diff mbox series

[FFmpeg-devel,v2,15/23] lavc/hevcdec: implement decoding MV-HEVC

Message ID 20240916093211.10441-2-anton@khirnov.net
State New
Headers show
Series None | expand

Commit Message

Anton Khirnov Sept. 16, 2024, 9:28 a.m. UTC
At most two layers are supported.

Aspects of this work were sponsored by Vimeo and Meta.
---
Now not using AV_FRAME_FLAG_DISCARD to discard views that are decoded
but not output.
---
 Changelog                 |   1 +
 doc/decoders.texi         |  45 +++++
 libavcodec/hevc/hevcdec.c | 334 ++++++++++++++++++++++++++++++++++----
 libavcodec/hevc/hevcdec.h |  35 +++-
 libavcodec/hevc/refs.c    |  71 ++++++--
 5 files changed, 429 insertions(+), 57 deletions(-)

 
+    ref->base_layer_frame = (l != &s->layers[0] && s->layers[0].cur_frame) ?
+                            s->layers[0].cur_frame - s->layers[0].DPB : -1;
+
     if (s->sh.pic_output_flag)
         ref->flags = HEVC_FRAME_FLAG_OUTPUT | HEVC_FRAME_FLAG_SHORT_REF;
     else
@@ -176,33 +199,49 @@ static void unref_missing_refs(HEVCLayerContext *l)
     }
 }
 
-int ff_hevc_output_frames(HEVCContext *s, HEVCLayerContext *l,
+int ff_hevc_output_frames(HEVCContext *s,
+                          unsigned layers_active_decode, unsigned layers_active_output,
                           unsigned max_output, unsigned max_dpb, int discard)
 {
     while (1) {
-        int nb_dpb    = 0;
+        int nb_dpb[HEVC_VPS_MAX_LAYERS] = { 0 };
         int nb_output = 0;
         int min_poc   = INT_MAX;
-        int i, min_idx, ret = 0;
+        int min_layer = -1;
+        int min_idx, ret = 0;
 
-        for (i = 0; i < FF_ARRAY_ELEMS(l->DPB); i++) {
-            HEVCFrame *frame = &l->DPB[i];
-            if (frame->flags & HEVC_FRAME_FLAG_OUTPUT) {
-                nb_output++;
-                if (frame->poc < min_poc || nb_output == 1) {
-                    min_poc = frame->poc;
-                    min_idx = i;
+        for (int layer = 0; layer < FF_ARRAY_ELEMS(s->layers); layer++) {
+            HEVCLayerContext *l = &s->layers[layer];
+
+            if (!(layers_active_decode & (1 << layer)))
+                continue;
+
+            for (int i = 0; i < FF_ARRAY_ELEMS(l->DPB); i++) {
+                HEVCFrame *frame = &l->DPB[i];
+                if (frame->flags & HEVC_FRAME_FLAG_OUTPUT) {
+                    // nb_output counts AUs with an output-pending frame
+                    // in at least one layer
+                    if (!(frame->base_layer_frame >= 0 &&
+                          (s->layers[0].DPB[frame->base_layer_frame].flags & HEVC_FRAME_FLAG_OUTPUT)))
+                        nb_output++;
+                    if (min_layer < 0 || frame->poc < min_poc) {
+                        min_poc = frame->poc;
+                        min_idx = i;
+                        min_layer = layer;
+                    }
                 }
+                nb_dpb[layer] += !!frame->flags;
             }
-            nb_dpb += !!frame->flags;
         }
 
         if (nb_output > max_output ||
-            (nb_output && nb_dpb > max_dpb)) {
-            HEVCFrame *frame = &l->DPB[min_idx];
+            (nb_output &&
+             (nb_dpb[0] > max_dpb || nb_dpb[1] > max_dpb))) {
+            HEVCFrame *frame = &s->layers[min_layer].DPB[min_idx];
             AVFrame *f = frame->needs_fg ? frame->frame_grain : frame->f;
+            int output = !discard && (layers_active_output & (1 << min_layer));
 
-            if (!discard) {
+            if (output) {
                 f->pkt_dts = s->pkt_dts;
                 ret = ff_container_fifo_write(s->output_fifo, f);
             }
@@ -210,8 +249,8 @@ int ff_hevc_output_frames(HEVCContext *s, HEVCLayerContext *l,
             if (ret < 0)
                 return ret;
 
-            av_log(s->avctx, AV_LOG_DEBUG, "%s frame with POC %d.\n",
-                   discard ? "Discarded" : "Output", frame->poc);
+            av_log(s->avctx, AV_LOG_DEBUG, "%s frame with POC %d/%d.\n",
+                   output ? "Output" : "Discarded", min_layer, frame->poc);
             continue;
         }
         return 0;
diff mbox series

Patch

diff --git a/Changelog b/Changelog
index b6f91d7c8c..ff5d1b1bc8 100644
--- a/Changelog
+++ b/Changelog
@@ -21,6 +21,7 @@  version <next>:
 - MediaCodec AAC/AMR-NB/AMR-WB/MP3 decoding
 - YUV colorspace negotiation for codecs and filters, obsoleting the
   YUVJ pixel format
+- MV-HEVC decoding
 
 
 version 7.0:
diff --git a/doc/decoders.texi b/doc/decoders.texi
index 2fcc761d2f..17bb361ffa 100644
--- a/doc/decoders.texi
+++ b/doc/decoders.texi
@@ -38,6 +38,51 @@  Select an operating point of a scalable AV1 bitstream (0 - 31). Default is 0.
 
 @end table
 
+@section hevc
+HEVC (AKA ITU-T H.265 or ISO/IEC 23008-2) decoder.
+
+The decoder supports MV-HEVC multiview streams with at most two views. Views to
+be output are selected by supplying a list of view IDs to the decoder (the
+@option{view_ids} option). This option may be set either statically before
+decoder init, or from the @code{get_format()} callback - useful for the case
+when the view count or IDs change dynamically during decoding.
+
+Only the base layer is decoded by default.
+
+Note that if you are using the @code{ffmpeg} CLI tool, you should be using view
+specifiers as documented in its manual, rather than the options documented here.
+
+@subsection Options
+
+@table @option
+
+@item view_ids (MV-HEVC)
+Specify a list of view IDs that should be output. This option can also be set to
+a single '-1', which will cause all views defined in the VPS to be decoded and
+output.
+
+@item view_ids_available (MV-HEVC)
+This option may be read by the caller to retrieve an array of view IDs available
+in the active VPS. The array is empty for single-layer video.
+
+The value of this option is guaranteed to be accurate when read from the
+@code{get_format()} callback. It may also be set at other times (e.g. after
+opening the decoder), but the value is informational only and may be incorrect
+(e.g. when the stream contains multiple distinct VPS NALUs).
+
+@item view_pos_available (MV-HEVC)
+This option may be read by the caller to retrieve an array of view positions
+(left, right, or unspecified) available in the active VPS, as
+@code{AVStereo3DView} values. When the array is available, its elements apply to
+the corresponding elements of @option{view_ids_available}, i.e.
+@code{view_pos_available[i]} contains the position of view with ID
+@code{view_ids_available[i]}.
+
+Same validity restrictions as for @option{view_ids_available} apply to
+this option.
+
+@end table
+
 @section rawvideo
 
 Raw video decoder.
diff --git a/libavcodec/hevc/hevcdec.c b/libavcodec/hevc/hevcdec.c
index cbf763b8be..15828ef9eb 100644
--- a/libavcodec/hevc/hevcdec.c
+++ b/libavcodec/hevc/hevcdec.c
@@ -34,6 +34,7 @@ 
 #include "libavutil/mem.h"
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
+#include "libavutil/stereo3d.h"
 #include "libavutil/timecode.h"
 
 #include "aom_film_grain.h"
@@ -417,6 +418,109 @@  static int export_stream_params_from_sei(HEVCContext *s)
     return 0;
 }
 
+static int export_multilayer(HEVCContext *s, const HEVCVPS *vps)
+{
+    const HEVCSEITDRDI *tdrdi = &s->sei.tdrdi;
+
+    av_freep(&s->view_ids_available);
+    s->nb_view_ids_available = 0;
+    av_freep(&s->view_pos_available);
+    s->nb_view_pos_available = 0;
+
+    // don't export anything in the trivial case (1 layer, view id=0)
+    if (vps->nb_layers < 2 && !vps->view_id[0])
+        return 0;
+
+    s->view_ids_available = av_calloc(vps->nb_layers, sizeof(*s->view_ids_available));
+    if (!s->view_ids_available)
+        return AVERROR(ENOMEM);
+
+    if (tdrdi->num_ref_displays) {
+        s->view_pos_available = av_calloc(vps->nb_layers, sizeof(*s->view_pos_available));
+        if (!s->view_pos_available)
+            return AVERROR(ENOMEM);
+    }
+
+    for (int i = 0; i < vps->nb_layers; i++) {
+        s->view_ids_available[i] = vps->view_id[i];
+
+        if (s->view_pos_available) {
+            s->view_pos_available[i] = vps->view_id[i] == tdrdi->left_view_id[0]  ?
+                                       AV_STEREO3D_VIEW_LEFT                      :
+                                       vps->view_id[i] == tdrdi->right_view_id[0] ?
+                                       AV_STEREO3D_VIEW_RIGHT : AV_STEREO3D_VIEW_UNSPEC;
+        }
+    }
+    s->nb_view_ids_available = vps->nb_layers;
+    s->nb_view_pos_available = s->view_pos_available ? vps->nb_layers : 0;
+
+    return 0;
+}
+
+static int setup_multilayer(HEVCContext *s, const HEVCVPS *vps)
+{
+    unsigned layers_active_output = 0, highest_layer;
+
+    s->layers_active_output = 1;
+    s->layers_active_decode = 1;
+
+    // nothing requested - decode base layer only
+    if (!s->nb_view_ids)
+        return 0;
+
+    if (s->nb_view_ids == 1 && s->view_ids[0] == -1) {
+        layers_active_output = (1 << vps->nb_layers) - 1;
+    } else {
+        for (int i = 0; i < s->nb_view_ids; i++) {
+            int view_id   = s->view_ids[i];
+            int layer_idx = -1;
+
+            if (view_id < 0) {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Invalid view ID requested: %d\n", view_id);
+                return AVERROR(EINVAL);
+            }
+
+            for (int j = 0; j < vps->nb_layers; j++) {
+                if (vps->view_id[j] == view_id) {
+                    layer_idx = j;
+                    break;
+                }
+            }
+            if (layer_idx < 0) {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "View ID %d not present in VPS\n", view_id);
+                return AVERROR(EINVAL);
+            }
+            layers_active_output |= 1 << layer_idx;
+        }
+    }
+
+    if (!layers_active_output) {
+        av_log(s->avctx, AV_LOG_ERROR, "No layers selected\n");
+        return AVERROR_BUG;
+    }
+
+    highest_layer = ff_log2(layers_active_output);
+    if (highest_layer >= FF_ARRAY_ELEMS(s->layers)) {
+        av_log(s->avctx, AV_LOG_ERROR,
+               "Too many layers requested: %u\n", layers_active_output);
+        return AVERROR(EINVAL);
+    }
+
+    /* Assume a higher layer depends on all the lower ones.
+     * This is enforced in VPS parsing currently, this logic will need
+     * to be changed if we want to support more complex dependency structures.
+     */
+    s->layers_active_decode = (1 << (highest_layer + 1)) - 1;
+    s->layers_active_output = layers_active_output;
+
+    av_log(s->avctx, AV_LOG_DEBUG, "decode/output layers: %x/%x\n",
+           s->layers_active_decode, s->layers_active_output);
+
+    return 0;
+}
+
 static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps)
 {
 #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + \
@@ -428,6 +532,7 @@  static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps)
                      CONFIG_HEVC_VDPAU_HWACCEL + \
                      CONFIG_HEVC_VULKAN_HWACCEL)
     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
+    int ret;
 
     switch (sps->pix_fmt) {
     case AV_PIX_FMT_YUV420P:
@@ -547,7 +652,23 @@  static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps)
     *fmt++ = sps->pix_fmt;
     *fmt = AV_PIX_FMT_NONE;
 
-    return ff_get_format(s->avctx, pix_fmts);
+    // export multilayer information from active VPS to the caller,
+    // so it is available in get_format()
+    ret = export_multilayer(s, sps->vps);
+    if (ret < 0)
+        return ret;
+
+    ret = ff_get_format(s->avctx, pix_fmts);
+    if (ret < 0)
+        return ret;
+    s->avctx->pix_fmt = ret;
+
+    // set up multilayer decoding, if requested by caller
+    ret = setup_multilayer(s, sps->vps);
+    if (ret < 0)
+        return ret;
+
+    return 0;
 }
 
 static int set_sps(HEVCContext *s, HEVCLayerContext *l, const HEVCSPS *sps)
@@ -2948,13 +3069,60 @@  static int set_side_data(HEVCContext *s)
     return 0;
 }
 
-static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
+static int find_finish_setup_nal(const HEVCContext *s)
+{
+    int nal_idx = 0;
+
+    for (int i = nal_idx; i < s->pkt.nb_nals; i++) {
+        const H2645NAL *nal = &s->pkt.nals[i];
+        const int  layer_id = nal->nuh_layer_id;
+        GetBitContext    gb = nal->gb;
+
+        if (layer_id > HEVC_MAX_NUH_LAYER_ID || s->vps->layer_idx[layer_id] < 0 ||
+            !(s->layers_active_decode & (1 << s->vps->layer_idx[layer_id])))
+            continue;
+
+        switch (nal->type) {
+        case HEVC_NAL_TRAIL_R:
+        case HEVC_NAL_TRAIL_N:
+        case HEVC_NAL_TSA_N:
+        case HEVC_NAL_TSA_R:
+        case HEVC_NAL_STSA_N:
+        case HEVC_NAL_STSA_R:
+        case HEVC_NAL_BLA_W_LP:
+        case HEVC_NAL_BLA_W_RADL:
+        case HEVC_NAL_BLA_N_LP:
+        case HEVC_NAL_IDR_W_RADL:
+        case HEVC_NAL_IDR_N_LP:
+        case HEVC_NAL_CRA_NUT:
+        case HEVC_NAL_RADL_N:
+        case HEVC_NAL_RADL_R:
+        case HEVC_NAL_RASL_N:
+        case HEVC_NAL_RASL_R:
+            if (!get_bits1(&gb)) // first_slice_segment_in_pic_flag
+                continue;
+        case HEVC_NAL_VPS:
+        case HEVC_NAL_SPS:
+        case HEVC_NAL_PPS:
+            nal_idx = i;
+            break;
+        }
+    }
+
+    return nal_idx;
+}
+
+static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l,
+                            unsigned nal_idx)
 {
     const HEVCPPS *const pps = s->ps.pps_list[s->sh.pps_id];
     const HEVCSPS *const sps = pps->sps;
     int pic_size_in_ctb  = ((sps->width  >> sps->log2_min_cb_size) + 1) *
                            ((sps->height >> sps->log2_min_cb_size) + 1);
-    int new_sequence = IS_IDR(s) || IS_BLA(s) || s->last_eos;
+    int new_sequence = (l == &s->layers[0]) &&
+                       (IS_IDR(s) || IS_BLA(s) || s->last_eos);
+    int prev_layers_active_decode = s->layers_active_decode;
+    int prev_layers_active_output = s->layers_active_output;
     int ret;
 
     if (sps->vps != s->vps && l != &s->layers[0]) {
@@ -2965,7 +3133,32 @@  static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
 
     ff_refstruct_replace(&s->pps, pps);
     if (l->sps != sps) {
-        enum AVPixelFormat pix_fmt;
+        const HEVCSPS *sps_base = s->layers[0].sps;
+        enum AVPixelFormat pix_fmt = sps->pix_fmt;
+
+        if (l != &s->layers[0]) {
+            if (!sps_base) {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Access unit starts with a non-base layer frame\n");
+                return AVERROR_INVALIDDATA;
+            }
+
+            // Files produced by Vision Pro lack VPS extension VUI,
+            // so the secondary layer has no range information.
+            // This check avoids failing in such a case.
+            if (sps_base->pix_fmt == AV_PIX_FMT_YUVJ420P &&
+                sps->pix_fmt == AV_PIX_FMT_YUV420P       &&
+                !sps->vui.common.video_signal_type_present_flag)
+                pix_fmt = sps_base->pix_fmt;
+
+            if (pix_fmt     != sps_base->pix_fmt ||
+                sps->width  != sps_base->width   ||
+                sps->height != sps_base->height) {
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Base/non-base layer SPS have unsupported parameter combination\n");
+                return AVERROR(ENOSYS);
+            }
+        }
 
         ff_hevc_clear_refs(l);
 
@@ -2973,14 +3166,17 @@  static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
         if (ret < 0)
             return ret;
 
-        export_stream_params(s, sps);
+        if (l == &s->layers[0]) {
+            export_stream_params(s, sps);
 
-        pix_fmt = get_format(s, sps);
-        if (pix_fmt < 0)
-            return pix_fmt;
-        s->avctx->pix_fmt = pix_fmt;
+            ret = get_format(s, sps);
+            if (ret < 0) {
+                set_sps(s, l, NULL);
+                return ret;
+            }
 
-        new_sequence = 1;
+            new_sequence = 1;
+        }
     }
 
     memset(l->horizontal_bs, 0, l->bs_width * l->bs_height);
@@ -3015,7 +3211,8 @@  static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
         s->local_ctx[0].end_of_tiles_x = pps->column_width[0] << sps->log2_ctb_size;
 
     if (new_sequence) {
-        ret = ff_hevc_output_frames(s, l, 0, 0, s->sh.no_output_of_prior_pics_flag);
+        ret = ff_hevc_output_frames(s, prev_layers_active_decode, prev_layers_active_output,
+                                    0, 0, s->sh.no_output_of_prior_pics_flag);
         if (ret < 0)
             return ret;
     }
@@ -3072,7 +3269,8 @@  static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
 
     s->cur_frame->f->pict_type = 3 - s->sh.slice_type;
 
-    ret = ff_hevc_output_frames(s, l, sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics,
+    ret = ff_hevc_output_frames(s, s->layers_active_decode, s->layers_active_output,
+                                sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics,
                                 sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering, 0);
     if (ret < 0)
         goto fail;
@@ -3083,13 +3281,21 @@  static int hevc_frame_start(HEVCContext *s, HEVCLayerContext *l)
             goto fail;
     }
 
-    ff_thread_finish_setup(s->avctx);
+    // after starting the base-layer frame we know which layers will be decoded,
+    // so we can now figure out which NALUs to wait for before we can call
+    // ff_thread_finish_setup()
+    if (l == &s->layers[0])
+        s->finish_setup_nal_idx = find_finish_setup_nal(s);
+
+    if (nal_idx >= s->finish_setup_nal_idx)
+        ff_thread_finish_setup(s->avctx);
 
     return 0;
 
 fail:
-    if (s->cur_frame)
-        ff_hevc_unref_frame(s->cur_frame, ~0);
+    if (l->cur_frame)
+        ff_hevc_unref_frame(l->cur_frame, ~0);
+    l->cur_frame = NULL;
     s->cur_frame = s->collocated_ref = NULL;
     s->slice_initialized = 0;
     return ret;
@@ -3164,9 +3370,9 @@  static int verify_md5(HEVCContext *s, AVFrame *frame)
     return err;
     }
 
-static int hevc_frame_end(HEVCContext *s)
+static int hevc_frame_end(HEVCContext *s, HEVCLayerContext *l)
 {
-    HEVCFrame *out = s->cur_frame;
+    HEVCFrame *out = l->cur_frame;
     const AVFilmGrainParams *fgp;
     av_unused int ret;
 
@@ -3198,23 +3404,32 @@  static int hevc_frame_end(HEVCContext *s)
     } else {
         if (s->avctx->err_recognition & AV_EF_CRCCHECK &&
             s->sei.picture_hash.is_md5) {
-            ret = verify_md5(s, s->cur_frame->f);
+            ret = verify_md5(s, out->f);
             if (ret < 0 && s->avctx->err_recognition & AV_EF_EXPLODE)
                 return ret;
         }
     }
     s->sei.picture_hash.is_md5 = 0;
 
-    av_log(s->avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
+    av_log(s->avctx, AV_LOG_DEBUG, "Decoded frame with POC %zu/%d.\n",
+           l - s->layers, s->poc);
 
     return 0;
 }
 
-static int decode_slice(HEVCContext *s, HEVCLayerContext *l,
-                        const H2645NAL *nal, GetBitContext *gb)
+static int decode_slice(HEVCContext *s, unsigned nal_idx, GetBitContext *gb)
 {
+    const int layer_idx = s->vps ? s->vps->layer_idx[s->nuh_layer_id] : 0;
+    HEVCLayerContext *l;
     int ret;
 
+    // skip layers not requested to be decoded
+    // layers_active_decode can only change while decoding a base-layer frame,
+    // so we can check it for non-base layers
+    if (layer_idx < 0 ||
+        (s->nuh_layer_id > 0 && !(s->layers_active_decode & (1 << layer_idx))))
+        return 0;
+
     ret = hls_slice_header(&s->sh, s, gb);
     if (ret < 0) {
         // hls_slice_header() does not cleanup on failure thus the state now is inconsistant so we cannot use it on depandant slices
@@ -3230,16 +3445,25 @@  static int decode_slice(HEVCContext *s, HEVCLayerContext *l,
         return 0;
     }
 
+    // switching to a new layer, mark previous layer's frame (if any) as done
+    if (s->cur_layer != layer_idx &&
+        s->layers[s->cur_layer].cur_frame &&
+        s->avctx->active_thread_type == FF_THREAD_FRAME)
+        ff_progress_frame_report(&s->layers[s->cur_layer].cur_frame->tf, INT_MAX);
+
+    s->cur_layer = layer_idx;
+    l = &s->layers[s->cur_layer];
+
     if (s->sh.first_slice_in_pic_flag) {
-        if (s->cur_frame) {
+        if (l->cur_frame) {
             av_log(s->avctx, AV_LOG_ERROR, "Two slices reporting being the first in the same frame.\n");
             return AVERROR_INVALIDDATA;
         }
 
-        ret = hevc_frame_start(s, l);
+        ret = hevc_frame_start(s, l, nal_idx);
         if (ret < 0)
             return ret;
-    } else if (!s->cur_frame) {
+    } else if (!l->cur_frame) {
         av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
         return AVERROR_INVALIDDATA;
     }
@@ -3251,16 +3475,16 @@  static int decode_slice(HEVCContext *s, HEVCLayerContext *l,
         return AVERROR_INVALIDDATA;
     }
 
-    ret = decode_slice_data(s, l, nal, gb);
+    ret = decode_slice_data(s, l, &s->pkt.nals[nal_idx], gb);
     if (ret < 0)
         return ret;
 
     return 0;
 }
 
-static int decode_nal_unit(HEVCContext *s, const H2645NAL *nal)
+static int decode_nal_unit(HEVCContext *s, unsigned nal_idx)
 {
-    HEVCLayerContext  *l = &s->layers[0];
+    H2645NAL *nal = &s->pkt.nals[nal_idx];
     GetBitContext     gb = nal->gb;
     int ret;
 
@@ -3319,7 +3543,7 @@  static int decode_nal_unit(HEVCContext *s, const H2645NAL *nal)
     case HEVC_NAL_RADL_R:
     case HEVC_NAL_RASL_N:
     case HEVC_NAL_RASL_R:
-        ret = decode_slice(s, l, nal, &gb);
+        ret = decode_slice(s, nal_idx, &gb);
         if (ret < 0)
             goto fail;
         break;
@@ -3420,11 +3644,10 @@  static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
         H2645NAL *nal = &s->pkt.nals[i];
 
         if (s->avctx->skip_frame >= AVDISCARD_ALL ||
-            (s->avctx->skip_frame >= AVDISCARD_NONREF
-            && ff_hevc_nal_is_nonref(nal->type)) || nal->nuh_layer_id > 0)
+            (s->avctx->skip_frame >= AVDISCARD_NONREF && ff_hevc_nal_is_nonref(nal->type)))
             continue;
 
-        ret = decode_nal_unit(s, nal);
+        ret = decode_nal_unit(s, i);
         if (ret < 0) {
             av_log(s->avctx, AV_LOG_WARNING,
                    "Error parsing NAL unit #%d.\n", i);
@@ -3433,12 +3656,17 @@  static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
     }
 
 fail:
-    if (s->cur_frame) {
+    for (int i = 0; i < FF_ARRAY_ELEMS(s->layers); i++) {
+        HEVCLayerContext *l = &s->layers[i];
+
+        if (!l->cur_frame)
+            continue;
+
         if (ret >= 0)
-            ret = hevc_frame_end(s);
+            ret = hevc_frame_end(s, l);
 
         if (s->avctx->active_thread_type == FF_THREAD_FRAME)
-            ff_progress_frame_report(&s->cur_frame->tf, INT_MAX);
+            ff_progress_frame_report(&l->cur_frame->tf, INT_MAX);
     }
 
     return ret;
@@ -3459,6 +3687,11 @@  static int hevc_decode_extradata(HEVCContext *s, uint8_t *buf, int length, int f
         if (first && s->ps.sps_list[i]) {
             const HEVCSPS *sps = s->ps.sps_list[i];
             export_stream_params(s, sps);
+
+            ret = export_multilayer(s, sps->vps);
+            if (ret < 0)
+                return ret;
+
             break;
         }
     }
@@ -3489,7 +3722,8 @@  static int hevc_receive_frame(AVCodecContext *avctx, AVFrame *frame)
     av_packet_unref(avpkt);
     ret = ff_decode_get_packet(avctx, avpkt);
     if (ret == AVERROR_EOF) {
-        ret = ff_hevc_output_frames(s, &s->layers[0], 0, 0, 0);
+        ret = ff_hevc_output_frames(s, s->layers_active_decode,
+                                    s->layers_active_output, 0, 0, 0);
         if (ret < 0)
             return ret;
         goto do_output;
@@ -3555,6 +3789,8 @@  static int hevc_ref_frame(HEVCFrame *dst, const HEVCFrame *src)
     dst->ctb_count  = src->ctb_count;
     dst->flags      = src->flags;
 
+    dst->base_layer_frame = src->base_layer_frame;
+
     ff_refstruct_replace(&dst->hwaccel_picture_private,
                           src->hwaccel_picture_private);
 
@@ -3690,9 +3926,24 @@  static int hevc_update_thread_context(AVCodecContext *dst,
 
     s->is_nalff        = s0->is_nalff;
     s->nal_length_size = s0->nal_length_size;
+    s->layers_active_decode = s0->layers_active_decode;
+    s->layers_active_output = s0->layers_active_output;
 
     s->film_grain_warning_shown = s0->film_grain_warning_shown;
 
+    if (s->nb_view_ids != s0->nb_view_ids ||
+        memcmp(s->view_ids, s0->view_ids, sizeof(*s->view_ids) * s->nb_view_ids)) {
+        av_freep(&s->view_ids);
+        s->nb_view_ids = 0;
+
+        if (s0->nb_view_ids) {
+            s->view_ids = av_memdup(s0->view_ids, s0->nb_view_ids * sizeof(*s0->view_ids));
+            if (!s->view_ids)
+                return AVERROR(ENOMEM);
+            s->nb_view_ids = s0->nb_view_ids;
+        }
+    }
+
     ret = ff_h2645_sei_ctx_replace(&s->sei.common, &s0->sei.common);
     if (ret < 0)
         return ret;
@@ -3787,6 +4038,19 @@  static const AVOption options[] = {
         AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, PAR },
     { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
         AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, PAR },
+    { "view_ids", "Array of view IDs that should be decoded and output; a single -1 to decode all views",
+        .offset = OFFSET(view_ids), .type = AV_OPT_TYPE_INT | AV_OPT_TYPE_FLAG_ARRAY,
+        .min = -1, .max = INT_MAX, .flags = PAR },
+    { "view_ids_available", "Array of available view IDs is exported here",
+        .offset = OFFSET(view_ids_available), .type = AV_OPT_TYPE_UINT | AV_OPT_TYPE_FLAG_ARRAY,
+        .flags = PAR | AV_OPT_FLAG_EXPORT | AV_OPT_FLAG_READONLY },
+    { "view_pos_available", "Array of view positions for view_ids_available is exported here, as AVStereo3DView",
+        .offset = OFFSET(view_pos_available), .type = AV_OPT_TYPE_UINT | AV_OPT_TYPE_FLAG_ARRAY,
+        .flags = PAR | AV_OPT_FLAG_EXPORT | AV_OPT_FLAG_READONLY, .unit = "view_pos" },
+        { "unspecified", .type = AV_OPT_TYPE_CONST, .default_val = { .i64 = AV_STEREO3D_VIEW_UNSPEC }, .unit = "view_pos" },
+        { "left",        .type = AV_OPT_TYPE_CONST, .default_val = { .i64 = AV_STEREO3D_VIEW_LEFT },   .unit = "view_pos" },
+        { "right",       .type = AV_OPT_TYPE_CONST, .default_val = { .i64 = AV_STEREO3D_VIEW_RIGHT },  .unit = "view_pos" },
+
     { NULL },
 };
 
diff --git a/libavcodec/hevc/hevcdec.h b/libavcodec/hevc/hevcdec.h
index 57bf5aa599..6ba2ca3887 100644
--- a/libavcodec/hevc/hevcdec.h
+++ b/libavcodec/hevc/hevcdec.h
@@ -375,6 +375,10 @@  typedef struct HEVCFrame {
 
     void *hwaccel_picture_private; ///< RefStruct reference
 
+    // for secondary-layer frames, this is the DPB index of the base-layer frame
+    // from the same AU, if it exists, otherwise -1
+    int base_layer_frame;
+
     /**
      * A combination of HEVC_FRAME_FLAG_*
      */
@@ -487,9 +491,13 @@  typedef struct HEVCContext {
     HEVCLocalContext     *local_ctx;
     unsigned           nb_local_ctx;
 
