_buffer(param, svt_enc);

> +

> +    return ret;

> +}

> +

> +static void read_in_data(EB_H265_ENC_CONFIGURATION *config,

> +                         const AVFrame *frame,

> +                         EB_BUFFERHEADERTYPE *header_ptr) {

> +    uint8_t is16bit = config->encoderBitDepth > 8;

> +    uint64_t luma_size =

> +        (uint64_t)config->sourceWidth * config->sourceHeight<< is16bit;

> +    EB_H265_ENC_INPUT *in_data = (EB_H265_ENC_INPUT 

> +*)header_ptr->pBuffer;

> +

> +    // support yuv420p and yuv420p010


Probably you can give an assert here. Comment for limited format can't avoid any run-time issue.
[Jing] There will be no run-time issue, for we have declared in ff_libsvt_hevc_encoder that what we can support:
    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P,
                                                    AV_PIX_FMT_YUV420P10,
                                                    AV_PIX_FMT_NONE },
I agree with you that the comment here is useless and I am removing it.

> +    in_data->luma = frame->data[0];

> +    in_data->cb   = frame->data[1];

> +    in_data->cr   = frame->data[2];

> +

> +    // stride info

> +    in_data->yStride  = frame->linesize[0] >> is16bit;

> +    in_data->cbStride = frame->linesize[1] >> is16bit;

> +    in_data->crStride = frame->linesize[2] >> is16bit;

> +

> +    header_ptr->nFilledLen   += luma_size * 3/2u;

> +}

> +

> +static av_cold int eb_enc_init(AVCodecContext *avctx) {

> +    SvtContext   *svt_enc = avctx->priv_data;

> +    EB_ERRORTYPE svt_ret;

> +

> +    svt_enc->eos_flag = EOS_NOT_REACHED;

> +

> +    svt_ret = EbInitHandle(&svt_enc->svt_handle, svt_enc,

> &svt_enc->enc_params);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error init encoder handle\n");

> +        goto failed;

> +    }

> +

> +    svt_ret = config_enc_params(&svt_enc->enc_params, avctx);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error configure encoder

> parameters\n");

> +        goto failed_init_handle;

> +    }

> +

> +    svt_ret = EbH265EncSetParameter(svt_enc->svt_handle,

> &svt_enc->enc_params);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error setting encoder

> parameters\n");

> +        goto failed_init_handle;

> +    }

> +

> +    svt_ret = EbInitEncoder(svt_enc->svt_handle);

> +    if (svt_ret != EB_ErrorNone) {

> +        av_log(avctx, AV_LOG_ERROR, "Error init encoder\n");

> +        goto failed_init_handle;

> +    }

> +

> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {

> +        EB_BUFFERHEADERTYPE *header_ptr = NULL;

> +

> +        svt_ret = EbH265EncStreamHeader(svt_enc->svt_handle,

> &header_ptr);

> +        if (svt_ret != EB_ErrorNone) {

> +            av_log(avctx, AV_LOG_ERROR, "Error when build stream

> header.\n");

> +            goto failed_init_enc;

> +        }

> +

> +        avctx->extradata_size = header_ptr->nFilledLen;

> +        avctx->extradata = av_mallocz(avctx->extradata_size +

> AV_INPUT_BUFFER_PADDING_SIZE);


The PADDING memory should be memset to zero.
[Jing] To be modified.

> +        if (!avctx->extradata) {

> +            av_log(avctx, AV_LOG_ERROR,

> +                   "Cannot allocate HEVC header of size %d.\n",

> avctx->extradata_size);

> +            svt_ret = EB_ErrorInsufficientResources;

> +            goto failed_init_enc;

> +        }

> +        memcpy(avctx->extradata, header_ptr->pBuffer,

> avctx->extradata_size);

> +    }

> +

> +    return 0;

> +

> +failed_init_enc:

> +    EbDeinitEncoder(svt_enc->svt_handle);

> +failed_init_handle:

> +    EbDeinitHandle(svt_enc->svt_handle);

> +failed:

> +    free_buffer(svt_enc);

> +    svt_enc->svt_handle = NULL;

> +    svt_enc = NULL;

> +    return error_mapping(svt_ret);

> +}

> +

> +static int eb_send_frame(AVCodecContext *avctx, const AVFrame *frame) {

> +    SvtContext           *svt_enc = avctx->priv_data;

> +    EB_BUFFERHEADERTYPE  *header_ptr = &svt_enc->in_buf;

> +

> +    if (!frame) {

> +        EB_BUFFERHEADERTYPE header_ptr_last;

> +        header_ptr_last.nAllocLen   = 0;

> +        header_ptr_last.nFilledLen  = 0;

> +        header_ptr_last.nTickCount  = 0;

> +        header_ptr_last.pAppPrivate = NULL;

> +        header_ptr_last.pBuffer     = NULL;

> +        header_ptr_last.nFlags      = EB_BUFFERFLAG_EOS;

> +

> +        EbH265EncSendPicture(svt_enc->svt_handle, &header_ptr_last);

> +        svt_enc->eos_flag = EOS_REACHED;

> +        av_log(avctx, AV_LOG_DEBUG, "Finish sending frames!!!\n");

> +        return 0;

> +    }

> +

> +    read_in_data(&svt_enc->enc_params, frame, header_ptr);

> +

> +    header_ptr->nFlags       = 0;

> +    header_ptr->pAppPrivate  = NULL;

> +    header_ptr->pts          = frame->pts;

> +    switch (frame->pict_type) {

> +    case AV_PICTURE_TYPE_I:

> +        header_ptr->sliceType = svt_enc->forced_idr > 0 ?

> EB_IDR_PICTURE : EB_I_PICTURE;

> +        break;

> +    case AV_PICTURE_TYPE_P:

> +        header_ptr->sliceType = EB_P_PICTURE;

> +        break;

> +    case AV_PICTURE_TYPE_B:

> +        header_ptr->sliceType = EB_B_PICTURE;

> +        break;

> +    default:

> +        header_ptr->sliceType = EB_INVALID_PICTURE;

> +        break;

> +    }

> +    EbH265EncSendPicture(svt_enc->svt_handle, header_ptr);

> +

> +    return 0;

> +}

> +

> +static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) {

> +    SvtContext  *svt_enc = avctx->priv_data;

> +    EB_BUFFERHEADERTYPE   *header_ptr = NULL;

> +    EB_ERRORTYPE          svt_ret;

> +    int ret = 0;

> +

> +    if (EOS_TOTRIGGER == svt_enc->eos_flag) {

> +        pkt = NULL;

> +        return AVERROR_EOF;

> +    }

> +

> +    svt_ret = EbH265GetPacket(svt_enc->svt_handle, &header_ptr,

> svt_enc->eos_flag);

> +    if (svt_ret == EB_NoErrorEmptyQueue)

> +        return AVERROR(EAGAIN);

> +

> +    if ((ret = ff_alloc_packet2(avctx, pkt, header_ptr->nFilledLen, 0)) < 0) {

> +        av_log(avctx, AV_LOG_ERROR, "Failed to allocate output

> packet.\n");

> +        EbH265ReleaseOutBuffer(&header_ptr);

> +        return ret;

> +    }

> +

> +    memcpy(pkt->data, header_ptr->pBuffer, header_ptr->nFilledLen);

> +

> +    pkt->size = header_ptr->nFilledLen;

> +    pkt->pts  = header_ptr->pts;

> +    pkt->dts  = header_ptr->dts;

> +    if ((header_ptr->sliceType == EB_IDR_PICTURE) ||

> +        (header_ptr->sliceType == EB_I_PICTURE))

> +        pkt->flags |= AV_PKT_FLAG_KEY;

> +    if (header_ptr->sliceType == EB_NON_REF_PICTURE)

> +        pkt->flags |= AV_PKT_FLAG_DISPOSABLE;

> +

> +    EbH265ReleaseOutBuffer(&header_ptr);

> +

> +    if (EB_BUFFERFLAG_EOS == header_ptr->nFlags)

> +        svt_enc->eos_flag = EOS_TOTRIGGER;

> +

> +    return 0;

> +}

> +

> +static av_cold int eb_enc_close(AVCodecContext *avctx) {

> +    SvtContext *svt_enc = avctx->priv_data;

> +

> +    if (svt_enc) {

> +        if (svt_enc->svt_handle) {

> +            EbDeinitEncoder(svt_enc->svt_handle);

