[FFmpeg-devel] [PATCH V3 2/2] avcodec: add av1 VAAPI decoder

Mark Thompson sw at jkqxz.net
Tue Sep 29 20:01:48 EEST 2020


On 09/09/2020 04:39, Fei Wang wrote:
> Example cmdline:
> ffmpeg -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -v verbose \
> -c:v av1 -i input.ivf -pix_fmt yuv420p -vsync passthrough -f md5     \
> -y out.md5
> 
> Signed-off-by: Fei Wang <fei.w.wang at intel.com>
> ---
>   Changelog                 |   1 +
>   configure                 |   3 +
>   libavcodec/Makefile       |   1 +
>   libavcodec/av1dec.c       |  23 ++-
>   libavcodec/hwaccels.h     |   1 +
>   libavcodec/vaapi_av1.c    | 285 ++++++++++++++++++++++++++++++++++++++
>   libavcodec/vaapi_decode.c |   8 ++
>   libavcodec/version.h      |   2 +-
>   8 files changed, 322 insertions(+), 2 deletions(-)
>   create mode 100644 libavcodec/vaapi_av1.c
> 
> diff --git a/Changelog b/Changelog
> index a89010d7e1..fb1fd55571 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -22,6 +22,7 @@ version <next>:
>   - MODS demuxer
>   - PhotoCD decoder
>   - AV1 decoder (Hardware acceleration used only)
> +- AV1 VAAPI decoder
>   
>   
>   version 4.3:
> diff --git a/configure b/configure
> index 5893ae3473..81f3c80e8e 100755
> --- a/configure
> +++ b/configure
> @@ -2917,6 +2917,8 @@ videotoolbox_hwaccel_deps="videotoolbox pthreads"
>   videotoolbox_hwaccel_extralibs="-framework QuartzCore"
>   xvmc_deps="X11_extensions_XvMClib_h"
>   
> +av1_vaapi_hwaccel_deps="vaapi_1_8"
> +av1_vaapi_hwaccel_select="av1_decoder"
>   h263_vaapi_hwaccel_deps="vaapi"
>   h263_vaapi_hwaccel_select="h263_decoder"
>   h263_videotoolbox_hwaccel_deps="videotoolbox"
> @@ -6680,6 +6682,7 @@ if enabled vaapi; then
>       fi
>   
>       check_cpp_condition vaapi_1 "va/va.h" "VA_CHECK_VERSION(1, 0, 0)"
> +    check_cpp_condition vaapi_1_8 "va/va.h" "VA_CHECK_VERSION(1, 8, 0)"

Add a check for the AV1 structure needed like the other codecs below rather than doing this.  The VAAPI_1 version macro exists for the API incompatibilities between VAAPI versions 0 and 1.

>   
>       check_type "va/va.h va/va_dec_hevc.h" "VAPictureParameterBufferHEVC"
>       check_struct "va/va.h" "VADecPictureParameterBufferVP9" bit_depth
> diff --git a/libavcodec/Makefile b/libavcodec/Makefile
> index 5729cdd8e7..4d54ddc684 100644
> --- a/libavcodec/Makefile
> +++ b/libavcodec/Makefile
> @@ -954,6 +954,7 @@ OBJS-$(CONFIG_VP9_NVDEC_HWACCEL)          += nvdec_vp9.o
>   OBJS-$(CONFIG_VP9_VAAPI_HWACCEL)          += vaapi_vp9.o
>   OBJS-$(CONFIG_VP9_VDPAU_HWACCEL)          += vdpau_vp9.o
>   OBJS-$(CONFIG_VP8_QSV_HWACCEL)            += qsvdec_other.o
> +OBJS-$(CONFIG_AV1_VAAPI_HWACCEL)          += vaapi_av1.o

This should be in alphabetical order, so put it at the top of the hwaccel section.

