[FFmpeg-devel] avcodec: Add librav1e encoder

Uses the crav1e C bindings for rav1e.

Missing version bump and changelog entry.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
---
Hoping to get some eyes on this, and maybe help some people who want to
test out rav1e without having to use Y4Ms or pipes.

Some points:
    * The C bindings for rav1e currently live in the crav1e repo. This will
      be merged into rav1e's repo Soon(TM), when stuff stabalizes.
    * I have been told that I could ditch the frame queue if I used the 'new'
      send_frame / recieve_packet API. I tried this and ran into odd issues
      like it trying to flush too often (mid stream, etc.). I may have bungled
      it, but it didn't seem to work, and no non-hw encoder uses this API yet.
      Since libaomenc also has this sot of queue sytem, I decided to play it
      safe.
---
 configure              |   4 +
 doc/encoders.texi      |  29 +++
 doc/general.texi       |   7 +
 libavcodec/Makefile    |   1 +
 libavcodec/allcodecs.c |   1 +
 libavcodec/librav1e.c  | 451 +++++++++++++++++++++++++++++++++++++++++
 6 files changed, 493 insertions(+)
 create mode 100644 libavcodec/librav1e.c

On 5/28/2019 2:29 PM, Derek Buitenhuis wrote:
> Uses the crav1e C bindings for rav1e.
> 
> Missing version bump and changelog entry.
> 
> Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
> ---
> Hoping to get some eyes on this, and maybe help some people who want to
> test out rav1e without having to use Y4Ms or pipes.
> 
> Some points:
>     * The C bindings for rav1e currently live in the crav1e repo. This will
>       be merged into rav1e's repo Soon(TM), when stuff stabalizes.
>     * I have been told that I could ditch the frame queue if I used the 'new'
>       send_frame / recieve_packet API. I tried this and ran into odd issues
>       like it trying to flush too often (mid stream, etc.). I may have bungled
>       it, but it didn't seem to work, and no non-hw encoder uses this API yet.
>       Since libaomenc also has this sot of queue sytem, I decided to play it
>       safe.
> ---
>  configure              |   4 +
>  doc/encoders.texi      |  29 +++
>  doc/general.texi       |   7 +
>  libavcodec/Makefile    |   1 +
>  libavcodec/allcodecs.c |   1 +
>  libavcodec/librav1e.c  | 451 +++++++++++++++++++++++++++++++++++++++++
>  6 files changed, 493 insertions(+)
>  create mode 100644 libavcodec/librav1e.c
> 
> diff --git a/configure b/configure
> index 32fc26356c..e61644b012 100755
> --- a/configure
> +++ b/configure
> @@ -254,6 +254,7 @@ External library support:
>    --enable-libopenmpt      enable decoding tracked files via libopenmpt [no]
>    --enable-libopus         enable Opus de/encoding via libopus [no]
>    --enable-libpulse        enable Pulseaudio input via libpulse [no]
> +  --enable-librav1e        enable AV1 encoding via rav1e [no]
>    --enable-librsvg         enable SVG rasterization via librsvg [no]
>    --enable-librubberband   enable rubberband needed for rubberband filter [no]
>    --enable-librtmp         enable RTMP[E] support via librtmp [no]
> @@ -1778,6 +1779,7 @@ EXTERNAL_LIBRARY_LIST="
>      libopenmpt
>      libopus
>      libpulse
> +    librav1e
>      librsvg
>      librtmp
>      libshine
> @@ -3173,6 +3175,7 @@ libopenmpt_demuxer_deps="libopenmpt"
>  libopus_decoder_deps="libopus"
>  libopus_encoder_deps="libopus"
>  libopus_encoder_select="audio_frame_queue"
> +librav1e_encoder_deps="librav1e"
>  librsvg_decoder_deps="librsvg"
>  libshine_encoder_deps="libshine"
>  libshine_encoder_select="audio_frame_queue"
> @@ -6199,6 +6202,7 @@ enabled libopus           && {
>      }
>  }
>  enabled libpulse          && require_pkg_config libpulse libpulse pulse/pulseaudio.h pa_context_new
> +enabled librav1e          && require_pkg_config librav1e rav1e rav1e.h rav1e_context_new
>  enabled librsvg           && require_pkg_config librsvg librsvg-2.0 librsvg-2.0/librsvg/rsvg.h rsvg_handle_render_cairo
>  enabled librtmp           && require_pkg_config librtmp librtmp librtmp/rtmp.h RTMP_Socket
>  enabled librubberband     && require_pkg_config librubberband "rubberband >= 1.8.1" rubberband/rubberband-c.h rubberband_new -lstdc++ && append librubberband_extralibs "-lstdc++"
> diff --git a/doc/encoders.texi b/doc/encoders.texi
> index eefd124751..2ed7053838 100644
> --- a/doc/encoders.texi
> +++ b/doc/encoders.texi
> @@ -1378,6 +1378,35 @@ makes it possible to store non-rgb pix_fmts.
>  
>  @end table
>  
> +@section librav1e
> +
> +rav1e AV1 encoder wrapper.
> +
> +Requires the presence of the rav1e headers and library from crav1e
> +during configuration. You need to explicitly configue the build with
