From patchwork Wed Sep 9 03:39:55 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Fei Wang X-Patchwork-Id: 22211 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 0DC78449FD0 for ; Wed, 9 Sep 2020 06:43:14 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id DF65268B74B; Wed, 9 Sep 2020 06:43:13 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mga07.intel.com (mga07.intel.com [134.134.136.100]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 9C28B68B63E for ; Wed, 9 Sep 2020 06:43:06 +0300 (EEST) IronPort-SDR: hkJzPqOFy7lOG1IL+DnpiaLAj79Cxx1nA0Nn4jroB8nP9nYv1/7uKGRWSglwXlTQtBlFgG/jvB qGc0ChE4zRaQ== X-IronPort-AV: E=McAfee;i="6000,8403,9738"; a="222468200" X-IronPort-AV: E=Sophos;i="5.76,408,1592895600"; d="scan'208";a="222468200" X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga004.fm.intel.com ([10.253.24.48]) by orsmga105.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 08 Sep 2020 20:43:03 -0700 IronPort-SDR: Y2ee3dURcXHyoHut0DCzIyNvks9bbKDDbWsnOi2tFdLptdYdEILfxQJ7DIqxpLvN4t11hexfY+ cuj1S9VVZQ3A== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.76,408,1592895600"; d="scan'208";a="328743722" Received: from f.sh.intel.com ([10.239.159.122]) by fmsmga004.fm.intel.com with ESMTP; 08 Sep 2020 20:43:01 -0700 From: Fei Wang To: ffmpeg-devel@ffmpeg.org Date: Wed, 9 Sep 2020 11:39:55 +0800 Message-Id: <20200909033956.27248-1-fei.w.wang@intel.com> X-Mailer: git-send-email 2.17.1 Subject: [FFmpeg-devel] [PATCH V3 1/2] avcodec: add av1 hardware acceleration decoder X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Fei Wang MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" This av1 decoder is now only used for av1 hardware acceleration decoder. Consider it can be extend to a local decoder like hevc or vp9 in the future, so define its name as "av1" and put it into external libraries codec list. Signed-off-by: Fei Wang --- update: 1. reuse buffer for tile_group_info. 2. set avctx/frame props from sequence and frame header. 3. refine ref/unref for s->seq_ref/s->header_ref. 4. refine av1_decode_frame to reduce duplicated code. Changelog | 1 + configure | 1 + libavcodec/Makefile | 1 + libavcodec/allcodecs.c | 1 + libavcodec/av1dec.c | 860 +++++++++++++++++++++++++++++++++++++++++ libavcodec/av1dec.h | 70 ++++ libavcodec/version.h | 2 +- 7 files changed, 935 insertions(+), 1 deletion(-) create mode 100644 libavcodec/av1dec.c create mode 100644 libavcodec/av1dec.h diff --git a/Changelog b/Changelog index ff9ff2fcb8..a89010d7e1 100644 --- a/Changelog +++ b/Changelog @@ -21,6 +21,7 @@ version : - MOFLEX demuxer - MODS demuxer - PhotoCD decoder +- AV1 decoder (Hardware acceleration used only) version 4.3: diff --git a/configure b/configure index f13d1a8c9a..5893ae3473 100755 --- a/configure +++ b/configure @@ -2685,6 +2685,7 @@ atrac3al_decoder_select="mdct" atrac3p_decoder_select="mdct sinewin" atrac3pal_decoder_select="mdct sinewin" atrac9_decoder_select="mdct" +av1_decoder_select="cbs_av1" avrn_decoder_select="exif jpegtables" bink_decoder_select="blockdsp hpeldsp" binkaudio_dct_decoder_select="mdct rdft dct sinewin wma_freqs" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 98f31e246b..5729cdd8e7 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -768,6 +768,7 @@ OBJS-$(CONFIG_ZLIB_DECODER) += lcldec.o OBJS-$(CONFIG_ZLIB_ENCODER) += lclenc.o OBJS-$(CONFIG_ZMBV_DECODER) += zmbv.o OBJS-$(CONFIG_ZMBV_ENCODER) += zmbvenc.o +OBJS-$(CONFIG_AV1_DECODER) += av1dec.o # (AD)PCM decoders/encoders OBJS-$(CONFIG_PCM_ALAW_DECODER) += pcm.o diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index f3572a47e3..1e374c99f2 100644 --- a/libavcodec/allcodecs.c +++ b/libavcodec/allcodecs.c @@ -766,6 +766,7 @@ extern AVCodec ff_idf_decoder; * above is available */ extern AVCodec ff_aac_mf_encoder; extern AVCodec ff_ac3_mf_encoder; +extern AVCodec ff_av1_decoder; extern AVCodec ff_h263_v4l2m2m_encoder; extern AVCodec ff_libaom_av1_decoder; extern AVCodec ff_libopenh264_encoder; diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c new file mode 100644 index 0000000000..eeeec2ca32 --- /dev/null +++ b/libavcodec/av1dec.c @@ -0,0 +1,860 @@ +/* + * AV1video decoder + * + * 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 "get_bits.h" +#include "hwconfig.h" +#include "internal.h" +#include "profiles.h" +#include "libavutil/pixdesc.h" +#include "av1dec.h" + +static void setup_past_independence(AV1Frame *f) +{ + f->loop_filter_delta_enabled = 1; + + f->loop_filter_ref_deltas[AV1_REF_FRAME_INTRA] = 1; + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST] = 0; + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST2] = 0; + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST3] = 0; + f->loop_filter_ref_deltas[AV1_REF_FRAME_GOLDEN] = -1; + f->loop_filter_ref_deltas[AV1_REF_FRAME_BWDREF] = 0; + f->loop_filter_ref_deltas[AV1_REF_FRAME_ALTREF2] = -1; + f->loop_filter_ref_deltas[AV1_REF_FRAME_ALTREF] = -1; + + f->loop_filter_mode_deltas[0] = 0; + f->loop_filter_mode_deltas[1] = 0; +} + +static void load_previous_and_update(AV1DecContext *s) +{ + uint8_t primary_frame, prev_frame; + + primary_frame = s->raw_frame_header->primary_ref_frame; + prev_frame = s->raw_frame_header->ref_frame_idx[primary_frame]; + memcpy(s->cur_frame.loop_filter_ref_deltas, + s->ref[prev_frame].loop_filter_ref_deltas, + AV1_NUM_REF_FRAMES * sizeof(int8_t)); + memcpy(s->cur_frame.loop_filter_mode_deltas, + s->ref[prev_frame].loop_filter_mode_deltas, + 2 * sizeof(int8_t)); + + if (s->raw_frame_header->loop_filter_delta_update) { + for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { + if (s->raw_frame_header->update_ref_delta[i]) + s->cur_frame.loop_filter_ref_deltas[i] = + s->raw_frame_header->loop_filter_ref_deltas[i]; + } + + for (int i = 0; i < 2; i++) { + if (s->raw_frame_header->update_mode_delta[i]) + s->cur_frame.loop_filter_mode_deltas[i] = + s->raw_frame_header->loop_filter_mode_deltas[i]; + } + } + + s->cur_frame.loop_filter_delta_enabled = + s->raw_frame_header->loop_filter_delta_enabled; +} + +static uint32_t inverse_recenter(int r, uint32_t v) +{ + if (v > 2 * r) + return v; + else if (v & 1) + return r - ((v + 1) >> 1); + else + return r + (v >> 1); +} + +static uint32_t decode_unsigned_subexp_with_ref(uint32_t sub_exp, + int mx, int r) +{ + if ((r << 1) <= mx) { + return inverse_recenter(r, sub_exp); + } else { + return mx - 1 - inverse_recenter(mx - 1 - r, sub_exp); + } +} + +static int32_t decode_signed_subexp_with_ref(uint32_t sub_exp, int low, + int high, int r) +{ + int32_t x = decode_unsigned_subexp_with_ref(sub_exp, high - low, r - low); + return x + low; +} + +static void read_global_param(AV1DecContext *s, int type, int ref, int idx) +{ + uint8_t primary_frame, prev_frame; + uint32_t abs_bits, prec_bits, round, prec_diff, sub, mx; + int32_t r; + + primary_frame = s->raw_frame_header->primary_ref_frame; + prev_frame = s->raw_frame_header->ref_frame_idx[primary_frame]; + abs_bits = AV1_GM_ABS_ALPHA_BITS; + prec_bits = AV1_GM_ALPHA_PREC_BITS; + + if (idx < 2) { + if (type == AV1_WARP_MODEL_TRANSLATION) { + abs_bits = AV1_GM_ABS_TRANS_ONLY_BITS - + !s->raw_frame_header->allow_high_precision_mv; + prec_bits = AV1_GM_TRANS_ONLY_PREC_BITS - + !s->raw_frame_header->allow_high_precision_mv; + } else { + abs_bits = AV1_GM_ABS_TRANS_BITS; + prec_bits = AV1_GM_TRANS_PREC_BITS; + } + } + round = (idx % 3) == 2 ? (1 << AV1_WARPEDMODEL_PREC_BITS) : 0; + prec_diff = AV1_WARPEDMODEL_PREC_BITS - prec_bits; + sub = (idx % 3) == 2 ? (1 << prec_bits) : 0; + mx = 1 << abs_bits; + r = (s->ref[prev_frame].gm_params[ref][idx] >> prec_diff) - sub; + + s->cur_frame.gm_params[ref][idx] = + (decode_signed_subexp_with_ref(s->raw_frame_header->gm_params[ref][idx], + -mx, mx + 1, r) << prec_diff) + round; +} + +/** +* update gm type/params, since cbs already implemented part of this funcation, +* so we don't need to full implement spec. +*/ +static void global_motion_params(AV1DecContext *s) +{ + const AV1RawFrameHeader *header = s->raw_frame_header; + int type, ref; + + for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) { + s->cur_frame.gm_type[ref] = AV1_WARP_MODEL_IDENTITY; + for (int i = 0; i < 6; i++) + s->cur_frame.gm_params[ref][i] = (i % 3 == 2) ? + 1 << AV1_WARPEDMODEL_PREC_BITS : 0; + } + if (header->frame_type == AV1_FRAME_KEY || + header->frame_type == AV1_FRAME_INTRA_ONLY) + return; + + for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) { + if (header->is_global[ref]) { + if (header->is_rot_zoom[ref]) { + type = AV1_WARP_MODEL_ROTZOOM; + } else { + type = header->is_translation[ref] ? AV1_WARP_MODEL_TRANSLATION + : AV1_WARP_MODEL_AFFINE; + } + } else { + type = AV1_WARP_MODEL_IDENTITY; + } + s->cur_frame.gm_type[ref] = type; + + if (type >= AV1_WARP_MODEL_ROTZOOM) { + read_global_param(s, type, ref, 2); + read_global_param(s, type, ref, 3); + if (type == AV1_WARP_MODEL_AFFINE) { + read_global_param(s, type, ref, 4); + read_global_param(s, type, ref, 5); + } else { + s->cur_frame.gm_params[ref][4] = -s->cur_frame.gm_params[ref][3]; + s->cur_frame.gm_params[ref][5] = s->cur_frame.gm_params[ref][2]; + } + } + if (type >= AV1_WARP_MODEL_TRANSLATION) { + read_global_param(s, type, ref, 0); + read_global_param(s, type, ref, 1); + } + } +} + +static int free_tile_data(AV1DecContext *s) +{ + if (s->tile_group_info) + av_freep(&s->tile_group_info); + return 0; +} + +static int init_tile_data(AV1DecContext *s) + +{ + uint16_t cur_tile_num = + s->raw_frame_header->tile_cols * s->raw_frame_header->tile_rows; + if (s->tile_num < cur_tile_num) { + s->tile_group_info = av_realloc_array(s->tile_group_info, cur_tile_num, + sizeof(TileGroupInfo)); + if (!s->tile_group_info) { + return AVERROR(ENOMEM); + } + s->tile_num = cur_tile_num; + } + + return 0; +} + +static int get_tiles_info(AVCodecContext *avctx, AV1RawTileGroup *tile_group) +{ + AV1DecContext *s = avctx->priv_data; + GetBitContext gb; + uint16_t tile_num, tile_row, tile_col; + uint32_t size = 0, size_bytes = 0, offset = 0; + int ret; + + if ((ret = init_get_bits8(&gb, + tile_group->tile_data.data, + tile_group->tile_data.data_size)) < 0) { + av_log(avctx, AV_LOG_ERROR, "Failed to initialize bitstream reader.