>   
>   # libavformat dependencies
>   OBJS-$(CONFIG_ISO_MEDIA)               += mpeg4audio.o mpegaudiodata.o
> diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c
> index eeeec2ca32..4098cbb655 100644
> --- a/libavcodec/av1dec.c
> +++ b/libavcodec/av1dec.c
> @@ -262,7 +262,7 @@ static int get_pixel_format(AVCodecContext *avctx)
>       uint8_t bit_depth;
>       int ret;
>       enum AVPixelFormat pix_fmt = AV_PIX_FMT_NONE;
> -#define HWACCEL_MAX (0)
> +#define HWACCEL_MAX (CONFIG_AV1_VAAPI_HWACCEL)
>       enum AVPixelFormat pix_fmts[HWACCEL_MAX + 1], *fmtp = pix_fmts;
>   
>       if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
> @@ -323,6 +323,24 @@ static int get_pixel_format(AVCodecContext *avctx)
>           return -1;
>       s->pix_fmt = pix_fmt;
>   
> +    switch (s->pix_fmt) {
> +    case AV_PIX_FMT_YUV420P:
> +#if CONFIG_AV1_VAAPI_HWACCEL
> +        *fmtp++ = AV_PIX_FMT_VAAPI;
> +#endif
> +        break;
> +    case AV_PIX_FMT_YUV420P10:
> +#if CONFIG_AV1_VAAPI_HWACCEL
> +        *fmtp++ = AV_PIX_FMT_VAAPI;
> +#endif
> +        break;
> +    case AV_PIX_FMT_YUV420P12:
> +#if CONFIG_AV1_VAAPI_HWACCEL
> +        *fmtp++ = AV_PIX_FMT_VAAPI;
> +#endif
> +        break;
> +    }

Are you sure this combination is right?  It doesn't match any profile (4:2:0 8- and 10-bit are profile 0, but 12-bit requires profile 2), so checking for support will be confusing.

> +
>       *fmtp = AV_PIX_FMT_NONE;
>       avctx->sw_pix_fmt = s->pix_fmt;
>       ret = ff_thread_get_format(avctx, pix_fmts);
> @@ -855,6 +873,9 @@ AVCodec ff_av1_decoder = {
>       .flush                 = av1_decode_flush,
>       .profiles              = NULL_IF_CONFIG_SMALL(ff_av1_profiles),
>       .hw_configs            = (const AVCodecHWConfigInternal * []) {
> +#if CONFIG_AV1_VAAPI_HWACCEL
> +        HWACCEL_VAAPI(av1),
> +#endif
>           NULL
>       },
>   };
> diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h
> index 6109c89bd6..18e9079c55 100644
> --- a/libavcodec/hwaccels.h
> +++ b/libavcodec/hwaccels.h
> @@ -21,6 +21,7 @@
>   
>   #include "avcodec.h"
>   
> +extern const AVHWAccel ff_av1_vaapi_hwaccel;
>   extern const AVHWAccel ff_h263_vaapi_hwaccel;
>   extern const AVHWAccel ff_h263_videotoolbox_hwaccel;
>   extern const AVHWAccel ff_h264_d3d11va_hwaccel;
> diff --git a/libavcodec/vaapi_av1.c b/libavcodec/vaapi_av1.c
> new file mode 100644
> index 0000000000..c5aeaf8b21
> --- /dev/null
> +++ b/libavcodec/vaapi_av1.c
> @@ -0,0 +1,285 @@
> +/*
> + * AV1 HW decode acceleration through VA API
> + *
> + * 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 "libavutil/pixdesc.h"
> +#include "hwconfig.h"
> +#include "vaapi_decode.h"
> +#include "av1dec.h"
> +
> +static VASurfaceID vaapi_av1_surface_id(AV1Frame *vf)
> +{
> +    if (vf)
> +        return ff_vaapi_get_surface_id(vf->tf.f);
> +    else
> +        return VA_INVALID_SURFACE;
> +}
> +
> +static VAAV1TransformationType vaapi_av1_get_gm_type(uint8_t type)
> +{
> +    if (type == AV1_WARP_MODEL_IDENTITY)
> +       return VAAV1TransformationIdentity;
> +    else if (type == AV1_WARP_MODEL_TRANSLATION)
> +       return VAAV1TransformationTranslation;
> +    else if (type == AV1_WARP_MODEL_ROTZOOM)
> +       return VAAV1TransformationRotzoom;
> +    else if (type == AV1_WARP_MODEL_AFFINE)
> +       return VAAV1TransformationAffine;
> +    else
> +       return VAAV1TransformationIdentity;
> +}