> +@code{--enable-librav1e}.
> +
> +@subsection Options
> +
> +@table @option
> +@item max-quantizer
> +Sets the maximum qauntizer (floor) to use when using bitrate mode.
> +
> +@item quantizer
> +Uses quantizers mode to encode at the given quantizer.
> +
> +@item rav1e-params
> +Set rav1e options using a list of @var{key}=@var{value} couples separated
> +by ":". See @command{rav1e --help} for a list of options.
> +
> +For example to specify librav1e encoding options with @option{-rav1e-params}:
> +
> +@example
> +ffmpeg -i input -c:v librav1e -rav1e-params speed=5:low_latency=true output.mp4
> +@end example
> +
> +@end table
> +
>  @section libaom-av1
>  
>  libaom AV1 encoder wrapper.
> diff --git a/doc/general.texi b/doc/general.texi
> index ec437230e3..6a1cd11a22 100644
> --- a/doc/general.texi
> +++ b/doc/general.texi
> @@ -243,6 +243,13 @@ FFmpeg can use the OpenJPEG libraries for decoding/encoding J2K videos.  Go to
>  instructions.  To enable using OpenJPEG in FFmpeg, pass @code{--enable-libopenjpeg} to
>  @file{./configure}.
>  
> +@section rav1e
> +
> +FFmpeg can make use of rav1e (Rust AV1 Encoder) via its C bindings to encode videos.
> +Go to @url{https://github.com/lu-zero/crav1e/} and @url{https://github.com/xiph/rav1e/}
> +and follow the instructions. To enable using rav1e in FFmpeg, pass @code{--enable-librav1e}
> +to @file{./configure}.
> +
>  @section TwoLAME
>  
>  FFmpeg can make use of the TwoLAME library for MP2 encoding.
> diff --git a/libavcodec/Makefile b/libavcodec/Makefile
> index edccd73037..dc589ce35d 100644
> --- a/libavcodec/Makefile
> +++ b/libavcodec/Makefile
> @@ -988,6 +988,7 @@ OBJS-$(CONFIG_LIBOPUS_DECODER)            += libopusdec.o libopus.o     \
>                                               vorbis_data.o
>  OBJS-$(CONFIG_LIBOPUS_ENCODER)            += libopusenc.o libopus.o     \
>                                               vorbis_data.o
> +OBJS-$(CONFIG_LIBRAV1E_ENCODER)           += librav1e.o
>  OBJS-$(CONFIG_LIBSHINE_ENCODER)           += libshine.o
>  OBJS-$(CONFIG_LIBSPEEX_DECODER)           += libspeexdec.o
>  OBJS-$(CONFIG_LIBSPEEX_ENCODER)           += libspeexenc.o
> diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
> index 6178d31b5c..8a59f39a90 100644
> --- a/libavcodec/allcodecs.c
> +++ b/libavcodec/allcodecs.c
> @@ -702,6 +702,7 @@ extern AVCodec ff_libopenjpeg_encoder;
>  extern AVCodec ff_libopenjpeg_decoder;
>  extern AVCodec ff_libopus_encoder;
>  extern AVCodec ff_libopus_decoder;
> +extern AVCodec ff_librav1e_encoder;
>  extern AVCodec ff_librsvg_decoder;
>  extern AVCodec ff_libshine_encoder;
>  extern AVCodec ff_libspeex_encoder;
> diff --git a/libavcodec/librav1e.c b/libavcodec/librav1e.c
> new file mode 100644
> index 0000000000..30ca7e509e
> --- /dev/null
> +++ b/libavcodec/librav1e.c
> @@ -0,0 +1,451 @@
> +/*
> + * librav1e encoder
> + *
> + * Copyright (c) 2019 Derek Buitenhuis
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +#include <rav1e.h>
> +
> +#include "libavutil/internal.h"
> +#include "libavutil/common.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/pixdesc.h"
> +#include "avcodec.h"
> +#include "internal.h"
> +
> +typedef struct PacketList {
> +    RaPacket *pkt;
> +    struct PacketList *next;
> +} PacketList;
> +
> +typedef struct librav1eContext {
> +    const AVClass *class;
> +
> +    RaContext *ctx;
> +    AVBSFContext *bsf;
> +    PacketList *pktlist;
> +    char *rav1e_opts;
> +    int max_quantizer;
> +    int quantizer;
> +    int done;
> +} librav1eContext;
> +
> +static inline RaPixelRange range_map(enum AVColorRange range)
> +
> +{
> +    switch (range) {
> +    case AVCOL_RANGE_MPEG:
> +        return RA_PIXEL_RANGE_LIMITED;
> +    case AVCOL_RANGE_JPEG:
> +        return RA_PIXEL_RANGE_FULL;
> +    default:
> +        return RA_PIXEL_RANGE_UNSPECIFIED;
> +    }
> +}
> +
> +static inline RaChromaSampling pix_fmt_map(enum AVPixelFormat pix_fmt)
> +{
> +    switch (pix_fmt) {
> +    case AV_PIX_FMT_YUV420P:
> +    case AV_PIX_FMT_YUV420P10:
> +    case AV_PIX_FMT_YUV420P12:
> +        return RA_CHROMA_SAMPLING_CS420;
> +    case AV_PIX_FMT_YUV422P:
> +    case AV_PIX_FMT_YUV422P10:
> +    case AV_PIX_FMT_YUV422P12:
> +        return RA_CHROMA_SAMPLING_CS422;
> +    case AV_PIX_FMT_YUV444P:
> +    case AV_PIX_FMT_YUV444P10:
> +    case AV_PIX_FMT_YUV444P12:
> +        return RA_CHROMA_SAMPLING_CS444;
> +    case AV_PIX_FMT_GRAY8:
> +    case AV_PIX_FMT_GRAY10:
> +    case AV_PIX_FMT_GRAY12:
> +        return RA_CHROMA_SAMPLING_CS400;
> +    default:
> +        // This should be impossible
> +        return (RaChromaSampling) -1;