\n"); + return ret; + } + + s->tg_start = tile_group->tg_start; + s->tg_end = tile_group->tg_end; + + for (tile_num = tile_group->tg_start; tile_num <= tile_group->tg_end; tile_num++) { + tile_row = tile_num / s->raw_frame_header->tile_cols; + tile_col = tile_num % s->raw_frame_header->tile_cols; + + if (tile_num == tile_group->tg_end) { + s->tile_group_info[tile_num].tile_size = get_bits_left(&gb) / 8; + s->tile_group_info[tile_num].tile_offset = offset; + s->tile_group_info[tile_num].tile_row = tile_row; + s->tile_group_info[tile_num].tile_column = tile_col; + return 0; + } + size_bytes = s->raw_frame_header->tile_size_bytes_minus1 + 1; + size = get_bits_le(&gb, size_bytes * 8) + 1; + skip_bits(&gb, size * 8); + + offset += size_bytes; + + s->tile_group_info[tile_num].tile_size = size; + s->tile_group_info[tile_num].tile_offset = offset; + s->tile_group_info[tile_num].tile_row = tile_row; + s->tile_group_info[tile_num].tile_column = tile_col; + + offset += size; + } + + return 0; + +} + +static int get_pixel_format(AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + const AV1RawSequenceHeader *seq = s->raw_seq; + uint8_t bit_depth; + int ret; + enum AVPixelFormat pix_fmt = AV_PIX_FMT_NONE; +#define HWACCEL_MAX (0) + enum AVPixelFormat pix_fmts[HWACCEL_MAX + 1], *fmtp = pix_fmts; + + 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, + "Unknow av1 profile %d.\n", seq->seq_profile); + return -1; + } + + if (!seq->color_config.mono_chrome) { + // 4:4:4 x:0 y:0, 4:2:2 x:1 y:0, 4:2:0 x:1 y:1 + if (seq->color_config.subsampling_x == 0 && + seq->color_config.subsampling_y == 0) { + if (bit_depth == 8) + pix_fmt = AV_PIX_FMT_YUV444P; + else if (bit_depth == 10) + pix_fmt = AV_PIX_FMT_YUV444P10; + else if (bit_depth == 12) + pix_fmt = AV_PIX_FMT_YUV444P12; + else + av_log(avctx, AV_LOG_WARNING, "Unknow av1 pixel format.\n"); + } else if (seq->color_config.subsampling_x == 1 && + seq->color_config.subsampling_y == 0) { + if (bit_depth == 8) + pix_fmt = AV_PIX_FMT_YUV422P; + else if (bit_depth == 10) + pix_fmt = AV_PIX_FMT_YUV422P10; + else if (bit_depth == 12) + pix_fmt = AV_PIX_FMT_YUV422P12; + else + av_log(avctx, AV_LOG_WARNING, "Unknow av1 pixel format.\n"); + } else if (seq->color_config.subsampling_x == 1 && + seq->color_config.subsampling_y == 1) { + if (bit_depth == 8) + pix_fmt = AV_PIX_FMT_YUV420P; + else if (bit_depth == 10) + pix_fmt = AV_PIX_FMT_YUV420P10; + else if (bit_depth == 12) + pix_fmt = AV_PIX_FMT_YUV420P12; + else + av_log(avctx, AV_LOG_WARNING, "Unknow av1 pixel format.\n"); + } + } else { + if (seq->color_config.subsampling_x == 1 && + seq->color_config.subsampling_y == 1) + pix_fmt = AV_PIX_FMT_YUV440P; + else + av_log(avctx, AV_LOG_WARNING, "Unknow av1 pixel format.\n"); + } + + av_log(avctx, AV_LOG_DEBUG, "Av1 decode get format: %s.\n", + av_get_pix_fmt_name(pix_fmt)); + + if (pix_fmt == AV_PIX_FMT_NONE) + return -1; + s->pix_fmt = pix_fmt; + + *fmtp = AV_PIX_FMT_NONE; + avctx->sw_pix_fmt = s->pix_fmt; + ret = ff_thread_get_format(avctx, pix_fmts); + if (ret < 0) + return ret; + + /** + * check if the HW accel is inited correctly. If not, return un-implemented. + * Since now the av1 decoder doesn't support native decode, if it will be + * implemented in the future, need remove this check. + */ + if (!avctx->hwaccel) { + av_log(avctx, AV_LOG_ERROR, "Your platform doesn't suppport" + " hardware acceleration AV1 decode.