Integration of VP9 packetization.

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

Index: webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc
diff --git a/webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc b/webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc
index a44d0f66bfc69869f926ffc4757dd3dbed599654..f7ff3545a0c9e4ea8d89e1193e9d25b754b96255 100644
--- a/webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc
+++ b/webrtc/modules/video_coding/codecs/vp9/vp9_impl.cc
@@ -56,7 +56,11 @@ VP9EncoderImpl::VP9EncoderImpl()
       rc_max_intra_target_(0),
       encoder_(NULL),
       config_(NULL),
-      raw_(NULL) {
+      raw_(NULL),
+      input_image_(NULL),
+      tl0_pic_idx_(0),
+      number_of_temporal_layers_(0),
+      number_of_spatial_layers_(0) {
   memset(&codec_, 0, sizeof(codec_));
   uint32_t seed = static_cast<uint32_t>(TickTime::MillisecondTimestamp());
   srand(seed);
@@ -90,6 +94,59 @@ int VP9EncoderImpl::Release() {
   return WEBRTC_VIDEO_CODEC_OK;
 }
 
+int VP9EncoderImpl::SetSvcRates() {
+  float sRateRatio[VPX_MAX_LAYERS] = {0};

[stefan-webrtc, 2015/07/09 14:48:59]

No camelCase.

[åsapersson, 2015/07/29 12:10:12]

Done.

+  float total = 0;
+  uint8_t i = 0;
+
+  if (number_of_spatial_layers_ > 1) {
+    for (i = 0; i < number_of_spatial_layers_; ++i) {
+      if (svc_internal_.svc_params.scaling_factor_num[i] <= 0 ||
+          !svc_internal_.svc_params.scaling_factor_den[i]) {

[stefan-webrtc, 2015/07/09 14:48:59]

svc_internal_.svc_params.scaling_factor_den[i] <= 0?

[åsapersson, 2015/07/29 12:10:12]

Done.

+        return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
+      }
+      sRateRatio[i] =
+          (float)(svc_internal_.svc_params.scaling_factor_num[i] * 1.0 /

[stefan-webrtc, 2015/07/09 14:48:59]

static_cast. I don't think the * 1.0 matters.

[åsapersson, 2015/07/29 12:10:12]

Changed parenthesis -> removed 1.0.

+                  svc_internal_.svc_params.scaling_factor_den[i]);
+      total += sRateRatio[i];
+    }
+  } else {
+    sRateRatio[0] = total = 1;
+  }
+

[stefan-webrtc, 2015/07/09 14:48:59]

Assert that total = 1?

[åsapersson, 2015/07/29 12:10:12]

added assert >=1?

+  for (i = 0; i < number_of_spatial_layers_; ++i) {
+    config_->ss_target_bitrate[i] =
+        (unsigned int)(config_->rc_target_bitrate * sRateRatio[i] / total);

[stefan-webrtc, 2015/07/09 14:48:59]

static_cast

[åsapersson, 2015/07/29 12:10:12]

Done.

+    if (number_of_temporal_layers_ == 1) {
+      config_->layer_target_bitrate[0] = config_->ss_target_bitrate[i];
+    } else if (number_of_temporal_layers_ == 2) {
+      config_->layer_target_bitrate[i * number_of_temporal_layers_] =
+          config_->ss_target_bitrate[i] * 2 / 3;
+      config_->layer_target_bitrate[i * number_of_temporal_layers_ + 1] =
+          config_->ss_target_bitrate[i];
+    } else {
+      if (number_of_temporal_layers_ != 3) {
+        return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
+      }
+      config_->layer_target_bitrate[i * number_of_temporal_layers_] =
+          config_->ss_target_bitrate[i] / 2;
+      config_->layer_target_bitrate[i * number_of_temporal_layers_ + 1] =
+          config_->layer_target_bitrate[i * number_of_temporal_layers_] +
+          (config_->ss_target_bitrate[i] / 4);
+      config_->layer_target_bitrate[i * number_of_temporal_layers_ + 2] =
+          config_->ss_target_bitrate[i];
+    }
+  }
+
+  if (number_of_spatial_layers_ == 1) {
+    for (i = 0; i < VPX_TS_MAX_LAYERS; ++i) {
+      config_->ts_target_bitrate[i] = config_->layer_target_bitrate[i];

[stefan-webrtc, 2015/07/09 14:48:59]

Are all VPX_TS_MAX_LAYERS elements of layer_target_bitrate set? Shouldn't there
be a length field we can read instead?

[åsapersson, 2015/07/29 12:10:12]

Changed to number of temporal layers.