> +            EbDeinitHandle(svt_enc->svt_handle);

> +            svt_enc->svt_handle = NULL;

> +        }

> +

> +        free_buffer(svt_enc);

> +        svt_enc = NULL;

> +    }

> +

> +    return 0;

> +}

> +

> +#define OFFSET(x) offsetof(SvtContext, x) #define VE 

> +AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM static

> const

> +AVOption options[] = {

> +    { "vui", "Enable Video Usability Info", OFFSET(vui_info),

> +      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },

> +

> +    { "aud", "Include Access Unit Delimiter", OFFSET(aud),

> +      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },

> +

> +    { "hielevel", "Hierarchical prediction levels setting",

> OFFSET(hierarchical_level),

> +      AV_OPT_TYPE_INT, { .i64 = 4 }, 1, 4, VE , "hielevel"},

> +        { "flat",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },

> INT_MIN, INT_MAX, VE, "hielevel" },

> +        { "2level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 },  

> + INT_MIN,

> INT_MAX, VE, "hielevel" },

> +        { "3level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 },  

> + INT_MIN,

> INT_MAX, VE, "hielevel" },

> +        { "4level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 4 },

> INT_MIN,

> + INT_MAX, VE, "hielevel" },

> +

> +    { "la_depth", "Look ahead distance [0, 256]", OFFSET(la_depth),

> +      AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 256, VE },

> +

> +    { "preset", "Encoding preset [0, 12] (e,g, for subjective quality 

> + tuning

> mode and >=4k resolution), [0, 10] (for >= 1080p resolution), [0, 9] 

> (for all resolution and modes)",

> +      OFFSET(enc_mode), AV_OPT_TYPE_INT, { .i64 = 9 }, 0, 12, VE },

> +

> +    { "profile", "Profile setting, Main Still Picture Profile not 

> + supported",

> OFFSET(profile),

> +      AV_OPT_TYPE_INT, { .i64 = FF_PROFILE_HEVC_MAIN_10 },


I am quite surprised why you set main10 as default profile. 
You added a debug message for 10bit input, but not added for 8bit case, doesn't mean 8bit case is more common? 

[Jing] Main 10 allows bit depth of 8 to 10, and it's not for 10-bit only. I chose it as the default for SVT HEVC do that as well. The default value should be the same.

> + FF_PROFILE_HEVC_MAIN, FF_PROFILE_HEVC_MAIN_10, VE, "profile"},

> +

> +    { "tier", "Set tier (general_tier_flag)", OFFSET(tier),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE, "tier" },

> +        { "main", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, VE,

> "tier" },

> +        { "high", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, VE, 

> + "tier" },

> +

> +    { "level", "Set level (level_idc)", OFFSET(level),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 0xff, VE, "level" },

> +

> +    { "rc", "Bit rate control mode", OFFSET(rc_mode),

> +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE , "rc"},

> +        { "cqp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN,

> INT_MAX, VE, "rc" },

> +        { "vbr", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, 

> + INT_MAX, VE, "rc" },


Hm. VBR mode can be support but I haven't seen any parameter to set max_rate/vbv buffer. 
Does this mean the bitrate fluctuation is totally out of control?

[Jing] SVT HEVC's VBR is controlled by the max allowed QP, min allowed QP and the bitrate.

> +    { "qp", "QP value for intra frames", OFFSET(qp),

> +      AV_OPT_TYPE_INT, { .i64 = 32 }, 0, 51, VE },

> +

> +    { "sc_detection", "Scene change detection", OFFSET(scd),

> +      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },

> +

> +    { "tune", "Quality tuning mode", OFFSET(tune), AV_OPT_TYPE_INT,

> { .i64 = 1 }, 0, 2, VE, "tune" },

> +        { "sq", "Visually optimized mode", 0,

> +          AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN, INT_MAX, VE,

> "tune" },

> +        { "oq",  "PSNR / SSIM optimized mode",  0,

> +          AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, INT_MAX, VE,

> "tune" },

> +        { "vmaf", "VMAF optimized mode", 0,

> +          AV_OPT_TYPE_CONST, { .i64 = 2 },  INT_MIN, INT_MAX, VE, 

> + "tune" },

> +

> +    { "bl_mode", "Random Access Prediction Structure type setting",

> OFFSET(base_layer_switch_mode),

> +      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },

> +

> +    { "forced-idr", "If forcing keyframes, force them as IDR 

> + frames.",

> OFFSET(forced_idr),

> +      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, -1, 1, VE },

> +

> +    { "hdr", "High dynamic range input", OFFSET(hdr),

> +      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, 0, 1, VE },

> +

> +    {NULL},

> +};

> +

> +static const AVClass class = {

> +    .class_name = "libsvt_hevc",

> +    .item_name  = av_default_item_name,

> +    .option     = options,

> +    .version    = LIBAVUTIL_VERSION_INT,

> +};

> +

> +static const AVCodecDefault eb_enc_defaults[] = {

> +    { "b",         "7M"    },

> +    { "flags",     "+cgop" },

> +    { "qmin",      "10"    },

> +    { "qmax",      "48"    },

> +    { "g",         "-2"    },

> +    { NULL },

> +};

> +

> +AVCodec ff_libsvt_hevc_encoder = {

> +    .name           = "libsvt_hevc",

> +    .long_name      = NULL_IF_CONFIG_SMALL("SVT-HEVC(Scalable

> Video Technology for HEVC) encoder"),

> +    .priv_data_size = sizeof(SvtContext),

> +    .type           = AVMEDIA_TYPE_VIDEO,

> +    .id             = AV_CODEC_ID_HEVC,

> +    .init           = eb_enc_init,

> +    .send_frame     = eb_send_frame,

> +    .receive_packet = eb_receive_packet,

> +    .close          = eb_enc_close,

> +    .capabilities   = AV_CODEC_CAP_DELAY |

> AV_CODEC_CAP_AUTO_THREADS,

> +    .pix_fmts       = (const enum

> AVPixelFormat[]){ AV_PIX_FMT_YUV420P,

> +

> AV_PIX_FMT_YUV420P10,

> +

> AV_PIX_FMT_NONE },

> +    .priv_class     = &class,

> +    .defaults       = eb_enc_defaults,

> +    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP,

> +    .wrapper_name   = "libsvt_hevc",

> +};

> --

> 1.8.3.1
Zhong Li April 26, 2019, 6:18 a.m. UTC | #3
> From: Sun, Jing A

> Sent: Tuesday, April 2, 2019 2:02 PM

> To: Li, Zhong <zhong.li@intel.com>; FFmpeg development discussions and

> patches <ffmpeg-devel@ffmpeg.org>

> Cc: Huang, Zhengxu <zhengxu.huang@intel.com>; Jun Zhao

> <jun.zhao@intel.com>; Tmar, Hassene <hassene.tmar@intel.com>

> Subject: RE: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc

> encoder wrapper

> 

> -----Original Message-----

> From: Li, Zhong

> Sent: Friday, March 29, 2019 6:59 PM

> To: FFmpeg development discussions and patches

> <ffmpeg-devel@ffmpeg.org>

> Cc: Sun, Jing A <jing.a.sun@intel.com>; Huang, Zhengxu

> <zhengxu.huang@intel.com>; Jun Zhao <jun.zhao@intel.com>; Tmar,

> Hassene <hassene.tmar@intel.com>

> Subject: RE: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc

> encoder wrapper

> 

> > From: ffmpeg-devel [mailto:ffmpeg-devel-bounces@ffmpeg.org] On

> Behalf

> > Of Jing Sun

> > Sent: Friday, March 29, 2019 4:22 PM

> > To: ffmpeg-devel@ffmpeg.org

> > Cc: Sun, Jing A <jing.a.sun@intel.com>; Huang, Zhengxu

> > <zhengxu.huang@intel.com>; Jun Zhao <jun.zhao@intel.com>; Tmar,

> > Hassene <hassene.tmar@intel.com>

> > Subject: [FFmpeg-devel] [PATCH v10 1/2] lavc/svt_hevc: add libsvt hevc

> > encoder wrapper

> >

> > Signed-off-by: Zhengxu Huang <zhengxu.huang@intel.com>

> > Signed-off-by: Hassene Tmar <hassene.tmar@intel.com>

> > Signed-off-by: Jun Zhao <jun.zhao@intel.com>

> > Signed-off-by: Jing Sun <jing.a.sun@intel.com>

> > ---

> >  configure                |   4 +

> >  libavcodec/Makefile      |   1 +

> >  libavcodec/allcodecs.c   |   1 +

> >  libavcodec/libsvt_hevc.c | 500

> > +++++++++++++++++++++++++++++++++++++++++++++++

> >  4 files changed, 506 insertions(+)