Does this identity function really need to exist?

> +
> +static uint8_t vaapi_av1_get_bit_depth_idx(AVCodecContext *avctx)
> +{
> +    AV1DecContext *s = avctx->priv_data;
> +    const AV1RawSequenceHeader *seq = s->raw_seq;
> +    uint8_t bit_depth = 8;
> +
> +    if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
> +        bit_depth = seq->color_config.twelve_bit ? 12 : 10;
> +    else if (seq->seq_profile <= 2)
> +        bit_depth = seq->color_config.high_bitdepth ? 10 : 8;
> +    else {
> +        av_log(avctx, AV_LOG_ERROR,
> +               "Couldn't get bit depth from profile:%d.\n", seq->seq_profile);

I think you want to eliminate this case earlier to avoid it.  If you do get here then you can't continue usefully, so this would need to return an error.

> +    }
> +    return bit_depth == 8 ? 0 : bit_depth == 10 ? 1 : 2;
> +}
> +
> +static int vaapi_av1_start_frame(AVCodecContext *avctx,
> +                                 av_unused const uint8_t *buffer,
> +                                 av_unused uint32_t size)
> +{
> +    AV1DecContext *s = avctx->priv_data;
> +    const AV1RawSequenceHeader *seq = s->raw_seq;
> +    const AV1RawFrameHeader *frame_header = s->raw_frame_header;
> +    VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
> +    VADecPictureParameterBufferAV1 pic_param;
> +    int err = 0;
> +    uint8_t lr_type[4] = {AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ};

This is the remap_lr_type[] table from the standard?  Maybe use that name.

> +
> +    pic->output_surface = vaapi_av1_surface_id(&s->cur_frame);
> +
> +    memset(&pic_param, 0, sizeof(VADecPictureParameterBufferAV1));
> +    pic_param = (VADecPictureParameterBufferAV1) {
> +        .profile = seq->seq_profile,
> +        .order_hint_bits_minus_1 = seq->order_hint_bits_minus_1,
> +        .bit_depth_idx = vaapi_av1_get_bit_depth_idx(avctx),
> +        .seq_info_fields.fields = {
> +            .still_picture = seq->still_picture,
> +            .use_128x128_superblock = seq->use_128x128_superblock,
> +            .enable_filter_intra = seq->enable_filter_intra,
> +            .enable_intra_edge_filter = seq->enable_intra_edge_filter,
> +            .enable_interintra_compound = seq->enable_interintra_compound,
> +            .enable_masked_compound = seq->enable_masked_compound,
> +            .enable_dual_filter = seq->enable_dual_filter,
> +            .enable_order_hint = seq->enable_order_hint,
> +            .enable_jnt_comp = seq->enable_jnt_comp,
> +            .enable_cdef = seq->enable_cdef,
> +            .mono_chrome = seq->color_config.mono_chrome,
> +            .color_range = seq->color_config.color_range,
> +            .subsampling_x = seq->color_config.subsampling_x,
> +            .subsampling_y = seq->color_config.subsampling_y,
> +            .chroma_sample_position = seq->color_config.chroma_sample_position,
> +            .film_grain_params_present  = seq->film_grain_params_present,

These tables would be easier to read with some horizontal alignment at the '=' (see other vaapi_codec.c files).

