[FFmpeg-devel] [PATCH v2 7/9] avcodec/nvenc: add MV-HEVC encoding support
James Almer
jamrial at gmail.com
Fri Jan 31 05:26:00 EET 2025
On 1/30/2025 8:23 PM, Timo Rothenpieler wrote:
> From: Diego de Souza <ddesouza at nvidia.com>
>
> Added support for MV-HEVC encoding for stereoscopic videos (2 views
> only). Compatible with the framepack filter when using the
> AV_STEREO3D_FRAMESEQUENCE format.
>
> Signed-off-by: Diego de Souza <ddesouza at nvidia.com>
> ---
> libavcodec/nvenc.c | 70 +++++++++++++++++++++++++++++++++++++++++
> libavcodec/nvenc.h | 8 +++++
> libavcodec/nvenc_hevc.c | 5 ++-
> 3 files changed, 82 insertions(+), 1 deletion(-)
>
> diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c
> index 3403fa8996..ad53eaaf98 100644
> --- a/libavcodec/nvenc.c
> +++ b/libavcodec/nvenc.c
> @@ -35,6 +35,7 @@
> #include "libavutil/mem.h"
> #include "libavutil/pixdesc.h"
> #include "libavutil/mathematics.h"
> +#include "libavutil/stereo3d.h"
> #include "atsc_a53.h"
> #include "codec_desc.h"
> #include "encode.h"
> @@ -656,6 +657,14 @@ static int nvenc_check_capabilities(AVCodecContext *avctx)
>
> ctx->support_dyn_bitrate = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_DYN_BITRATE_CHANGE);
>
> +#ifdef NVENC_HAVE_MVHEVC
> + ctx->multiview_supported = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_MVHEVC_ENCODE) > 0;
> + if(ctx->profile == NV_ENC_HEVC_PROFILE_MULTIVIEW_MAIN && !ctx->multiview_supported) {
> + av_log(avctx, AV_LOG_WARNING, "Multiview not supported by the device\n");
> + return AVERROR(ENOSYS);
> + }
> +#endif
> +
> return 0;
> }
>
> @@ -1475,6 +1484,13 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
>
> hevc->outputPictureTimingSEI = 1;
>
> +#ifdef NVENC_HAVE_MVHEVC
> + if (ctx->multiview_supported && ctx->profile == NV_ENC_HEVC_PROFILE_MAIN &&
> + (av_frame_side_data_get(avctx->decoded_side_data, avctx->nb_decoded_side_data, AV_FRAME_DATA_STEREO3D) ||
Stereo 3D side data being present in the encoder global side data array
is not guarantee that it's there to signal stereoscopic video as it
could have a type of AV_STEREO3D_2D, for example.
You need to ensure it's either AV_STEREO3D_UNSPEC, or
AV_STEREO3D_FRAMESEQUENCE (which afaik we don't export at the global
level, only for frames containing one of the two views).
> + av_frame_side_data_get(avctx->decoded_side_data, avctx->nb_decoded_side_data, AV_FRAME_DATA_VIEW_ID)))
VIEW_ID is not a global type, so it should not be there are at all.
