Reason for revert:
There seems an error made in this patch.
Hi Henrik,
I think the bit shift returned by CrossCorrelationWithAutoShift may be wrongly used by DotProduct.
We'd better revert this CL.
Doing another fix (and future fixes) will be paintful. I will work on a easy-to-modify bitexactness test first.
Original issue's description:
> Avoiding overflow in cross correlation in NetEq.
>
> BUG=
TBR=henrik.lundin@webrtc.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=
Review-Url: https://codereview.webrtc.org/1925053002
Cr-Commit-Position: refs/heads/master@{#12543}
Specifically, this change replaces peak_period_counter_ms_ with
peak_period_stopwatch_ which obtains a Stopwatch object from
TickTimer. Necessary plumbing to get the TickTimer through to the
DelayPeakDetector is also included.
BUG=webrtc:5608
NOTRY=True
Review-Url: https://codereview.webrtc.org/1921163003
Cr-Commit-Position: refs/heads/master@{#12542}
Changed DecoderDatabase::IsComfortNoise to do one, rather than four,
lookups of the rtp payload type. IsComfortNoise is called more frequently
since CNG was changed to not be an AudioDecoder.
BUG=606765
Review URL: https://codereview.webrtc.org/1923763003
Cr-Commit-Position: refs/heads/master@{#12533}
With this change, the NetEqImpl constructor takes a struct
(NetEqImpl::Dependencies) as input instead of a collection of
individual dependencies. The NetEqImpl unit test fixture is modified
to make better used of unique_ptrs.
Review URL: https://codereview.webrtc.org/1921243002
Cr-Commit-Position: refs/heads/master@{#12514}
This change makes use of the TickTimer::Stopwatch in Packets. When a
packet is inserted into the PacketBuffer, a Stopwatch object is
attached to it. When the packet is extracted from the buffer, the
Stopwatch is read to know how long the packet waited in the buffer.
BUG=webrtc:5608
Review URL: https://codereview.webrtc.org/1917913002
Cr-Commit-Position: refs/heads/master@{#12508}
The TickTimer is incremented on each call to GetAudioInternal(). Other
than that, the new object is not used yet.
Also adding a unit test in NetEqImplTest to verify that the tick timer
is incremented in the call to NetEq::GetAudio.
BUG=webrtc:5608
Review URL: https://codereview.webrtc.org/1903153005
Cr-Commit-Position: refs/heads/master@{#12493}
Broke out CNG from AudioDecoder as they didn't really share an interface.
Converted the CNG code to C++, to make initialization and resource handling easier. This includes several changes to the behavior, favoring RTC_CHECKs over returning error codes.
Review URL: https://codereview.webrtc.org/1868143002
Cr-Commit-Position: refs/heads/master@{#12491}
The new class is intended to be used as a central time-keeping object
inside NetEq. The actual use of the class will come in subsequent
changes.
BUG=webrtc:5608
Review URL: https://codereview.webrtc.org/1910523003
Cr-Commit-Position: refs/heads/master@{#12477}
The fs_hz member variable is going away too, being replaced by a
method in the AudioDecoder interface. If we ever end up needing the
RTP sample rate here, a method ought to be the right solution for that
too.
BUG=webrtc:5801
Review URL: https://codereview.webrtc.org/1907183002
Cr-Commit-Position: refs/heads/master@{#12462}
By eliminating one of the two constructors, handling decoder ownership
with a unique_ptr instead of a raw pointer, and making all member
variables const (except one, which is made private instead).
BUG=webrtc:5801
Review URL: https://codereview.webrtc.org/1899733002
Cr-Commit-Position: refs/heads/master@{#12425}
With this change, the return value from NetEq::GetPlayoutTimestamp is
empty if the latest call to NetEq::GetAudio resulted in comfort noise
(codec-internal or external) being played out. This is because the
playout timestamp is not updated during CNG, and can therefore not be
trusted.
A few unit tests were updated to reflect the change.