If it's not meant to happen, then it should probably be an av_assert0.

> +    }
> +}
> +
> +static inline RaChromaSamplePosition chroma_loc_map(enum AVChromaLocation chroma_loc)
> +{
> +    switch (chroma_loc) {
> +    case AVCHROMA_LOC_LEFT:
> +        return RA_CHROMA_SAMPLE_POSITION_VERTICAL;
> +    case AVCHROMA_LOC_TOPLEFT:
> +        return RA_CHROMA_SAMPLE_POSITION_COLOCATED;
> +    default:
> +        return RA_CHROMA_SAMPLE_POSITION_UNKNOWN;
> +    }
> +}
> +
> +static int add_packet(PacketList **list, RaPacket *pkt)
> +{
> +    PacketList *cur = *list;
> +    PacketList *newentry = av_mallocz(sizeof(PacketList));
> +    if (!newentry)
> +        return AVERROR(ENOMEM);
> +
> +    newentry->pkt = pkt;
> +
> +    if (!cur) {
> +        *list = newentry;
> +        return 0;
> +    }
> +
> +    /*
> +     * Just use a simple linear search, since the reoroder buffer in
> +     * AV1 is capped to something fairly low.
> +     */
> +    while (cur->next)
> +        cur = cur->next;
> +
> +    cur->next = newentry;
> +
> +    return 0;
> +}
> +
> +static RaPacket *get_packet(PacketList **list)
> +{
> +    PacketList *head = *list;
> +    RaPacket *ret;
> +
> +    if (!head)
> +        return NULL;
> +
> +    ret = head->pkt;
> +
> +    *list = head->next;
> +    av_free(head);
> +
> +    return ret;
> +}
> +
> +static av_cold int librav1e_encode_close(AVCodecContext *avctx)
> +{
> +    librav1eContext *ctx = avctx->priv_data;
> +
> +    if (ctx->ctx) {
> +        rav1e_context_unref(ctx->ctx);
> +        ctx->ctx = NULL;
> +    }
> +
> +    while (ctx->pktlist) {
> +        RaPacket *rpkt = get_packet(&ctx->pktlist);
> +        rav1e_packet_unref(rpkt);
> +    }
> +
> +    av_bsf_free(&ctx->bsf);
> +
> +    return 0;
> +}
> +
> +static av_cold int librav1e_encode_init(AVCodecContext *avctx)
> +{
> +    librav1eContext *ctx = avctx->priv_data;
> +    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
> +    RaConfig *cfg = NULL;
> +    int rret;
> +    int ret = 0;

Why two ret variables?

