zqz/platform-external-webrtc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Reland of Work on flexible mode and screen sharing. (patchset #1 id:1 of https://codereview.webrtc.org/1438543002/ )

Browse Source 
Reason for revert:
Failed test not related to this CL (test fails on
master at an earlier date), re-landing original CL..

(This time from my @webrtc account.)

Original issue's description:
> Revert of Work on flexible mode and screen sharing. (patchset #28 id:520001 of https://codereview.webrtc.org/1328113004/ )
>
> Reason for revert:
> Seems to break VideoSendStreamTest.ReconfigureBitratesSetsEncoderBitratesCorrectly on Linux Memcheck buildbot.
>
> Original issue's description:
> > Work on flexible mode and screen sharing.
> >
> > Implement VP8 style screensharing but with spatial layers.
> > Implement flexible mode.
> >
> > Files from other patches:
> > generic_encoder.cc
> > layer_filtering_transport.cc
> >
> > BUG=webrtc:4914
> >
> > Committed: https://crrev.com/77ccfb4d16c148e61a316746bb5d9705e8b39f4a
> > Cr-Commit-Position: refs/heads/master@{#10572}
>
> TBR=sprang@webrtc.org,stefan@webrtc.org,philipel@google.com,asapersson@webrtc.org,mflodman@webrtc.org,philipel@webrtc.org
> NOPRESUBMIT=true
> NOTREECHECKS=true
> NOTRY=true
> BUG=webrtc:4914
>
> Committed: https://crrev.com/0be8f1d347bdb171462df89c2a4c69b3f3eb7519
> Cr-Commit-Position: refs/heads/master@{#10578}

TBR=sprang@webrtc.org,stefan@webrtc.org,philipel@google.com,asapersson@webrtc.org,mflodman@webrtc.org,terelius@webrtc.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=webrtc:4914

Review URL: https://codereview.webrtc.org/1431283002

Cr-Commit-Position: refs/heads/master@{#10581}
This commit is contained in:
philipel
2015-11-10 07:17:23 -08:00
committed by Commit bot
parent c95c366f5a
commit cfc319be1d
20 changed files with 1120 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
webrtc/modules/video_coding/main/source/decoding_state.cc
View File