> +        },
> +        .current_frame = pic->output_surface,
> +        .current_display_picture = pic->output_surface,
> +        .frame_width_minus1 = frame_header->frame_width_minus_1,
> +        .frame_height_minus1 = frame_header->frame_height_minus_1,
> +        .primary_ref_frame = frame_header->primary_ref_frame,
> +        .order_hint = frame_header->order_hint,
> +        .seg_info.segment_info_fields.bits = {
> +            .enabled = frame_header->segmentation_enabled,
> +            .update_map = frame_header->segmentation_update_map,
> +            .temporal_update = frame_header->segmentation_temporal_update,
> +            .update_data = frame_header->segmentation_update_data,
> +        },
> +        .film_grain_info.film_grain_info_fields.bits = {
> +            .apply_grain = frame_header->apply_grain,
> +            .chroma_scaling_from_luma = frame_header->chroma_scaling_from_luma,
> +            .grain_scaling_minus_8 = frame_header->grain_scaling_minus_8,
> +            .ar_coeff_lag = frame_header->ar_coeff_lag,
> +            .ar_coeff_shift_minus_6 = frame_header->ar_coeff_shift_minus_6,
> +            .grain_scale_shift = frame_header->grain_scale_shift,
> +            .overlap_flag = frame_header->overlap_flag,
> +            .clip_to_restricted_range = frame_header->clip_to_restricted_range,
> +        },
> +        .tile_cols = frame_header->tile_cols,
> +        .tile_rows = frame_header->tile_rows,
> +        .context_update_tile_id = frame_header->context_update_tile_id,
> +        .pic_info_fields.bits = {
> +            .frame_type = frame_header->frame_type,
> +            .show_frame = frame_header->show_frame,
> +            .showable_frame = frame_header->showable_frame,
> +            .error_resilient_mode = frame_header->error_resilient_mode,
> +            .disable_cdf_update = frame_header->disable_cdf_update,
> +            .allow_screen_content_tools = frame_header->allow_screen_content_tools,
> +            .force_integer_mv = frame_header->force_integer_mv,
> +            .allow_intrabc = frame_header->allow_intrabc,
> +            .use_superres = frame_header->use_superres,
> +            .allow_high_precision_mv = frame_header->allow_high_precision_mv,
> +            .is_motion_mode_switchable = frame_header->is_motion_mode_switchable,
> +            .use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
> +            .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
> +            .uniform_tile_spacing_flag = frame_header->uniform_tile_spacing_flag,
> +            .allow_warped_motion = frame_header->allow_warped_motion,
> +        },
> +        .interp_filter = frame_header->interpolation_filter,
> +        .filter_level[0] = frame_header->loop_filter_level[0],
> +        .filter_level[1] = frame_header->loop_filter_level[1],
> +        .filter_level_u = frame_header->loop_filter_level[2],
> +        .filter_level_v = frame_header->loop_filter_level[3],
> +        .loop_filter_info_fields.bits = {
> +            .sharpness_level = frame_header->loop_filter_sharpness,
> +            .mode_ref_delta_enabled = s->cur_frame.loop_filter_delta_enabled,
> +            .mode_ref_delta_update = frame_header->loop_filter_delta_update,
> +        },
> +        .base_qindex = frame_header->base_q_idx,
> +        .qmatrix_fields.bits.using_qmatrix = frame_header->using_qmatrix,
> +        .mode_control_fields.bits = {
> +            .delta_q_present_flag = frame_header->delta_q_present,
> +            .log2_delta_q_res = frame_header->delta_q_res,
> +            .tx_mode = frame_header->tx_mode,
> +            .reference_select = frame_header->reference_select,
> +            .reduced_tx_set_used = frame_header->reduced_tx_set,
> +            .skip_mode_present = frame_header->skip_mode_present,
> +        },
> +        .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
> +        .cdef_bits = frame_header->cdef_bits,
> +        .loop_restoration_fields.bits = {
> +            .yframe_restoration_type = lr_type[frame_header->lr_type[0]],
> +            .cbframe_restoration_type = lr_type[frame_header->lr_type[1]],
> +            .crframe_restoration_type = lr_type[frame_header->lr_type[2]],
> +            .lr_unit_shift = frame_header->lr_unit_shift,
> +            .lr_uv_shift = frame_header->lr_uv_shift,
> +        },
> +    };
> +
> +    for (int i = 0; i < 8; i++) {

The magic number would be clearer as AV1_NUM_REF_FRAMES.