> +
> +    cfg = rav1e_config_default();
> +    if (!cfg) {
> +        av_log(avctx, AV_LOG_ERROR, "Could not allocate rav1e config.\n");
> +        ret = AVERROR_EXTERNAL;
> +        goto end;
> +    }
> +
> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
> +         const AVBitStreamFilter *filter = av_bsf_get_by_name("extract_extradata");

See
https://github.com/dwbuiten/FFmpeg/commit/e146749232928a1cebe637338189938644237894#r33622810

You don't need to extract the Sequence Header from encoded frames since
rav1e provides a function for this purpose.
So does libaom, for that matter, but it's buggy and that's why it's not
being used.

> +         int bret;
> +
> +         if (!filter) {
> +            av_log(avctx, AV_LOG_ERROR, "extract_extradata bitstream filter "
> +                   "not found. This is a bug, please report it.\n");
> +            ret = AVERROR_BUG;
> +            goto end;
> +         }
> +
> +         bret = av_bsf_alloc(filter, &ctx->bsf);
> +         if (bret < 0) {
> +             ret = bret;
> +             goto end;
> +         }
> +
> +         bret = avcodec_parameters_from_context(ctx->bsf->par_in, avctx);
> +         if (bret < 0) {
> +             ret = bret;
> +             goto end;
> +         }
> +
> +         bret = av_bsf_init(ctx->bsf);
> +         if (bret < 0) {
> +             ret = bret;
> +             goto end;
> +         }
> +    }
> +
> +    if (ctx->rav1e_opts) {
> +        AVDictionary *dict    = NULL;
> +        AVDictionaryEntry *en = NULL;
> +
> +        if (!av_dict_parse_string(&dict, ctx->rav1e_opts, "=", ":", 0)) {
> +            while (en = av_dict_get(dict, "", en, AV_DICT_IGNORE_SUFFIX)) {
> +                int parse_ret = rav1e_config_parse(cfg, en->key, en->value);
> +                if (parse_ret < 0)
> +                    av_log(avctx, AV_LOG_WARNING, "Invalid value for %s: %s.\n", en->key, en->value);
> +            }
> +            av_dict_free(&dict);
> +        }
> +    }
> +
> +    rret = rav1e_config_parse_int(cfg, "width", avctx->width);
> +    if (rret < 0) {
> +        av_log(avctx, AV_LOG_ERROR, "Invalid width passed to rav1e.\n");
> +        ret = AVERROR_INVALIDDATA;
> +        goto end;
> +    }
> +
> +    rret = rav1e_config_parse_int(cfg, "height", avctx->height);
> +    if (rret < 0) {
> +        av_log(avctx, AV_LOG_ERROR, "Invalid width passed to rav1e.\n");
> +        ret = AVERROR_INVALIDDATA;
> +        goto end;
> +    }
> +
> +    rret = rav1e_config_parse_int(cfg, "threads", avctx->thread_count);
> +    if (rret < 0)
> +        av_log(avctx, AV_LOG_WARNING, "Invalid number of threads, defaulting to auto.\n");
> +
> +    if (avctx->bit_rate && ctx->quantizer < 0) {
> +        rret = rav1e_config_parse_int(cfg, "quantizer", ctx->max_quantizer);
> +        if (rret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "Could not set max quantizer.\n");
> +            ret = AVERROR_EXTERNAL;
> +            goto end;
> +        }
> +        rret = rav1e_config_parse_int(cfg, "bitrate", avctx->bit_rate);
> +        if (rret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "Could not set bitrate.\n");
> +            ret = AVERROR_INVALIDDATA;
> +            goto end;
> +        }
> +    } else if (ctx->quantizer >= 0) {
> +        rret = rav1e_config_parse_int(cfg, "quantizer", ctx->quantizer);
> +        if (rret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "Could not set quantizer.\n");
> +            ret = AVERROR_EXTERNAL;
> +            goto end;
> +        }
> +    }
> +
> +    rret = rav1e_config_set_pixel_format(cfg, desc->comp[0].depth,
> +                                         pix_fmt_map(avctx->pix_fmt),
> +                                         chroma_loc_map(avctx->chroma_sample_location),
> +                                         range_map(avctx->color_range));
> +    if (rret < 0) {
> +        av_log(avctx, AV_LOG_ERROR, "Failed to set pixel format properties.\n");
> +        ret = AVERROR_INVALIDDATA;