> >  create mode 100644 libavcodec/libsvt_hevc.c

> >

> > diff --git a/configure b/configure

> > index 938ff10..2aabac4 100755

> > --- a/configure

> > +++ b/configure

> > @@ -264,6 +264,7 @@ External library support:

> >    --enable-libspeex        enable Speex de/encoding via libspeex [no]

> >    --enable-libsrt          enable Haivision SRT protocol via libsrt [no]

> >    --enable-libssh          enable SFTP protocol via libssh [no]

> > +  --enable-libsvthevc      enable HEVC encoding via svt [no]

> >    --enable-libtensorflow   enable TensorFlow as a DNN module

> backend

> >                             for DNN based filters like sr [no]

> >    --enable-libtesseract    enable Tesseract, needed for ocr filter [no]

> > @@ -1784,6 +1785,7 @@ EXTERNAL_LIBRARY_LIST="

> >      libspeex

> >      libsrt

> >      libssh

> > +    libsvthevc

> >      libtensorflow

> >      libtesseract

> >      libtheora

> > @@ -3173,6 +3175,7 @@ libshine_encoder_select="audio_frame_queue"

> >  libspeex_decoder_deps="libspeex"

> >  libspeex_encoder_deps="libspeex"

> >  libspeex_encoder_select="audio_frame_queue"

> > +libsvt_hevc_encoder_deps="libsvthevc"

> >  libtheora_encoder_deps="libtheora"

> >  libtwolame_encoder_deps="libtwolame"

> >  libvo_amrwbenc_encoder_deps="libvo_amrwbenc"

> > @@ -6209,6 +6212,7 @@ enabled libsoxr           && require libsoxr

> > soxr.h soxr_create -lsoxr

> >  enabled libssh            && require_pkg_config libssh libssh

> > libssh/sftp.h sftp_init

> >  enabled libspeex          && require_pkg_config libspeex speex

> > speex/speex.h speex_decoder_init

> >  enabled libsrt            && require_pkg_config libsrt "srt >= 1.3.0"

> > srt/srt.h srt_socket

> > +enabled libsvthevc        && require_pkg_config libsvthevc

> SvtHevcEnc

> > EbApi.h EbInitHandle

> >  enabled libtensorflow     && require libtensorflow

> tensorflow/c/c_api.h

> > TF_Version -ltensorflow

> >  enabled libtesseract      && require_pkg_config libtesseract tesseract

> > tesseract/capi.h TessBaseAPICreate

> >  enabled libtheora         && require libtheora theora/theoraenc.h

> > th_info_init -ltheoraenc -ltheoradec -logg diff --git

> > a/libavcodec/Makefile b/libavcodec/Makefile index

> > 15c43a8..c93e545 100644

> > --- a/libavcodec/Makefile

> > +++ b/libavcodec/Makefile

> > @@ -987,6 +987,7 @@ OBJS-$(CONFIG_LIBOPUS_ENCODER)

> > += libopusenc.o libopus.o     \

> >  OBJS-$(CONFIG_LIBSHINE_ENCODER)           += libshine.o

> >  OBJS-$(CONFIG_LIBSPEEX_DECODER)           += libspeexdec.o

> >  OBJS-$(CONFIG_LIBSPEEX_ENCODER)           += libspeexenc.o

> > +OBJS-$(CONFIG_LIBSVT_HEVC_ENCODER)        += libsvt_hevc.o

> >  OBJS-$(CONFIG_LIBTHEORA_ENCODER)          += libtheoraenc.o

> >  OBJS-$(CONFIG_LIBTWOLAME_ENCODER)         += libtwolame.o

> >  OBJS-$(CONFIG_LIBVO_AMRWBENC_ENCODER)     +=

> > libvo-amrwbenc.o

> > diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index

> > b26aeca..e93f66f 100644

> > --- a/libavcodec/allcodecs.c

> > +++ b/libavcodec/allcodecs.c

> > @@ -703,6 +703,7 @@ extern AVCodec ff_librsvg_decoder;  extern

> AVCodec

> > ff_libshine_encoder;  extern AVCodec ff_libspeex_encoder; extern

> > AVCodec ff_libspeex_decoder;

> > +extern AVCodec ff_libsvt_hevc_encoder;

> >  extern AVCodec ff_libtheora_encoder;

> >  extern AVCodec ff_libtwolame_encoder;  extern AVCodec

> > ff_libvo_amrwbenc_encoder; diff --git a/libavcodec/libsvt_hevc.c

> > b/libavcodec/libsvt_hevc.c new file mode 100644 index 0000000..4ef06a8

> > --- /dev/null

> > +++ b/libavcodec/libsvt_hevc.c

> > @@ -0,0 +1,500 @@

> > +/*

> > +* Scalable Video Technology for HEVC encoder library plugin

> > +*

> > +* Copyright (c) 2018 Intel Corporation

> 

> It is 2019 now.

> 

> [Jing] To be modified.

> 

> > +* 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 this program; if not, write to the Free Software

> > +* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

> > +02110-1301 USA */

> > +

> > +#include "EbErrorCodes.h"

> > +#include "EbTime.h"

> > +#include "EbApi.h"

> > +

> > +#include "libavutil/common.h"

> > +#include "libavutil/frame.h"

> > +#include "libavutil/opt.h"

> > +

> > +#include "internal.h"

> > +#include "avcodec.h"

> > +

> > +typedef enum eos_status {

> > +    EOS_NOT_REACHED = 0,

> > +    EOS_REACHED,

> > +    EOS_TOTRIGGER

> > +}EOS_STATUS;

> > +

> > +typedef struct SvtContext {

> > +    AVClass *class;

> > +

> > +    EB_H265_ENC_CONFIGURATION enc_params;

> > +    EB_COMPONENTTYPE *svt_handle;

> > +    EB_BUFFERHEADERTYPE in_buf;

> > +    EOS_STATUS eos_flag;

> > +

> > +    // User options.

> > +    int vui_info;

> > +    int hierarchical_level;

> > +    int la_depth;

> > +    int enc_mode;

> > +    int rc_mode;

> > +    int scd;

> > +    int tune;

> > +    int qp;

> > +    int hdr;

> > +

> > +    int forced_idr;

> > +

> > +    int aud;

> > +

> > +    int profile;

> > +    int tier;

> > +    int level;

> > +

> > +    int base_layer_switch_mode;

> > +} SvtContext;

> > +

> > +static int error_mapping(EB_ERRORTYPE svt_ret) {

> > +    int err;

> > +

> > +    switch (svt_ret) {

> > +    case EB_ErrorInsufficientResources:

> > +        err = AVERROR(ENOMEM);

> > +        break;

> > +

> > +    case EB_ErrorUndefined:

> > +    case EB_ErrorInvalidComponent:

> > +    case EB_ErrorBadParameter:

> > +        err = AVERROR(EINVAL);

> > +        break;

> > +

> > +    case EB_ErrorDestroyThreadFailed:

> > +    case EB_ErrorSemaphoreUnresponsive:

> > +    case EB_ErrorDestroySemaphoreFailed:

> > +    case EB_ErrorCreateMutexFailed:

> > +    case EB_ErrorMutexUnresponsive:

> > +    case EB_ErrorDestroyMutexFailed:

> > +        err = AVERROR_EXTERNAL;

> > +            break;

> > +

> > +    case EB_NoErrorEmptyQueue:

> > +        err = AVERROR(EAGAIN);

> > +

> > +    case EB_ErrorNone:

> > +        err = 0;

> > +        break;

> > +

> > +    default:

> > +        err = AVERROR_UNKNOWN;

> > +    }

> 

> How about add some error message for each case instead of just return?