@ -24,7 +24,9 @@ VCMDecodingState::VCMDecodingState()
temporal_id_(kNoTemporalIdx),
tl0_pic_id_(kNoTl0PicIdx),
full_sync_(true),
in_initial_state_(true) {}
in_initial_state_(true) {
memset(frame_decoded_, 0, sizeof(frame_decoded_));
}
VCMDecodingState::~VCMDecodingState() {}
@ -37,6 +39,7 @@ void VCMDecodingState::Reset() {
tl0_pic_id_ = kNoTl0PicIdx;
full_sync_ = true;
in_initial_state_ = true;
memset(frame_decoded_, 0, sizeof(frame_decoded_));
}
uint32_t VCMDecodingState::time_stamp() const {
@ -63,12 +66,33 @@ bool VCMDecodingState::IsOldPacket(const VCMPacket* packet) const {
void VCMDecodingState::SetState(const VCMFrameBuffer* frame) {
assert(frame != NULL && frame->GetHighSeqNum() >= 0);
UpdateSyncState(frame);
if (!UsingFlexibleMode(frame))
UpdateSyncState(frame);
sequence_num_ = static_cast<uint16_t>(frame->GetHighSeqNum());
time_stamp_ = frame->TimeStamp();
picture_id_ = frame->PictureId();
temporal_id_ = frame->TemporalId();
tl0_pic_id_ = frame->Tl0PicId();
if (UsingFlexibleMode(frame)) {
uint16_t frame_index = picture_id_ % kFrameDecodedLength;
if (in_initial_state_) {
frame_decoded_cleared_to_ = frame_index;
} else if (frame->FrameType() == kVideoFrameKey) {
memset(frame_decoded_, 0, sizeof(frame_decoded_));
frame_decoded_cleared_to_ = frame_index;
} else {
if (AheadOfFramesDecodedClearedTo(frame_index)) {
while (frame_decoded_cleared_to_ != frame_index) {
frame_decoded_cleared_to_ =
(frame_decoded_cleared_to_ + 1) % kFrameDecodedLength;
frame_decoded_[frame_decoded_cleared_to_] = false;
}
}
}
frame_decoded_[frame_index] = true;
}
in_initial_state_ = false;
}
@ -80,6 +104,8 @@ void VCMDecodingState::CopyFrom(const VCMDecodingState& state) {
tl0_pic_id_ = state.tl0_pic_id_;
full_sync_ = state.full_sync_;
in_initial_state_ = state.in_initial_state_;
frame_decoded_cleared_to_ = state.frame_decoded_cleared_to_;
memcpy(frame_decoded_, state.frame_decoded_, sizeof(frame_decoded_));
}
bool VCMDecodingState::UpdateEmptyFrame(const VCMFrameBuffer* frame) {
@ -173,7 +199,11 @@ bool VCMDecodingState::ContinuousFrame(const VCMFrameBuffer* frame) const {
if (!full_sync_ && !frame->LayerSync())
return false;
if (UsingPictureId(frame)) {
return ContinuousPictureId(frame->PictureId());
if (UsingFlexibleMode(frame)) {
return ContinuousFrameRefs(frame);
} else {
return ContinuousPictureId(frame->PictureId());
}
} else {
return ContinuousSeqNum(static_cast<uint16_t>(frame->GetLowSeqNum()));
}
@ -216,8 +246,41 @@ bool VCMDecodingState::ContinuousLayer(int temporal_id,
return (static_cast<uint8_t>(tl0_pic_id_ + 1) == tl0_pic_id);
}
bool VCMDecodingState::ContinuousFrameRefs(const VCMFrameBuffer* frame) const {
uint8_t num_refs = frame->CodecSpecific()->codecSpecific.VP9.num_ref_pics;
for (uint8_t r = 0; r < num_refs; ++r) {
uint16_t frame_ref = frame->PictureId() -
frame->CodecSpecific()->codecSpecific.VP9.p_diff[r];
uint16_t frame_index = frame_ref % kFrameDecodedLength;
if (AheadOfFramesDecodedClearedTo(frame_index) ||
!frame_decoded_[frame_index]) {
return false;
}
}
return true;
}
bool VCMDecodingState::UsingPictureId(const VCMFrameBuffer* frame) const {
return (frame->PictureId() != kNoPictureId && picture_id_ != kNoPictureId);
}
bool VCMDecodingState::UsingFlexibleMode(const VCMFrameBuffer* frame) const {
return frame->CodecSpecific()->codecType == kVideoCodecVP9 &&
frame->CodecSpecific()->codecSpecific.VP9.flexible_mode;
}
// TODO(philipel): change how check work, this check practially
// limits the max p_diff to 64.
bool VCMDecodingState::AheadOfFramesDecodedClearedTo(uint16_t index) const {
// No way of knowing for sure if we are actually ahead of
// frame_decoded_cleared_to_. We just make the assumption
// that we are not trying to reference back to a very old
// index, but instead are referencing a newer index.
uint16_t diff =
index > frame_decoded_cleared_to_
? kFrameDecodedLength - (index - frame_decoded_cleared_to_)
: frame_decoded_cleared_to_ - index;
return diff > kFrameDecodedLength / 2;
}
} // namespace webrtc

12
webrtc/modules/video_coding/main/source/decoding_state.h
View File

@ -21,6 +21,11 @@ class VCMPacket;
class VCMDecodingState {
public:
// The max number of bits used to reference back
// to a previous frame when using flexible mode.
static const uint16_t kNumRefBits = 7;
static const uint16_t kFrameDecodedLength = 1 << kNumRefBits;
VCMDecodingState();
~VCMDecodingState();
// Check for old frame
@ -52,7 +57,10 @@ class VCMDecodingState {
bool ContinuousPictureId(int picture_id) const;
bool ContinuousSeqNum(uint16_t seq_num) const;
bool ContinuousLayer(int temporal_id, int tl0_pic_id) const;
bool ContinuousFrameRefs(const VCMFrameBuffer* frame) const;
bool UsingPictureId(const VCMFrameBuffer* frame) const;
bool UsingFlexibleMode(const VCMFrameBuffer* frame) const;
bool AheadOfFramesDecodedClearedTo(uint16_t index) const;
// Keep state of last decoded frame.
// TODO(mikhal/stefan): create designated classes to handle these types.
@ -63,6 +71,10 @@ class VCMDecodingState {
int tl0_pic_id_;
bool full_sync_; // Sync flag when temporal layers are used.
bool in_initial_state_;
// Used to check references in flexible mode.
bool frame_decoded_[kFrameDecodedLength];
uint16_t frame_decoded_cleared_to_;
};
} // namespace webrtc