> +        goto end;
> +    }
> +
> +    /* rav1e's colorspace enums match standard values. */
> +    rret = rav1e_config_set_color_description(cfg, (RaMatrixCoefficients) avctx->colorspace,
> +                                              (RaColorPrimaries) avctx->color_primaries,
> +                                              (RaTransferCharacteristics) avctx->color_trc);
> +    if (rret < 0) {
> +        av_log(avctx, AV_LOG_WARNING, "Failed to set color properties.\n");
> +        if (avctx->err_recognition & AV_EF_EXPLODE) {
> +            ret = AVERROR_INVALIDDATA;
> +            goto end;
> +        }
> +    }
> +
> +    ctx->ctx = rav1e_context_new(cfg);
> +    if (!ctx->ctx) {
> +        av_log(avctx, AV_LOG_ERROR, "Failed to create rav1e encode context.\n");
> +        ret = AVERROR_EXTERNAL;
> +        goto end;
> +    }
> +
> +    ret = 0;
> +
> +end:
> +    if (cfg)
> +        rav1e_config_unref(cfg);
> +
> +    if (ret)
> +        librav1e_encode_close(avctx);
> +
> +    return ret;
> +}
> +
> +static int librav1e_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
> +                                 const AVFrame *pic, int *got_packet)
> +{
> +    librav1eContext *ctx = avctx->priv_data;
> +    RaPacket *rpkt = NULL;
> +    RaFrame *rframe = NULL;
> +    RaEncoderStatus ret;
> +    int pret;
> +
> +    if (pic) {
> +        const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pic->format);
> +
> +        rframe = rav1e_frame_new(ctx->ctx);
> +        if (!rframe) {
> +            av_log(avctx, AV_LOG_ERROR, "Could not allocate new rav1e frame.\n");
> +            return AVERROR(ENOMEM);
> +        }
> +
> +        for (int i = 0; i < 3; i++) {
> +            int shift = i ? desc->log2_chroma_h : 0;
> +            int bytes = desc->comp[0].depth == 8 ? 1 : 2;
> +            rav1e_frame_fill_plane(rframe, i, pic->data[i],
> +                                   (pic->height >> shift) * pic->linesize[i],
> +                                   pic->linesize[i], bytes);
> +        }
> +    }
> +
> +    ret = rav1e_send_frame(ctx->ctx, rframe);
> +    if (rframe)
> +         rav1e_frame_unref(rframe); /* No need to unref if flushing. */
> +    if (ret < 0) {
> +        av_log(avctx, AV_LOG_ERROR, "Could not send frame.\n");
> +        return AVERROR_EXTERNAL;
> +    } else if (ret == RA_ENCODER_STATUS_ENOUGH_DATA) {
> +        av_log(avctx, AV_LOG_WARNING, "rav1e encode queue is full. Frames may be dropped.\n");
> +    }
> +
> +    while (!ctx->done) {
> +        ret = rav1e_receive_packet(ctx->ctx, &rpkt);
> +        if (ret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "Could not encode frame.\n");
> +            return AVERROR_EXTERNAL;
> +        } else if (ret == RA_ENCODER_STATUS_NEED_MORE_DATA || ret == RA_ENCODER_STATUS_ENCODED) {
> +            break;
> +        } else if (ret == RA_ENCODER_STATUS_LIMIT_REACHED) {
> +            /* We're done. Nothing else to flush, so stop tryng. */
> +            ctx->done = 1;
> +            break;
> +        } else if (ret == RA_ENCODER_STATUS_SUCCESS) {
> +            /*
> +             * Since we must drain the encoder of packets when we finally successfully
> +             * receive one, add it to a packet queue and output as need be. Since this
> +             * is only due to the frame "reordering" (alt-ref) internal to the encoder,
> +             * it should never get very big before all being output. This is is similar
> +             * to what is done in libaomenc.c.
> +             */
> +            int aret = add_packet(&ctx->pktlist, rpkt);
> +            if (aret < 0) {
> +                rav1e_packet_unref(rpkt);
> +                return aret;
> +            }
> +            rpkt = NULL;
> +        } else {
> +            av_log(avctx, AV_LOG_ERROR, "Unknown return code from ra1ve_receive_packet.\n");
> +            return AVERROR_UNKNOWN;
> +        }
> +    }
> +
> +    rpkt = get_packet(&ctx->pktlist);
> +    if (!rpkt)
> +        return 0;
> +

> +    pret = ff_alloc_packet2(avctx, pkt, rpkt->len, rpkt->len);
> +    if (pret < 0) {
> +        rav1e_packet_unref(rpkt);
> +        av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
> +        return ret;
> +    }
> +
> +    memcpy(pkt->data, rpkt->data, rpkt->len);
> +
> +    if (rpkt->frame_type == RA_FRAME_TYPE_KEY)
> +        pkt->flags |= AV_PKT_FLAG_KEY;
> +
> +    pkt->pts = pkt->dts = rpkt->number * avctx->ticks_per_frame;
> +
> +    rav1e_packet_unref(rpkt);

You can avoid the data copy if you wrap the RaPacket into the
AVBufferRef used in the AVPacket, but for this you need to use the
send/receive API, given the encode2 one allows the caller to use their
own allocated packet.

> +
> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
> +        pret = av_bsf_send_packet(ctx->bsf, pkt);
> +        if (pret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "extradata extraction send failed.\n");
> +            return pret;
> +        }
> +
> +        pret = av_bsf_receive_packet(ctx->bsf, pkt);
> +        if (pret < 0) {
> +            av_log(avctx, AV_LOG_ERROR, "extradata extraction receive failed.\n");
> +            return pret;
> +        }
> +    }
> +
> +    *got_packet = 1;
> +
> +    return 0;
> +}
> +
> +#define OFFSET(x) offsetof(librav1eContext, x)
> +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
> +
> +static const AVOption options[] = {
> +    { "quantizer", "use constant quantizer mode", OFFSET(quantizer), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, VE },
> +    { "max-quantizer", "max quantizer when using bitrate mode", OFFSET(max_quantizer), AV_OPT_TYPE_INT, { .i64 = 255 }, 1, 255, VE },

This should be mapped to qmax lavc option instead.

> +    { "rav1e-params", "set the rav1e configuration using a :-separated list of key=value parameters", OFFSET(rav1e_opts), AV_OPT_TYPE_STRING, { 0 }, 0, 0, VE },
> +    { NULL }
> +};
> +
> +static const AVClass class = {
> +    .class_name = "librav1e",
> +    .item_name  = av_default_item_name,
> +    .option     = options,
> +    .version    = LIBAVUTIL_VERSION_INT,
> +};
> +
> +AVCodec ff_librav1e_encoder = {
> +    .name           = "librav1e",
> +    .long_name      = NULL_IF_CONFIG_SMALL("librav1e AV1"),
> +    .type           = AVMEDIA_TYPE_VIDEO,
> +    .id             = AV_CODEC_ID_AV1,
> +    .init           = librav1e_encode_init,
> +    .encode2        = librav1e_encode_frame,
> +    .close          = librav1e_encode_close,
> +    .priv_data_size = sizeof(librav1eContext),
> +    .priv_class     = &class,
> +    .pix_fmts       = (const enum AVPixelFormat[]) {
> +        AV_PIX_FMT_YUV420P,
> +        AV_PIX_FMT_YUV420P10,
> +        AV_PIX_FMT_YUV420P12,
> +        AV_PIX_FMT_YUV422P,
> +        AV_PIX_FMT_YUV422P10,
> +        AV_PIX_FMT_YUV422P12,
> +        AV_PIX_FMT_YUV444P,
> +        AV_PIX_FMT_YUV444P10,
> +        AV_PIX_FMT_YUV444P12,
> +        AV_PIX_FMT_GRAY8,
> +        AV_PIX_FMT_GRAY10,
> +        AV_PIX_FMT_GRAY12,
> +        AV_PIX_FMT_NONE
> +    },
> +    .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AUTO_THREADS,
> +    .wrapper_name   = "librav1e",
> +};
>

On 28/05/2019 20:32, James Almer wrote:
>> +    default:
>> +        // This should be impossible
>> +        return (RaChromaSampling) -1;
> 
> If it's not meant to happen, then it should probably be an av_assert0.

Yeah, that makes more sense. Will change.

>> +    int rret;
>> +    int ret = 0;
> 
> Why two ret variables?

Will fix.

>> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
>> +         const AVBitStreamFilter *filter = av_bsf_get_by_name("extract_extradata");
> 
> See
> https://github.com/dwbuiten/FFmpeg/commit/e146749232928a1cebe637338189938644237894#r33622810
> 
> You don't need to extract the Sequence Header from encoded frames since
> rav1e provides a function for this purpose.
> So does libaom, for that matter, but it's buggy and that's why it's not
> being used.

As discussed on IRC, that function does not return the config OBUs at the end of
the ISOBMFF-style data, so this way is still preferred until it does.

I'll switch over if/once rav1e's API does return that.


>> +    memcpy(pkt->data, rpkt->data, rpkt->len);
>> +
>> +    if (rpkt->frame_type == RA_FRAME_TYPE_KEY)
>> +        pkt->flags |= AV_PKT_FLAG_KEY;
>> +
>> +    pkt->pts = pkt->dts = rpkt->number * avctx->ticks_per_frame;
>> +
>> +    rav1e_packet_unref(rpkt);
> 
> You can avoid the data copy if you wrap the RaPacket into the
> AVBufferRef used in the AVPacket, but for this you need to use the
> send/receive API, given the encode2 one allows the caller to use their
> own allocated packet.

As discussed on IRC / noted in the top of this email, I'll see how
you fare switching to send/recieve. :)

>> +static const AVOption options[] = {
>> +    { "quantizer", "use constant quantizer mode", OFFSET(quantizer), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, VE },
>> +    { "max-quantizer", "max quantizer when using bitrate mode", OFFSET(max_quantizer), AV_OPT_TYPE_INT, { .i64 = 255 }, 1, 255, VE },
> 
> This should be mapped to qmax lavc option instead.

OK. Is there a matching option for plain old quantizer?
I couldn't find an obvious one.

- Derk

On 5/28/2019 4:49 PM, Derek Buitenhuis wrote:
> On 28/05/2019 20:32, James Almer wrote:
>>> +    default:
>>> +        // This should be impossible
>>> +        return (RaChromaSampling) -1;
>>
>> If it's not meant to happen, then it should probably be an av_assert0.
> 
> Yeah, that makes more sense. Will change.
> 
>>> +    int rret;
>>> +    int ret = 0;
>>
>> Why two ret variables?
> 
> Will fix.
> 
>>> +    if (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
>>> +         const AVBitStreamFilter *filter = av_bsf_get_by_name("extract_extradata");
>>
>> See
>> https://github.com/dwbuiten/FFmpeg/commit/e146749232928a1cebe637338189938644237894#r33622810
>>
>> You don't need to extract the Sequence Header from encoded frames since
>> rav1e provides a function for this purpose.
>> So does libaom, for that matter, but it's buggy and that's why it's not
>> being used.
> 
> As discussed on IRC, that function does not return the config OBUs at the end of
> the ISOBMFF-style data, so this way is still preferred until it does.
> 
> I'll switch over if/once rav1e's API does return that.
> 
> 
>>> +    memcpy(pkt->data, rpkt->data, rpkt->len);
>>> +
>>> +    if (rpkt->frame_type == RA_FRAME_TYPE_KEY)
>>> +        pkt->flags |= AV_PKT_FLAG_KEY;
>>> +
>>> +    pkt->pts = pkt->dts = rpkt->number * avctx->ticks_per_frame;
>>> +
>>> +    rav1e_packet_unref(rpkt);
>>
>> You can avoid the data copy if you wrap the RaPacket into the
>> AVBufferRef used in the AVPacket, but for this you need to use the
>> send/receive API, given the encode2 one allows the caller to use their
>> own allocated packet.
> 
> As discussed on IRC / noted in the top of this email, I'll see how
> you fare switching to send/recieve. :)
> 
>>> +static const AVOption options[] = {
>>> +    { "quantizer", "use constant quantizer mode", OFFSET(quantizer), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, VE },
>>> +    { "max-quantizer", "max quantizer when using bitrate mode", OFFSET(max_quantizer), AV_OPT_TYPE_INT, { .i64 = 255 }, 1, 255, VE },
>>
>> This should be mapped to qmax lavc option instead.
> 
> OK. Is there a matching option for plain old quantizer?
> I couldn't find an obvious one.

I think x26* and vpx/aom call it crf? It's not in option_tables.h in any
case.

> 
> - Derk
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>

On 28/05/2019 20:49, Derek Buitenhuis wrote:
>>> +static const AVOption options[] = {
>>> +    { "quantizer", "use constant quantizer mode", OFFSET(quantizer), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, VE },
>>> +    { "max-quantizer", "max quantizer when using bitrate mode", OFFSET(max_quantizer), AV_OPT_TYPE_INT, { .i64 = 255 }, 1, 255, VE },
>> This should be mapped to qmax lavc option instead.
> OK. Is there a matching option for plain old quantizer?
> I couldn't find an obvious one.