\n"); + return AVERROR(ENOSYS); + } + + return 0; +} + +static void av1_frame_unref(AVCodecContext *avctx, AV1Frame *f) +{ + ff_thread_release_buffer(avctx, &f->tf); + av_buffer_unref(&f->hwaccel_priv_buf); + f->hwaccel_picture_private = NULL; +} + +static int av1_frame_ref(AVCodecContext *avctx, AV1Frame *dst, AV1Frame *src) +{ + int ret; + + ret = ff_thread_ref_frame(&dst->tf, &src->tf); + if (ret < 0) + return ret; + + if (src->hwaccel_picture_private) { + dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf); + if (!dst->hwaccel_priv_buf) + goto fail; + dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data; + } + + dst->loop_filter_delta_enabled = src->loop_filter_delta_enabled; + memcpy(dst->loop_filter_ref_deltas, + src->loop_filter_ref_deltas, + AV1_NUM_REF_FRAMES * sizeof(int8_t)); + memcpy(dst->loop_filter_mode_deltas, + src->loop_filter_mode_deltas, + 2 * sizeof(int8_t)); + memcpy(dst->gm_type, + src->gm_type, + AV1_NUM_REF_FRAMES * sizeof(uint8_t)); + memcpy(dst->gm_params, + src->gm_params, + AV1_NUM_REF_FRAMES * 6 * sizeof(int32_t)); + + return 0; + +fail: + av1_frame_unref(avctx, dst); + return AVERROR(ENOMEM); +} + +static av_cold int av1_decode_free(AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) { + if (s->ref[i].tf.f->buf[0]) + av1_frame_unref(avctx, &s->ref[i]); + av_frame_free(&s->ref[i].tf.f); + } + if (s->cur_frame.tf.f->buf[0]) + av1_frame_unref(avctx, &s->cur_frame); + av_frame_free(&s->cur_frame.tf.f); + + if (s->seq_ref) { + av_buffer_unref(&s->seq_ref); + s->raw_seq = NULL; + } + + if (s->header_ref) { + av_buffer_unref(&s->header_ref); + s->raw_frame_header = NULL; + } + + free_tile_data(s); + ff_cbs_fragment_free(&s->current_obu); + ff_cbs_close(&s->cbc); + + return 0; +} + +static int set_context_with_sequence(AVCodecContext *avctx, + AV1RawSequenceHeader *seq) +{ + int ret; + + avctx->profile = seq->seq_profile; + avctx->level = seq->seq_level_idx[0]; + + avctx->color_range = + seq->color_config.color_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG; + avctx->color_primaries = seq->color_config.color_primaries; + avctx->colorspace = seq->color_config.color_primaries; + avctx->color_trc = seq->color_config.transfer_characteristics; + + if (seq->timing_info.num_units_in_display_tick && + seq->timing_info.time_scale) { + ret = av_reduce(&avctx->framerate.den, &avctx->framerate.num, + seq->timing_info.num_units_in_display_tick, + seq->timing_info.time_scale, + INT_MAX); + if (ret < 0) + return ret; + } + + return 0; +} + +static int update_context_with_frame_header(AVCodecContext *avctx, + AV1RawFrameHeader *header) +{ + AVRational aspect_ratio; + int width, height, r_width, r_height; + int ret; + + width = header->frame_width_minus_1 + 1; + height = header->frame_height_minus_1 + 1; + r_width = header->render_width_minus_1 + 1; + r_height = header->render_height_minus_1 + 1; + + if (avctx->width != width || avctx->height != height) { + ret = ff_set_dimensions(avctx, width, height); + if (ret < 0) + return ret; + } + + ret = av_reduce(&aspect_ratio.num, &aspect_ratio.den, + height * r_width, width * r_height, INT_MAX); + if (ret < 0) + return ret; + + if (avctx->sample_aspect_ratio.num != aspect_ratio.num || + avctx->sample_aspect_ratio.den != aspect_ratio.den) { + ret = ff_set_sar(avctx, aspect_ratio); + if (ret < 0) + return ret; + } + + return 0; +} + +static av_cold int av1_decode_init(AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + AV1RawSequenceHeader *seq; + int ret; + + s->avctx = avctx; + s->pix_fmt = AV_PIX_FMT_NONE; + s->seq_ref = NULL; + s->header_ref = NULL; + s->raw_frame_header = NULL; + s->raw_seq = NULL; + s->tile_group_info = NULL; + s->tile_num = 0; + + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) { + s->ref[i].tf.f = av_frame_alloc(); + if (!s->ref[i].tf.f) { + av1_decode_free(avctx); + av_log(avctx, AV_LOG_ERROR, + "Failed to allocate reference frame buffer %d.