> 

> [Jing] When an error happens, logs are dumped before this func is called,

> such as:

>         ret = EbH265EncStreamHeader(svt_enc->svt_handle, &headerPtr);

>         if (ret != EB_ErrorNone) {

>             av_log(avctx, AV_LOG_ERROR, "Error when build stream

> header.\n");

>             av_freep(&headerPtr.pBuffer);

>             goto failed_init_enc;

>         }


I saw it but what I mean was to provide an error message to every error type. 
E.g. User can't know the error is due to EB_ErrorDestroyThreadFailed or EB_ErrorCreateMutexFailed
QSV has an example can be taken as reference : https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/qsv.c#L90 


> > +    return err;

> > +

> > +static void free_buffer(SvtContext *svt_enc) {

> > +    uint8_t *in_data = svt_enc->in_buf.pBuffer;

> > +

> > +    av_freep(&in_data);

> > +}

> > +

> > +static int alloc_buffer(EB_H265_ENC_CONFIGURATION *config,

> SvtContext

> > +*svt_enc) {

> > +    EB_H265_ENC_INPUT *in_data;

> > +

> > +    // allocate buffer for in and out

> > +    in_data  = av_mallocz(sizeof(*in_data));

> > +    if (!in_data)

> > +        goto failed;

> > +    svt_enc->in_buf.pBuffer  = (unsigned char *)in_data;

> > +

> > +    svt_enc->in_buf.nSize        = sizeof(svt_enc->in_buf);

> > +    svt_enc->in_buf.pAppPrivate  = NULL;

> > +

> > +    return 0;

> > +

> > +failed:

> > +    free_buffer(svt_enc);

> > +    return AVERROR(ENOMEM);

> > +}

> > +

> > +static int config_enc_params(EB_H265_ENC_CONFIGURATION *param,

> > +                             AVCodecContext *avctx) {

> > +    SvtContext *svt_enc = avctx->priv_data;

> > +    int ret;

> > +

> > +    param->sourceWidth = avctx->width;

> > +    param->sourceHeight = avctx->height;

> > +    param->encoderBitDepth = 8;

> > +

> > +    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P10LE) {

> > +        av_log(avctx, AV_LOG_DEBUG, "Encoder 10 bits depth

> input\n");

> > +

> > +        param->encoderBitDepth = 10;

> > +    }

> 

> Get the bitdepth informat for pix_fmt should be easier to maintain (e.g: 12bit

> support).

> Here is an example:

> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/vaapi_encode

> _h265.c#L295

> 

> [Jing] It requires an extra step av_pix_fmt_desc_get(). Currently only 420 is

> supported, and checking avctx->pix_fmt is more straightforward. When 422

> and 444 are supported later, I will change to the suggested method.


A good design should be that you can consider the further extension as possible at the very beginning.

> > +    param->encoderColorFormat = EB_YUV420;

> > +

> > +    // Update param from options

> > +    param->hierarchicalLevels   = svt_enc->hierarchical_level - 1;

> > +    param->encMode              = svt_enc->enc_mode;

> > +    param->profile              = svt_enc->profile;

> > +    param->tier                 = svt_enc->tier;

> > +    param->level                = svt_enc->level;

> > +    param->rateControlMode      = svt_enc->rc_mode;

> > +    param->sceneChangeDetection = svt_enc->scd;

> > +    param->tune                 = svt_enc->tune;

> > +    param->baseLayerSwitchMode  =

> > svt_enc->base_layer_switch_mode;

> > +    param->qp                   = svt_enc->qp;

> > +    param->accessUnitDelimiter  = svt_enc->aud;

> > +

> > +    param->targetBitRate = avctx->bit_rate;

> > +    if (avctx->gop_size > 0)

> > +        param->intraPeriodLength = avctx->gop_size - 1;

> > +

> > +    if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {

> > +        param->frameRateNumerator = avctx->framerate.num;

> > +        param->frameRateDenominator =

> > +            avctx->framerate.den * avctx->ticks_per_frame;

> > +    } else {

> > +        param->frameRateNumerator = avctx->time_base.den;

> > +        param->frameRateDenominator =

> > +            avctx->time_base.num * avctx->ticks_per_frame;

> > +    }

> > +

> > +    if (param->rateControlMode) {

> > +        param->maxQpAllowed = avctx->qmax;

> > +        param->minQpAllowed = avctx->qmin;

> > +    }

> > +

> > +    param->intraRefreshType =

> > +        !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) + 1;

> > +

> > +    // is it repeat headers for MP4 or Annex-b

> > +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)

> > +        param->codeVpsSpsPps = 0;

> > +    else

> > +        param->codeVpsSpsPps = 1;

> > +

> > +    param->codeEosNal = 1;

> > +

> > +    if (svt_enc->hdr) {

> > +        av_log(avctx, AV_LOG_WARNING, "Color properties not

> > supported

> > + yet\n");

> > +

> > +        svt_enc->vui_info = 1;

> > +        param->highDynamicRangeInput = svt_enc->hdr;

> > +    }

> > +

> > +    if (svt_enc->vui_info)

> > +        param->videoUsabilityInfo = svt_enc->vui_info;

> > +

> > +    if (svt_enc->la_depth != -1)

> 

> Though the option la_depth is limited to [-1, 256], still would be better to be

> la_dapth >= 0.

> 

> [Jing] That range is defined by the SVT-HEVC lib, and -1 means no setting and

> the encoding lib will decide it.


Yes, I know. My suggestion is just to replace "!= -1" to be ">=0". That is easier to read and don't need to care about 
SVT internal implementation. La_depath is invalid to be a negative value.

> > +        param->lookAheadDistance = svt_enc->la_depth;

> > +

> > +    ret = alloc_buffer(param, svt_enc);

> > +

> > +    return ret;

> > +}

> > +

> > +static void read_in_data(EB_H265_ENC_CONFIGURATION *config,

> > +                         const AVFrame *frame,

> > +                         EB_BUFFERHEADERTYPE *header_ptr) {

> > +    uint8_t is16bit = config->encoderBitDepth > 8;

> > +    uint64_t luma_size =

> > +        (uint64_t)config->sourceWidth * config->sourceHeight<<

> is16bit;

> > +    EB_H265_ENC_INPUT *in_data = (EB_H265_ENC_INPUT

> > +*)header_ptr->pBuffer;

> > +

> > +    // support yuv420p and yuv420p010

> 

> Probably you can give an assert here. Comment for limited format can't

> avoid any run-time issue.

> [Jing] There will be no run-time issue, for we have declared in

> ff_libsvt_hevc_encoder that what we can support:

>     .pix_fmts       = (const enum

> AVPixelFormat[]){ AV_PIX_FMT_YUV420P,

> 

> AV_PIX_FMT_YUV420P10,

> 

> AV_PIX_FMT_NONE }, I agree with you that the comment here is useless and

> I am removing it.

> 

> > +    in_data->luma = frame->data[0];

> > +    in_data->cb   = frame->data[1];

> > +    in_data->cr   = frame->data[2];

> > +

> > +    // stride info

> > +    in_data->yStride  = frame->linesize[0] >> is16bit;

> > +    in_data->cbStride = frame->linesize[1] >> is16bit;

> > +    in_data->crStride = frame->linesize[2] >> is16bit;

> > +

> > +    header_ptr->nFilledLen   += luma_size * 3/2u;

> > +}

> > +

> > +static av_cold int eb_enc_init(AVCodecContext *avctx) {

> > +    SvtContext   *svt_enc = avctx->priv_data;

> > +    EB_ERRORTYPE svt_ret;

> > +

> > +    svt_enc->eos_flag = EOS_NOT_REACHED;

> > +

> > +    svt_ret = EbInitHandle(&svt_enc->svt_handle, svt_enc,

> > &svt_enc->enc_params);

> > +    if (svt_ret != EB_ErrorNone) {

> > +        av_log(avctx, AV_LOG_ERROR, "Error init encoder handle\n");

> > +        goto failed;

> > +    }

> > +

> > +    svt_ret = config_enc_params(&svt_enc->enc_params, avctx);

> > +    if (svt_ret != EB_ErrorNone) {

> > +        av_log(avctx, AV_LOG_ERROR, "Error configure encoder

> > parameters\n");

> > +        goto failed_init_handle;

> > +    }

> > +

> > +    svt_ret = EbH265EncSetParameter(svt_enc->svt_handle,

> > &svt_enc->enc_params);

> > +    if (svt_ret != EB_ErrorNone) {

> > +        av_log(avctx, AV_LOG_ERROR, "Error setting encoder

> > parameters\n");

> > +        goto failed_init_handle;

> > +    }

> > +

> > +    svt_ret = EbInitEncoder(svt_enc->svt_handle);

> > +    if (svt_ret != EB_ErrorNone) {

> > +        av_log(avctx, AV_LOG_ERROR, "Error init encoder\n");

> > +        goto failed_init_handle;

> > +    }

> > +

> > +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {

> > +        EB_BUFFERHEADERTYPE *header_ptr = NULL;

> > +

> > +        svt_ret = EbH265EncStreamHeader(svt_enc->svt_handle,

> > &header_ptr);

> > +        if (svt_ret != EB_ErrorNone) {

> > +            av_log(avctx, AV_LOG_ERROR, "Error when build stream

> > header.\n");

> > +            goto failed_init_enc;

> > +        }

> > +

> > +        avctx->extradata_size = header_ptr->nFilledLen;

> > +        avctx->extradata = av_mallocz(avctx->extradata_size +

> > AV_INPUT_BUFFER_PADDING_SIZE);

> 

> The PADDING memory should be memset to zero.

