Handling the case when encoder drops only the higher layer.
Added options to screenshare loopback test to discard high temporal or spatial layers (to view the lower layers).
Review URL: https://codereview.webrtc.org/1287643002
Cr-Commit-Position: refs/heads/master@{#9883}
The easy way also happens to be more efficient if we have to
reallocate, but that's a minor concern here.
Review URL: https://codereview.webrtc.org/1327053002
Cr-Commit-Position: refs/heads/master@{#9876}
This CL contains major modifications of the audio output parts for WebRTC on iOS:
- general code cleanup
- improves thread handling (added thread checks, remove critical section, atomic ops etc.)
- reduces loopback latency of iPhone 6 from ~90ms to ~60ms ;-)
- improves selection of audio parameters on iOS
- reduces complexity by removing complex and redundant delay estimates
- now instead uses fixed delay estimates if for some reason the SW EAC must be used
- adds AudioFineBuffer to compensate for differences in native output buffer size and
the 10ms size used by WebRTC. Same class as is used today on Android and we have unit tests for
this class (the old code was buggy and we have several issue reports of crashes related to it)
Similar improvements will be done for the recording sid as well in a separate CL.
I will also add support for 48kHz in an upcoming CL since that will improve Opus performance.
BUG=webrtc:4796,webrtc:4817,webrtc:4954, webrtc:4212
TEST=AppRTC demo and iOS modules_unittests using --gtest_filter=AudioDevice*
R=pbos@webrtc.org, tkchin@webrtc.org
Review URL: https://codereview.webrtc.org/1254883002 .
Cr-Commit-Position: refs/heads/master@{#9875}
DesktopFrameWin::Create() may return nullptr when it fails to allocate
windows bitmap. ScreenCapturerWinGdi wasn't handling that case properly.
BUG=527660
Review URL: https://codereview.webrtc.org/1309143007
Cr-Commit-Position: refs/heads/master@{#9865}
Telling the encoder to adjust the parameters for the screen content.
Also, telling the encoder to skip the encoding of very flat/low content blocks. For now only for screensharing. (number 8 in VP8E_SET_STATIC_THRESHOLD is correct)
Review URL: https://codereview.webrtc.org/1308753006
Cr-Commit-Position: refs/heads/master@{#9860}
This will be used for the send side bitrate estimation. Storing various
meta-data about packets that can be retreived when arrival time feeback
arrives.
BUG=webrtc:4173
Review URL: https://codereview.webrtc.org/1288033008
Cr-Commit-Position: refs/heads/master@{#9859}
Value was incorrectly truncated to 16 bits when serializing the message.
Fixed, with added regression tests.
BUG=
Review URL: https://codereview.webrtc.org/1294393002
Cr-Commit-Position: refs/heads/master@{#9858}
Pitfalls:
* Left shift of signed integer has undefined behavior
* Right-shift of signed integer has platform-specific behavior is value is negative
* Cast from unsigned to signed has undefined behavior if value is negative
BUG=webrtc:4824
Review URL: https://codereview.webrtc.org/1226993003
Cr-Commit-Position: refs/heads/master@{#9854}
Hopefully, this will make it easier to figure out what's wrong the
next time this happens.
BUG=526478
Review URL: https://codereview.webrtc.org/1313073008
Cr-Commit-Position: refs/heads/master@{#9844}
In some cases, the number of samples (per channel) in NetEq's sync
buffer could fall below the allowed minimum (5 samples for narrowband,
scaling for other rates). If the number of samples extracted from the
buffer was smaller than the desired number, an error is
returned. However, if the decoder returns fewer samples than expected,
it could happen that the sync buffer level falls under the minimum,
but enough samples are extracted. This triggered an assert. With this
change, the minimum level of the sync buffer is always enforced.
A test is implemented to trigger the problem. It made the assert fire
without this fix, but it now passes.
BUG=webrtc:4840
R=minyue@webrtc.org
Review URL: https://codereview.webrtc.org/1324453002 .
Cr-Commit-Position: refs/heads/master@{#9828}
Cleaning AudioConferenceMixer APIs to match Chromium style guide.
Main changes:
1. change all mutable references to pointers
2. add const to all non-mutable references
3. add const to as many methods as possible
BUG=
R=andrew@webrtc.org
Review URL: https://codereview.webrtc.org/1311733003 .
Cr-Commit-Position: refs/heads/master@{#9821}
Replaced the wav file dumping functionality in aec_core.c with the newly added corresponding macros
Added macros for logging of AEC internal data
BUG=
Review URL: https://codereview.webrtc.org/1272403003
Cr-Commit-Position: refs/heads/master@{#9808}
These asserts cover error cases that are also handled by the code
after the assert. Should not have both assert and error handling.