\n", i); + return AVERROR(ENOMEM); + } + } + + s->cur_frame.tf.f = av_frame_alloc(); + if (!s->cur_frame.tf.f) { + av1_decode_free(avctx); + av_log(avctx, AV_LOG_ERROR, + "Failed to allocate current frame buffer.\n"); + return AVERROR(ENOMEM); + } + + ret = ff_cbs_init(&s->cbc, AV_CODEC_ID_AV1, avctx); + if (ret < 0) + return ret; + + if (avctx->extradata && avctx->extradata_size) { + ret = ff_cbs_read(s->cbc, &s->current_obu, avctx->extradata, + avctx->extradata_size); + if (ret < 0) { + av_log(avctx, AV_LOG_WARNING, "Failed to read extradata.\n"); + goto end; + } + + seq = ((CodedBitstreamAV1Context *)(s->cbc->priv_data))->sequence_header; + if (!seq) { + av_log(avctx, AV_LOG_WARNING, "No sequence header available.\n"); + goto end; + } + + ret = set_context_with_sequence(avctx, seq); + if (ret < 0) { + av_log(avctx, AV_LOG_WARNING, "Failed to set context.\n"); + goto end; + } + + end: + ff_cbs_fragment_reset(&s->current_obu); + } + + return ret; +} + +static int av1_frame_alloc(AVCodecContext *avctx, AV1Frame *f) +{ + AV1DecContext *s = avctx->priv_data; + AV1RawFrameHeader *header= s->raw_frame_header; + AVFrame *frame; + int ret; + + if ((ret = ff_thread_get_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF)) < 0) + return ret; + + frame = f->tf.f; + frame->key_frame = + header->frame_type == AV1_FRAME_KEY && header->show_frame == 1; + + switch (header->frame_type) { + case AV1_FRAME_KEY: + case AV1_FRAME_INTRA_ONLY: + frame->pict_type = AV_PICTURE_TYPE_I; + break; + case AV1_FRAME_INTER: + frame->pict_type = AV_PICTURE_TYPE_P; + break; + case AV1_FRAME_SWITCH: + frame->pict_type = AV_PICTURE_TYPE_SP; + break; + } + + if (avctx->hwaccel) { + const AVHWAccel *hwaccel = avctx->hwaccel; + if (hwaccel->frame_priv_data_size) { + f->hwaccel_priv_buf = + av_buffer_allocz(hwaccel->frame_priv_data_size); + if (!f->hwaccel_priv_buf) + goto fail; + f->hwaccel_picture_private = f->hwaccel_priv_buf->data; + } + } + return 0; + +fail: + av1_frame_unref(avctx, f); + return AVERROR(ENOMEM); +} + +static int set_output_frame(AVCodecContext *avctx, AVFrame *frame, + AVPacket *pkt, int *got_frame) +{ + AV1DecContext *s = avctx->priv_data; + AV1RawFrameHeader *header= s->raw_frame_header; + AVFrame *srcframe; + int ret; + + if (header->show_existing_frame) + srcframe = s->ref[header->frame_to_show_map_idx].tf.f; + else + srcframe = s->cur_frame.tf.f; + + if ((ret = av_frame_ref(frame, srcframe)) < 0) { + return ret; + } + + frame->pts = pkt->pts; + frame->pkt_dts = pkt->dts; + + *got_frame = 1; + + return 0; +} + +static int update_reference_list (AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + AV1RawFrameHeader *header= s->raw_frame_header; + int ret; + + for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { + if (header->refresh_frame_flags & (1 << i)) { + if (s->ref[i].tf.f->buf[0]) + av1_frame_unref(avctx, &s->ref[i]); + if ((ret = av1_frame_ref(avctx, &s->ref[i], &s->cur_frame)) < 0) { + av_log(avctx, AV_LOG_DEBUG, "Ref frame error:%d.\n", ret); + return ret; + } + } + } + return 0; +} + +static int get_current_frame(AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + int ret; + + if (s->cur_frame.tf.f->buf[0]) + av1_frame_unref(avctx, &s->cur_frame); + + ret = av1_frame_alloc(avctx, &s->cur_frame); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, + "Failed to allocate space for current frame.\n"); + return ret; + } + + ret = init_tile_data(s); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "Failed to init tile data.