250
webrtc/modules/video_coding/main/source/decoding_state_unittest.cc
View File

@ -446,4 +446,254 @@ TEST(TestDecodingState, PictureIdRepeat) {
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
}
TEST(TestDecodingState, FrameContinuityFlexibleModeKeyFrame) {
VCMDecodingState dec_state;
VCMFrameBuffer frame;
VCMPacket packet;
packet.isFirstPacket = true;
packet.timestamp = 1;
packet.seqNum = 0xffff;
uint8_t data[] = "I need a data pointer for this test!";
packet.sizeBytes = sizeof(data);
packet.dataPtr = data;
packet.codecSpecificHeader.codec = kRtpVideoVp9;
RTPVideoHeaderVP9& vp9_hdr = packet.codecSpecificHeader.codecHeader.VP9;
vp9_hdr.picture_id = 10;
vp9_hdr.flexible_mode = true;
FrameData frame_data;
frame_data.rtt_ms = 0;
frame_data.rolling_average_packets_per_frame = -1;
// Key frame as first frame
packet.frameType = kVideoFrameKey;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Key frame again
vp9_hdr.picture_id = 11;
frame.Reset();
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to 11, continuous
frame.Reset();
packet.frameType = kVideoFrameDelta;
vp9_hdr.picture_id = 12;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 1;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
}
TEST(TestDecodingState, FrameContinuityFlexibleModeOutOfOrderFrames) {
VCMDecodingState dec_state;
VCMFrameBuffer frame;
VCMPacket packet;
packet.isFirstPacket = true;
packet.timestamp = 1;
packet.seqNum = 0xffff;
uint8_t data[] = "I need a data pointer for this test!";
packet.sizeBytes = sizeof(data);
packet.dataPtr = data;
packet.codecSpecificHeader.codec = kRtpVideoVp9;
RTPVideoHeaderVP9& vp9_hdr = packet.codecSpecificHeader.codecHeader.VP9;
vp9_hdr.picture_id = 10;
vp9_hdr.flexible_mode = true;
FrameData frame_data;
frame_data.rtt_ms = 0;
frame_data.rolling_average_packets_per_frame = -1;
// Key frame as first frame
packet.frameType = kVideoFrameKey;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to 10, continuous
frame.Reset();
packet.frameType = kVideoFrameDelta;
vp9_hdr.picture_id = 15;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 5;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Out of order, last id 15, this id 12, ref to 10, continuous
frame.Reset();
vp9_hdr.picture_id = 12;
vp9_hdr.pid_diff[0] = 2;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref 10, 12, 15, continuous
frame.Reset();
vp9_hdr.picture_id = 20;
vp9_hdr.num_ref_pics = 3;
vp9_hdr.pid_diff[0] = 10;
vp9_hdr.pid_diff[1] = 8;
vp9_hdr.pid_diff[2] = 5;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
}
TEST(TestDecodingState, FrameContinuityFlexibleModeGeneral) {
VCMDecodingState dec_state;
VCMFrameBuffer frame;
VCMPacket packet;
packet.isFirstPacket = true;
packet.timestamp = 1;
packet.seqNum = 0xffff;
uint8_t data[] = "I need a data pointer for this test!";
packet.sizeBytes = sizeof(data);
packet.dataPtr = data;
packet.codecSpecificHeader.codec = kRtpVideoVp9;
RTPVideoHeaderVP9& vp9_hdr = packet.codecSpecificHeader.codecHeader.VP9;
vp9_hdr.picture_id = 10;
vp9_hdr.flexible_mode = true;
FrameData frame_data;
frame_data.rtt_ms = 0;
frame_data.rolling_average_packets_per_frame = -1;
// Key frame as first frame
packet.frameType = kVideoFrameKey;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
// Delta frame as first frame
frame.Reset();
packet.frameType = kVideoFrameDelta;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
// Key frame then delta frame
frame.Reset();
packet.frameType = kVideoFrameKey;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
dec_state.SetState(&frame);
frame.Reset();
packet.frameType = kVideoFrameDelta;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.picture_id = 15;
vp9_hdr.pid_diff[0] = 5;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to 11, not continuous
frame.Reset();
vp9_hdr.picture_id = 16;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
// Ref to 15, continuous
frame.Reset();
vp9_hdr.picture_id = 16;
vp9_hdr.pid_diff[0] = 1;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to 11 and 15, not continuous
frame.Reset();
vp9_hdr.picture_id = 20;
vp9_hdr.num_ref_pics = 2;
vp9_hdr.pid_diff[0] = 9;
vp9_hdr.pid_diff[1] = 5;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
// Ref to 10, 15 and 16, continuous
frame.Reset();
vp9_hdr.picture_id = 22;
vp9_hdr.num_ref_pics = 3;
vp9_hdr.pid_diff[0] = 12;
vp9_hdr.pid_diff[1] = 7;
vp9_hdr.pid_diff[2] = 6;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Key Frame, continuous
frame.Reset();
packet.frameType = kVideoFrameKey;
vp9_hdr.picture_id = VCMDecodingState::kFrameDecodedLength - 2;
vp9_hdr.num_ref_pics = 0;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Frame at last index, ref to KF, continuous
frame.Reset();
packet.frameType = kVideoFrameDelta;
vp9_hdr.picture_id = VCMDecodingState::kFrameDecodedLength - 1;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 1;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Frame after wrapping buffer length, ref to last index, continuous
frame.Reset();
vp9_hdr.picture_id = 0;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 1;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Frame after wrapping start frame, ref to 0, continuous
frame.Reset();
vp9_hdr.picture_id = 20;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 20;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Frame after wrapping start frame, ref to 10, not continuous
frame.Reset();
vp9_hdr.picture_id = 23;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 13;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
// Key frame, continuous
frame.Reset();
packet.frameType = kVideoFrameKey;
vp9_hdr.picture_id = 25;
vp9_hdr.num_ref_pics = 0;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to KF, continuous
frame.Reset();
packet.frameType = kVideoFrameDelta;
vp9_hdr.picture_id = 26;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 1;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_TRUE(dec_state.ContinuousFrame(&frame));
dec_state.SetState(&frame);
// Ref to frame previous to KF, not continuous
frame.Reset();
vp9_hdr.picture_id = 30;
vp9_hdr.num_ref_pics = 1;
vp9_hdr.pid_diff[0] = 30;
EXPECT_LE(0, frame.InsertPacket(packet, 0, kNoErrors, frame_data));
EXPECT_FALSE(dec_state.ContinuousFrame(&frame));
}
} // namespace webrtc