+    }
+  }
+
+  return WEBRTC_VIDEO_CODEC_OK;

[stefan-webrtc, 2015/07/09 14:48:59]

Make this method return bool instead.

[åsapersson, 2015/07/29 12:10:12]

Done.

+}
+
 int VP9EncoderImpl::SetRates(uint32_t new_bitrate_kbit,
                              uint32_t new_framerate) {
   if (!inited_) {
@@ -107,6 +164,11 @@ int VP9EncoderImpl::SetRates(uint32_t new_bitrate_kbit,
   }
   config_->rc_target_bitrate = new_bitrate_kbit;
   codec_.maxFramerate = new_framerate;
+
+  int ret = SetSvcRates();
+  if (ret != WEBRTC_VIDEO_CODEC_OK)
+    return ret;
+
   // Update encoder context
   if (vpx_codec_enc_config_set(encoder_, config_)) {
     return WEBRTC_VIDEO_CODEC_ERROR;
@@ -147,6 +209,16 @@ int VP9EncoderImpl::InitEncode(const VideoCodec* inst,
   if (&codec_ != inst) {
     codec_ = *inst;
   }
+
+  // Member codec_ has access to VideoCodecVP9 (common_types.h), which should
+  // have number of spatial layers and temporal structure mode set.

[stefan-webrtc, 2015/07/09 14:48:59]

Not sure if this comment is useful?

[åsapersson, 2015/07/29 12:10:12]

Removed.

+  number_of_temporal_layers_ = inst->codecSpecific.VP9.numberOfTemporalLayers;
+  number_of_spatial_layers_ = inst->codecSpecific.VP9.numberOfSpatialLayers;
+  // For now, only support 1 spatial layer.
+  if (number_of_spatial_layers_ != 1) {
+    return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
+  }
+
   // Random start 16 bits is enough.
   picture_id_ = static_cast<uint16_t>(rand()) & 0x7FFF;
   // Allocate memory for encoded image
@@ -199,12 +271,53 @@ int VP9EncoderImpl::InitEncode(const VideoCodec* inst,
   config_->g_threads = NumberOfThreads(config_->g_w,
                                        config_->g_h,
                                        number_of_cores);
+
+  if (number_of_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 (number_of_temporal_layers_ == 2) {
+    gof_.SetGofInfoVP9(kTemporalStructureMode2);
+    config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_0101;
+    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;
+  } else if (number_of_temporal_layers_ == 3) {
+    gof_.SetGofInfoVP9(kTemporalStructureMode3);
+    config_->temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_0212;
+    config_->ts_number_layers = 3;
+    config_->ts_rate_decimator[0] = 4;
+    config_->ts_rate_decimator[1] = 2;
+    config_->ts_rate_decimator[2] = 1;
+    config_->ts_periodicity = 4;
+    config_->ts_layer_id[0] = 0;
+    config_->ts_layer_id[1] = 2;
+    config_->ts_layer_id[2] = 1;
+    config_->ts_layer_id[3] = 2;
+  } else {
+    return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
+  }
+
+  tl0_pic_idx_ = static_cast<uint8_t>(rand());
+
   return InitAndSetControlSettings(inst);
 }
 
 int VP9EncoderImpl::NumberOfThreads(int width,
                                     int height,
                                     int number_of_cores) {
+  // For the current libvpx library, only 1 thread is supported when SVC is
+  // turned on.
+  if (number_of_temporal_layers_ > 1 || number_of_spatial_layers_ > 1) {
+    return 1;
+  }
+
   // Keep the number of encoder threads equal to the possible number of column
   // tiles, which is (1, 2, 4, 8). See comments below for VP9E_SET_TILE_COLUMNS.
   if (width * height >= 1280 * 720 && number_of_cores > 4) {
@@ -218,6 +331,26 @@ int VP9EncoderImpl::NumberOfThreads(int width,
 }
 
 int VP9EncoderImpl::InitAndSetControlSettings(const VideoCodec* inst) {
+
+  config_->ss_number_layers = number_of_spatial_layers_;
+
+  if (number_of_spatial_layers_ > 1) {
+    config_->rc_min_quantizer = 0;
+    config_->rc_max_quantizer = 63;
+  }
+  for (uint8_t i = 0; i < number_of_spatial_layers_; i++) {
+    svc_internal_.svc_params.max_quantizers[i] = config_->rc_max_quantizer;
+    svc_internal_.svc_params.min_quantizers[i] = config_->rc_min_quantizer;
+    /* 1:2 scaling in each dimension */

[stefan-webrtc, 2015/07/09 14:48:59]

Fix this comment.