> + ctx->profile = NV_ENC_HEVC_PROFILE_MULTIVIEW_MAIN;
> +#endif
> +
> switch (ctx->profile) {
> case NV_ENC_HEVC_PROFILE_MAIN:
> cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
> @@ -1488,6 +1504,16 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
> cc->profileGUID = NV_ENC_HEVC_PROFILE_FREXT_GUID;
> avctx->profile = AV_PROFILE_HEVC_REXT;
> break;
> +#ifdef NVENC_HAVE_MVHEVC
> + case NV_ENC_HEVC_PROFILE_MULTIVIEW_MAIN:
> + cc->profileGUID = NV_ENC_HEVC_PROFILE_MAIN_GUID;
> + avctx->profile = AV_PROFILE_HEVC_MULTIVIEW_MAIN;
> + ctx->multiview = 1;
> +
> + hevc->enableMVHEVC = 1;
> + hevc->outputHevc3DReferenceDisplayInfo = 1;
> + break;
> +#endif
> }
>
> // force setting profile as main10 if input is 10 bit or if it should be encoded as 10 bit
> @@ -1502,6 +1528,13 @@ static av_cold int nvenc_setup_hevc_config(AVCodecContext *avctx)
> avctx->profile = AV_PROFILE_HEVC_REXT;
> }
>
> +#ifdef NVENC_HAVE_MVHEVC
> + if (ctx->multiview && avctx->profile != AV_PROFILE_HEVC_MULTIVIEW_MAIN) {
> + av_log(avctx, AV_LOG_ERROR, "Multiview encoding only works for Main profile content.\n");
> + return AVERROR(EINVAL);
> + }
> +#endif
> +
> hevc->chromaFormatIDC = IS_YUV444(ctx->data_pix_fmt) ? 3 : IS_YUV422(ctx->data_pix_fmt) ? 2 : 1;
>
> #ifdef NVENC_HAVE_NEW_BIT_DEPTH_API
> @@ -2469,6 +2502,9 @@ static int nvenc_set_timestamp(AVCodecContext *avctx,
>
> // This can be more than necessary, but we don't know the real reorder delay.
> delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0);
> +#ifdef NVENC_HAVE_MVHEVC
> + delay *= ctx->multiview ? 2 : 1;
> +#endif
> if (ctx->output_frame_num >= delay) {
> pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list);
> ctx->output_frame_num++;
> @@ -2875,6 +2911,9 @@ static int nvenc_send_frame(AVCodecContext *avctx, const AVFrame *frame)
> int res, res2;
> int sei_count = 0;
> int i;
> +#ifdef NVENC_HAVE_MVHEVC
> + HEVC_3D_REFERENCE_DISPLAY_INFO ref_disp_info = { 0 };
> +#endif
>
> NvencContext *ctx = avctx->priv_data;
> NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
> @@ -2952,6 +2991,37 @@ static int nvenc_send_frame(AVCodecContext *avctx, const AVFrame *frame)
> if (res < 0)
> return res;
>
> +#ifdef NVENC_HAVE_MVHEVC
> + if (ctx->multiview) {
> + const AVFrameSideData *sd_stereo3d = av_frame_get_side_data(frame, AV_FRAME_DATA_STEREO3D);
> + const AVFrameSideData *sd_view_id = av_frame_get_side_data(frame, AV_FRAME_DATA_VIEW_ID);
> + int view_ids_reversed = 0;
> +
> + if (sd_view_id)
> + ctx->next_view_id = *(int*)sd_view_id->data;
> +
> + if (sd_stereo3d) {
> + const AVStereo3D *stereo = (const AVStereo3D*)sd_stereo3d->data;
> +
> + if (stereo->type == AV_STEREO3D_FRAMESEQUENCE) {
> + if (!sd_view_id)
> + ctx->next_view_id = (stereo->view == AV_STEREO3D_VIEW_LEFT) ? 0 : 1;
> + if (stereo->view == AV_STEREO3D_VIEW_LEFT && ctx->next_view_id)
> + view_ids_reversed = 1;
> + } else
> + av_log(avctx, AV_LOG_ERROR, "Stereo format %d not supported! Only AV_STEREO3D_FRAMESEQUENCE is supported!\n", stereo->type);
> + }
> +
> + pic_params.codecPicParams.hevcPicParams.viewId = ctx->next_view_id;
> + ctx->next_view_id ^= 1;
> +
> + ref_disp_info.precRefDisplayWidth = 31;
> + ref_disp_info.leftViewId[0] = view_ids_reversed ? 1 : 0;