> +        if (pic_param.pic_info_fields.bits.frame_type == AV1_FRAME_KEY)
> +            pic_param.ref_frame_map[i] = VA_INVALID_ID;
> +        else
> +            pic_param.ref_frame_map[i] = vaapi_av1_surface_id(&s->ref[i]);
> +        av_log(avctx, AV_LOG_DEBUG, "Vaapi set ref surface id: %d - 0x:%x\n",
> +               i, pic_param.ref_frame_map[i]);
> +    }
> +    for (int i = 0; i < 7; i++) {

AV_REFS_PER_FRAME

> +        pic_param.ref_frame_idx[i] = frame_header->ref_frame_idx[i];
> +    }
> +    for (int i = 0; i < 8; i++) {

AV_TOTAL_REFS_PER_FRAME

> +        pic_param.ref_deltas[i] = s->cur_frame.loop_filter_ref_deltas[i];
> +    }
> +    for (int i = 0; i < 2; i++) {
> +        pic_param.mode_deltas[i] = s->cur_frame.loop_filter_mode_deltas[i];
> +    }
> +    for (int i = 0; i < 8; i++) {

1 << cdef_bits?  The other values aren't set.

> +        pic_param.cdef_y_strengths[i] =
> +            (frame_header->cdef_y_pri_strength[i] << 2) +
> +                frame_header->cdef_y_sec_strength[i];
> +        pic_param.cdef_uv_strengths[i] =
> +            (frame_header->cdef_uv_pri_strength[i] << 2) +
> +                frame_header->cdef_uv_sec_strength[i];
> +    }
> +    for (int i = 0; i < frame_header->tile_cols; i++) {
> +        pic_param.width_in_sbs_minus_1[i] =
> +            frame_header->width_in_sbs_minus_1[i];
> +    }
> +    for (int i = 0; i < frame_header->tile_rows; i++) {
> +        pic_param.height_in_sbs_minus_1[i] =
> +            frame_header->height_in_sbs_minus_1[i];
> +    }
> +    for (int i = AV1_REF_FRAME_LAST; i <= AV1_REF_FRAME_ALTREF; i++) {
> +        pic_param.wm[i - 1].wmtype =
> +            vaapi_av1_get_gm_type(s->cur_frame.gm_type[i]);
> +        for (int j=0; j<6; j++)
> +            pic_param.wm[i - 1].wmmat[j] = s->cur_frame.gm_params[i][j];
> +    }
> +    err = ff_vaapi_decode_make_param_buffer(avctx, pic,
> +                                            VAPictureParameterBufferType,
> +                                            &pic_param, sizeof(pic_param));
> +    if (err < 0)
> +        goto fail;
> +
> +    return 0;
> +
> +fail:
> +    ff_vaapi_decode_cancel(avctx, pic);
> +    return err;
> +}
> +
> +static int vaapi_av1_end_frame(AVCodecContext *avctx)
> +{
> +    const AV1DecContext *s = avctx->priv_data;
> +    VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
> +    return ff_vaapi_decode_issue(avctx, pic);
> +}
> +
> +static int vaapi_av1_decode_slice(AVCodecContext *avctx,
> +                                  const uint8_t *buffer,
> +                                  uint32_t size)
> +{
> +    const AV1DecContext *s = avctx->priv_data;
> +    VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
> +    VASliceParameterBufferAV1 slice_param;
> +    uint32_t offset;
> +    int err = 0;
> +
> +    for (int i = s->tg_start; i <= s->tg_end; i++) {
> +        memset(&slice_param, 0, sizeof(VASliceParameterBufferAV1));
> +        offset = s->tile_group_info[i].tile_offset;
> +
> +        slice_param = (VASliceParameterBufferAV1) {
> +            .slice_data_size = s->tile_group_info[i].tile_size,
> +            .slice_data_offset = 0,
> +            .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
> +            .tile_row = s->tile_group_info[i].tile_row,
> +            .tile_column = s->tile_group_info[i].tile_column,
> +            .tg_start = s->tg_start,
> +            .tg_end = s->tg_end,