[åsapersson, 2015/07/29 12:10:12]

Done.

+    svc_internal_.svc_params.scaling_factor_num[i] =
+        256 >> (number_of_spatial_layers_ - i - 1);

[stefan-webrtc, 2015/07/09 14:48:59]

Can't we divide instead of shifting? I think it makes it a lot easier to read.

[åsapersson, 2015/07/29 12:10:12]

Done.

+    svc_internal_.svc_params.scaling_factor_den[i] = 256;
+  }
+
+  int ret = SetSvcRates();
+  if (ret != WEBRTC_VIDEO_CODEC_OK)
+    return ret;
+
   if (vpx_codec_enc_init(encoder_, vpx_codec_vp9_cx(), config_, 0)) {
     return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
   }
@@ -231,6 +364,20 @@ int VP9EncoderImpl::InitAndSetControlSettings(const VideoCodec* inst) {
                     rc_max_intra_target_);
   vpx_codec_control(encoder_, VP9E_SET_AQ_MODE,
                     inst->codecSpecific.VP9.adaptiveQpMode ? 3 : 0);
+
+  vpx_codec_control(
+      encoder_, VP9E_SET_SVC,
+      (number_of_temporal_layers_ > 1 || number_of_spatial_layers_ > 1) ? 1
+                                                                        : 0);
+  if (number_of_spatial_layers_ > 1) {
+    vpx_codec_control(encoder_, VP9E_SET_SVC_PARAMETERS,
+                      &svc_internal_.svc_params);
+  }
+  // Register callback for getting each spatial layer
+  vpx_codec_priv_output_cx_pkt_cb_pair_t cbp = {
+      VP9EncoderImpl::EncocderOutputCodedPacketCb, (void*)(this)};
+  vpx_codec_control(encoder_, VP9E_REGISTER_CX_CALLBACK, (void*)(&cbp));
+
   // Control function to set the number of column tiles in encoding a frame, in
   // log2 unit: e.g., 0 = 1 tile column, 1 = 2 tile columns, 2 = 4 tile columns.
   // The number tile columns will be capped by the encoder based on image size
@@ -242,6 +389,7 @@ int VP9EncoderImpl::InitAndSetControlSettings(const VideoCodec* inst) {
   vpx_codec_control(encoder_, VP9E_SET_NOISE_SENSITIVITY,
                     inst->codecSpecific.VP9.denoisingOn ? 1 : 0);
 #endif
+
   inited_ = true;
   return WEBRTC_VIDEO_CODEC_OK;
 }
@@ -280,6 +428,13 @@ int VP9EncoderImpl::Encode(const VideoFrame& input_image,
   }
   DCHECK_EQ(input_image.width(), static_cast<int>(raw_->d_w));
   DCHECK_EQ(input_image.height(), static_cast<int>(raw_->d_h));
+
+  // Set input image for use in the callback.
+  // This was necessary since you need some information from input_image.
+  // You can save only the necessary information (such as timestamp) instead of
+  // doing this.
+  input_image_ = &input_image;
+
   // 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));
@@ -302,7 +457,8 @@ int VP9EncoderImpl::Encode(const VideoFrame& input_image,
     return WEBRTC_VIDEO_CODEC_ERROR;
   }
   timestamp_ += duration;
-  return GetEncodedPartitions(input_image);
+
+  return WEBRTC_VIDEO_CODEC_OK;
 }
 
 void VP9EncoderImpl::PopulateCodecSpecific(CodecSpecificInfo* codec_specific,
@@ -311,20 +467,81 @@ void VP9EncoderImpl::PopulateCodecSpecific(CodecSpecificInfo* codec_specific,
   assert(codec_specific != NULL);
   codec_specific->codecType = kVideoCodecVP9;
   CodecSpecificInfoVP9 *vp9_info = &(codec_specific->codecSpecific.VP9);
+  // TODO(asapersson): Set correct values.
+  vp9_info->interPicPredicted =
+      (pkt.data.frame.flags & VPX_FRAME_IS_KEY) ? false : true;
+  vp9_info->flexibleMode = codec_.codecSpecific.VP9.flexibleMode;
+  vp9_info->beginningOfFrame = true;
+  vp9_info->endOfFrame = true;
+  vp9_info->ssDataAvailable =
+      (pkt.data.frame.flags & VPX_FRAME_IS_KEY) ? true : false;
+
+  vpx_svc_layer_id_t lyr = {0};

[stefan-webrtc, 2015/07/09 14:48:59]

layer_id

[åsapersson, 2015/07/29 12:10:12]

Done.

