[FFmpeg-devel] [PATCH V1 1/2] lavc/svt_hevc: add libsvt hevc encoder wrapper.
mypopy at gmail.com
mypopy at gmail.com
Mon Dec 3 03:59:46 EET 2018
On Sun, Dec 2, 2018 at 10:19 PM James Almer <jamrial at gmail.com> wrote:
>
> On 12/2/2018 2:38 AM, Jun Zhao wrote:
> > base on patch by Huang, Zhengxu from https://github.com/intel/SVT-HEVC
> >
> > Signed-off-by: Huang, Zhengxu <zhengxu.huang at intel.com>
> > Signed-off-by: hassene <hassene.tmar at intel.com>
> > Signed-off-by: Jun Zhao <jun.zhao at intel.com>
> > ---
> > configure | 4 +
> > libavcodec/Makefile | 1 +
> > libavcodec/allcodecs.c | 1 +
> > libavcodec/libsvt_hevc.c | 370
++++++++++++++++++++++++++++++++++++++++++++++
> > 4 files changed, 376 insertions(+), 0 deletions(-)
> > create mode 100644 libavcodec/libsvt_hevc.c
> >
> > diff --git a/configure b/configure
> > index 54b7e11..c3dca73 100755
> > --- a/configure
> > +++ b/configure
> > @@ -263,6 +263,7 @@ 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-libsvt 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]
> > @@ -1665,6 +1666,7 @@ EXTERNAL_LIBRARY_GPL_LIST="
> > libcdio
> > libdavs2
> > librubberband
> > + libsvt
> > libvidstab
> > libx264
> > libx265
> > @@ -3087,6 +3089,7 @@ pcm_mulaw_at_encoder_select="audio_frame_queue"
> > chromaprint_muxer_deps="chromaprint"
> > h264_videotoolbox_encoder_deps="pthreads"
> > h264_videotoolbox_encoder_select="videotoolbox_encoder"
> > +hevc_svt_encoder_deps="libsvt"
>
> Since this is an external library based encoder, the correct name would
> be libsvt_encoder.
> You can also make it libsvt_hevc_encoder instead, which is what you used
> as AVCodec.name.
>
I perfer libsvt_hevc_encoder than libsvt_encoder, because maybe we will
have libsvt_{av1/vp9/xxx}_encoder in the future.
> > hevc_videotoolbox_encoder_deps="pthreads"
> > hevc_videotoolbox_encoder_select="videotoolbox_encoder"
> > libaom_av1_decoder_deps="libaom"
>
>
> [...]
>
> > +static av_cold int eb_enc_init(AVCodecContext *avctx)
> > +{
> > + SvtContext *q = avctx->priv_data;
> > + SvtEncoder *svt_enc = NULL;
> > + EB_ERRORTYPE ret = EB_ErrorNone;
> > +
> > + q->svt_enc = av_mallocz(sizeof(*q->svt_enc));
> > + if (!q->svt_enc)
> > + return AVERROR(ENOMEM);
> > + svt_enc = q->svt_enc;
> > +
> > + q->eos_flag = 0;
> > +
> > + ret = EbInitHandle(&svt_enc->svt_handle, q, &svt_enc->enc_params);
> > + if (ret != EB_ErrorNone)
> > + goto failed_init;
> > +
> > + ret = config_enc_params(&svt_enc->enc_params, avctx);
> > + if (ret != EB_ErrorNone)
> > + goto failed_init;
> > +
> > + ret = EbH265EncSetParameter(svt_enc->svt_handle,
&svt_enc->enc_params);
> > + if (ret != EB_ErrorNone)
> > + goto failed_init;
> > +
> > + ret = EbInitEncoder(svt_enc->svt_handle);
> > + if (ret != EB_ErrorNone)
> > + goto failed_init;
> > +
> > + if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
> > +
> > + EB_BUFFERHEADERTYPE headerPtr;
> > + headerPtr.nSize = sizeof(EB_BUFFERHEADERTYPE);
> > + headerPtr.nFilledLen = 0;
> > + headerPtr.pBuffer = av_malloc(10 * 1024 * 1024);
> > + headerPtr.nAllocLen = (10 * 1024 * 1024);
> > +
> > + if (!headerPtr.pBuffer)
> > + return AVERROR(ENOMEM);
> > +
> > + ret = EbH265EncStreamHeader(svt_enc->svt_handle, &headerPtr);
> > + if (ret != EB_ErrorNone) {
> > + av_freep(&headerPtr.pBuffer);
> > + goto failed_init;
> > + }
> > + avctx->extradata_size = headerPtr.nFilledLen;
> > + avctx->extradata = av_malloc(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);
> > + return AVERROR(ENOMEM);
> > + }
> > + memcpy(avctx->extradata, headerPtr.pBuffer,
avctx->extradata_size);
> > + av_freep(&headerPtr.pBuffer);
> > + }
> > + return 0;
> > +
> > +failed_init:
> > + return error_mapping(ret);
>
> If it's only this line, just use it directly where needed instead of a
goto.