> + ref_disp_info.rightViewId[0] = view_ids_reversed ? 0 : 1;
> + pic_params.codecPicParams.hevcPicParams.p3DReferenceDisplayInfo = &ref_disp_info;
> + }
> +#endif
> +
> nv_status = p_nvenc->nvEncEncodePicture(ctx->nvencoder, &pic_params);
>
> for (i = 0; i < sei_count; i++)
> diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h
> index 6f7f8d4e7f..d085b08260 100644
> --- a/libavcodec/nvenc.h
> +++ b/libavcodec/nvenc.h
> @@ -99,6 +99,7 @@ typedef void ID3D11Device;
> #define NVENC_HAVE_422_SUPPORT
> #define NVENC_HAVE_AV1_UHQ_TUNING
> #define NVENC_HAVE_H264_AND_AV1_TEMPORAL_FILTER
> +#define NVENC_HAVE_MVHEVC
> #endif
>
> typedef struct NvencSurface
> @@ -172,6 +173,11 @@ enum {
> NV_ENC_HEVC_PROFILE_MAIN,
> NV_ENC_HEVC_PROFILE_MAIN_10,
> NV_ENC_HEVC_PROFILE_REXT,
> +#ifdef NVENC_HAVE_MVHEVC
> + NV_ENC_HEVC_PROFILE_MULTIVIEW_MAIN,
> +#endif
> +
> + NV_ENC_HEVC_PROFILE_COUNT
> };
>
> enum {
> @@ -245,6 +251,7 @@ typedef struct NvencContext
> void *nvencoder;
>
> uint32_t frame_idx_counter;
> + uint32_t next_view_id;
>
> int preset;
> int profile;
> @@ -299,6 +306,7 @@ typedef struct NvencContext
> int lookahead_level;
> int unidir_b;
> int split_encode_mode;
> + int multiview, multiview_supported;
> } NvencContext;
>
> int ff_nvenc_encode_init(AVCodecContext *avctx);
> diff --git a/libavcodec/nvenc_hevc.c b/libavcodec/nvenc_hevc.c
> index 5696e14dd4..3c08563c1f 100644
> --- a/libavcodec/nvenc_hevc.c
> +++ b/libavcodec/nvenc_hevc.c
> @@ -60,10 +60,13 @@ static const AVOption options[] = {
> { "ull", "Ultra low latency", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_TUNING_INFO_ULTRA_LOW_LATENCY }, 0, 0, VE, .unit = "tune" },
> { "lossless", "Lossless", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_TUNING_INFO_LOSSLESS }, 0, 0, VE, .unit = "tune" },
> #endif
> - { "profile", "Set the encoding profile", OFFSET(profile), AV_OPT_TYPE_INT, { .i64 = NV_ENC_HEVC_PROFILE_MAIN }, NV_ENC_HEVC_PROFILE_MAIN, AV_PROFILE_HEVC_REXT, VE, .unit = "profile" },
> + { "profile", "Set the encoding profile", OFFSET(profile), AV_OPT_TYPE_INT, { .i64 = NV_ENC_HEVC_PROFILE_MAIN }, NV_ENC_HEVC_PROFILE_MAIN, NV_ENC_HEVC_PROFILE_COUNT - 1, VE, .unit = "profile" },
> { "main", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_MAIN }, 0, 0, VE, .unit = "profile" },
> { "main10", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_MAIN_10 }, 0, 0, VE, .unit = "profile" },
> { "rext", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_REXT }, 0, 0, VE, .unit = "profile" },
> +#ifdef NVENC_HAVE_MVHEVC
> + { "mv", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_HEVC_PROFILE_MULTIVIEW_MAIN }, 0, 0, VE, .unit = "profile" },
> +#endif
> { "level", "Set the encoding level restriction", OFFSET(level), AV_OPT_TYPE_INT, { .i64 = NV_ENC_LEVEL_AUTOSELECT }, NV_ENC_LEVEL_AUTOSELECT, NV_ENC_LEVEL_HEVC_62, VE, .unit = "level" },
> { "auto", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_LEVEL_AUTOSELECT }, 0, 0, VE, .unit = "level" },
> { "1", "", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_LEVEL_HEVC_1 }, 0, 0, VE, .unit = "level" },
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20250131/1679f5e5/attachment.sig>
More information about the ffmpeg-devel
mailing list