\n"); + return ret; + } + + if (s->raw_frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE) + setup_past_independence(&s->cur_frame); + else + load_previous_and_update(s); + + global_motion_params(s); + + return ret; +} + +static int av1_decode_frame(AVCodecContext *avctx, void *frame, + int *got_frame, AVPacket *pkt) +{ + AV1DecContext *s = avctx->priv_data; + AV1RawTileGroup *raw_tile_group = NULL; + int ret; + + ret = ff_cbs_read_packet(s->cbc, &s->current_obu, pkt); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "Failed to read packet.\n"); + goto end; + } + av_log(avctx, AV_LOG_DEBUG, "Total obu for this frame:%d.\n", + s->current_obu.nb_units); + + for (int i = 0; i < s->current_obu.nb_units; i++) { + CodedBitstreamUnit *unit = &s->current_obu.units[i]; + AV1RawOBU *obu = unit->content; + av_log(avctx, AV_LOG_DEBUG, "Obu idx:%d, obu type:%d.\n", i, unit->type); + + switch (unit->type) { + case AV1_OBU_SEQUENCE_HEADER: + av_buffer_unref(&s->seq_ref); + s->seq_ref = av_buffer_ref(unit->content_ref); + if (!s->seq_ref) + return AVERROR(ENOMEM); + + s->raw_seq = &obu->obu.sequence_header; + + ret = set_context_with_sequence(avctx, s->raw_seq); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "Failed to set context.\n"); + goto end; + } + + if (s->pix_fmt == AV_PIX_FMT_NONE) { + ret = get_pixel_format(avctx); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, + "Failed to get pixel format.\n"); + goto end; + } + } + + if (avctx->hwaccel && avctx->hwaccel->decode_params) { + ret = avctx->hwaccel->decode_params(avctx, unit->type, unit->data, + unit->data_size); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "HW accel decode params fail.\n"); + return ret; + } + } + break; + case AV1_OBU_REDUNDANT_FRAME_HEADER: + if (s->raw_frame_header) + break; + case AV1_OBU_FRAME: + case AV1_OBU_FRAME_HEADER: + av_buffer_unref(&s->header_ref); + s->header_ref = av_buffer_ref(unit->content_ref); + if (!s->header_ref) + return AVERROR(ENOMEM); + + if (unit->type == AV1_OBU_FRAME) + s->raw_frame_header = &obu->obu.frame.header; + else + s->raw_frame_header = &obu->obu.frame_header; + + ret = update_context_with_frame_header(avctx, s->raw_frame_header); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, + "Update context with frame header fail:%d.\n", ret); + goto end; + } + + if (s->raw_frame_header->show_existing_frame) { + ret = set_output_frame(avctx, frame, pkt, got_frame); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "Set output frame error:%d.\n", + ret); + goto end; + } + goto end; + } + + ret = get_current_frame(avctx); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "Get current frame error:%d.\n", + ret); + goto end; + } + + if (avctx->hwaccel) { + ret = avctx->hwaccel->start_frame(avctx, unit->data, + unit->data_size); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "HW accel start frame fail.\n"); + return ret; + } + } + if (unit->type != AV1_OBU_FRAME) + break; + case AV1_OBU_TILE_GROUP: + if (!s->raw_frame_header) { + ret = AVERROR_INVALIDDATA; + goto end; + } + + if (unit->type == AV1_OBU_FRAME) + raw_tile_group = &obu->obu.frame.tile_group; + else + raw_tile_group = &obu->obu.tile_group; + + get_tiles_info(avctx, raw_tile_group); + + if (avctx->hwaccel) { + ret = avctx->hwaccel->decode_slice(avctx, + raw_tile_group->tile_data.data, + raw_tile_group->tile_data.data_size); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, + "HW accel decode slice fail.\n"); + return ret; + } + } + break; + case AV1_OBU_TILE_LIST: + case AV1_OBU_TEMPORAL_DELIMITER: + case AV1_OBU_PADDING: + case AV1_OBU_METADATA: + break; + default: + av_log(avctx, AV_LOG_DEBUG, + "Unknow obu type: %d (%zd bits).