Will double-check the fail handle, tks.
>
> [...]
>
> > +#define OFFSET(x) offsetof(SvtContext, x)
> > +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
> > +static const AVOption options[] = {
> > + {"vui", "Enable vui info", OFFSET(svt_param.vui_info),
AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
> > + {"hielevel", "Hierarchical Prediction Levels [0,3]",
OFFSET(svt_param.hierarchical_level), AV_OPT_TYPE_INT, { .i64 = 3 }, 0, 3,
VE },
> > + {"la_depth", "Look Ahead Distance [0,256]",
OFFSET(svt_param.la_depth), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 256, VE },
> > + {"intra_ref_type", "Intra Refresh Type 0: No intra refresh 1: CRA
(Open GOP) 2: IDR", OFFSET(svt_param.intra_ref_type), AV_OPT_TYPE_INT, {
.i64 = 1 }, 1, 2, VE },
> > + {"enc_p", "Encoding preset [0,12] (for tune 0 and >=4k
resolution), [0,10] (for >= 1080p resolution), [0,9] (for all resolution
and modes)", OFFSET(svt_param.enc_mode), AV_OPT_TYPE_INT, { .i64 = 9 }, 0,
12, VE },
>
> Use preset instead, which is more in line with similar encoders.
Ok, will follow
>
> > + {"profile", "Profile now support[1,2], Main Still Picture Profile
not supported", OFFSET(svt_param.profile), AV_OPT_TYPE_INT, { .i64 = 2 },
1, 2, VE },
> > + {"rc", "RC mode 0: CQP 1: VBR", OFFSET(svt_param.rc_mode),
AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
> > + {"q", "QP value for intra frames", OFFSET(svt_param.qp),
AV_OPT_TYPE_INT, { .i64 = 32 }, 0, 51, VE },
>
> qp.
Will follow the name to avoid surprise
>
> > + {"scd", "Scene change detection", OFFSET(svt_param.scd),
AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
> > + {"tune", "Tune mode: SQ/OQ[0,1]", OFFSET(svt_param.tune),
AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
> > + {"bl_mode", "Random Access Prediction Structure Type",
OFFSET(svt_param.base_layer_switch_mode), 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" },
> > + { "refs", "0" },
> > + { "g", "64" },
> > + { "flags", "+cgop" },
> > + { NULL },
> > +};
> > +
> > +AVCodec ff_hevc_svt_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,
>
> Keep in mind that, at least for now and unlike with the decoupled
> input-output decode API, if you only provide a send/receive callback
> then the encoder will not work with avcodec_encode_video2(), only with
> avcodec_send_frame() and avcodec_receive_packet().
>
Because avcodec_encode_video2() flags with deprecated, so I think is Ok if
we don't implement the encode2 callback.
BTW: What's the reason we don't use avcodec_send_frame() and
avcodec_receive_packet() in avcodec_encode_video2() like use
avcodec_send_packet() and avcodec_receive_frame() in avcodec_decode_video2()
?
> > + .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",
> > +};
> >
>
More information about the ffmpeg-devel
mailing list