/* GStreamer * Copyright (C) 2020 Collabora * Author: Daniel Almeida * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Library General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library 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 * Library General Public License for more details. * * You should have received a copy of the GNU Library General Public * License along with this library; if not, write to the0 * Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, * Boston, MA 02110-1301, USA. */ #ifdef HAVE_CONFIG_H #include "config.h" #endif #include "gstv4l2codecallocator.h" #include "gstv4l2codecav1dec.h" #include "gstv4l2codecpool.h" #include "gstv4l2format.h" #include "linux/v4l2-controls.h" #define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c)) #define V4L2_MIN_KERNEL_VER_MAJOR 6 #define V4L2_MIN_KERNEL_VER_MINOR 7 #define V4L2_MIN_KERNEL_VERSION KERNEL_VERSION(V4L2_MIN_KERNEL_VER_MAJOR, V4L2_MIN_KERNEL_VER_MINOR, 0) GST_DEBUG_CATEGORY_STATIC (v4l2_av1dec_debug); #define GST_CAT_DEFAULT v4l2_av1dec_debug /* Used to mark picture that have been outputted */ #define FLAG_PICTURE_HOLDS_BUFFER GST_MINI_OBJECT_FLAG_LAST enum { PROP_0, PROP_LAST = PROP_0 }; static GstStaticPadTemplate sink_template = GST_STATIC_PAD_TEMPLATE (GST_VIDEO_DECODER_SINK_NAME, GST_PAD_SINK, GST_PAD_ALWAYS, GST_STATIC_CAPS ("video/x-av1, alignment=frame")); #define SRC_CAPS \ GST_VIDEO_CAPS_MAKE (GST_V4L2_DEFAULT_VIDEO_FORMATS) static GstStaticCaps static_src_caps = GST_STATIC_CAPS (SRC_CAPS); static GstStaticPadTemplate src_template = GST_STATIC_PAD_TEMPLATE (GST_VIDEO_DECODER_SRC_NAME, GST_PAD_SRC, GST_PAD_ALWAYS, GST_STATIC_CAPS (SRC_CAPS)); struct _GstV4l2CodecAV1Dec { GstAV1Decoder parent; GstV4l2Decoder *decoder; GstVideoCodecState *output_state; GstVideoInfo vinfo; GstV4l2CodecAllocator *sink_allocator; GstV4l2CodecAllocator *src_allocator; GstV4l2CodecPool *src_pool; gint min_pool_size; gboolean has_videometa; gboolean streaming; gboolean copy_frames; gint frame_width; gint frame_height; gint render_width; gint render_height; guint bit_depth; guint profile; guint16 operating_point_idc; struct v4l2_ctrl_av1_sequence v4l2_sequence; struct v4l2_ctrl_av1_frame v4l2_frame; struct v4l2_ctrl_av1_film_grain v4l2_film_grain; gboolean need_sequence; GArray *tile_group_entries; gboolean fill_film_grain; GstMemory *bitstream; GstMapInfo bitstream_map; }; G_DEFINE_ABSTRACT_TYPE (GstV4l2CodecAV1Dec, gst_v4l2_codec_av1_dec, GST_TYPE_AV1_DECODER); #define parent_class gst_v4l2_codec_av1_dec_parent_class static GstFlowReturn gst_v4l2_codec_av1_dec_ensure_bitstream (GstV4l2CodecAV1Dec * self) { if (self->bitstream) goto done; self->bitstream = gst_v4l2_codec_allocator_alloc (self->sink_allocator); if (!self->bitstream) { GST_ELEMENT_ERROR (self, RESOURCE, NO_SPACE_LEFT, ("Not enough memory to decode AV1 stream."), (NULL)); return GST_FLOW_ERROR; } if (!gst_memory_map (self->bitstream, &self->bitstream_map, GST_MAP_WRITE)) { GST_ELEMENT_ERROR (self, RESOURCE, WRITE, ("Could not access bitstream memory for writing"), (NULL)); g_clear_pointer (&self->bitstream, gst_memory_unref); return GST_FLOW_ERROR; } done: /* We use this field to track how much we have written */ self->bitstream_map.size = 0; return GST_FLOW_OK; } static void gst_v4l2_codec_av1_reset_bitstream (GstV4l2CodecAV1Dec * self) { if (self->bitstream) { if (self->bitstream_map.memory) gst_memory_unmap (self->bitstream, &self->bitstream_map); g_clear_pointer (&self->bitstream, gst_memory_unref); self->bitstream_map = (GstMapInfo) GST_MAP_INFO_INIT; } } static gboolean gst_v4l2_decoder_av1_api_check (GstV4l2Decoder * decoder) { guint i, ret_size; /* *INDENT-OFF* */ #define SET_ID(cid) .id = (cid), .name = #cid struct { const gchar *name; unsigned int id; unsigned int size; gboolean optional; } controls[] = { { SET_ID (V4L2_CID_STATELESS_AV1_FRAME), .size = sizeof(struct v4l2_ctrl_av1_frame), }, { SET_ID (V4L2_CID_STATELESS_AV1_TILE_GROUP_ENTRY), .size = sizeof(struct v4l2_ctrl_av1_tile_group_entry), }, { SET_ID (V4L2_CID_STATELESS_AV1_SEQUENCE), .size = sizeof(struct v4l2_ctrl_av1_sequence), }, { SET_ID (V4L2_CID_STATELESS_AV1_FILM_GRAIN), .size = sizeof(struct v4l2_ctrl_av1_film_grain), .optional = TRUE, } }; #undef SET_ID /* *INDENT-ON* */ /* * Compatibility check: make sure the pointer controls are * the right size. */ for (i = 0; i < G_N_ELEMENTS (controls); i++) { gboolean control_found; control_found = gst_v4l2_decoder_query_control_size (decoder, controls[i].id, &ret_size); if (!controls[i].optional && !control_found) { GST_WARNING ("Driver is missing %s support.", controls[i].name); return FALSE; } if (control_found && ret_size != controls[i].size) { GST_WARNING ("%s control size mismatch: got %d bytes but %d expected.", controls[i].name, ret_size, controls[i].size); return FALSE; } } return TRUE; } static gboolean gst_v4l2_codec_av1_dec_open (GstVideoDecoder * decoder) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); if (!gst_v4l2_decoder_open (self->decoder)) { GST_ELEMENT_ERROR (self, RESOURCE, OPEN_READ_WRITE, ("Failed to open AV1 decoder"), ("gst_v4l2_decoder_open() failed: %s", g_strerror (errno))); return FALSE; } self->fill_film_grain = gst_v4l2_decoder_query_control_size (self->decoder, V4L2_CID_STATELESS_AV1_FILM_GRAIN, NULL); return TRUE; } static gboolean gst_v4l2_codec_av1_dec_close (GstVideoDecoder * decoder) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); gst_v4l2_decoder_close (self->decoder); return TRUE; } static void gst_v4l2_codec_av1_dec_streamoff (GstV4l2CodecAV1Dec * self) { if (self->streaming) { gst_v4l2_decoder_streamoff (self->decoder, GST_PAD_SINK); gst_v4l2_decoder_streamoff (self->decoder, GST_PAD_SRC); self->streaming = FALSE; } } static void gst_v4l2_codec_av1_dec_reset_allocation (GstV4l2CodecAV1Dec * self) { if (self->sink_allocator) { gst_v4l2_codec_allocator_detach (self->sink_allocator); g_clear_object (&self->sink_allocator); } if (self->src_allocator) { gst_v4l2_codec_allocator_detach (self->src_allocator); g_clear_object (&self->src_allocator); g_clear_object (&self->src_pool); } } static gboolean gst_v4l2_codec_av1_dec_stop (GstVideoDecoder * decoder) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); gst_v4l2_decoder_streamoff (self->decoder, GST_PAD_SINK); gst_v4l2_decoder_streamoff (self->decoder, GST_PAD_SRC); gst_v4l2_codec_av1_dec_reset_allocation (self); if (self->output_state) gst_video_codec_state_unref (self->output_state); self->output_state = NULL; return GST_VIDEO_DECODER_CLASS (parent_class)->stop (decoder); } static gboolean gst_v4l2_codec_av1_dec_negotiate (GstVideoDecoder * decoder) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GstAV1Decoder *av1dec = GST_AV1_DECODER (decoder); /* *INDENT-OFF* */ struct v4l2_ext_control control[] = { { .id = V4L2_CID_STATELESS_AV1_SEQUENCE, .ptr = &self->v4l2_sequence, .size = sizeof (self->v4l2_sequence), }, }; /* *INDENT-ON* */ GstCaps *filter, *caps; /* Ignore downstream renegotiation request. */ if (self->streaming) goto done; GST_DEBUG_OBJECT (self, "Negotiate"); gst_v4l2_codec_av1_dec_reset_allocation (self); if (!gst_v4l2_decoder_set_sink_fmt (self->decoder, V4L2_PIX_FMT_AV1_FRAME, self->frame_width, self->frame_height, self->bit_depth)) { GST_ELEMENT_ERROR (self, CORE, NEGOTIATION, ("Failed to configure AV1 decoder"), ("gst_v4l2_decoder_set_sink_fmt() failed: %s", g_strerror (errno))); gst_v4l2_decoder_close (self->decoder); return FALSE; } if (!gst_v4l2_decoder_set_controls (self->decoder, NULL, control, G_N_ELEMENTS (control))) { GST_ELEMENT_ERROR (decoder, RESOURCE, WRITE, ("Driver does not support the selected stream."), (NULL)); return FALSE; } filter = gst_v4l2_decoder_enum_src_formats (self->decoder, &static_src_caps); if (!filter) { GST_ELEMENT_ERROR (self, CORE, NEGOTIATION, ("No supported decoder output formats"), (NULL)); return FALSE; } GST_DEBUG_OBJECT (self, "Supported output formats: %" GST_PTR_FORMAT, filter); caps = gst_pad_peer_query_caps (decoder->srcpad, filter); gst_caps_unref (filter); GST_DEBUG_OBJECT (self, "Peer supported formats: %" GST_PTR_FORMAT, caps); if (!gst_v4l2_decoder_select_src_format (self->decoder, caps, &self->vinfo)) { GST_ELEMENT_ERROR (self, CORE, NEGOTIATION, ("Unsupported pixel format"), ("No support for %ux%u", self->frame_width, self->frame_height)); gst_caps_unref (caps); return FALSE; } gst_caps_unref (caps); done: if (self->output_state) gst_video_codec_state_unref (self->output_state); self->output_state = gst_video_decoder_set_output_state (GST_VIDEO_DECODER (self), self->vinfo.finfo->format, self->render_width, self->render_height, av1dec->input_state); self->output_state->caps = gst_video_info_to_caps (&self->output_state->info); if (GST_VIDEO_DECODER_CLASS (parent_class)->negotiate (decoder)) { if (self->streaming) return TRUE; if (!gst_v4l2_decoder_streamon (self->decoder, GST_PAD_SINK)) { GST_ELEMENT_ERROR (self, RESOURCE, FAILED, ("Could not enable the decoder driver."), ("VIDIOC_STREAMON(SINK) failed: %s", g_strerror (errno))); return FALSE; } if (!gst_v4l2_decoder_streamon (self->decoder, GST_PAD_SRC)) { GST_ELEMENT_ERROR (self, RESOURCE, FAILED, ("Could not enable the decoder driver."), ("VIDIOC_STREAMON(SRC) failed: %s", g_strerror (errno))); return FALSE; } self->streaming = TRUE; return TRUE; } return FALSE; } static gboolean gst_v4l2_codec_av1_dec_decide_allocation (GstVideoDecoder * decoder, GstQuery * query) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); guint min = 0, num_bitstream; /* If we are streaming here, then it means there is nothing allocation * related in the new state and allocation can be ignored */ if (self->streaming) goto no_internal_changes; self->has_videometa = gst_query_find_allocation_meta (query, GST_VIDEO_META_API_TYPE, NULL); g_clear_object (&self->src_pool); g_clear_object (&self->src_allocator); if (gst_query_get_n_allocation_pools (query) > 0) gst_query_parse_nth_allocation_pool (query, 0, NULL, NULL, &min, NULL); min = MAX (2, min); num_bitstream = 1 + MAX (1, gst_v4l2_decoder_get_render_delay (self->decoder)); self->sink_allocator = gst_v4l2_codec_allocator_new (self->decoder, GST_PAD_SINK, num_bitstream); if (!self->sink_allocator) { GST_ELEMENT_ERROR (self, RESOURCE, NO_SPACE_LEFT, ("Not enough memory to allocate sink buffers."), (NULL)); return FALSE; } self->src_allocator = gst_v4l2_codec_allocator_new (self->decoder, GST_PAD_SRC, self->min_pool_size + min); if (!self->src_allocator) { GST_ELEMENT_ERROR (self, RESOURCE, NO_SPACE_LEFT, ("Not enough memory to allocate source buffers."), (NULL)); g_clear_object (&self->sink_allocator); return FALSE; } self->src_pool = gst_v4l2_codec_pool_new (self->src_allocator, &self->vinfo); no_internal_changes: /* Our buffer pool is internal, we will let the base class create a video * pool, and use it if we are running out of buffers or if downstream does * not support GstVideoMeta */ return GST_VIDEO_DECODER_CLASS (parent_class)->decide_allocation (decoder, query); } static guint gst_v4l2_codec_av1_dec_get_preferred_output_delay (GstAV1Decoder * decoder, gboolean live) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); guint delay; if (live) delay = 0; else /* Just one for now, perhaps we can make this configurable in the future. */ delay = 1; gst_v4l2_decoder_set_render_delay (self->decoder, delay); return delay; } static void gst_v4l2_codec_av1_dec_set_flushing (GstV4l2CodecAV1Dec * self, gboolean flushing) { if (self->sink_allocator) gst_v4l2_codec_allocator_set_flushing (self->sink_allocator, flushing); if (self->src_allocator) gst_v4l2_codec_allocator_set_flushing (self->src_allocator, flushing); } static gboolean gst_v4l2_codec_av1_dec_flush (GstVideoDecoder * decoder) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GST_DEBUG_OBJECT (self, "Flushing decoder state."); gst_v4l2_decoder_flush (self->decoder); gst_v4l2_codec_av1_dec_set_flushing (self, FALSE); return GST_VIDEO_DECODER_CLASS (parent_class)->flush (decoder); } static gboolean gst_v4l2_codec_av1_dec_sink_event (GstVideoDecoder * decoder, GstEvent * event) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); switch (GST_EVENT_TYPE (event)) { case GST_EVENT_FLUSH_START: GST_DEBUG_OBJECT (self, "flush start"); gst_v4l2_codec_av1_dec_set_flushing (self, TRUE); break; default: break; } return GST_VIDEO_DECODER_CLASS (parent_class)->sink_event (decoder, event); } static GstStateChangeReturn gst_v4l2_codec_av1_dec_change_state (GstElement * element, GstStateChange transition) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (element); if (transition == GST_STATE_CHANGE_PAUSED_TO_READY) gst_v4l2_codec_av1_dec_set_flushing (self, TRUE); return GST_ELEMENT_CLASS (parent_class)->change_state (element, transition); } static void gst_v4l2_codec_av1_dec_set_property (GObject * object, guint prop_id, const GValue * value, GParamSpec * pspec) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (object); GObject *dec = G_OBJECT (self->decoder); switch (prop_id) { default: gst_v4l2_decoder_set_property (dec, prop_id - PROP_LAST, value, pspec); break; } } static void gst_v4l2_codec_av1_dec_get_property (GObject * object, guint prop_id, GValue * value, GParamSpec * pspec) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (object); GObject *dec = G_OBJECT (self->decoder); switch (prop_id) { default: gst_v4l2_decoder_get_property (dec, prop_id - PROP_LAST, value, pspec); break; } } static void gst_v4l2_codec_av1_dec_fill_sequence_params (GstV4l2CodecAV1Dec * self, const GstAV1SequenceHeaderOBU * seq_hdr) { /* *INDENT-OFF* */ self->v4l2_sequence = (struct v4l2_ctrl_av1_sequence) { .flags = (seq_hdr->still_picture ? V4L2_AV1_SEQUENCE_FLAG_STILL_PICTURE : 0) | (seq_hdr->use_128x128_superblock ? V4L2_AV1_SEQUENCE_FLAG_USE_128X128_SUPERBLOCK : 0) | (seq_hdr->enable_filter_intra ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_FILTER_INTRA : 0) | (seq_hdr->enable_intra_edge_filter ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_INTRA_EDGE_FILTER : 0) | (seq_hdr->enable_interintra_compound ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_INTERINTRA_COMPOUND : 0) | (seq_hdr->enable_masked_compound ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_MASKED_COMPOUND : 0) | (seq_hdr->enable_warped_motion ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_WARPED_MOTION : 0) | (seq_hdr->enable_dual_filter ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_DUAL_FILTER : 0) | (seq_hdr->enable_order_hint ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_ORDER_HINT : 0) | (seq_hdr->enable_jnt_comp ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_JNT_COMP : 0) | (seq_hdr->enable_ref_frame_mvs ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_REF_FRAME_MVS : 0) | (seq_hdr->enable_superres ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_SUPERRES : 0) | (seq_hdr->enable_cdef ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_CDEF : 0) | (seq_hdr->enable_restoration ? V4L2_AV1_SEQUENCE_FLAG_ENABLE_RESTORATION : 0) | (seq_hdr->color_config.mono_chrome ? V4L2_AV1_SEQUENCE_FLAG_MONO_CHROME : 0) | (seq_hdr->color_config.color_range ? V4L2_AV1_SEQUENCE_FLAG_COLOR_RANGE : 0) | (seq_hdr->color_config.subsampling_x ? V4L2_AV1_SEQUENCE_FLAG_SUBSAMPLING_X : 0) | (seq_hdr->color_config.subsampling_y ? V4L2_AV1_SEQUENCE_FLAG_SUBSAMPLING_Y : 0) | (seq_hdr->film_grain_params_present ? V4L2_AV1_SEQUENCE_FLAG_FILM_GRAIN_PARAMS_PRESENT : 0) | (seq_hdr->color_config.separate_uv_delta_q ? V4L2_AV1_SEQUENCE_FLAG_SEPARATE_UV_DELTA_Q : 0), .seq_profile = seq_hdr->seq_profile, .order_hint_bits = seq_hdr->order_hint_bits, .bit_depth = seq_hdr->bit_depth, .max_frame_width_minus_1 = seq_hdr->max_frame_width_minus_1, .max_frame_height_minus_1 = seq_hdr->max_frame_height_minus_1, }; /* *INDENT-ON* */ } static void gst_v4l2_codec_av1_fill_refs (GstV4l2CodecAV1Dec * self, const GstAV1FrameHeaderOBU * frame_hdr, const GstAV1Dpb * reference_frames) { gint i; G_STATIC_ASSERT (G_N_ELEMENTS (self->v4l2_frame.reference_frame_ts) == G_N_ELEMENTS (reference_frames->pic_list)); G_STATIC_ASSERT (sizeof (self->v4l2_frame.ref_frame_idx) == sizeof (frame_hdr->ref_frame_idx)); g_return_if_fail (reference_frames != NULL); for (i = 0; i < G_N_ELEMENTS (reference_frames->pic_list); i++) { GstAV1Picture *ref_pic = reference_frames->pic_list[i]; /* the decoder might not have filled all slots in the first few frames */ self->v4l2_frame.reference_frame_ts[i] = ref_pic ? GST_CODEC_PICTURE_TS_NS (ref_pic) : 0; } memcpy (self->v4l2_frame.ref_frame_idx, frame_hdr->ref_frame_idx, sizeof (frame_hdr->ref_frame_idx)); } static void gst_v4l2_codec_av1_fill_tile_info (GstV4l2CodecAV1Dec * self, const GstAV1TileInfo * ti) { struct v4l2_av1_tile_info *v4l2_ti = &self->v4l2_frame.tile_info; G_STATIC_ASSERT (sizeof (v4l2_ti->mi_col_starts) == sizeof (ti->mi_col_starts)); G_STATIC_ASSERT (sizeof (v4l2_ti->mi_row_starts) == sizeof (ti->mi_row_starts)); G_STATIC_ASSERT (sizeof (v4l2_ti->width_in_sbs_minus_1) == sizeof (ti->width_in_sbs_minus_1)); G_STATIC_ASSERT (sizeof (v4l2_ti->height_in_sbs_minus_1) == sizeof (ti->height_in_sbs_minus_1)); memcpy (v4l2_ti->mi_col_starts, ti->mi_col_starts, sizeof (v4l2_ti->mi_col_starts)); memcpy (v4l2_ti->mi_row_starts, ti->mi_row_starts, sizeof (v4l2_ti->mi_row_starts)); memcpy (v4l2_ti->width_in_sbs_minus_1, ti->width_in_sbs_minus_1, sizeof (v4l2_ti->width_in_sbs_minus_1)); memcpy (v4l2_ti->height_in_sbs_minus_1, ti->height_in_sbs_minus_1, sizeof (v4l2_ti->height_in_sbs_minus_1)); } static void gst_v4l2_codec_av1_fill_loop_filter (GstV4l2CodecAV1Dec * self, const GstAV1LoopFilterParams * lf) { struct v4l2_av1_loop_filter *v4l2_lf = &self->v4l2_frame.loop_filter; G_STATIC_ASSERT (sizeof (v4l2_lf->level) == sizeof (lf->loop_filter_level)); G_STATIC_ASSERT (sizeof (v4l2_lf->ref_deltas) == sizeof (lf->loop_filter_ref_deltas)); G_STATIC_ASSERT (sizeof (v4l2_lf->mode_deltas) == sizeof (lf->loop_filter_mode_deltas)); memcpy (v4l2_lf->level, lf->loop_filter_level, sizeof (v4l2_lf->level)); memcpy (v4l2_lf->ref_deltas, lf->loop_filter_ref_deltas, sizeof (v4l2_lf->ref_deltas)); memcpy (v4l2_lf->mode_deltas, lf->loop_filter_mode_deltas, sizeof (v4l2_lf->mode_deltas)); } static void gst_v4l2_codec_av1_fill_segmentation (GstV4l2CodecAV1Dec * self, const GstAV1SegmenationParams * seg) { struct v4l2_av1_segmentation *v4l2_seg = &self->v4l2_frame.segmentation; guint32 i; guint32 j; G_STATIC_ASSERT (sizeof (v4l2_seg->feature_data) == sizeof (seg->feature_data)); for (i = 0; i < G_N_ELEMENTS (v4l2_seg->feature_enabled); i++) for (j = 0; j < V4L2_AV1_SEG_LVL_MAX; j++) v4l2_seg->feature_enabled[i] |= (seg->feature_enabled[i][j] << j); memcpy (v4l2_seg->feature_data, seg->feature_data, sizeof (v4l2_seg->feature_data)); } static void gst_v4l2_codec_av1_fill_cdef (GstV4l2CodecAV1Dec * self, const GstAV1CDEFParams * cdef) { struct v4l2_av1_cdef *v4l2_cdef = &self->v4l2_frame.cdef; G_STATIC_ASSERT (sizeof (v4l2_cdef->y_pri_strength) == sizeof (cdef->cdef_y_pri_strength)); G_STATIC_ASSERT (sizeof (v4l2_cdef->y_sec_strength) == sizeof (cdef->cdef_y_sec_strength)); G_STATIC_ASSERT (sizeof (v4l2_cdef->uv_pri_strength) == sizeof (cdef->cdef_uv_pri_strength)); G_STATIC_ASSERT (sizeof (v4l2_cdef->uv_sec_strength) == sizeof (cdef->cdef_uv_sec_strength)); memcpy (v4l2_cdef->y_pri_strength, cdef->cdef_y_pri_strength, sizeof (v4l2_cdef->y_pri_strength)); memcpy (v4l2_cdef->y_sec_strength, cdef->cdef_y_sec_strength, sizeof (v4l2_cdef->y_sec_strength)); memcpy (v4l2_cdef->uv_pri_strength, cdef->cdef_uv_pri_strength, sizeof (v4l2_cdef->uv_pri_strength)); memcpy (v4l2_cdef->uv_sec_strength, cdef->cdef_uv_sec_strength, sizeof (v4l2_cdef->uv_sec_strength)); } static void gst_v4l2_codec_av1_fill_loop_restoration (GstV4l2CodecAV1Dec * self, const GstAV1LoopRestorationParams * lr) { struct v4l2_av1_loop_restoration *v4l2_lr = &self->v4l2_frame.loop_restoration; G_STATIC_ASSERT (sizeof (v4l2_lr->loop_restoration_size) == sizeof (lr->loop_restoration_size)); memcpy (v4l2_lr->loop_restoration_size, lr->loop_restoration_size, sizeof (v4l2_lr->loop_restoration_size)); } static void gst_v4l2_codec_av1_fill_global_motion (GstV4l2CodecAV1Dec * self, const GstAV1GlobalMotionParams * gm) { struct v4l2_av1_global_motion *v4l2_gm = &self->v4l2_frame.global_motion; gint i; G_STATIC_ASSERT (sizeof (v4l2_gm->type) == sizeof (gm->gm_type)); G_STATIC_ASSERT (sizeof (v4l2_gm->params) == sizeof (gm->gm_params)); G_STATIC_ASSERT (G_N_ELEMENTS (v4l2_gm->flags) == GST_AV1_SEG_LVL_MAX); for (i = 0; i < G_N_ELEMENTS (v4l2_gm->flags); i++) { v4l2_gm->flags[i] = (gm->is_global[i] ? V4L2_AV1_GLOBAL_MOTION_FLAG_IS_GLOBAL : 0) | (gm->is_rot_zoom[i] ? V4L2_AV1_GLOBAL_MOTION_FLAG_IS_ROT_ZOOM : 0) | (gm->is_translation[i] ? V4L2_AV1_GLOBAL_MOTION_FLAG_IS_TRANSLATION : 0); switch (gm->gm_type[i]) { case GST_AV1_WARP_MODEL_IDENTITY: v4l2_gm->type[i] = V4L2_AV1_WARP_MODEL_IDENTITY; break; case GST_AV1_WARP_MODEL_TRANSLATION: v4l2_gm->type[i] = V4L2_AV1_WARP_MODEL_TRANSLATION; break; case GST_AV1_WARP_MODEL_ROTZOOM: v4l2_gm->type[i] = V4L2_AV1_WARP_MODEL_ROTZOOM; break; case GST_AV1_WARP_MODEL_AFFINE: v4l2_gm->type[i] = V4L2_AV1_WARP_MODEL_AFFINE; break; } v4l2_gm->invalid |= (gm->invalid[i] << i); } memcpy (v4l2_gm->type, gm->gm_type, sizeof (v4l2_gm->type)); memcpy (v4l2_gm->params, gm->gm_params, sizeof (v4l2_gm->params)); } static void gst_v4l2_codec_av1_fill_film_grain (GstV4l2CodecAV1Dec * self, const GstAV1FilmGrainParams * fg) { struct v4l2_ctrl_av1_film_grain *v4l2_fg = &self->v4l2_film_grain; *v4l2_fg = (struct v4l2_ctrl_av1_film_grain) { .flags = (fg->apply_grain ? V4L2_AV1_FILM_GRAIN_FLAG_APPLY_GRAIN : 0) | (fg->update_grain ? V4L2_AV1_FILM_GRAIN_FLAG_UPDATE_GRAIN : 0) | (fg->chroma_scaling_from_luma ? V4L2_AV1_FILM_GRAIN_FLAG_CHROMA_SCALING_FROM_LUMA : 0) | (fg->overlap_flag ? V4L2_AV1_FILM_GRAIN_FLAG_OVERLAP : 0) | (fg->clip_to_restricted_range ? V4L2_AV1_FILM_GRAIN_FLAG_CLIP_TO_RESTRICTED_RANGE : 0), .grain_seed = fg->grain_seed, .film_grain_params_ref_idx = fg->film_grain_params_ref_idx, .num_y_points = fg->num_y_points, .num_cb_points = fg->num_cb_points, .num_cr_points = fg->num_cr_points, .grain_scaling_minus_8 = fg->grain_scaling_minus_8, .ar_coeff_lag = fg->ar_coeff_lag, .ar_coeff_shift_minus_6 = fg->ar_coeff_shift_minus_6, .grain_scale_shift = fg->grain_scale_shift, .cb_mult = fg->cb_mult, .cb_luma_mult = fg->cb_luma_mult, .cb_offset = fg->cb_offset, .cr_mult = fg->cr_mult, .cr_luma_mult = fg->cr_luma_mult, .cr_offset = fg->cr_offset }; G_STATIC_ASSERT (sizeof (v4l2_fg->point_y_value) == sizeof (fg->point_y_value)); G_STATIC_ASSERT (sizeof (v4l2_fg->point_y_scaling) == sizeof (fg->point_y_scaling)); G_STATIC_ASSERT (sizeof (v4l2_fg->point_cb_value) == sizeof (fg->point_cb_value)); G_STATIC_ASSERT (sizeof (v4l2_fg->point_cb_scaling) == sizeof (fg->point_cb_scaling)); G_STATIC_ASSERT (sizeof (v4l2_fg->point_cr_value) == sizeof (fg->point_cr_value)); G_STATIC_ASSERT (sizeof (v4l2_fg->point_cr_scaling) == sizeof (fg->point_cr_scaling)); G_STATIC_ASSERT (sizeof (v4l2_fg->ar_coeffs_y_plus_128) == sizeof (fg->ar_coeffs_y_plus_128)); G_STATIC_ASSERT (sizeof (v4l2_fg->ar_coeffs_cb_plus_128) == sizeof (fg->ar_coeffs_cb_plus_128)); G_STATIC_ASSERT (sizeof (v4l2_fg->ar_coeffs_cr_plus_128) == sizeof (fg->ar_coeffs_cr_plus_128)); memcpy (v4l2_fg->point_y_value, fg->point_y_value, sizeof (v4l2_fg->point_y_value)); memcpy (v4l2_fg->point_y_scaling, fg->point_y_scaling, sizeof (v4l2_fg->point_y_scaling)); memcpy (v4l2_fg->point_cb_value, fg->point_cb_value, sizeof (v4l2_fg->point_cb_value)); memcpy (v4l2_fg->point_cb_scaling, fg->point_cb_scaling, sizeof (v4l2_fg->point_cb_scaling)); memcpy (v4l2_fg->point_cr_value, fg->point_cr_value, sizeof (v4l2_fg->point_cr_value)); memcpy (v4l2_fg->point_cr_scaling, fg->point_cr_scaling, sizeof (v4l2_fg->point_cr_scaling)); memcpy (v4l2_fg->ar_coeffs_y_plus_128, fg->ar_coeffs_y_plus_128, sizeof (v4l2_fg->ar_coeffs_y_plus_128)); memcpy (v4l2_fg->ar_coeffs_cb_plus_128, fg->ar_coeffs_cb_plus_128, sizeof (v4l2_fg->ar_coeffs_cb_plus_128)); memcpy (v4l2_fg->ar_coeffs_cr_plus_128, fg->ar_coeffs_cr_plus_128, sizeof (v4l2_fg->ar_coeffs_cr_plus_128)); } static void gst_v4l2_codec_av1_dec_fill_frame_hdr (GstV4l2CodecAV1Dec * self, const GstAV1Picture * pic, const GstAV1Dpb * reference_frames) { const GstAV1FrameHeaderOBU *f = &pic->frame_hdr; const GstAV1TileInfo *ti = &f->tile_info; const GstAV1QuantizationParams *q = &f->quantization_params; const GstAV1SegmenationParams *seg = &f->segmentation_params; /* FIXME: send patch upstream to fix spelling on the parser s/segmenation/segmentation */ const GstAV1LoopFilterParams *lf = &f->loop_filter_params; const GstAV1LoopRestorationParams *lr = &f->loop_restoration_params; guint i; G_STATIC_ASSERT (sizeof (self->v4l2_frame. loop_restoration.frame_restoration_type) == sizeof (lr->frame_restoration_type)); G_STATIC_ASSERT (sizeof (self->v4l2_frame.buffer_removal_time) == sizeof (f->buffer_removal_time)); G_STATIC_ASSERT (sizeof (self->v4l2_frame.order_hints) == sizeof (f->order_hints)); G_STATIC_ASSERT (sizeof (self->v4l2_frame.skip_mode_frame) == sizeof (f->skip_mode_frame)); self->v4l2_frame = (struct v4l2_ctrl_av1_frame) { /* *INDENT-OFF* */ .flags = (f->show_frame ? V4L2_AV1_FRAME_FLAG_SHOW_FRAME : 0) | (f->showable_frame ? V4L2_AV1_FRAME_FLAG_SHOWABLE_FRAME : 0) | (f->error_resilient_mode ? V4L2_AV1_FRAME_FLAG_ERROR_RESILIENT_MODE : 0) | (f->disable_cdf_update ? V4L2_AV1_FRAME_FLAG_DISABLE_CDF_UPDATE : 0) | (f->allow_screen_content_tools ? V4L2_AV1_FRAME_FLAG_ALLOW_SCREEN_CONTENT_TOOLS : 0) | (f->force_integer_mv ? V4L2_AV1_FRAME_FLAG_FORCE_INTEGER_MV : 0) | (f->allow_intrabc ? V4L2_AV1_FRAME_FLAG_ALLOW_INTRABC : 0) | (f->use_superres ? V4L2_AV1_FRAME_FLAG_USE_SUPERRES : 0) | (f->allow_high_precision_mv ? V4L2_AV1_FRAME_FLAG_ALLOW_HIGH_PRECISION_MV : 0) | (f->is_motion_mode_switchable ? V4L2_AV1_FRAME_FLAG_IS_MOTION_MODE_SWITCHABLE : 0) | (f->use_ref_frame_mvs ? V4L2_AV1_FRAME_FLAG_USE_REF_FRAME_MVS : 0) | (f->disable_frame_end_update_cdf ? V4L2_AV1_FRAME_FLAG_DISABLE_FRAME_END_UPDATE_CDF : 0) | (f->allow_warped_motion ? V4L2_AV1_FRAME_FLAG_ALLOW_WARPED_MOTION : 0) | (f->reference_select ? V4L2_AV1_FRAME_FLAG_REFERENCE_SELECT : 0) | (f->reduced_tx_set ? V4L2_AV1_FRAME_FLAG_REDUCED_TX_SET : 0) | (f->skip_mode_frame[0] > 0 ? V4L2_AV1_FRAME_FLAG_SKIP_MODE_ALLOWED : 0) | (f->skip_mode_present ? V4L2_AV1_FRAME_FLAG_SKIP_MODE_PRESENT : 0) | (f->frame_size_override_flag ? V4L2_AV1_FRAME_FLAG_FRAME_SIZE_OVERRIDE : 0) | (f->buffer_removal_time_present_flag ? V4L2_AV1_FRAME_FLAG_BUFFER_REMOVAL_TIME_PRESENT : 0) | (f->frame_refs_short_signaling ? V4L2_AV1_FRAME_FLAG_FRAME_REFS_SHORT_SIGNALING : 0), .order_hint = f->order_hint, .superres_denom = f->superres_denom, .upscaled_width = f->upscaled_width, .frame_width_minus_1 = f->frame_width - 1, .frame_height_minus_1 = f->frame_height - 1, .render_width_minus_1 = f->render_width - 1, .render_height_minus_1 = f->render_height - 1, .current_frame_id = f->current_frame_id, .primary_ref_frame = f->primary_ref_frame, .refresh_frame_flags = f->refresh_frame_flags, .tile_info = (struct v4l2_av1_tile_info) { .flags = (f->tile_info.uniform_tile_spacing_flag ? V4L2_AV1_TILE_INFO_FLAG_UNIFORM_TILE_SPACING : 0), .tile_size_bytes = f->tile_info.tile_size_bytes, .context_update_tile_id = f->tile_info.context_update_tile_id, .tile_cols = f->tile_info.tile_cols, .tile_rows = f->tile_info.tile_rows, }, .quantization = (struct v4l2_av1_quantization) { .flags = (q->diff_uv_delta ? V4L2_AV1_QUANTIZATION_FLAG_DIFF_UV_DELTA : 0) | (q->using_qmatrix ? V4L2_AV1_QUANTIZATION_FLAG_USING_QMATRIX : 0) | (q->delta_q_present ? V4L2_AV1_QUANTIZATION_FLAG_DELTA_Q_PRESENT : 0), .base_q_idx = q->base_q_idx, .delta_q_y_dc = q->delta_q_y_dc, .delta_q_u_dc = q->delta_q_u_dc, .delta_q_u_ac = q->delta_q_u_ac, .delta_q_v_dc = q->delta_q_v_dc, .delta_q_v_ac = q->delta_q_v_ac, .qm_y = q->qm_y, .qm_u = q->qm_u, .qm_v = q->qm_v, .delta_q_res = q->delta_q_res, }, .segmentation = (struct v4l2_av1_segmentation) { .flags = (seg->segmentation_enabled ? V4L2_AV1_SEGMENTATION_FLAG_ENABLED : 0) | (seg->segmentation_update_map ? V4L2_AV1_SEGMENTATION_FLAG_UPDATE_MAP : 0) | (seg->segmentation_temporal_update ? V4L2_AV1_SEGMENTATION_FLAG_TEMPORAL_UPDATE : 0) | (seg->segmentation_update_data ? V4L2_AV1_SEGMENTATION_FLAG_UPDATE_DATA : 0) | (seg->seg_id_pre_skip ? V4L2_AV1_SEGMENTATION_FLAG_SEG_ID_PRE_SKIP : 0), .last_active_seg_id = seg->last_active_seg_id, }, .loop_filter = (struct v4l2_av1_loop_filter) { .flags = (lf->loop_filter_delta_enabled ? V4L2_AV1_LOOP_FILTER_FLAG_DELTA_ENABLED : 0) | (lf->loop_filter_delta_update ? V4L2_AV1_LOOP_FILTER_FLAG_DELTA_UPDATE : 0) | (lf->delta_lf_present ? V4L2_AV1_LOOP_FILTER_FLAG_DELTA_LF_PRESENT : 0) | (lf->delta_lf_multi ? V4L2_AV1_LOOP_FILTER_FLAG_DELTA_LF_MULTI : 0), .sharpness = lf->loop_filter_sharpness, .delta_lf_res = lf->delta_lf_res, }, .cdef = (struct v4l2_av1_cdef) { .damping_minus_3 = f->cdef_params.cdef_damping - 3, /* FIXME: is minus 3 really needed? */ .bits = f->cdef_params.cdef_bits, }, .loop_restoration = (struct v4l2_av1_loop_restoration) { .flags = (lr->uses_lr ? V4L2_AV1_LOOP_RESTORATION_FLAG_USES_LR : 0) | (lr->frame_restoration_type[1] ? V4L2_AV1_LOOP_RESTORATION_FLAG_USES_CHROMA_LR : 0), .lr_unit_shift = lr->lr_unit_shift, .lr_uv_shift = lr->lr_uv_shift, } /* *INDENT-ON* */ }; switch (f->frame_type) { case GST_AV1_KEY_FRAME: self->v4l2_frame.frame_type = V4L2_AV1_KEY_FRAME; break; case GST_AV1_INTER_FRAME: self->v4l2_frame.frame_type = V4L2_AV1_INTER_FRAME; break; case GST_AV1_INTRA_ONLY_FRAME: self->v4l2_frame.frame_type = V4L2_AV1_INTRA_ONLY_FRAME; break; case GST_AV1_SWITCH_FRAME: self->v4l2_frame.frame_type = V4L2_AV1_SWITCH_FRAME; break; } switch (f->interpolation_filter) { case GST_AV1_INTERPOLATION_FILTER_EIGHTTAP: self->v4l2_frame.interpolation_filter = V4L2_AV1_INTERPOLATION_FILTER_EIGHTTAP; break; case GST_AV1_INTERPOLATION_FILTER_EIGHTTAP_SMOOTH: self->v4l2_frame.interpolation_filter = V4L2_AV1_INTERPOLATION_FILTER_EIGHTTAP_SMOOTH; break; case GST_AV1_INTERPOLATION_FILTER_EIGHTTAP_SHARP: self->v4l2_frame.interpolation_filter = V4L2_AV1_INTERPOLATION_FILTER_EIGHTTAP_SHARP; break; case GST_AV1_INTERPOLATION_FILTER_BILINEAR: self->v4l2_frame.interpolation_filter = V4L2_AV1_INTERPOLATION_FILTER_BILINEAR; break; case GST_AV1_INTERPOLATION_FILTER_SWITCHABLE: self->v4l2_frame.interpolation_filter = V4L2_AV1_INTERPOLATION_FILTER_SWITCHABLE; break; } switch (f->tx_mode) { case GST_AV1_TX_MODE_ONLY_4x4: self->v4l2_frame.tx_mode = V4L2_AV1_TX_MODE_ONLY_4X4; break; case GST_AV1_TX_MODE_LARGEST: self->v4l2_frame.tx_mode = V4L2_AV1_TX_MODE_LARGEST; break; case GST_AV1_TX_MODE_SELECT: self->v4l2_frame.tx_mode = V4L2_AV1_TX_MODE_SELECT; break; } for (i = 0; i < V4L2_AV1_NUM_PLANES_MAX; i++) switch (lr->frame_restoration_type[i]) { case GST_AV1_FRAME_RESTORE_NONE: self->v4l2_frame.loop_restoration.frame_restoration_type[i] = V4L2_AV1_FRAME_RESTORE_NONE; break; case GST_AV1_FRAME_RESTORE_WIENER: self->v4l2_frame.loop_restoration.frame_restoration_type[i] = V4L2_AV1_FRAME_RESTORE_WIENER; break; case GST_AV1_FRAME_RESTORE_SGRPROJ: self->v4l2_frame.loop_restoration.frame_restoration_type[i] = V4L2_AV1_FRAME_RESTORE_SGRPROJ; break; case GST_AV1_FRAME_RESTORE_SWITCHABLE: self->v4l2_frame.loop_restoration.frame_restoration_type[i] = V4L2_AV1_FRAME_RESTORE_SWITCHABLE; break; } gst_v4l2_codec_av1_fill_refs (self, f, reference_frames); gst_v4l2_codec_av1_fill_tile_info (self, ti); gst_v4l2_codec_av1_fill_segmentation (self, seg); gst_v4l2_codec_av1_fill_loop_filter (self, lf); gst_v4l2_codec_av1_fill_cdef (self, &f->cdef_params); gst_v4l2_codec_av1_fill_loop_restoration (self, &f->loop_restoration_params); gst_v4l2_codec_av1_fill_global_motion (self, &f->global_motion_params); if (self->fill_film_grain) gst_v4l2_codec_av1_fill_film_grain (self, &f->film_grain_params); memcpy (self->v4l2_frame.buffer_removal_time, f->buffer_removal_time, sizeof (self->v4l2_frame.buffer_removal_time)); memcpy (self->v4l2_frame.order_hints, f->order_hints, sizeof (self->v4l2_frame.order_hints)); memcpy (self->v4l2_frame.skip_mode_frame, f->skip_mode_frame, sizeof (self->v4l2_frame.skip_mode_frame)); } static GstFlowReturn gst_v4l2_codec_av1_dec_new_sequence (GstAV1Decoder * decoder, const GstAV1SequenceHeaderOBU * seq_hdr, gint max_dpb_size) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); /* we'll use this as a hint to allow smaller resolution to be decoded * without sending new caps, but some better signalling would be nice from * the base class. */ self->operating_point_idc = seq_hdr->operating_points[0].idc; gst_v4l2_codec_av1_dec_fill_sequence_params (self, seq_hdr); self->min_pool_size = max_dpb_size; self->need_sequence = TRUE; return GST_FLOW_OK; } static void gst_v4l2_codec_av1_dec_reset_picture (GstV4l2CodecAV1Dec * self) { gst_v4l2_codec_av1_reset_bitstream (self); g_array_set_size (self->tile_group_entries, 0); } static GstFlowReturn gst_v4l2_codec_av1_dec_new_picture (GstAV1Decoder * decoder, GstVideoCodecFrame * frame, GstAV1Picture * picture) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GstAV1FrameHeaderOBU *frame_hdr = &picture->frame_hdr; struct v4l2_ctrl_av1_sequence *seq_hdr = &self->v4l2_sequence; gboolean negotiation_needed = FALSE; gint max_width; gint max_height; max_width = seq_hdr->max_frame_width_minus_1 + 1; max_height = seq_hdr->max_frame_height_minus_1 + 1; if (self->vinfo.finfo->format == GST_VIDEO_FORMAT_UNKNOWN) negotiation_needed = TRUE; /* FIXME the base class could signal this, but let's assume that when we * have spatial layers, that smaller resolution will never be shown, and * that the max size can be assumed to be render size. */ if ((self->operating_point_idc >> 8)) { if (self->frame_width != max_width || self->frame_height != max_height) { self->frame_width = self->render_width = max_width; self->frame_height = self->render_height = max_height; negotiation_needed = TRUE; GST_INFO_OBJECT (self, "max {width|height} changed to %dx%d", self->frame_width, self->frame_height); } if (self->frame_height < frame_hdr->frame_height || self->frame_width < frame_hdr->upscaled_width) { GST_ELEMENT_ERROR (self, CORE, NEGOTIATION, ("SVC AV1 frame is larger then sequence max size."), ("Frame have size %dx%d but the max size is %dx%d", frame_hdr->upscaled_width, frame_hdr->frame_height, max_width, max_height)); return GST_FLOW_ERROR; } } else if (self->frame_width != frame_hdr->upscaled_width || self->frame_height != frame_hdr->frame_height || self->render_width != frame_hdr->render_width || self->render_height != frame_hdr->render_height) { self->frame_width = frame_hdr->upscaled_width; self->frame_height = frame_hdr->frame_height; self->render_width = frame_hdr->render_width; self->render_height = frame_hdr->render_height; negotiation_needed = TRUE; GST_INFO_OBJECT (self, "frame {width|height} changed to %dx%d", self->frame_width, self->frame_height); GST_INFO_OBJECT (self, "render {width|height} changed to %dx%d", self->render_width, self->render_height); } if (self->bit_depth != seq_hdr->bit_depth) { GST_DEBUG_OBJECT (self, "bit-depth changed from %d to %d", self->bit_depth, seq_hdr->bit_depth); self->bit_depth = seq_hdr->bit_depth; negotiation_needed = TRUE; } if (self->profile != GST_AV1_PROFILE_UNDEFINED && seq_hdr->seq_profile != self->profile) { GST_DEBUG_OBJECT (self, "profile changed from %d to %d", self->profile, seq_hdr->seq_profile); self->profile = seq_hdr->seq_profile; negotiation_needed = TRUE; } if (seq_hdr->bit_depth != self->bit_depth) { GST_DEBUG_OBJECT (self, "bit-depth changed from %d to %d", self->bit_depth, seq_hdr->bit_depth); self->bit_depth = seq_hdr->bit_depth; negotiation_needed = TRUE; } if (negotiation_needed) { if (frame_hdr->frame_type != GST_AV1_KEY_FRAME) { GST_ERROR_OBJECT (self, "Inter-frame resolution changes are not yet supported in v4l2"); return GST_FLOW_ERROR; } gst_v4l2_codec_av1_dec_streamoff (self); if (!gst_video_decoder_negotiate (GST_VIDEO_DECODER (self))) { GST_ERROR_OBJECT (self, "Failed to negotiate with downstream"); return GST_FLOW_ERROR; } /* Check if we can zero-copy buffers */ if (!self->has_videometa) { GstVideoInfo ref_vinfo; gint i; gst_video_info_set_format (&ref_vinfo, GST_VIDEO_INFO_FORMAT (&self->vinfo), self->render_width, self->render_height); for (i = 0; i < GST_VIDEO_INFO_N_PLANES (&self->vinfo); i++) { if (self->vinfo.stride[i] != ref_vinfo.stride[i] || self->vinfo.offset[i] != ref_vinfo.offset[i]) { GST_WARNING_OBJECT (self, "GstVideoMeta support required, copying frames."); self->copy_frames = TRUE; break; } } } else { self->copy_frames = FALSE; } } /* * if for any reason the base class has dropped the frame midway through * decoding, then make sure we start off with a clean slate and that the * GstMemory is unmapped. */ gst_v4l2_codec_av1_dec_reset_picture (self); return gst_v4l2_codec_av1_dec_ensure_bitstream (self); } static GstAV1Picture * gst_v4l2_codec_av1_dec_duplicate_picture (GstAV1Decoder * decoder, GstVideoCodecFrame * frame, GstAV1Picture * picture) { GstAV1Picture *new_picture; GST_DEBUG_OBJECT (decoder, "Duplicate picture %u", GST_CODEC_PICTURE_FRAME_NUMBER (picture)); new_picture = gst_av1_picture_new (); new_picture->frame_hdr = picture->frame_hdr; GST_CODEC_PICTURE_COPY_FRAME_NUMBER (new_picture, picture); if (GST_MINI_OBJECT_FLAG_IS_SET (picture, FLAG_PICTURE_HOLDS_BUFFER)) { GstBuffer *output_buffer = gst_av1_picture_get_user_data (picture); if (output_buffer) { frame->output_buffer = gst_buffer_ref (output_buffer); gst_av1_picture_set_user_data (new_picture, gst_buffer_ref (frame->output_buffer), (GDestroyNotify) gst_buffer_unref); } GST_MINI_OBJECT_FLAG_SET (new_picture, FLAG_PICTURE_HOLDS_BUFFER); } else { GstV4l2Request *request = gst_av1_picture_get_user_data (picture); gst_av1_picture_set_user_data (new_picture, gst_v4l2_request_ref (request), (GDestroyNotify) gst_v4l2_request_unref); frame->output_buffer = gst_v4l2_request_dup_pic_buf (request); } return new_picture; } static GstFlowReturn gst_v4l2_codec_av1_dec_start_picture (GstAV1Decoder * decoder, GstAV1Picture * picture, GstAV1Dpb * dpb) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); gst_v4l2_codec_av1_dec_fill_frame_hdr (self, picture, dpb); return GST_FLOW_OK; } static GstFlowReturn _copy_into_bitstream_buffer (GstV4l2CodecAV1Dec * self, guint8 * src, gsize len) { guint8 *bitstream_ptr = self->bitstream_map.data + self->bitstream_map.size; if (self->bitstream_map.size + len > self->bitstream_map.maxsize) { GST_ELEMENT_ERROR (self, RESOURCE, NO_SPACE_LEFT, ("Not enough space left on the bitstream buffer."), (NULL)); gst_v4l2_codec_av1_dec_reset_picture (self); return GST_FLOW_ERROR; } memcpy (bitstream_ptr, src, len); self->bitstream_map.size += len; return GST_FLOW_OK; } static GstFlowReturn gst_v4l2_codec_av1_dec_decode_tile (GstAV1Decoder * decoder, GstAV1Picture * picture, GstAV1Tile * tile) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GstAV1TileGroupOBU *tile_group = &tile->tile_group; struct v4l2_ctrl_av1_tile_group_entry v4l2_tile_group_entry; guint32 obu_offset = self->bitstream_map.size; gint i; for (i = tile_group->tg_start; i <= tile_group->tg_end; i++) { v4l2_tile_group_entry = (struct v4l2_ctrl_av1_tile_group_entry) { .tile_offset = tile_group->entry[i].tile_offset + obu_offset, .tile_size = tile_group->entry[i].tile_size, .tile_row = tile_group->entry[i].tile_row, .tile_col = tile_group->entry[i].tile_col, }; GST_DEBUG_OBJECT (self, "Decoded tile group entry %d of size %d at offset %d, rows: %d, cols %d", self->tile_group_entries->len, v4l2_tile_group_entry.tile_size, v4l2_tile_group_entry.tile_offset, v4l2_tile_group_entry.tile_row, v4l2_tile_group_entry.tile_col); g_array_append_val (self->tile_group_entries, v4l2_tile_group_entry); } return _copy_into_bitstream_buffer (self, tile->obu.data, tile->obu.obu_size); } static GstFlowReturn gst_v4l2_codec_av1_dec_end_picture (GstAV1Decoder * decoder, GstAV1Picture * picture) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GstVideoCodecFrame *frame; GstV4l2Request *request; GstFlowReturn flow_ret; guint count = 1; gsize bytesused; struct v4l2_ctrl_av1_tile_group_entry tge = { }; /* *INDENT-OFF* */ struct v4l2_ext_control decode_params_control[] = { { .id = V4L2_CID_STATELESS_AV1_FRAME, .ptr = &self->v4l2_frame, .size = sizeof(self->v4l2_frame), }, {}, /* tile groups */ {}, /* tile group entries */ {}, /* sequence */ {}, /* film grain */ }; /* *INDENT-ON* */ if (self->tile_group_entries->len > 0) { decode_params_control[count++] = (struct v4l2_ext_control) { .id = V4L2_CID_STATELESS_AV1_TILE_GROUP_ENTRY, .ptr = self->tile_group_entries->data, .size = g_array_get_element_size (self->tile_group_entries) * self->tile_group_entries->len, }; } else { decode_params_control[count++] = (struct v4l2_ext_control) { .id = V4L2_CID_STATELESS_AV1_TILE_GROUP_ENTRY, .ptr = &tge, .size = sizeof (tge), }; } if (self->need_sequence) { decode_params_control[count++] = (struct v4l2_ext_control) { .id = V4L2_CID_STATELESS_AV1_SEQUENCE, .ptr = &self->v4l2_sequence, .size = sizeof (self->v4l2_sequence), }; self->need_sequence = FALSE; } if (self->fill_film_grain) { decode_params_control[count++] = (struct v4l2_ext_control) { .id = V4L2_CID_STATELESS_AV1_FILM_GRAIN, .ptr = &self->v4l2_film_grain, .size = sizeof (self->v4l2_film_grain), }; } bytesused = self->bitstream_map.size; gst_memory_unmap (self->bitstream, &self->bitstream_map); self->bitstream_map = (GstMapInfo) GST_MAP_INFO_INIT; gst_memory_resize (self->bitstream, 0, bytesused); frame = gst_video_decoder_get_frame (GST_VIDEO_DECODER (self), GST_CODEC_PICTURE_FRAME_NUMBER (picture)); g_return_val_if_fail (frame, FALSE); flow_ret = gst_buffer_pool_acquire_buffer (GST_BUFFER_POOL (self->src_pool), &frame->output_buffer, NULL); if (flow_ret != GST_FLOW_OK) { if (flow_ret == GST_FLOW_FLUSHING) GST_DEBUG_OBJECT (self, "Frame decoding aborted, we are flushing."); else GST_ELEMENT_ERROR (self, RESOURCE, WRITE, ("No more picture buffer available."), (NULL)); goto fail; } request = gst_v4l2_decoder_alloc_request (self->decoder, GST_CODEC_PICTURE_FRAME_NUMBER (picture), self->bitstream, frame->output_buffer); gst_video_codec_frame_unref (frame); if (!request) { GST_ELEMENT_ERROR (self, RESOURCE, NO_SPACE_LEFT, ("Failed to allocate a media request object."), (NULL)); goto fail; } if (!gst_v4l2_decoder_set_controls (self->decoder, request, decode_params_control, count)) { GST_ELEMENT_ERROR (self, RESOURCE, WRITE, ("Driver did not accept the bitstream parameters."), (NULL)); goto fail; } if (!gst_v4l2_request_queue (request, 0)) { GST_ELEMENT_ERROR (self, RESOURCE, WRITE, ("Driver did not accept the decode request."), (NULL)); goto fail; } gst_av1_picture_set_user_data (picture, request, (GDestroyNotify) gst_v4l2_request_unref); gst_v4l2_codec_av1_dec_reset_picture (self); return GST_FLOW_OK; fail: gst_v4l2_codec_av1_dec_reset_picture (self); return GST_FLOW_ERROR; } static gboolean gst_v4l2_codec_av1_dec_copy_output_buffer (GstV4l2CodecAV1Dec * self, GstVideoCodecFrame * codec_frame) { GstVideoFrame src_frame; GstVideoFrame dest_frame; GstVideoInfo dest_vinfo; GstBuffer *buffer; gst_video_info_set_format (&dest_vinfo, GST_VIDEO_INFO_FORMAT (&self->vinfo), self->render_width, self->render_height); buffer = gst_video_decoder_allocate_output_buffer (GST_VIDEO_DECODER (self)); if (!buffer) goto fail; if (!gst_video_frame_map (&src_frame, &self->vinfo, codec_frame->output_buffer, GST_MAP_READ)) goto fail; if (!gst_video_frame_map (&dest_frame, &dest_vinfo, buffer, GST_MAP_WRITE)) { gst_video_frame_unmap (&dest_frame); goto fail; } /* gst_video_frame_copy can crop this, but does not know, so let make it * think it's all right */ GST_VIDEO_INFO_WIDTH (&src_frame.info) = self->render_width; GST_VIDEO_INFO_HEIGHT (&src_frame.info) = self->render_height; if (!gst_video_frame_copy (&dest_frame, &src_frame)) { gst_video_frame_unmap (&src_frame); gst_video_frame_unmap (&dest_frame); goto fail; } gst_video_frame_unmap (&src_frame); gst_video_frame_unmap (&dest_frame); gst_buffer_replace (&codec_frame->output_buffer, buffer); gst_buffer_unref (buffer); return TRUE; fail: GST_ERROR_OBJECT (self, "Failed copy output buffer."); return FALSE; } static GstFlowReturn gst_v4l2_codec_av1_dec_output_picture (GstAV1Decoder * decoder, GstVideoCodecFrame * frame, GstAV1Picture * picture) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (decoder); GstVideoDecoder *vdec = GST_VIDEO_DECODER (decoder); GstV4l2Request *request = NULL; GstCodecPicture *codec_picture = GST_CODEC_PICTURE (picture); gint ret; if (codec_picture->discont_state) { if (!gst_video_decoder_negotiate (vdec)) { GST_ERROR_OBJECT (vdec, "Could not re-negotiate with updated state"); return FALSE; } } GST_DEBUG_OBJECT (self, "Output picture %u", codec_picture->system_frame_number); if (!GST_MINI_OBJECT_FLAG_IS_SET (picture, FLAG_PICTURE_HOLDS_BUFFER)) request = gst_av1_picture_get_user_data (picture); if (request) { ret = gst_v4l2_request_set_done (request); if (ret == 0) { GST_ELEMENT_ERROR (self, STREAM, DECODE, ("Decoding frame took too long"), (NULL)); goto error; } else if (ret < 0) { GST_ELEMENT_ERROR (self, STREAM, DECODE, ("Decoding request failed: %s", g_strerror (errno)), (NULL)); goto error; } if (gst_v4l2_request_failed (request)) { GST_ELEMENT_ERROR (self, STREAM, DECODE, ("Failed to decode frame %u", codec_picture->system_frame_number), (NULL)); goto error; } /* Hold on reference buffers for the rest of the picture lifetime */ gst_av1_picture_set_user_data (picture, gst_buffer_ref (frame->output_buffer), (GDestroyNotify) gst_buffer_unref); GST_MINI_OBJECT_FLAG_SET (picture, FLAG_PICTURE_HOLDS_BUFFER); } /* This may happen if we duplicate a picture witch failed to decode */ if (!frame->output_buffer) { GST_ELEMENT_ERROR (self, STREAM, DECODE, ("Failed to decode frame %u", codec_picture->system_frame_number), (NULL)); goto error; } if (self->copy_frames) gst_v4l2_codec_av1_dec_copy_output_buffer (self, frame); gst_av1_picture_unref (picture); return gst_video_decoder_finish_frame (vdec, frame); error: gst_video_decoder_drop_frame (vdec, frame); gst_av1_picture_unref (picture); return GST_FLOW_ERROR; } static void gst_v4l2_codec_av1_dec_dispose (GObject * object) { GstV4l2CodecAV1Dec *self = GST_V4L2_CODEC_AV1_DEC (object); g_clear_object (&self->decoder); g_clear_pointer (&self->tile_group_entries, g_array_unref); G_OBJECT_CLASS (parent_class)->dispose (object); } static void gst_v4l2_codec_av1_dec_subclass_init (GstV4l2CodecAV1DecClass * klass, GstV4l2CodecDevice * device) { GObjectClass *gobject_class = G_OBJECT_CLASS (klass); GstElementClass *element_class = GST_ELEMENT_CLASS (klass); GstVideoDecoderClass *decoder_class = GST_VIDEO_DECODER_CLASS (klass); GstAV1DecoderClass *av1decoder_class = GST_AV1_DECODER_CLASS (klass); gobject_class->set_property = gst_v4l2_codec_av1_dec_set_property; gobject_class->get_property = gst_v4l2_codec_av1_dec_get_property; gobject_class->dispose = gst_v4l2_codec_av1_dec_dispose; gst_element_class_set_static_metadata (element_class, "V4L2 Stateless AV1 Video Decoder", "Codec/Decoder/Video/Hardware", "A V4L2 based AV1 video decoder", "Daniel Almeida "); gst_element_class_add_static_pad_template (element_class, &sink_template); gst_element_class_add_static_pad_template (element_class, &src_template); element_class->change_state = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_change_state); decoder_class->open = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_open); decoder_class->close = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_close); decoder_class->stop = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_stop); decoder_class->negotiate = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_negotiate); decoder_class->decide_allocation = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_decide_allocation); decoder_class->flush = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_flush); decoder_class->sink_event = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_sink_event); av1decoder_class->new_sequence = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_new_sequence); av1decoder_class->new_picture = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_new_picture); av1decoder_class->start_picture = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_start_picture); av1decoder_class->decode_tile = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_decode_tile); av1decoder_class->end_picture = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_end_picture); av1decoder_class->output_picture = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_output_picture); av1decoder_class->duplicate_picture = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_duplicate_picture); av1decoder_class->get_preferred_output_delay = GST_DEBUG_FUNCPTR (gst_v4l2_codec_av1_dec_get_preferred_output_delay); klass->device = device; gst_v4l2_decoder_install_properties (gobject_class, PROP_LAST, device); } static void gst_v4l2_codec_av1_dec_subinit (GstV4l2CodecAV1Dec * self, GstV4l2CodecAV1DecClass * klass) { self->decoder = gst_v4l2_decoder_new (klass->device); gst_video_info_init (&self->vinfo); self->tile_group_entries = g_array_new (FALSE, TRUE, sizeof (struct v4l2_ctrl_av1_tile_group_entry)); } static void gst_v4l2_codec_av1_dec_class_init (GstV4l2CodecAV1DecClass * klass) { } static void gst_v4l2_codec_av1_dec_init (GstV4l2CodecAV1Dec * self) { } void gst_v4l2_codec_av1_dec_register (GstPlugin * plugin, GstV4l2Decoder * decoder, GstV4l2CodecDevice * device, guint rank) { GstCaps *src_caps; GST_DEBUG_CATEGORY_INIT (v4l2_av1dec_debug, "v4l2codecs-av1dec", 0, "V4L2 stateless AV1 decoder"); if (!gst_v4l2_decoder_set_sink_fmt (decoder, V4L2_PIX_FMT_AV1_FRAME, 320, 240, 8)) return; src_caps = gst_v4l2_decoder_enum_src_formats (decoder, &static_src_caps); if (gst_caps_is_empty (src_caps)) { GST_WARNING ("Not registering AV1 decoder since it produces no " "supported format"); goto done; } /* TODO uncomment this when AV1 get included in Linus tree */ #if 0 version = gst_v4l2_decoder_get_version (decoder); if (version < V4L2_MIN_KERNEL_VERSION) GST_WARNING ("V4L2 API v%u.%u too old, at least v%u.%u required", (version >> 16) & 0xff, (version >> 8) & 0xff, V4L2_MIN_KERNEL_VER_MAJOR, V4L2_MIN_KERNEL_VER_MINOR); #endif if (!gst_v4l2_decoder_av1_api_check (decoder)) { GST_WARNING ("Not registering AV1 decoder as it failed ABI check."); goto done; } gst_v4l2_decoder_register (plugin, GST_TYPE_V4L2_CODEC_AV1_DEC, (GClassInitFunc) gst_v4l2_codec_av1_dec_subclass_init, gst_mini_object_ref (GST_MINI_OBJECT (device)), (GInstanceInitFunc) gst_v4l2_codec_av1_dec_subinit, "v4l2sl%sav1dec", device, rank, NULL); done: gst_caps_unref (src_caps); }