+  vpx_codec_control(encoder_, VP9E_GET_SVC_LAYER_ID, &lyr);
+
+  assert(number_of_temporal_layers_ > 0);
+  assert(number_of_spatial_layers_ > 0);
+  if (number_of_temporal_layers_ == 1) {
+    assert(lyr.temporal_layer_id == 0);
+    vp9_info->temporalIdx = kNoTemporalIdx;
+  } else {
+    vp9_info->temporalIdx = lyr.temporal_layer_id;
+  }
+  if (number_of_spatial_layers_ == 1) {
+    assert(lyr.spatial_layer_id == 0);
+    vp9_info->spatialIdx = kNoSpatialIdx;
+  } else {
+    vp9_info->spatialIdx = lyr.spatial_layer_id;
+  }
+  if (lyr.spatial_layer_id != 0) {
+    vp9_info->ssDataAvailable = false;
+  }
+
+  if (vp9_info->flexibleMode) {
+    vp9_info->gofIdx = kNoGofIdx;
+  } else {
+    vp9_info->gofIdx = lyr.temporal_layer_id;
+  }
+
+  // TODO(asapersson): this info has to be obtained from the encoder.
+  vp9_info->temporalUpSwitch = true;
+
+  if (lyr.spatial_layer_id == 0) {
+    picture_id_ = (picture_id_ + 1) & 0x7FFF;
+    // TODO(asapersson): this info has to be obtained from the encoder.
+    vp9_info->interLayerPredicted = false;
+  } else {
+    // TODO(asapersson): this info has to be obtained from the encoder.
+    vp9_info->interLayerPredicted = true;
+  }
+
   vp9_info->pictureId = picture_id_;
-  vp9_info->keyIdx = kNoKeyIdx;
-  vp9_info->nonReference = (pkt.data.frame.flags & VPX_FRAME_IS_DROPPABLE) != 0;
-  // TODO(marpan): Temporal layers are supported in the current VP9 version,
-  // but for now use 1 temporal layer encoding. Will update this when temporal
-  // layer support for VP9 is added in webrtc.
-  vp9_info->temporalIdx = kNoTemporalIdx;
-  vp9_info->layerSync = false;
-  vp9_info->tl0PicIdx = kNoTl0PicIdx;
-  picture_id_ = (picture_id_ + 1) & 0x7FFF;
+
+  if (!vp9_info->flexibleMode) {
+    if (lyr.temporal_layer_id == 0 && lyr.spatial_layer_id == 0) {
+      tl0_pic_idx_++;
+    }
+    vp9_info->tl0PicIdx = tl0_pic_idx_;
+  }
+
+  if (vp9_info->ssDataAvailable) {
+    vp9_info->numSpatialLayers = number_of_spatial_layers_;
+    vp9_info->spatialLayerResolutionPresent = true;
+    for (uint8_t i = 0; i < vp9_info->numSpatialLayers; 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];
+    }
+    if (!vp9_info->flexibleMode) {
+      vp9_info->gof.CopyGofInfoVP9(gof_);
+    }
+  }
 }
 
-int VP9EncoderImpl::GetEncodedPartitions(const VideoFrame& input_image) {
-  vpx_codec_iter_t iter = NULL;
+int VP9EncoderImpl::GetEncodedLayerFrame(const vpx_codec_cx_pkt* pkt) {
   encoded_image_._length = 0;
   encoded_image_._frameType = kDeltaFrame;
   RTPFragmentationHeader frag_info;
@@ -333,44 +550,34 @@ int VP9EncoderImpl::GetEncodedPartitions(const VideoFrame& input_image) {
   frag_info.VerifyAndAllocateFragmentationHeader(1);
   int part_idx = 0;
   CodecSpecificInfo codec_specific;
-  const vpx_codec_cx_pkt_t *pkt = NULL;
-  while ((pkt = vpx_codec_get_cx_data(encoder_, &iter)) != NULL) {
-    switch (pkt->kind) {
-      case 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);
-        assert(encoded_image_._length <= encoded_image_._size);
-        break;
-      }
-      default: {
-        break;
-      }
-    }
-    // End of frame.
-    if ((pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT) == 0) {
-      // 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());
-      break;
+
+  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);
+  assert(encoded_image_._length <= encoded_image_._size);
+
+  // End of frame.
+  if ((pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT) == 0) {
+    // 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_._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);
+                                        &frag_info);
   }
   return WEBRTC_VIDEO_CODEC_OK;
 }