-    HEVCLayerContext      layers[1];
-    // index in layers of the layer currently being decoded
+    // per-layer decoding state, addressed by VPS layer indices
+    HEVCLayerContext      layers[HEVC_VPS_MAX_LAYERS];
+    // VPS index of the layer currently being decoded
     unsigned              cur_layer;
+    // bitmask of layer indices that are active for decoding/output
+    unsigned              layers_active_decode;
+    unsigned              layers_active_output;
 
     /** 1 if the independent slice segment header was successfully parsed */
     uint8_t slice_initialized;
@@ -539,11 +547,24 @@  typedef struct HEVCContext {
     H2645Packet pkt;
     // type of the first VCL NAL of the current frame
     enum HEVCNALUnitType first_nal_type;
+    // index in pkt.nals of the NAL unit after which we can call
+    // ff_thread_finish_setup()
+    unsigned finish_setup_nal_idx;
 
     int is_nalff;           ///< this flag is != 0 if bitstream is encapsulated
                             ///< as a format defined in 14496-15
     int apply_defdispwin;
 
+    // multi-layer AVOptions
+    int         *view_ids;
+    unsigned  nb_view_ids;
+
+    unsigned    *view_ids_available;
+    unsigned  nb_view_ids_available;
+
+    unsigned    *view_pos_available;
+    unsigned  nb_view_pos_available;
+
     int nal_length_size;    ///< Number of bytes used for nal length (1, 2 or 4)
     int nuh_layer_id;
 
@@ -644,12 +665,14 @@  static av_always_inline int ff_hevc_nal_is_nonref(enum HEVCNALUnitType type)
  * Find frames in the DPB that are ready for output and either write them to the
  * output FIFO or drop their output flag, depending on the value of discard.
  *
- * @param max_output maximum number of output-pending frames that can be
- *                   present in the DPB before output is triggered
+ * @param max_output maximum number of AUs with an output-pending frame in at
+ *                   least one layer that can be present in the DPB before output
+ *                   is triggered
  * @param max_dpb maximum number of any frames that can be present in the DPB
- *                before output is triggered
+ *                for any layer before output is triggered
  */
-int ff_hevc_output_frames(HEVCContext *s, HEVCLayerContext *l,
+int ff_hevc_output_frames(HEVCContext *s,
+                          unsigned layers_active_decode, unsigned layers_active_output,
                           unsigned max_output, unsigned max_dpb, int discard);
 
 void ff_hevc_unref_frame(HEVCFrame *frame, int flags);
diff --git a/libavcodec/hevc/refs.c b/libavcodec/hevc/refs.c
index 625ac68aaa..71f07c798c 100644
--- a/libavcodec/hevc/refs.c
+++ b/libavcodec/hevc/refs.c
@@ -80,12 +80,32 @@  void ff_hevc_flush_dpb(HEVCContext *s)
 
 static HEVCFrame *alloc_frame(HEVCContext *s, HEVCLayerContext *l)
 {
+    const HEVCVPS *vps = l->sps->vps;
+    const int  view_id = vps->view_id[s->cur_layer];
     int i, j, ret;
     for (i = 0; i < FF_ARRAY_ELEMS(l->DPB); i++) {
         HEVCFrame *frame = &l->DPB[i];
         if (frame->f)
             continue;
 
+        ret = ff_progress_frame_alloc(s->avctx, &frame->tf);
+        if (ret < 0)
+            return NULL;
+
+        // add view ID side data if it's nontrivial
+        if (vps->nb_layers > 1 || view_id) {
+            AVFrameSideData *sd = av_frame_side_data_new(&frame->f->side_data,
+                                                         &frame->f->nb_side_data,
+                                                         AV_FRAME_DATA_VIEW_ID,
+                                                         sizeof(int), 0);
+            if (!sd)
+                goto fail;
+            *(int*)sd->data = view_id;
+        }
+
         ret = ff_progress_frame_get_buffer(s->avctx, &frame->tf,
                                            AV_GET_BUFFER_FLAG_REF);
         if (ret < 0)
@@ -152,6 +172,9 @@  int ff_hevc_set_new_ref(HEVCContext *s, HEVCLayerContext *l, int poc)
     l->cur_frame = ref;
     s->collocated_ref = NULL;