> [Jing] To be modified.

> 

> > +        if (!avctx->extradata) {

> > +            av_log(avctx, AV_LOG_ERROR,

> > +                   "Cannot allocate HEVC header of size %d.\n",

> > avctx->extradata_size);

> > +            svt_ret = EB_ErrorInsufficientResources;

> > +            goto failed_init_enc;

> > +        }

> > +        memcpy(avctx->extradata, header_ptr->pBuffer,

> > avctx->extradata_size);

> > +    }

> > +

> > +    return 0;

> > +

> > +failed_init_enc:

> > +    EbDeinitEncoder(svt_enc->svt_handle);

> > +failed_init_handle:

> > +    EbDeinitHandle(svt_enc->svt_handle);

> > +failed:

> > +    free_buffer(svt_enc);

> > +    svt_enc->svt_handle = NULL;

> > +    svt_enc = NULL;

> > +    return error_mapping(svt_ret);

> > +}

> > +

> > +static int eb_send_frame(AVCodecContext *avctx, const AVFrame *frame)

> {

> > +    SvtContext           *svt_enc = avctx->priv_data;

> > +    EB_BUFFERHEADERTYPE  *header_ptr = &svt_enc->in_buf;

> > +

> > +    if (!frame) {

> > +        EB_BUFFERHEADERTYPE header_ptr_last;

> > +        header_ptr_last.nAllocLen   = 0;

> > +        header_ptr_last.nFilledLen  = 0;

> > +        header_ptr_last.nTickCount  = 0;

> > +        header_ptr_last.pAppPrivate = NULL;

> > +        header_ptr_last.pBuffer     = NULL;

> > +        header_ptr_last.nFlags      = EB_BUFFERFLAG_EOS;

> > +

> > +        EbH265EncSendPicture(svt_enc->svt_handle,

> &header_ptr_last);

> > +        svt_enc->eos_flag = EOS_REACHED;

> > +        av_log(avctx, AV_LOG_DEBUG, "Finish sending frames!!!\n");

> > +        return 0;

> > +    }

> > +

> > +    read_in_data(&svt_enc->enc_params, frame, header_ptr);

> > +

> > +    header_ptr->nFlags       = 0;

> > +    header_ptr->pAppPrivate  = NULL;

> > +    header_ptr->pts          = frame->pts;

> > +    switch (frame->pict_type) {

> > +    case AV_PICTURE_TYPE_I:

> > +        header_ptr->sliceType = svt_enc->forced_idr > 0 ?

> > EB_IDR_PICTURE : EB_I_PICTURE;

> > +        break;

> > +    case AV_PICTURE_TYPE_P:

> > +        header_ptr->sliceType = EB_P_PICTURE;

> > +        break;

> > +    case AV_PICTURE_TYPE_B:

> > +        header_ptr->sliceType = EB_B_PICTURE;

> > +        break;

> > +    default:

> > +        header_ptr->sliceType = EB_INVALID_PICTURE;

> > +        break;

> > +    }

> > +    EbH265EncSendPicture(svt_enc->svt_handle, header_ptr);

> > +

> > +    return 0;

> > +}

> > +

> > +static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) {

> > +    SvtContext  *svt_enc = avctx->priv_data;

> > +    EB_BUFFERHEADERTYPE   *header_ptr = NULL;

> > +    EB_ERRORTYPE          svt_ret;

> > +    int ret = 0;

> > +

> > +    if (EOS_TOTRIGGER == svt_enc->eos_flag) {

> > +        pkt = NULL;

> > +        return AVERROR_EOF;

> > +    }

> > +

> > +    svt_ret = EbH265GetPacket(svt_enc->svt_handle, &header_ptr,

> > svt_enc->eos_flag);

> > +    if (svt_ret == EB_NoErrorEmptyQueue)

> > +        return AVERROR(EAGAIN);

> > +

> > +    if ((ret = ff_alloc_packet2(avctx, pkt, header_ptr->nFilledLen, 0)) < 0)

> {

> > +        av_log(avctx, AV_LOG_ERROR, "Failed to allocate output

> > packet.\n");

> > +        EbH265ReleaseOutBuffer(&header_ptr);

> > +        return ret;

> > +    }

> > +

> > +    memcpy(pkt->data, header_ptr->pBuffer, header_ptr->nFilledLen);

> > +

> > +    pkt->size = header_ptr->nFilledLen;

> > +    pkt->pts  = header_ptr->pts;

> > +    pkt->dts  = header_ptr->dts;

> > +    if ((header_ptr->sliceType == EB_IDR_PICTURE) ||

> > +        (header_ptr->sliceType == EB_I_PICTURE))

> > +        pkt->flags |= AV_PKT_FLAG_KEY;

> > +    if (header_ptr->sliceType == EB_NON_REF_PICTURE)

> > +        pkt->flags |= AV_PKT_FLAG_DISPOSABLE;

> > +

> > +    EbH265ReleaseOutBuffer(&header_ptr);

> > +

> > +    if (EB_BUFFERFLAG_EOS == header_ptr->nFlags)

> > +        svt_enc->eos_flag = EOS_TOTRIGGER;

> > +

> > +    return 0;

> > +}

> > +

> > +static av_cold int eb_enc_close(AVCodecContext *avctx) {

> > +    SvtContext *svt_enc = avctx->priv_data;

> > +

> > +    if (svt_enc) {

> > +        if (svt_enc->svt_handle) {

> > +            EbDeinitEncoder(svt_enc->svt_handle);

> > +            EbDeinitHandle(svt_enc->svt_handle);

> > +            svt_enc->svt_handle = NULL;

> > +        }

> > +

> > +        free_buffer(svt_enc);

> > +        svt_enc = NULL;

> > +    }

> > +

> > +    return 0;

> > +}

> > +

> > +#define OFFSET(x) offsetof(SvtContext, x) #define VE

> > +AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM static

> > const

> > +AVOption options[] = {

> > +    { "vui", "Enable Video Usability Info", OFFSET(vui_info),

> > +      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },

> > +

> > +    { "aud", "Include Access Unit Delimiter", OFFSET(aud),

> > +      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },

> > +

> > +    { "hielevel", "Hierarchical prediction levels setting",

> > OFFSET(hierarchical_level),

> > +      AV_OPT_TYPE_INT, { .i64 = 4 }, 1, 4, VE , "hielevel"},

> > +        { "flat",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },

> > INT_MIN, INT_MAX, VE, "hielevel" },

> > +        { "2level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 },

> > + INT_MIN,

> > INT_MAX, VE, "hielevel" },

> > +        { "3level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 },

> > + INT_MIN,

> > INT_MAX, VE, "hielevel" },

> > +        { "4level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 4 },

> > INT_MIN,

> > + INT_MAX, VE, "hielevel" },

> > +

> > +    { "la_depth", "Look ahead distance [0, 256]", OFFSET(la_depth),

> > +      AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 256, VE },

> > +

> > +    { "preset", "Encoding preset [0, 12] (e,g, for subjective quality

> > + tuning

> > mode and >=4k resolution), [0, 10] (for >= 1080p resolution), [0, 9]

> > (for all resolution and modes)",

> > +      OFFSET(enc_mode), AV_OPT_TYPE_INT, { .i64 = 9 }, 0, 12, VE },

> > +

> > +    { "profile", "Profile setting, Main Still Picture Profile not

> > + supported",

> > OFFSET(profile),

> > +      AV_OPT_TYPE_INT, { .i64 = FF_PROFILE_HEVC_MAIN_10 },

> 

> I am quite surprised why you set main10 as default profile.

> You added a debug message for 10bit input, but not added for 8bit case,

> doesn't mean 8bit case is more common?

> 

> [Jing] Main 10 allows bit depth of 8 to 10, and it's not for 10-bit only. I chose

> it as the default for SVT HEVC do that as well. The default value should be

> the same.


Then it will be a big problem as I can see: A hevc decoder may declare it just support main profile
main10 profile is not supported. Then the video encoded by SVT-HEVC can't be decoded, but actually it is 8bits video that 
should can be decoded if you declare it is main profile. 

And I have a quick check for x265, it is using main profile for 8bit video.

Probably you can set a UNKNOWN profile in FFmpeg level, and then let svt-hevc library to decide as input.
But still it shouldn't encode a 8 bit video as main10 profile. 

> > + FF_PROFILE_HEVC_MAIN, FF_PROFILE_HEVC_MAIN_10, VE, "profile"},

> > +

> > +    { "tier", "Set tier (general_tier_flag)", OFFSET(tier),

> > +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE, "tier" },

> > +        { "main", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, VE,

> > "tier" },

> > +        { "high", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, VE,

> > + "tier" },

> > +

> > +    { "level", "Set level (level_idc)", OFFSET(level),

> > +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 0xff, VE, "level" },

> > +

> > +    { "rc", "Bit rate control mode", OFFSET(rc_mode),

> > +      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE , "rc"},

> > +        { "cqp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN,

> > INT_MAX, VE, "rc" },

> > +        { "vbr", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN,

> > + INT_MAX, VE, "rc" },

> 

> Hm. VBR mode can be support but I haven't seen any parameter to set

> max_rate/vbv buffer.

> Does this mean the bitrate fluctuation is totally out of control?

> 

> [Jing] SVT HEVC's VBR is controlled by the max allowed QP, min allowed QP

> and the bitrate.

>


QP range is not equal to bitrate range. VBV buffer setting and HRD mode would better to be supported (mostly speaking, it is must to be supported).
However, I am fine since here is just a plugin.
diff mbox