\n", + unit->type, unit->data_size); + } + + if (raw_tile_group && (s->tile_num == raw_tile_group->tg_end + 1)) { + if (avctx->hwaccel) { + ret = avctx->hwaccel->end_frame(avctx); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "HW accel end frame fail.\n"); + return ret; + } + } + + ret = update_reference_list(avctx); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "Failed to update reference list.\n"); + return ret; + } + + if (s->raw_frame_header->show_frame) { + ret = set_output_frame(avctx, frame, pkt, got_frame); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "Set output frame error:%d.\n", + ret); + goto end; + } + } + raw_tile_group = NULL; + s->raw_frame_header = NULL; + } + } + +end: + ff_cbs_fragment_reset(&s->current_obu); + return ret; +} + +static void av1_decode_flush(AVCodecContext *avctx) +{ + AV1DecContext *s = avctx->priv_data; + + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) + av1_frame_unref(avctx, &s->ref[i]); + + av1_frame_unref(avctx, &s->cur_frame); +} + +AVCodec ff_av1_decoder = { + .name = "av1", + .long_name = NULL_IF_CONFIG_SMALL("Alliance for Open Media AV1"), + .type = AVMEDIA_TYPE_VIDEO, + .id = AV_CODEC_ID_AV1, + .priv_data_size = sizeof(AV1DecContext), + .init = av1_decode_init, + .close = av1_decode_free, + .decode = av1_decode_frame, + .capabilities = AV_CODEC_CAP_DR1, + .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | + FF_CODEC_CAP_INIT_CLEANUP | + FF_CODEC_CAP_SETS_PKT_DTS, + .flush = av1_decode_flush, + .profiles = NULL_IF_CONFIG_SMALL(ff_av1_profiles), + .hw_configs = (const AVCodecHWConfigInternal * []) { + NULL + }, +}; diff --git a/libavcodec/av1dec.h b/libavcodec/av1dec.h new file mode 100644 index 0000000000..ab0bda4d3a --- /dev/null +++ b/libavcodec/av1dec.h @@ -0,0 +1,70 @@ +/* + * AV1 video decoder + * * + * 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 + */ + +#ifndef AVCODEC_AV1DEC_H +#define AVCODEC_AV1DEC_H + +#include "thread.h" +#include "cbs.h" +#include "cbs_av1.h" + +typedef struct AV1Frame { + ThreadFrame tf; + + AVBufferRef *hwaccel_priv_buf; + void *hwaccel_picture_private; + + uint8_t loop_filter_delta_enabled; + int8_t loop_filter_ref_deltas[AV1_NUM_REF_FRAMES]; + int8_t loop_filter_mode_deltas[2]; + uint8_t gm_type[AV1_NUM_REF_FRAMES]; + int32_t gm_params[AV1_NUM_REF_FRAMES][6]; +} AV1Frame; + +typedef struct TileGroupInfo { + uint32_t tile_offset; + uint32_t tile_size; + uint16_t tile_row; + uint16_t tile_column; +} TileGroupInfo; + +typedef struct AV1DecContext { + const AVClass *class; + AVCodecContext *avctx; + + enum AVPixelFormat pix_fmt; + CodedBitstreamContext *cbc; + CodedBitstreamFragment current_obu; + + AVBufferRef *seq_ref; + AV1RawSequenceHeader *raw_seq; + AVBufferRef *header_ref; + AV1RawFrameHeader *raw_frame_header; + TileGroupInfo *tile_group_info; + uint16_t tile_num; + uint16_t tg_start; + uint16_t tg_end; + + AV1Frame ref[AV1_NUM_REF_FRAMES]; + AV1Frame cur_frame; + +} AV1DecContext; + +#endif /* AVCODEC_AV1DEC_H */ diff --git a/libavcodec/version.h b/libavcodec/version.h index 4b221f96ad..5ce4ba55d9 100644 --- a/libavcodec/version.h +++ b/libavcodec/version.h @@ -28,7 +28,7 @@ #include "libavutil/version.h" #define LIBAVCODEC_VERSION_MAJOR 58 -#define LIBAVCODEC_VERSION_MINOR 105 +#define LIBAVCODEC_VERSION_MINOR 106 #define LIBAVCODEC_VERSION_MICRO 100 #define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \