Work on flexible mode and screen sharing.

webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc

 /*
  *  Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  *
  */
@@ skipping 58 matching lines @@
       inited_(false),
       timestamp_(0),
       picture_id_(0),
       cpu_speed_(3),
       rc_max_intra_target_(0),
       encoder_(NULL),
       config_(NULL),
       raw_(NULL),
       input_image_(NULL),
       tl0_pic_idx_(0),
-      gof_idx_(0),
+      frames_since_kf_(0),
       num_temporal_layers_(0),
-      num_spatial_layers_(0) {
+      num_spatial_layers_(0),
+      frames_encoded_(0) {
   memset(&codec_, 0, sizeof(codec_));
   uint32_t seed = static_cast<uint32_t>(TickTime::MillisecondTimestamp());
   srand(seed);
 }
 
 VP9EncoderImpl::~VP9EncoderImpl() {
   Release();
 }
 
 int VP9EncoderImpl::Release() {
@@ skipping 193 matching lines @@
   }
   // Determine number of threads based on the image size and #cores.
   config_->g_threads = NumberOfThreads(config_->g_w,
                                        config_->g_h,
                                        number_of_cores);
 
   cpu_speed_ = GetCpuSpeed(config_->g_w, config_->g_h);
 
   // TODO(asapersson): Check configuration of temporal switch up and increase
   // pattern length.
-  if (num_temporal_layers_ == 1) {
+  is_flexible_mode_ = inst->codecSpecific.VP9.flexibleMode;
+  if (is_flexible_mode_) {
+    config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_BYPASS;
+    config_->ts_number_layers = num_temporal_layers_;
+
+    if (codec_.mode == kScreensharing) {
+      // TODO(philipel): Using VP8 style screen sharing. Change conditions
+      // when flexible mode + spatial layers become available.
+      assert(inst->codecSpecific.VP9.numberOfTemporalLayers == 2);

[sprang_webrtc, 2015/09/07 14:39:05]

CHECK_EQ(2, ...)

[philipel, 2015/09/10 14:47:02]

Code removed, now use spatial layers instead. Hurra!

+      config_->ts_number_layers = 2;
+      config_->ts_rate_decimator[0] = 2;
+      config_->ts_rate_decimator[1] = 1;
+      config_->ts_periodicity = 2;
+      config_->ts_layer_id[0] = 0;
+      config_->ts_layer_id[1] = 1;
+
+      // TODO(philipel): The following should be configurable, not fixed:
+      temporal_layer_.ConfigureBitrate(1000);

[sprang_webrtc, 2015/09/07 14:39:05]

Why not set to start bitrate from VideoCodec paramter and update in SetRates()?

[philipel, 2015/09/10 14:47:02]

Done.

+    }
+  } else if (num_temporal_layers_ == 1) {
     gof_.SetGofInfoVP9(kTemporalStructureMode1);
     config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING;
     config_->ts_number_layers = 1;
     config_->ts_rate_decimator[0] = 1;
     config_->ts_periodicity = 1;
     config_->ts_layer_id[0] = 0;
   } else if (num_temporal_layers_ == 2) {
     gof_.SetGofInfoVP9(kTemporalStructureMode2);
     config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_0101;
     config_->ts_number_layers = 2;
@@ skipping 145 matching lines @@
 
   // Image in vpx_image_t format.
   // Input image is const. VPX's raw image is not defined as const.
   raw_->planes[VPX_PLANE_Y] = const_cast<uint8_t*>(input_image.buffer(kYPlane));
   raw_->planes[VPX_PLANE_U] = const_cast<uint8_t*>(input_image.buffer(kUPlane));
   raw_->planes[VPX_PLANE_V] = const_cast<uint8_t*>(input_image.buffer(kVPlane));
   raw_->stride[VPX_PLANE_Y] = input_image.stride(kYPlane);
   raw_->stride[VPX_PLANE_U] = input_image.stride(kUPlane);
   raw_->stride[VPX_PLANE_V] = input_image.stride(kVPlane);
 
-  int flags = 0;
+  vpx_enc_frame_flags_t flags = 0;
   bool send_keyframe = (frame_type == kKeyFrame);
   if (send_keyframe) {
     // Key frame request from caller.
     flags = VPX_EFLAG_FORCE_KF;
   }
+
+  if (is_flexible_mode_) {
+    vpx_svc_layer_id_t svc_layer;
+    svc_layer.spatial_layer_id = 0;
+    if (send_keyframe) {
+      flags = GenerateRefsAndFlags();
+    } else if (codec_.mode == kRealtimeVideo) {
+      // TODO(philipel): For now produce the 0-2-1-2 mod 8 pattern
+      switch ((frames_since_kf_ - 1) % 8) {
+        case 0: {
+          flags = GenerateRefsAndFlags(0, 0);
+          break;
+        }
+        case 1: {
+          flags = GenerateRefsAndFlags(2, 0);
+          break;
+        }
+        case 2: {
+          flags = GenerateRefsAndFlags(1, 0);
+          break;
+        }
+        case 3: {
+          flags = GenerateRefsAndFlags(2, 0, 1, 2);
+          break;
+        }
+        case 4: {
+          flags = GenerateRefsAndFlags(0, 0);
+          break;
+        }
+        case 5: {
+          flags = GenerateRefsAndFlags(2, 0, 1, 2);
+          break;
+        }
+        case 6: {
+          flags = GenerateRefsAndFlags(1, 0, 1);
+          break;
+        }
+        case 7: {
+          flags = GenerateRefsAndFlags(2, 0, 1, 2);
+          break;
+        }
+      }
+      static const int temporal_layers[4] = {0, 2, 1, 2};
+      svc_layer.temporal_layer_id = temporal_layers[(frames_since_kf_ - 1) % 4];
+    } else {
+      flags = GenerateRefsAndFlags(
+          temporal_layer_.BufferArguments(input_image.timestamp()));
+      svc_layer.temporal_layer_id = temporal_layer_.CurrentLayer();
+    }
+    vpx_codec_control(encoder_, VP9E_SET_SVC_LAYER_ID, &svc_layer);
+  }
+
   assert(codec_.maxFramerate > 0);
   uint32_t duration = 90000 / codec_.maxFramerate;
   if (vpx_codec_encode(encoder_, raw_, timestamp_, duration, flags,
                        VPX_DL_REALTIME)) {
     return WEBRTC_VIDEO_CODEC_ERROR;
   }
   timestamp_ += duration;
 
   return WEBRTC_VIDEO_CODEC_OK;
 }
 
 void VP9EncoderImpl::PopulateCodecSpecific(CodecSpecificInfo* codec_specific,
                                        const vpx_codec_cx_pkt& pkt,
                                        uint32_t timestamp) {
   assert(codec_specific != NULL);
   codec_specific->codecType = kVideoCodecVP9;
   CodecSpecificInfoVP9 *vp9_info = &(codec_specific->codecSpecific.VP9);
   // TODO(asapersson): Set correct values.
   vp9_info->inter_pic_predicted =
       (pkt.data.frame.flags & VPX_FRAME_IS_KEY) ? false : true;
   vp9_info->flexible_mode = codec_.codecSpecific.VP9.flexibleMode;
   vp9_info->ss_data_available = ((pkt.data.frame.flags & VPX_FRAME_IS_KEY) &&
                                  !codec_.codecSpecific.VP9.flexibleMode)
                                     ? true
                                     : false;
-  if (pkt.data.frame.flags & VPX_FRAME_IS_KEY) {
-    gof_idx_ = 0;
-  }
+  if (pkt.data.frame.flags & VPX_FRAME_IS_KEY)
+    frames_since_kf_ = 0;
 
   vpx_svc_layer_id_t layer_id = {0};
   vpx_codec_control(encoder_, VP9E_GET_SVC_LAYER_ID, &layer_id);
 
   assert(num_temporal_layers_ > 0);
   assert(num_spatial_layers_ > 0);
   if (num_temporal_layers_ == 1) {
     assert(layer_id.temporal_layer_id == 0);
     vp9_info->temporal_idx = kNoTemporalIdx;
   } else {
     vp9_info->temporal_idx = layer_id.temporal_layer_id;
   }
   if (num_spatial_layers_ == 1) {
     assert(layer_id.spatial_layer_id == 0);
     vp9_info->spatial_idx = kNoSpatialIdx;
   } else {
     vp9_info->spatial_idx = layer_id.spatial_layer_id;
   }
   if (layer_id.spatial_layer_id != 0) {
     vp9_info->ss_data_available = false;
   }
 
-  if (vp9_info->flexible_mode) {
-    vp9_info->gof_idx = kNoGofIdx;
-  } else {
-    vp9_info->gof_idx =
-        static_cast<uint8_t>(gof_idx_++ % gof_.num_frames_in_gof);
-  }
-
   // TODO(asapersson): this info has to be obtained from the encoder.
   vp9_info->temporal_up_switch = true;
 
   if (layer_id.spatial_layer_id == 0) {
     picture_id_ = (picture_id_ + 1) & 0x7FFF;
     // TODO(asapersson): this info has to be obtained from the encoder.
     vp9_info->inter_layer_predicted = false;
   } else {
     // TODO(asapersson): this info has to be obtained from the encoder.
     vp9_info->inter_layer_predicted = true;
   }
 
   vp9_info->picture_id = picture_id_;
 
   if (!vp9_info->flexible_mode) {
     if (layer_id.temporal_layer_id == 0 && layer_id.spatial_layer_id == 0) {
       tl0_pic_idx_++;
     }
     vp9_info->tl0_pic_idx = tl0_pic_idx_;
   }
 