6
webrtc/modules/video_coding/main/source/encoded_frame.cc
View File

@ -147,6 +147,12 @@ void VCMEncodedFrame::CopyCodecSpecific(const RTPVideoHeader* header)
header->codecHeader.VP9.inter_pic_predicted;
_codecSpecificInfo.codecSpecific.VP9.flexible_mode =
header->codecHeader.VP9.flexible_mode;
_codecSpecificInfo.codecSpecific.VP9.num_ref_pics =
header->codecHeader.VP9.num_ref_pics;
for (uint8_t r = 0; r < header->codecHeader.VP9.num_ref_pics; ++r) {
_codecSpecificInfo.codecSpecific.VP9.p_diff[r] =
header->codecHeader.VP9.pid_diff[r];
}
_codecSpecificInfo.codecSpecific.VP9.ss_data_available =
header->codecHeader.VP9.ss_data_available;
if (header->codecHeader.VP9.picture_id != kNoPictureId) {

8
webrtc/modules/video_coding/main/source/generic_encoder.cc
View File

@ -54,11 +54,9 @@ void CopyCodecSpecific(const CodecSpecificInfo* info, RTPVideoHeader* rtp) {
rtp->codecHeader.VP9.inter_layer_predicted =
info->codecSpecific.VP9.inter_layer_predicted;
rtp->codecHeader.VP9.gof_idx = info->codecSpecific.VP9.gof_idx;
// Packetizer needs to know the number of spatial layers to correctly set
// the marker bit, even when the number won't be written in the packet.
rtp->codecHeader.VP9.num_spatial_layers =
info->codecSpecific.VP9.num_spatial_layers;
if (info->codecSpecific.VP9.ss_data_available) {
rtp->codecHeader.VP9.spatial_layer_resolution_present =
info->codecSpecific.VP9.spatial_layer_resolution_present;
@ -71,6 +69,10 @@ void CopyCodecSpecific(const CodecSpecificInfo* info, RTPVideoHeader* rtp) {
}
rtp->codecHeader.VP9.gof.CopyGofInfoVP9(info->codecSpecific.VP9.gof);
}
rtp->codecHeader.VP9.num_ref_pics = info->codecSpecific.VP9.num_ref_pics;
for (int i = 0; i < info->codecSpecific.VP9.num_ref_pics; ++i)
rtp->codecHeader.VP9.pid_diff[i] = info->codecSpecific.VP9.p_diff[i];
return;
}
case kVideoCodecH264:

6
webrtc/modules/video_coding/main/source/jitter_buffer.cc
View File

@ -686,12 +686,6 @@ VCMFrameBufferEnum VCMJitterBuffer::InsertPacket(const VCMPacket& packet,
num_consecutive_old_packets_ = 0;
if (packet.codec == kVideoCodecVP9 &&
packet.codecSpecificHeader.codecHeader.VP9.flexible_mode) {
// TODO(asapersson): Add support for flexible mode.
return kGeneralError;
}
VCMFrameBuffer* frame;
FrameList* frame_list;
const VCMFrameBufferEnum error = GetFrame(packet, &frame, &frame_list);