Looking closer at this, it does not seem correct to use it.

As, per `ffmpeg -h full`:

      -qmax              <int>        E..V..... maximum video quantizer scale (VBR) (from -1 to 1024) (default 31)

The default of 31 is really bad for AV1. Further, rav1e's CLI default of 255
is probably the safest to use here.

- Derek

On 28/05/2019 20:58, James Almer wrote:
> I think x26* and vpx/aom call it crf? It's not in option_tables.h in any
> case.

They do not. This is a constant quantizer mode, not constant rate factor.

- Derek

On 2019-05-28 22:00, Derek Buitenhuis wrote:
> On 28/05/2019 20:58, James Almer wrote:
>> I think x26* and vpx/aom call it crf? It's not in option_tables.h in any
>> case.
> 
> They do not. This is a constant quantizer mode, not constant rate factor.

IIRC either qp or cqp

Am Di., 28. Mai 2019 um 19:29 Uhr schrieb Derek Buitenhuis
<derek.buitenhuis@gmail.com>:

>     * The C bindings for rav1e currently live in the crav1e repo. This will
>       be merged into rav1e's repo Soon(TM), when stuff stabalizes.

So is your patch meant to wait until this merge is done?
I would suggest so...

Carl Eugen

On 31/05/2019 23:08, Carl Eugen Hoyos wrote:
> So is your patch meant to wait until this merge is done?
> I would suggest so...

Soon has a (TM) beside it for a reason (there needs to be some
work done on Rust itself first, I think.) I would expect it to
take at least 1-2 months. The API itself won't change.

While I prefer to merge it sooner rather than later to aid in
peoples' use of rav1e, I'm open to others opinions here.

- Derek

Am So., 2. Juni 2019 um 19:54 Uhr schrieb Derek Buitenhuis
<derek.buitenhuis@gmail.com>:
>
> On 31/05/2019 23:08, Carl Eugen Hoyos wrote:
> > So is your patch meant to wait until this merge is done?
> > I would suggest so...
>
> Soon has a (TM) beside it for a reason (there needs to be some
> work done on Rust itself first, I think.) I would expect it to
> take at least 1-2 months.

Since a lot of unexpected changes can happen after months,
this sounds to me like a strong reason to wait.

Carl Eugen

On 6/2/19, Carl Eugen Hoyos <ceffmpeg@gmail.com> wrote:
> Am So., 2. Juni 2019 um 19:54 Uhr schrieb Derek Buitenhuis
> <derek.buitenhuis@gmail.com>:
>>
>> On 31/05/2019 23:08, Carl Eugen Hoyos wrote:
>> > So is your patch meant to wait until this merge is done?
>> > I would suggest so...
>>
>> Soon has a (TM) beside it for a reason (there needs to be some
>> work done on Rust itself first, I think.) I would expect it to
>> take at least 1-2 months.
>
> Since a lot of unexpected changes can happen after months,
> this sounds to me like a strong reason to wait.
>

 Why, could you elaborate?

On Sun, Jun 02, 2019 at 18:46:17 +0100, Derek Buitenhuis wrote:
> take at least 1-2 months. The API itself won't change.

https://github.com/lu-zero/crav1e says:

  Status
    The API is far from stable

Just sayin',
Moritz

On 02/06/2019 19:45, Moritz Barsnick wrote:
> https://github.com/lu-zero/crav1e says:
> 
>   Status
>     The API is far from stable

Yeah, that's not super true any more, and I'll remove it.

- Derek

On 02/06/2019 19:27, Carl Eugen Hoyos wrote:
> Since a lot of unexpected changes can happen after months,
> this sounds to me like a strong reason to wait.

This has nothing to do with API changes, it's a Rust toolchain (build
system, specifically) thing. These changes don't change API/ABI.

- Derek

On 02/06/2019 21:19, Derek Buitenhuis wrote:
> This has nothing to do with API changes, it's a Rust toolchain (build
> system, specifically) thing. These changes don't change API/ABI.

Also to add: We will probably be cutting a 'real' rav1e release soon (0.1.0),
I think. That could also be a place to settle on for merging the patch. I'll
bring it up in the weekly Xiph/Daala/rav1e meeting on Tuesday.

- Derek