+  vp9_info->num_ref_pics = 0;
+  if (vp9_info->flexible_mode) {
+    vp9_info->gof_idx = kNoGofIdx;
+    if (!(pkt.data.frame.flags & VPX_FRAME_IS_KEY)) {
+      vp9_info->num_ref_pics = num_refs_pics_;
+      for (int i = 0; i < num_refs_pics_; ++i) {
+        vp9_info->p_diff[i] = p_diff_[i];
+      }
+    }
+  } else {
+    vp9_info->gof_idx =
+        static_cast<uint8_t>(frames_since_kf_ % gof_.num_frames_in_gof);
+  }
+  ++frames_since_kf_;
+
   if (vp9_info->ss_data_available) {
     vp9_info->num_spatial_layers = num_spatial_layers_;
     vp9_info->spatial_layer_resolution_present = true;
     for (size_t i = 0; i < vp9_info->num_spatial_layers; ++i) {
       vp9_info->width[i] = codec_.width *
                            svc_internal_.svc_params.scaling_factor_num[i] /
                            svc_internal_.svc_params.scaling_factor_den[i];
       vp9_info->height[i] = codec_.height *
                             svc_internal_.svc_params.scaling_factor_num[i] /
                             svc_internal_.svc_params.scaling_factor_den[i];
@@ skipping 16 matching lines @@
 
   assert(pkt->kind == VPX_CODEC_CX_FRAME_PKT);
   memcpy(&encoded_image_._buffer[encoded_image_._length], pkt->data.frame.buf,
          pkt->data.frame.sz);
   frag_info.fragmentationOffset[part_idx] = encoded_image_._length;
   frag_info.fragmentationLength[part_idx] =
       static_cast<uint32_t>(pkt->data.frame.sz);
   frag_info.fragmentationPlType[part_idx] = 0;
   frag_info.fragmentationTimeDiff[part_idx] = 0;
   encoded_image_._length += static_cast<uint32_t>(pkt->data.frame.sz);
+  if (is_flexible_mode_ && codec_.mode == kScreensharing) {
+    temporal_layer_.FrameEncoded(encoded_image_._length);
+  }

[sprang_webrtc, 2015/09/07 14:39:05]

Remove {}

[philipel, 2015/09/10 14:47:02]

Done.

   assert(encoded_image_._length <= encoded_image_._size);
 
   // End of frame.
   // Check if encoded frame is a key frame.
   if (pkt->data.frame.flags & VPX_FRAME_IS_KEY) {
     encoded_image_._frameType = kKeyFrame;
   }
   PopulateCodecSpecific(&codec_specific, *pkt, input_image_->timestamp());
 
   if (encoded_image_._length > 0) {
     TRACE_COUNTER1("webrtc", "EncodedFrameSize", encoded_image_._length);
     encoded_image_._timeStamp = input_image_->timestamp();
     encoded_image_.capture_time_ms_ = input_image_->render_time_ms();
     encoded_image_._encodedHeight = raw_->d_h;
     encoded_image_._encodedWidth = raw_->d_w;
     encoded_complete_callback_->Encoded(encoded_image_, &codec_specific,
                                         &frag_info);
   }
   return WEBRTC_VIDEO_CODEC_OK;
 }
 
+vpx_enc_frame_flags_t VP9EncoderImpl::GenerateRefsAndFlags(int8_t upd_buffer,
+                                                           int8_t ref_buf1,
+                                                           int8_t ref_buf2,
+                                                           int8_t ref_buf3) {
+  // For now we only use 3 out of the 8 buffers available
+  DCHECK(upd_buffer < 3);

[sprang_webrtc, 2015/09/07 14:39:05]

DCHECK_LT

[philipel, 2015/09/10 14:47:02]

Code removed.

+
+  // BUF0 = LAST
+  // BUF1 = GF
+  // BUF2 = ARF
+  num_refs_pics_ = 0;
+  vpx_enc_frame_flags_t flags = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_LAST |
+                                VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_ARF |
+                                VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_LAST;
+  if (upd_buffer == -2) {

[sprang_webrtc, 2015/09/07 14:39:05]

Have a named constant for -2?

[philipel, 2015/09/10 14:47:02]

Code removed.

+#ifdef NDEBUG

[sprang_webrtc, 2015/09/07 14:39:05]

Can't just always do this? memset it?

[philipel, 2015/09/10 14:47:02]

Done.

+    // Used later on to make sure we don't make any invalid references.
+    for (int i = 0; i < 8; ++i)
+      buf_upd_at_frame_[i] = -1;
+#endif
+
+    // Keyframe, always stored in BUF0. Why? Because of line 161 in

[sprang_webrtc, 2015/09/07 14:39:05]

This comment may grow stale quickly :)
"Keyframe always stored in BUF0 by libvpx." is probably sufficient.

[philipel, 2015/09/10 14:47:02]

Changed so that the keyframes are stored in the buffer specified by the
parameters to the function.

+    // vpx_temporal_svc_encoder.c. I'm not sure if that comment applies here
+    buf_upd_at_frame_[0] = frames_encoded_;
+    flags = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
+  } else {
+    int8_t refs[3] = {ref_buf1, ref_buf2, ref_buf3};

[sprang_webrtc, 2015/09/07 14:39:05]

I'd probably add a named constant for 3 and use here and below.
const size_t kNumReferences = 3;
or so...

[philipel, 2015/09/10 14:47:01]

Now use the kMaxVp9RefPics instead of '3'.

+    for (int i = 0; i < 3; ++i) {
+      switch (refs[i]) {
+        case -1:
+          goto done;

[sprang_webrtc, 2015/09/07 14:39:05]

Ugh! How bout:

i = kNumReferences;
continue;

Perhaps a comment of when this would actually happen as well.

[philipel, 2015/09/10 14:47:02]

Code removed.

+        case 0: {
+          flags ^= VP8_EFLAG_NO_REF_LAST;
+          break;
+        }
+        case 1: {
+          flags ^= VP8_EFLAG_NO_REF_GF;
+          break;
+        }
+        case 2: {
+          flags ^= VP8_EFLAG_NO_REF_ARF;
+          break;
+        }
+        default:
+          DCHECK(false);

[sprang_webrtc, 2015/09/07 14:39:05]

RTC_NOTREACHED();

[philipel, 2015/09/10 14:47:02]

Code removed.

+      }
+
+      // Make sure this frame doesn't reference to an unavailable
+      // buffer, either because it has not yet been used or
+      // because a KF has occurred since it was used.
+      DCHECK(buf_upd_at_frame_[refs[i]] != -1);
+
+      p_diff_[i] = frames_encoded_ - buf_upd_at_frame_[refs[i]];
+      num_refs_pics_++;
+    }
+  // Everybody love gotos!

[sprang_webrtc, 2015/09/07 14:39:05]

No. :)

[philipel, 2015/09/10 14:47:02]

Yes!

+  done:
+
+    buf_upd_at_frame_[upd_buffer] = frames_encoded_;
+    switch (upd_buffer) {
+      case -1:
+        break;
+      case 0: {
+        flags ^= VP8_EFLAG_NO_UPD_LAST;
+        break;
+      }
+      case 1: {
+        flags ^= VP8_EFLAG_NO_UPD_GF;
+        break;
+      }
+      case 2: {
+        flags ^= VP8_EFLAG_NO_UPD_ARF;
+        break;
+      }
+      default:
+        DCHECK(false);

[sprang_webrtc, 2015/09/07 14:39:05]

RTC_NOTREACHED();

[philipel, 2015/09/10 14:47:01]

Code removed.

+    }
+  }
+
+  frames_encoded_++;
+  return flags;
+}
+
+vpx_enc_frame_flags_t VP9EncoderImpl::GenerateRefsAndFlags(
+    std::array<int8_t, 4> args) {
+  return GenerateRefsAndFlags(args[0], args[1], args[2], args[3]);
+}
+
 int VP9EncoderImpl::SetChannelParameters(uint32_t packet_loss, int64_t rtt) {
   return WEBRTC_VIDEO_CODEC_OK;
 }
 
 int VP9EncoderImpl::RegisterEncodeCompleteCallback(
     EncodedImageCallback* callback) {
   encoded_complete_callback_ = callback;
   return WEBRTC_VIDEO_CODEC_OK;
 }
 
@@ skipping 164 matching lines @@
     decoder_ = NULL;
   }
   // Releases buffers from the pool. Any buffers not in use are deleted. Buffers
   // still referenced externally are deleted once fully released, not returning
   // to the pool.
   frame_buffer_pool_.ClearPool();
   inited_ = false;
   return WEBRTC_VIDEO_CODEC_OK;
 }
 }  // namespace webrtc