[FFmpeg-devel] [PATCH] libavcodec/qsvenc: enable LowDelayBRC and MaxFrameSizeI/MaxFrameSizeP for more accurate bitrate control
"zhilizhao(赵志立)"
quinkblack at foxmail.com
Mon Mar 21 07:17:40 EET 2022
> On Mar 10, 2022, at 6:11 PM, He, Fan F <fan.f.he-at-intel.com at ffmpeg.org> wrote:
>
> Feature introduction of LowDelayBRC, MaxFrameSizeI and MaxFrameSizeP could be found here:
> https://github.com/Intel-Media-SDK/MediaSDK/blob/master/doc/mediasdk-man.md
>
> Signed-off-by: Dmitry Ermilov <dmitry.ermilov at intel.com>
> Signed-off-by: Fan F He <fan.f.he at intel.com>
> ---
> doc/encoders.texi | 26 ++++++++++++++++++++++++++
> libavcodec/qsvenc.c | 17 +++++++++++++++++
> libavcodec/qsvenc.h | 6 ++++++
> 3 files changed, 49 insertions(+)
>
> diff --git a/doc/encoders.texi b/doc/encoders.texi
> index 1bd38671ca..47c8577e09 100644
> --- a/doc/encoders.texi
> +++ b/doc/encoders.texi
> @@ -3264,6 +3264,14 @@ Enable rate distortion optimization.
> @item @var{max_frame_size}
> Maximum encoded frame size in bytes.
>
> + at item @var{max_frame_size_i}
> +Maximum encoded frame size for I frames in bytes. If this value is set as larger
> +than zero, then for I frames the value set by max_frame_size is ignored.
> +
> + at item @var{max_frame_size_p}
> +Maximum encoded frame size for P frames in bytes. If this value is set as larger
> +than zero, then for P frames the value set by max_frame_size is ignored.
> +
> @item @var{max_slice_size}
> Maximum encoded slice size in bytes.
>
> @@ -3280,6 +3288,11 @@ Setting this flag enables macroblock level bitrate control that generally
> improves subjective visual quality. Enabling this flag may have negative impact
> on performance and objective visual quality metric.
>
> + at item @var{low_delay_brc}
> +Setting this flag turns on or off LowDelayBRC feautre in qsv plugin, which provides
> +more accurate bitrate control to minimize the variance of bitstream size frame
> +by frame. Value: -1-default 0-off 1-on
> +
> @item @var{adaptive_i}
> This flag controls insertion of I frames by the QSV encoder. Turn ON this flag
> to allow changing of frame type from P and B to I.
> @@ -3392,6 +3405,14 @@ Enable rate distortion optimization.
> @item @var{max_frame_size}
> Maximum encoded frame size in bytes.
>
> + at item @var{max_frame_size_i}
> +Maximum encoded frame size for I frames in bytes. If this value is set as larger
> +than zero, then for I frames the value set by max_frame_size is ignored.
> +
> + at item @var{max_frame_size_p}
> +Maximum encoded frame size for P frames in bytes. If this value is set as larger
> +than zero, then for P frames the value set by max_frame_size is ignored.
> +
> @item @var{max_slice_size}
> Maximum encoded slice size in bytes.
>
> @@ -3400,6 +3421,11 @@ Setting this flag enables macroblock level bitrate control that generally
> improves subjective visual quality. Enabling this flag may have negative impact
> on performance and objective visual quality metric.
>
> + at item @var{low_delay_brc}
> +Setting this flag turns on or off LowDelayBRC feautre in qsv plugin, which provides
> +more accurate bitrate control to minimize the variance of bitstream size frame
> +by frame. Value: -1-default 0-off 1-on
> +
> @item @var{p_strategy}
> Enable P-pyramid: 0-default 1-simple 2-pyramid(bf need to be set to 0).