Patch

diff --git a/configure b/configure
index 938ff10..2aabac4 100755
--- a/configure
+++ b/configure
@@ -264,6 +264,7 @@  External library support:
   --enable-libspeex        enable Speex de/encoding via libspeex [no]
   --enable-libsrt          enable Haivision SRT protocol via libsrt [no]
   --enable-libssh          enable SFTP protocol via libssh [no]
+  --enable-libsvthevc      enable HEVC encoding via svt [no]
   --enable-libtensorflow   enable TensorFlow as a DNN module backend
                            for DNN based filters like sr [no]
   --enable-libtesseract    enable Tesseract, needed for ocr filter [no]
@@ -1784,6 +1785,7 @@  EXTERNAL_LIBRARY_LIST="
     libspeex
     libsrt
     libssh
+    libsvthevc
     libtensorflow
     libtesseract
     libtheora
@@ -3173,6 +3175,7 @@  libshine_encoder_select="audio_frame_queue"
 libspeex_decoder_deps="libspeex"
 libspeex_encoder_deps="libspeex"
 libspeex_encoder_select="audio_frame_queue"
+libsvt_hevc_encoder_deps="libsvthevc"
 libtheora_encoder_deps="libtheora"
 libtwolame_encoder_deps="libtwolame"
 libvo_amrwbenc_encoder_deps="libvo_amrwbenc"
@@ -6209,6 +6212,7 @@  enabled libsoxr           && require libsoxr soxr.h soxr_create -lsoxr
 enabled libssh            && require_pkg_config libssh libssh libssh/sftp.h sftp_init
 enabled libspeex          && require_pkg_config libspeex speex speex/speex.h speex_decoder_init
 enabled libsrt            && require_pkg_config libsrt "srt >= 1.3.0" srt/srt.h srt_socket
+enabled libsvthevc        && require_pkg_config libsvthevc SvtHevcEnc EbApi.h EbInitHandle
 enabled libtensorflow     && require libtensorflow tensorflow/c/c_api.h TF_Version -ltensorflow
 enabled libtesseract      && require_pkg_config libtesseract tesseract tesseract/capi.h TessBaseAPICreate
 enabled libtheora         && require libtheora theora/theoraenc.h th_info_init -ltheoraenc -ltheoradec -logg
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 15c43a8..c93e545 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -987,6 +987,7 @@  OBJS-$(CONFIG_LIBOPUS_ENCODER)            += libopusenc.o libopus.o     \
 OBJS-$(CONFIG_LIBSHINE_ENCODER)           += libshine.o
 OBJS-$(CONFIG_LIBSPEEX_DECODER)           += libspeexdec.o
 OBJS-$(CONFIG_LIBSPEEX_ENCODER)           += libspeexenc.o
+OBJS-$(CONFIG_LIBSVT_HEVC_ENCODER)        += libsvt_hevc.o
 OBJS-$(CONFIG_LIBTHEORA_ENCODER)          += libtheoraenc.o
 OBJS-$(CONFIG_LIBTWOLAME_ENCODER)         += libtwolame.o
 OBJS-$(CONFIG_LIBVO_AMRWBENC_ENCODER)     += libvo-amrwbenc.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index b26aeca..e93f66f 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -703,6 +703,7 @@  extern AVCodec ff_librsvg_decoder;
 extern AVCodec ff_libshine_encoder;
 extern AVCodec ff_libspeex_encoder;
 extern AVCodec ff_libspeex_decoder;
+extern AVCodec ff_libsvt_hevc_encoder;
 extern AVCodec ff_libtheora_encoder;
 extern AVCodec ff_libtwolame_encoder;
 extern AVCodec ff_libvo_amrwbenc_encoder;
diff --git a/libavcodec/libsvt_hevc.c b/libavcodec/libsvt_hevc.c
new file mode 100644
index 0000000..4ef06a8
--- /dev/null
+++ b/libavcodec/libsvt_hevc.c
@@ -0,0 +1,500 @@ 
+/*
+* Scalable Video Technology for HEVC encoder library plugin
+*
+* Copyright (c) 2018 Intel Corporation
+*
+* 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 this program; if not, write to the Free Software
+* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+*/
+
+#include "EbErrorCodes.h"
+#include "EbTime.h"
+#include "EbApi.h"
+
+#include "libavutil/common.h"
+#include "libavutil/frame.h"
+#include "libavutil/opt.h"
+
+#include "internal.h"
+#include "avcodec.h"
+
+typedef enum eos_status {
+    EOS_NOT_REACHED = 0,
+    EOS_REACHED,
+    EOS_TOTRIGGER
+}EOS_STATUS;
+
+typedef struct SvtContext {
+    AVClass *class;
+
+    EB_H265_ENC_CONFIGURATION enc_params;
+    EB_COMPONENTTYPE *svt_handle;
+    EB_BUFFERHEADERTYPE in_buf;
+    EOS_STATUS eos_flag;
+
+    // User options.
+    int vui_info;
+    int hierarchical_level;
+    int la_depth;
+    int enc_mode;
+    int rc_mode;
+    int scd;
+    int tune;
+    int qp;
+    int hdr;
+
+    int forced_idr;
+
+    int aud;
+
+    int profile;
+    int tier;
+    int level;
+
+    int base_layer_switch_mode;
+} SvtContext;
+
+static int error_mapping(EB_ERRORTYPE svt_ret)
+{
+    int err;
+
+    switch (svt_ret) {
+    case EB_ErrorInsufficientResources:
+        err = AVERROR(ENOMEM);
+        break;
+
+    case EB_ErrorUndefined:
+    case EB_ErrorInvalidComponent:
+    case EB_ErrorBadParameter:
+        err = AVERROR(EINVAL);
+        break;
+
+    case EB_ErrorDestroyThreadFailed:
+    case EB_ErrorSemaphoreUnresponsive:
+    case EB_ErrorDestroySemaphoreFailed:
+    case EB_ErrorCreateMutexFailed:
+    case EB_ErrorMutexUnresponsive:
+    case EB_ErrorDestroyMutexFailed:
+        err = AVERROR_EXTERNAL;
+            break;
+
+    case EB_NoErrorEmptyQueue:
+        err = AVERROR(EAGAIN);
+
+    case EB_ErrorNone:
+        err = 0;
+        break;
+
+    default:
+        err = AVERROR_UNKNOWN;
+    }
+
+    return err;
+}
+
+static void free_buffer(SvtContext *svt_enc)
+{
+    uint8_t *in_data = svt_enc->in_buf.pBuffer;
+
+    av_freep(&in_data);
+}
+
+static int alloc_buffer(EB_H265_ENC_CONFIGURATION *config, SvtContext *svt_enc)
+{
+    EB_H265_ENC_INPUT *in_data;
+
+    // allocate buffer for in and out
+    in_data  = av_mallocz(sizeof(*in_data));
+    if (!in_data)
+        goto failed;
+    svt_enc->in_buf.pBuffer  = (unsigned char *)in_data;
+
+    svt_enc->in_buf.nSize        = sizeof(svt_enc->in_buf);
+    svt_enc->in_buf.pAppPrivate  = NULL;
+
+    return 0;
+
+failed:
+    free_buffer(svt_enc);
+    return AVERROR(ENOMEM);
+}
+
+static int config_enc_params(EB_H265_ENC_CONFIGURATION *param,
+                             AVCodecContext *avctx)
+{
+    SvtContext *svt_enc = avctx->priv_data;
+    int ret;
+
+    param->sourceWidth = avctx->width;
+    param->sourceHeight = avctx->height;
+    param->encoderBitDepth = 8;
+
+    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P10LE) {
+        av_log(avctx, AV_LOG_DEBUG, "Encoder 10 bits depth input\n");
+
+        param->encoderBitDepth = 10;
+    }
+    param->encoderColorFormat = EB_YUV420;
+
+    // Update param from options
+    param->hierarchicalLevels   = svt_enc->hierarchical_level - 1;
+    param->encMode              = svt_enc->enc_mode;
+    param->profile              = svt_enc->profile;
+    param->tier                 = svt_enc->tier;
+    param->level                = svt_enc->level;
+    param->rateControlMode      = svt_enc->rc_mode;
+    param->sceneChangeDetection = svt_enc->scd;
+    param->tune                 = svt_enc->tune;
+    param->baseLayerSwitchMode  = svt_enc->base_layer_switch_mode;
+    param->qp                   = svt_enc->qp;
+    param->accessUnitDelimiter  = svt_enc->aud;
+
+    param->targetBitRate = avctx->bit_rate;
+    if (avctx->gop_size > 0)
+        param->intraPeriodLength = avctx->gop_size - 1;
+
+    if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {
+        param->frameRateNumerator = avctx->framerate.num;
+        param->frameRateDenominator =
+            avctx->framerate.den * avctx->ticks_per_frame;
+    } else {
+        param->frameRateNumerator = avctx->time_base.den;
+        param->frameRateDenominator =
+            avctx->time_base.num * avctx->ticks_per_frame;
+    }
+
+    if (param->rateControlMode) {
+        param->maxQpAllowed = avctx->qmax;
+        param->minQpAllowed = avctx->qmin;
+    }
+
+    param->intraRefreshType =
+        !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) + 1;
+
+    // is it repeat headers for MP4 or Annex-b
+    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)
+        param->codeVpsSpsPps = 0;
+    else
+        param->codeVpsSpsPps = 1;
+
+    param->codeEosNal = 1;
+
+    if (svt_enc->hdr) {
+        av_log(avctx, AV_LOG_WARNING, "Color properties not supported yet\n");
+
+        svt_enc->vui_info = 1;
+        param->highDynamicRangeInput = svt_enc->hdr;
+    }
+
+    if (svt_enc->vui_info)
+        param->videoUsabilityInfo = svt_enc->vui_info;
+
+    if (svt_enc->la_depth != -1)
+        param->lookAheadDistance = svt_enc->la_depth;
+
+    ret = alloc_buffer(param, svt_enc);
+
+    return ret;
+}
+
+static void read_in_data(EB_H265_ENC_CONFIGURATION *config,
+                         const AVFrame *frame,
+                         EB_BUFFERHEADERTYPE *header_ptr)
+{
+    uint8_t is16bit = config->encoderBitDepth > 8;
+    uint64_t luma_size =
+        (uint64_t)config->sourceWidth * config->sourceHeight<< is16bit;
+    EB_H265_ENC_INPUT *in_data = (EB_H265_ENC_INPUT *)header_ptr->pBuffer;
+
+    // support yuv420p and yuv420p010
+    in_data->luma = frame->data[0];
+    in_data->cb   = frame->data[1];
+    in_data->cr   = frame->data[2];
+
+    // stride info
+    in_data->yStride  = frame->linesize[0] >> is16bit;
+    in_data->cbStride = frame->linesize[1] >> is16bit;
+    in_data->crStride = frame->linesize[2] >> is16bit;
+
+    header_ptr->nFilledLen   += luma_size * 3/2u;
+}
+
+static av_cold int eb_enc_init(AVCodecContext *avctx)
+{
+    SvtContext   *svt_enc = avctx->priv_data;
+    EB_ERRORTYPE svt_ret;
+
+    svt_enc->eos_flag = EOS_NOT_REACHED;
+
+    svt_ret = EbInitHandle(&svt_enc->svt_handle, svt_enc, &svt_enc->enc_params);
+    if (svt_ret != EB_ErrorNone) {
+        av_log(avctx, AV_LOG_ERROR, "Error init encoder handle\n");
+        goto failed;
+    }
+
+    svt_ret = config_enc_params(&svt_enc->enc_params, avctx);
+    if (svt_ret != EB_ErrorNone) {
+        av_log(avctx, AV_LOG_ERROR, "Error configure encoder parameters\n");
