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.
  *
  */
 
 #include "webrtc/modules/video_coding/codecs/vp9/vp9_impl.h"
 
 #include <stdlib.h>
 #include <string.h>
 #include <time.h>
 #include <vector>
 
 #include "vpx/vpx_encoder.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx/vp8cx.h"
 #include "vpx/vp8dx.h"
 
 #include "webrtc/base/bind.h"
 #include "webrtc/base/checks.h"
 #include "webrtc/base/trace_event.h"
 #include "webrtc/common.h"
 #include "webrtc/common_video/libyuv/include/webrtc_libyuv.h"
 #include "webrtc/modules/include/module_common_types.h"
+#include "webrtc/modules/video_coding/codecs/vp9/screenshare_layers.h"
 #include "webrtc/system_wrappers/include/logging.h"
 #include "webrtc/system_wrappers/include/tick_util.h"
 
 namespace {
 
 // VP9DecoderImpl::ReturnFrame helper function used with WrappedI420Buffer.
 static void WrappedI420BufferNoLongerUsedCb(
     webrtc::Vp9FrameBufferPool::Vp9FrameBuffer* img_buffer) {
   img_buffer->Release();
 }
@@ skipping 29 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),
+      // Use two spatial when screensharing with flexible mode.
+      spatial_layer_(new ScreenshareLayersVP9(2)) {
   memset(&codec_, 0, sizeof(codec_));
   uint32_t seed = static_cast<uint32_t>(TickTime::MillisecondTimestamp());
   srand(seed);
 }
 
 VP9EncoderImpl::~VP9EncoderImpl() {
   Release();
 }
 
 int VP9EncoderImpl::Release() {
@@ skipping 109 matching lines @@
   }
   if (new_framerate < 1) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
   // Update bit rate
   if (codec_.maxBitrate > 0 && new_bitrate_kbit > codec_.maxBitrate) {
     new_bitrate_kbit = codec_.maxBitrate;
   }
   config_->rc_target_bitrate = new_bitrate_kbit;
   codec_.maxFramerate = new_framerate;
+  spatial_layer_->ConfigureBitrate(new_bitrate_kbit, 0);
 
   if (!SetSvcRates()) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
 
   // Update encoder context
   if (vpx_codec_enc_config_set(encoder_, config_)) {
     return WEBRTC_VIDEO_CODEC_ERROR;
   }
   return WEBRTC_VIDEO_CODEC_OK;
@@ skipping 18 matching lines @@
   if (number_of_cores < 1) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
   if (inst->codecSpecific.VP9.numberOfTemporalLayers > 3) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
   // libvpx currently supports only one or two spatial layers.
   if (inst->codecSpecific.VP9.numberOfSpatialLayers > 2) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
+
   int retVal = Release();
   if (retVal < 0) {
     return retVal;
   }
   if (encoder_ == NULL) {
     encoder_ = new vpx_codec_ctx_t;
   }
   if (config_ == NULL) {
     config_ = new vpx_codec_enc_cfg_t;
   }
@@ skipping 58 matching lines @@
       1 : 0;
   // 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)
+      spatial_layer_->ConfigureBitrate(inst->startBitrate, 0);
+  } 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 50 matching lines @@
       svc_internal_.svc_params.scaling_factor_den[i] = layer.scaling_factor_den;
     }
   } else {
     int scaling_factor_num = 256;
     for (int i = num_spatial_layers_ - 1; i >= 0; --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.
       svc_internal_.svc_params.scaling_factor_num[i] = scaling_factor_num;
       svc_internal_.svc_params.scaling_factor_den[i] = 256;
-      scaling_factor_num /= 2;
+      if (codec_.mode != kScreensharing)
+        scaling_factor_num /= 2;
     }
   }
 
   if (!SetSvcRates()) {
     return WEBRTC_VIDEO_CODEC_ERR_PARAMETER;
   }
 
   if (vpx_codec_enc_init(encoder_, vpx_codec_vp9_cx(), config_, 0)) {
     return WEBRTC_VIDEO_CODEC_UNINITIALIZED;
   }
@@ skipping 79 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 == kVideoFrameKey);
   if (send_keyframe) {
     // Key frame request from caller.
     flags = VPX_EFLAG_FORCE_KF;
   }
+
+  if (is_flexible_mode_) {
+    SuperFrameRefSettings settings;
+
+    // These structs are copied when calling vpx_codec_control,
+    // therefore it is ok for them to go out of scope.
+    vpx_svc_ref_frame_config enc_layer_conf;
+    vpx_svc_layer_id layer_id;
+
+    if (codec_.mode == kRealtimeVideo) {
+      // Real time video not yet implemented in flexible mode.
+      RTC_NOTREACHED();
+    } else {
+      settings = spatial_layer_->GetSuperFrameSettings(input_image.timestamp(),
+                                                       send_keyframe);
+    }
+    enc_layer_conf = GenerateRefsAndFlags(settings);
+    layer_id.temporal_layer_id = 0;
+    layer_id.spatial_layer_id = settings.start_layer;
+    vpx_codec_control(encoder_, VP9E_SET_SVC_LAYER_ID, &layer_id);
+    vpx_codec_control(encoder_, VP9E_SET_SVC_REF_FRAME_CONFIG, &enc_layer_conf);
+  }
+
   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) {
+  bool is_first_frame = false;
+  if (is_flexible_mode_) {
+    is_first_frame =
+        layer_id.spatial_layer_id == spatial_layer_->GetStartLayer();
+  } else {
+    is_first_frame = layer_id.spatial_layer_id == 0;
+  }
+
+  if (is_first_frame) {
     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_;
   }
 