>
> diff --git a/libavcodec/qsvenc.c b/libavcodec/qsvenc.c
> index 9b71487666..d5e2c5fb32 100644
> --- a/libavcodec/qsvenc.c
> +++ b/libavcodec/qsvenc.c
> @@ -374,6 +374,13 @@ static void dump_video_param(AVCodecContext *avctx, QSVEncContext *q,
> #if QSV_VERSION_ATLEAST(1, 16)
> av_log(avctx, AV_LOG_VERBOSE, "IntRefCycleDist: %"PRId16"\n", co3->IntRefCycleDist);
> #endif
> +#if QSV_VERSION_ATLEAST(1, 23)
> + av_log(avctx, AV_LOG_VERBOSE, "LowDelayBRC: %s\n", print_threestate(co3->LowDelayBRC));
> +#endif
> +#if QSV_VERSION_ATLEAST(1, 19)
> + av_log(avctx, AV_LOG_VERBOSE, "MaxFrameSizeI: %d; ", co3->MaxFrameSizeI);
> + av_log(avctx, AV_LOG_VERBOSE, "MaxFrameSizeP: %d\n", co3->MaxFrameSizeP);
> +#endif
> }
>
> static void dump_video_vp9_param(AVCodecContext *avctx, QSVEncContext *q,
> @@ -988,6 +995,16 @@ static int init_video_param(AVCodecContext *avctx, QSVEncContext *q)
> #if QSV_VERSION_ATLEAST(1, 16)
> if (q->int_ref_cycle_dist >= 0)
> q->extco3.IntRefCycleDist = q->int_ref_cycle_dist;
> +#endif
> +#if QSV_VERSION_ATLEAST(1, 23)
> + if (q->low_delay_brc >= 0)
> + q->extco3.LowDelayBRC = q->low_delay_brc ? MFX_CODINGOPTION_ON : MFX_CODINGOPTION_OFF;
> +#endif
> +#if QSV_VERSION_ATLEAST(1, 19)
> + if (q->max_frame_size_p >= 0)
> + q->extco3.MaxFrameSizeI = q->max_frame_size_i;
> + if (q->max_frame_size_p >= 0)
> + q->extco3.MaxFrameSizeP = q->max_frame_size_p;
> #endif
> }
>
> diff --git a/libavcodec/qsvenc.h b/libavcodec/qsvenc.h
> index 2bda858427..5e07580289 100644
> --- a/libavcodec/qsvenc.h
> +++ b/libavcodec/qsvenc.h
> @@ -91,6 +91,8 @@
> { "veryslow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_TARGETUSAGE_BEST_QUALITY }, INT_MIN, INT_MAX, VE, "preset" }, \
> { "rdo", "Enable rate distortion optimization", OFFSET(qsv.rdo), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 1, VE }, \
> { "max_frame_size", "Maximum encoded frame size in bytes", OFFSET(qsv.max_frame_size), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, UINT16_MAX, VE }, \
> +{ "max_frame_size_i", "Maximum encoded I frame size in bytes",OFFSET(qsv.max_frame_size_i), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, UINT16_MAX, VE }, \
> +{ "max_frame_size_p", "Maximum encoded P frame size in bytes",OFFSET(qsv.max_frame_size_p), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, UINT16_MAX, VE }, \
I can't find the threshold from 'mediasdk-man.md’.
Does `UINT16_MAX` comes from mediasdk, or just a
reasonable limitation set by this wrapper?
> { "max_slice_size", "Maximum encoded slice size in bytes", OFFSET(qsv.max_slice_size), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, UINT16_MAX, VE }, \
> { "bitrate_limit", "Toggle bitrate limitations", OFFSET(qsv.bitrate_limit), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 1, VE }, \
> { "mbbrc", "MB level bitrate control", OFFSET(qsv.mbbrc), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 1, VE }, \
> @@ -102,6 +104,7 @@
> { "forced_idr", "Forcing I frames as IDR frames", OFFSET(qsv.forced_idr), AV_OPT_TYPE_BOOL,{ .i64 = 0 }, 0, 1, VE }, \
> { "low_power", "enable low power mode(experimental: many limitations by mfx version, BRC modes, etc.)", OFFSET(qsv.low_power), AV_OPT_TYPE_BOOL, { .i64 = -1}, -1, 1, VE},\
> { "dblk_idc", "This option disable deblocking. It has value in range 0~2.", OFFSET(qsv.dblk_idc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 2, VE}, \
> +{ "low_delay_brc", "Allow to strictly obey avg frame size", OFFSET(qsv.low_delay_brc), AV_OPT_TYPE_BOOL,{ .i64 = -1 }, -1, 1, VE }, \
>
> extern const AVCodecHWConfigInternal *const ff_qsv_enc_hw_configs[];
>
> @@ -173,6 +176,8 @@ typedef struct QSVEncContext {
> int vcm;
> int rdo;
> int max_frame_size;
> + int max_frame_size_i;
> + int max_frame_size_p;
> int max_slice_size;
> int dblk_idc;
>
> @@ -212,6 +217,7 @@ typedef struct QSVEncContext {
> char *load_plugins;
> SetEncodeCtrlCB *set_encode_ctrl_cb;
> int forced_idr;
> + int low_delay_brc;
> } QSVEncContext;
>
> int ff_qsv_enc_init(AVCodecContext *avctx, QSVEncContext *q);
> --
> 2.25.1
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
More information about the ffmpeg-devel
mailing list