+        goto failed_init_handle;
+    }
+
+    svt_ret = EbH265EncSetParameter(svt_enc->svt_handle, &svt_enc->enc_params);
+    if (svt_ret != EB_ErrorNone) {
+        av_log(avctx, AV_LOG_ERROR, "Error setting encoder parameters\n");
+        goto failed_init_handle;
+    }
+
+    svt_ret = EbInitEncoder(svt_enc->svt_handle);
+    if (svt_ret != EB_ErrorNone) {
+        av_log(avctx, AV_LOG_ERROR, "Error init encoder\n");
+        goto failed_init_handle;
+    }
+
+    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
+        EB_BUFFERHEADERTYPE *header_ptr = NULL;
+
+        svt_ret = EbH265EncStreamHeader(svt_enc->svt_handle, &header_ptr);
+        if (svt_ret != EB_ErrorNone) {
+            av_log(avctx, AV_LOG_ERROR, "Error when build stream header.\n");
+            goto failed_init_enc;
+        }
+
+        avctx->extradata_size = header_ptr->nFilledLen;
+        avctx->extradata = av_mallocz(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
+        if (!avctx->extradata) {
+            av_log(avctx, AV_LOG_ERROR,
+                   "Cannot allocate HEVC header of size %d.\n", avctx->extradata_size);
+            svt_ret = EB_ErrorInsufficientResources;
+            goto failed_init_enc;
+        }
+        memcpy(avctx->extradata, header_ptr->pBuffer, avctx->extradata_size);
+    }
+
+    return 0;
+
+failed_init_enc:
+    EbDeinitEncoder(svt_enc->svt_handle);
+failed_init_handle:
+    EbDeinitHandle(svt_enc->svt_handle);
+failed:
+    free_buffer(svt_enc);
+    svt_enc->svt_handle = NULL;
+    svt_enc = NULL;
+    return error_mapping(svt_ret);
+}
+
+static int eb_send_frame(AVCodecContext *avctx, const AVFrame *frame)
+{
+    SvtContext           *svt_enc = avctx->priv_data;
+    EB_BUFFERHEADERTYPE  *header_ptr = &svt_enc->in_buf;
+
+    if (!frame) {
+        EB_BUFFERHEADERTYPE header_ptr_last;
+        header_ptr_last.nAllocLen   = 0;
+        header_ptr_last.nFilledLen  = 0;
+        header_ptr_last.nTickCount  = 0;
+        header_ptr_last.pAppPrivate = NULL;
+        header_ptr_last.pBuffer     = NULL;
+        header_ptr_last.nFlags      = EB_BUFFERFLAG_EOS;
+
+        EbH265EncSendPicture(svt_enc->svt_handle, &header_ptr_last);
+        svt_enc->eos_flag = EOS_REACHED;
+        av_log(avctx, AV_LOG_DEBUG, "Finish sending frames!!!\n");
+        return 0;
+    }
+
+    read_in_data(&svt_enc->enc_params, frame, header_ptr);
+
+    header_ptr->nFlags       = 0;
+    header_ptr->pAppPrivate  = NULL;
+    header_ptr->pts          = frame->pts;
+    switch (frame->pict_type) {
+    case AV_PICTURE_TYPE_I:
+        header_ptr->sliceType = svt_enc->forced_idr > 0 ? EB_IDR_PICTURE : EB_I_PICTURE;
+        break;
+    case AV_PICTURE_TYPE_P:
+        header_ptr->sliceType = EB_P_PICTURE;
+        break;
+    case AV_PICTURE_TYPE_B:
+        header_ptr->sliceType = EB_B_PICTURE;
+        break;
+    default:
+        header_ptr->sliceType = EB_INVALID_PICTURE;
+        break;
+    }
+    EbH265EncSendPicture(svt_enc->svt_handle, header_ptr);
+
+    return 0;
+}
+
+static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt)
+{
+    SvtContext  *svt_enc = avctx->priv_data;
+    EB_BUFFERHEADERTYPE   *header_ptr = NULL;
+    EB_ERRORTYPE          svt_ret;
+    int ret = 0;
+
+    if (EOS_TOTRIGGER == svt_enc->eos_flag) {
+        pkt = NULL;
+        return AVERROR_EOF;
+    }
+
+    svt_ret = EbH265GetPacket(svt_enc->svt_handle, &header_ptr, svt_enc->eos_flag);
+    if (svt_ret == EB_NoErrorEmptyQueue)
+        return AVERROR(EAGAIN);
+
+    if ((ret = ff_alloc_packet2(avctx, pkt, header_ptr->nFilledLen, 0)) < 0) {
+        av_log(avctx, AV_LOG_ERROR, "Failed to allocate output packet.\n");
+        EbH265ReleaseOutBuffer(&header_ptr);
+        return ret;
+    }
+
+    memcpy(pkt->data, header_ptr->pBuffer, header_ptr->nFilledLen);
+
+    pkt->size = header_ptr->nFilledLen;
+    pkt->pts  = header_ptr->pts;
+    pkt->dts  = header_ptr->dts;
+    if ((header_ptr->sliceType == EB_IDR_PICTURE) ||
+        (header_ptr->sliceType == EB_I_PICTURE))
+        pkt->flags |= AV_PKT_FLAG_KEY;
+    if (header_ptr->sliceType == EB_NON_REF_PICTURE)
+        pkt->flags |= AV_PKT_FLAG_DISPOSABLE;
+
+    EbH265ReleaseOutBuffer(&header_ptr);
+
+    if (EB_BUFFERFLAG_EOS == header_ptr->nFlags)
+        svt_enc->eos_flag = EOS_TOTRIGGER;
+
+    return 0;
+}
+
+static av_cold int eb_enc_close(AVCodecContext *avctx)
+{
+    SvtContext *svt_enc = avctx->priv_data;
+
+    if (svt_enc) {
+        if (svt_enc->svt_handle) {
+            EbDeinitEncoder(svt_enc->svt_handle);
+            EbDeinitHandle(svt_enc->svt_handle);
+            svt_enc->svt_handle = NULL;
+        }
+
+        free_buffer(svt_enc);
+        svt_enc = NULL;
+    }
+
+    return 0;
+}
+
+#define OFFSET(x) offsetof(SvtContext, x)
+#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
+static const AVOption options[] = {
+    { "vui", "Enable Video Usability Info", OFFSET(vui_info),
+      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },
+
+    { "aud", "Include Access Unit Delimiter", OFFSET(aud),
+      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
+
+    { "hielevel", "Hierarchical prediction levels setting", OFFSET(hierarchical_level),
+      AV_OPT_TYPE_INT, { .i64 = 4 }, 1, 4, VE , "hielevel"},
+        { "flat",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, INT_MAX, VE, "hielevel" },
+        { "2level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 },  INT_MIN, INT_MAX, VE, "hielevel" },
+        { "3level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 },  INT_MIN, INT_MAX, VE, "hielevel" },
+        { "4level", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 4 },  INT_MIN, INT_MAX, VE, "hielevel" },
+
+    { "la_depth", "Look ahead distance [0, 256]", OFFSET(la_depth),
+      AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 256, VE },
+
+    { "preset", "Encoding preset [0, 12] (e,g, for subjective quality tuning mode and >=4k resolution), [0, 10] (for >= 1080p resolution), [0, 9] (for all resolution and modes)",
+      OFFSET(enc_mode), AV_OPT_TYPE_INT, { .i64 = 9 }, 0, 12, VE },
+
+    { "profile", "Profile setting, Main Still Picture Profile not supported", OFFSET(profile),
+      AV_OPT_TYPE_INT, { .i64 = FF_PROFILE_HEVC_MAIN_10 }, FF_PROFILE_HEVC_MAIN, FF_PROFILE_HEVC_MAIN_10, VE, "profile"},
+
+    { "tier", "Set tier (general_tier_flag)", OFFSET(tier),
+      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE, "tier" },
+        { "main", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, VE, "tier" },
+        { "high", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, VE, "tier" },
+
+    { "level", "Set level (level_idc)", OFFSET(level),
+      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 0xff, VE, "level" },
+
+    { "rc", "Bit rate control mode", OFFSET(rc_mode),
+      AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE , "rc"},
+        { "cqp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN, INT_MAX, VE, "rc" },
+        { "vbr", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, INT_MAX, VE, "rc" },
+
+    { "qp", "QP value for intra frames", OFFSET(qp),
+      AV_OPT_TYPE_INT, { .i64 = 32 }, 0, 51, VE },
+
+    { "sc_detection", "Scene change detection", OFFSET(scd),
+      AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VE },
+
+    { "tune", "Quality tuning mode", OFFSET(tune), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 2, VE, "tune" },
+        { "sq", "Visually optimized mode", 0,
+          AV_OPT_TYPE_CONST, { .i64 = 0 },  INT_MIN, INT_MAX, VE, "tune" },
+        { "oq",  "PSNR / SSIM optimized mode",  0,
+          AV_OPT_TYPE_CONST, { .i64 = 1 },  INT_MIN, INT_MAX, VE, "tune" },
+        { "vmaf", "VMAF optimized mode", 0,
+          AV_OPT_TYPE_CONST, { .i64 = 2 },  INT_MIN, INT_MAX, VE, "tune" },
+
+    { "bl_mode", "Random Access Prediction Structure type setting", OFFSET(base_layer_switch_mode),
+      AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
+
+    { "forced-idr", "If forcing keyframes, force them as IDR frames.", OFFSET(forced_idr),
+      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, -1, 1, VE },
+
+    { "hdr", "High dynamic range input", OFFSET(hdr),
+      AV_OPT_TYPE_BOOL,   { .i64 = 0 }, 0, 1, VE },
+
+    {NULL},
+};
+
+static const AVClass class = {
+    .class_name = "libsvt_hevc",
+    .item_name  = av_default_item_name,
+    .option     = options,
+    .version    = LIBAVUTIL_VERSION_INT,
+};
+
+static const AVCodecDefault eb_enc_defaults[] = {
+    { "b",         "7M"    },
+    { "flags",     "+cgop" },
+    { "qmin",      "10"    },
+    { "qmax",      "48"    },
+    { "g",         "-2"    },
+    { NULL },
+};
+
+AVCodec ff_libsvt_hevc_encoder = {
+    .name           = "libsvt_hevc",
+    .long_name      = NULL_IF_CONFIG_SMALL("SVT-HEVC(Scalable Video Technology for HEVC) encoder"),
+    .priv_data_size = sizeof(SvtContext),
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = AV_CODEC_ID_HEVC,
+    .init           = eb_enc_init,
+    .send_frame     = eb_send_frame,
+    .receive_packet = eb_receive_packet,
+    .close          = eb_enc_close,
+    .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AUTO_THREADS,
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P,
+                                                    AV_PIX_FMT_YUV420P10,
+                                                    AV_PIX_FMT_NONE },
+    .priv_class     = &class,
+    .defaults       = eb_enc_defaults,
+    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP,
+    .wrapper_name   = "libsvt_hevc",
+};