BUG=webrtc:4840
Review URL: https://codereview.webrtc.org/1321023002
Cr-Commit-Position: refs/heads/master@{#9804}
This is a bug that was introduced in
https://codereview.webrtc.org/1230503003, where the variable "int
temp_bufsize" was changed to a size_t. If the packet buffer was
flushed while inserting a packet, temp_bufsize became negative, which
was tested later in an if-statement. Now, with size_t instead, it
would just become very large, and the if-statement would never see a
negative value. The effect was that the packet size in samples could
be updated with a very large positive number, causing an overflow
which triggered rtc::checked_cast in
StatisticsCalculator::GetNetworkStatistics, line 220.
Also adding a test to reproduce the crash. Without the fix, the test
results in the above mentioned checked_cast to trigger. With the fix,
everything works fine.
BUG=chromium:525260
Review URL: https://codereview.webrtc.org/1307893004
Cr-Commit-Position: refs/heads/master@{#9802}
This was not implemented before. It returns the current total delay
(packet buffer and sync buffer) of NetEq. This is the same information
that was already available in
NetEqNetworkStatistics::current_buffer_size_ms, that can be obtained
through NetEq::NetworkStatistics(). But, since the current delay is a
key metric of NetEq, it is convenient to have it available in a
simpler way.
This is a re-landing of r9359,
https://webrtc-codereview.appspot.com/51149004, which was reverted in
r9360. The refactoring made in r9669 facilitated the relanding.
TBR=minyue@webrtc.org
Review URL: https://codereview.webrtc.org/1313873003
Cr-Commit-Position: refs/heads/master@{#9801}
The Init() method was previously used to initialize and reset
decoders, and returned an error code. The new Reset() method is used
for reset only; the constructor is now responsible for fully
initializing the AudioDecoder.
Reset() doesn't return an error code; it turned out that none of the
functions it ended up calling could actually fail, so this CL removes
their error return codes as well.
R=henrik.lundin@webrtc.org
Review URL: https://codereview.webrtc.org/1319683002 .
Cr-Commit-Position: refs/heads/master@{#9798}
The Buffer is saved between calls, so after the initial allocation
it'll already be allocated and of the right size. The stack-allocated
array had the advantage of requiring no heap allocation at all, but
for most popular encoders it ended up allocating about 15 kB too much,
and now that we allow user-defined encoders there was also the
(remote) possibility that the buffer would actually be too small.
R=henrik.lundin@webrtc.org
Review URL: https://codereview.webrtc.org/1303413003 .
Cr-Commit-Position: refs/heads/master@{#9789}
VideoFrameBuffer currently has two overloaded data() functions for pixel access, one for const and one for non-const. Unfortunately, it will default to the non-const version, even when 'const scoped_refptr<VideoFrameBuffer>&' is used. This is a problem, because many subclasses use RTC_NOTREACHED() in the non-const version.
This CL makes the non-const version of data() explicit with a different, longer function name MutableData().
R=tommi@webrtc.org
Review URL: https://codereview.webrtc.org/1304143003 .
Cr-Commit-Position: refs/heads/master@{#9787}
The primary fix in this CL is to remove the dangling |thread_| pointer in AndroidVideoCapturerJni. That thread is not safe to use after Stop() has been called. Even after Stop() has been called, we must still be able to return late frames to Java in order to not leak them, so that path has been made thread safe instead. To make sure that we always return frames, the Java frame should be wrapped in a scoped_refptr as quickly as possible, so this CL moves the wrapping from AndroidVideoCapturer to AndroidVideoCapturerJni. This also removes the need for the interface function AndroidVideoCapturerDelegate::ReturnBuffer().
Some other minor changes are:
* Remove |valid_global_refs_| and all logic related to that. Now that rtc::Bind() captures method objects as scoped_refptr, the destructor of AndroidVideoCapturerJni will not be called before all frames are returned.
* Remove global ref |j_frame_observer_|. No need for this, we don’t call it and it is kept alive with standard Java memory management.
* Add helper function ShallowCenterCrop() for VideoFrameBuffers. This functionality already exists in the constructor of WrappedI420Buffer, but it’s more convenient to have it as a separate function.
BUG=webrtc:4742,webrtc:4909
R=glaznev@webrtc.org, tommi@webrtc.org
Review URL: https://codereview.webrtc.org/1307973002 .
Cr-Commit-Position: refs/heads/master@{#9784}
With this change, the aggregates for packet waiting times are
calculated in NetEq's StatisticsCalculator insead of in
AcmReceiver. This simplifies things somewhat, and avoids having to
copy the raw data on polling.
R=ivoc@webrtc.org, minyue@webrtc.org
Review URL: https://codereview.webrtc.org/1296633002 .
Cr-Commit-Position: refs/heads/master@{#9778}
By converting three raw pointers to scoped_ptrs, we can eliminate the
need for a manually-defined destructor, and generally sleep better at
night.
R=henrik.lundin@webrtc.org
Review URL: https://codereview.webrtc.org/1310213003 .
Cr-Commit-Position: refs/heads/master@{#9776}