Horizontal alignment would be nice again.

> +        };
> +
> +        err = ff_vaapi_decode_make_slice_buffer(avctx, pic, &slice_param,
> +                                                sizeof(VASliceParameterBufferAV1),
> +                                                buffer + offset,
> +                                                s->tile_group_info[i].tile_size);
> +        if (err) {
> +            ff_vaapi_decode_cancel(avctx, pic);
> +            return err;
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +const AVHWAccel ff_av1_vaapi_hwaccel = {
> +    .name                 = "av1_vaapi",
> +    .type                 = AVMEDIA_TYPE_VIDEO,
> +    .id                   = AV_CODEC_ID_AV1,
> +    .pix_fmt              = AV_PIX_FMT_VAAPI,
> +    .start_frame          = vaapi_av1_start_frame,
> +    .end_frame            = vaapi_av1_end_frame,
> +    .decode_slice         = vaapi_av1_decode_slice,
> +    .frame_priv_data_size = sizeof(VAAPIDecodePicture),
> +    .init                 = ff_vaapi_decode_init,
> +    .uninit               = ff_vaapi_decode_uninit,
> +    .frame_params         = ff_vaapi_common_frame_params,
> +    .priv_data_size       = sizeof(VAAPIDecodeContext),
> +    .caps_internal        = HWACCEL_CAP_ASYNC_SAFE,
> +};
> diff --git a/libavcodec/vaapi_decode.c b/libavcodec/vaapi_decode.c
> index 5e4f62baad..2a38910cf8 100644
> --- a/libavcodec/vaapi_decode.c
> +++ b/libavcodec/vaapi_decode.c
> @@ -410,6 +410,11 @@ static const struct {
>   #if VA_CHECK_VERSION(0, 39, 0)
>       MAP(VP9,         VP9_2,           VP9Profile2 ),
>   #endif
> +#if VA_CHECK_VERSION(1, 8, 0)
> +    MAP(AV1,         AV1_MAIN,        AV1Profile0),
> +    MAP(AV1,         AV1_HIGH,        AV1Profile1),

What about the 12-bit YUV 4:2:0 referenced above?  that requires profile 2.

> +#endif
> +
>   #undef MAP
>   };
>   
> @@ -580,6 +585,9 @@ static int vaapi_decode_make_config(AVCodecContext *avctx,
>           case AV_CODEC_ID_VP8:
>               frames->initial_pool_size += 3;
>               break;
> +        case AV_CODEC_ID_AV1:
> +            frames->initial_pool_size += 8;
> +            break;

It can be in the same case as VP9, since the reference structure is the same (like H.26[45]).

>           default:
>               frames->initial_pool_size += 2;
>           }
> diff --git a/libavcodec/version.h b/libavcodec/version.h
> index 5ce4ba55d9..90e48a2ece 100644
> --- a/libavcodec/version.h
> +++ b/libavcodec/version.h
> @@ -29,7 +29,7 @@
>   
>   #define LIBAVCODEC_VERSION_MAJOR  58
>   #define LIBAVCODEC_VERSION_MINOR 106
> -#define LIBAVCODEC_VERSION_MICRO 100
> +#define LIBAVCODEC_VERSION_MICRO 101
>   
>   #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
>                                                  LIBAVCODEC_VERSION_MINOR, \
> 

Thanks,

- Mark


More information about the ffmpeg-devel mailing list