   // Always populate this, so that the packetizer can properly set the marker
   // bit.
   vp9_info->num_spatial_layers = num_spatial_layers_;
+
+  vp9_info->num_ref_pics = 0;
+  if (vp9_info->flexible_mode) {
+    vp9_info->gof_idx = kNoGofIdx;
+    vp9_info->num_ref_pics = num_ref_pics_[layer_id.spatial_layer_id];
+    for (int i = 0; i < num_ref_pics_[layer_id.spatial_layer_id]; ++i) {
+      vp9_info->p_diff[i] = p_diff_[layer_id.spatial_layer_id][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->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 15 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);
+
+  vpx_svc_layer_id_t layer_id = {0};
+  vpx_codec_control(encoder_, VP9E_GET_SVC_LAYER_ID, &layer_id);
+  if (is_flexible_mode_ && codec_.mode == kScreensharing)
+    spatial_layer_->LayerFrameEncoded(
+        static_cast<unsigned int>(encoded_image_._length),
+        layer_id.spatial_layer_id);
+
   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 = kVideoFrameKey;
   }
   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_svc_ref_frame_config VP9EncoderImpl::GenerateRefsAndFlags(
+    const SuperFrameRefSettings& settings) {
+  static const vpx_enc_frame_flags_t kAllFlags =
+      VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_LAST |
+      VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF;
+  vpx_svc_ref_frame_config sf_conf = {};
+  if (settings.is_keyframe) {
+    // Used later on to make sure we don't make any invalid references.
+    memset(buffer_updated_at_frame_, -1, sizeof(buffer_updated_at_frame_));
+    for (int layer = settings.start_layer; layer <= settings.stop_layer;
+         ++layer) {
+      num_ref_pics_[layer] = 0;
+      buffer_updated_at_frame_[settings.layer[layer].upd_buf] = frames_encoded_;
+      // When encoding a keyframe only the alt_fb_idx is used
+      // to specify which layer ends up in which buffer.
+      sf_conf.alt_fb_idx[layer] = settings.layer[layer].upd_buf;
+    }
+  } else {
+    for (int layer_idx = settings.start_layer; layer_idx <= settings.stop_layer;
+         ++layer_idx) {
+      vpx_enc_frame_flags_t layer_flags = kAllFlags;
+      num_ref_pics_[layer_idx] = 0;
+      int8_t refs[3] = {settings.layer[layer_idx].ref_buf1,
+                        settings.layer[layer_idx].ref_buf2,
+                        settings.layer[layer_idx].ref_buf3};
+
+      for (unsigned int ref_idx = 0; ref_idx < kMaxVp9RefPics; ++ref_idx) {
+        if (refs[ref_idx] == -1)
+          continue;
+
+        RTC_DCHECK_GE(refs[ref_idx], 0);
+        RTC_DCHECK_LE(refs[ref_idx], 7);
+        // Easier to remove flags from all flags rather than having to
+        // build the flags from 0.
+        switch (num_ref_pics_[layer_idx]) {
+          case 0: {
+            sf_conf.lst_fb_idx[layer_idx] = refs[ref_idx];
+            layer_flags &= ~VP8_EFLAG_NO_REF_LAST;
+            break;
+          }
+          case 1: {
+            sf_conf.gld_fb_idx[layer_idx] = refs[ref_idx];
+            layer_flags &= ~VP8_EFLAG_NO_REF_GF;
+            break;
+          }
+          case 2: {
+            sf_conf.alt_fb_idx[layer_idx] = refs[ref_idx];
+            layer_flags &= ~VP8_EFLAG_NO_REF_ARF;
+            break;
+          }
+        }
+        // Make sure we don't reference a buffer that hasn't been
+        // used at all or hasn't been used since a keyframe.
+        RTC_DCHECK_NE(buffer_updated_at_frame_[refs[ref_idx]], -1);
+
+        p_diff_[layer_idx][num_ref_pics_[layer_idx]] =
+            frames_encoded_ - buffer_updated_at_frame_[refs[ref_idx]];
+        num_ref_pics_[layer_idx]++;
+      }
+
+      bool upd_buf_same_as_a_ref = false;
+      if (settings.layer[layer_idx].upd_buf != -1) {
+        for (unsigned int ref_idx = 0; ref_idx < kMaxVp9RefPics; ++ref_idx) {
+          if (settings.layer[layer_idx].upd_buf == refs[ref_idx]) {
+            switch (ref_idx) {
+              case 0: {
+                layer_flags &= ~VP8_EFLAG_NO_UPD_LAST;
+                break;
+              }
+              case 1: {
+                layer_flags &= ~VP8_EFLAG_NO_UPD_GF;
+                break;
+              }
+              case 2: {
+                layer_flags &= ~VP8_EFLAG_NO_UPD_ARF;
+                break;
+              }
+            }
+            upd_buf_same_as_a_ref = true;
+            break;
+          }
+        }
+        if (!upd_buf_same_as_a_ref) {
+          // If we have three references and a buffer is specified to be
+          // updated,
+          // then that buffer must be the same as one of the three references.
+          RTC_CHECK_LT(num_ref_pics_[layer_idx], kMaxVp9RefPics);
+
+          sf_conf.alt_fb_idx[layer_idx] = settings.layer[layer_idx].upd_buf;
+          layer_flags ^= VP8_EFLAG_NO_UPD_ARF;
+        }
+
+        int updated_buffer = settings.layer[layer_idx].upd_buf;
+        buffer_updated_at_frame_[updated_buffer] = frames_encoded_;
+        sf_conf.frame_flags[layer_idx] = layer_flags;
+      }
+    }
+  }
+  frames_encoded_++;
+  return sf_conf;
+}
+
 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