BUG=webrtc:5669
Review URL: https://codereview.webrtc.org/1861303002
Cr-Commit-Position: refs/heads/master@{#12268}
This was previously done in AcmReceiver, but belongs in NetEq where the
rest of the AudioFrame fields are populated.
BUG=webrtc:5669,webrtc:5607
Review URL: https://codereview.webrtc.org/1863993002
Cr-Commit-Position: refs/heads/master@{#12265}
This is in preparation for changes to when the playout timestamp is
valid.
BUG=webrtc:5669
Review URL: https://codereview.webrtc.org/1853183002
Cr-Commit-Position: refs/heads/master@{#12256}
This new unit test verifies the parameter fields (not the audio data
itself) written to the AudioFrame output by AcmReceiver::GetAudio.
Also corrected a few comments reflecting recent changes in the code.
BUG=webrtc:5669
Review URL: https://codereview.webrtc.org/1859953002
Cr-Commit-Position: refs/heads/master@{#12253}
- The interleaved_ field. Never set to anything but 'true'. AudioFrame data appears to always be treated as interleaved.
- The Append() method.
- operator-=().
BUG=
Review URL: https://codereview.webrtc.org/1830713003
Cr-Commit-Position: refs/heads/master@{#12152}
The test code created an AudioBuffer object inside the work loop. This
turned out to be expensive, since the AudioBuffer ctor implicitly
called memset on all of the audio data array. The obvious remedy is to
create the buffer outside of the loop. This does not have any impact
apart from the performance boost, since the output data from NetEq is
not even considered in the test.
BUG=chromium:592907,webrtc:5647
TBR=ivoc@webrtc.org
NOTRY=true
Review URL: https://codereview.webrtc.org/1782803002
Cr-Commit-Position: refs/heads/master@{#11940}
This copies the contents (unittest excluded) of base/numerics in
chromium to base/numerics in webrtc. Files added:
- safe_conversions.h
- safe_conversions_impl.h
- safe_math.h
- safe_math_impl.h
A really old version of safe_conversions[_impl].h previously existed in
base/, this has been deleted and sources using it have been updated
to include the new base/numerics/safe_converions.h.
This CL also adds a DEPS file to webrtc/base.
NOPRESUBMIT=True
BUG=webrtc:5548, webrtc:5623
Review URL: https://codereview.webrtc.org/1753293002
Cr-Commit-Position: refs/heads/master@{#11907}
The type is included in the AudioFrame output parameter.
Rename the type NetEqOutputType to just OutputType, since it is now
internal to NetEq.
BUG=webrtc:5607
Review URL: https://codereview.webrtc.org/1769883002
Cr-Commit-Position: refs/heads/master@{#11903}
This change essentially does two things:
1. Remove the VAD-related methods from AcmReceiver. These are
EnableVad(), DisableVad(), and vad_enabled(). None of them were used
outside of unit tests.
2. Move the functionality to set AudioFrame::speech_type_ and
AudioFrame::vad_activity_ inside NetEq. This was previously done in
AcmReceiver, but based on information inherently owned by NetEq.
With the change in 2, NetEq's GetAudio interface can be simplified by
removing the output type parameter. This will be done in a follow-up
CL.
BUG=webrtc:5607
Review URL: https://codereview.webrtc.org/1772583002
Cr-Commit-Position: refs/heads/master@{#11902}
Sparse macro replaced for all audio histograms that have a constant name.
BUG=webrtc:5283
Review URL: https://codereview.webrtc.org/1762863003
Cr-Commit-Position: refs/heads/master@{#11885}
With this change, NetEq now uses AudioFrame as output type, like the
surrounding functions in ACM and VoiceEngine already do.
The computational savings is probably slim, since one memcpy is
removed while another one is added (both in AcmReceiver::GetAudio).
More simplifications and clean-up will be done in
AcmReceiver::GetAudio in future CLs.
BUG=webrtc:5607
Review URL: https://codereview.webrtc.org/1750353002
Cr-Commit-Position: refs/heads/master@{#11874}
For backwards compatibility, I've added kept the old interface to
Encode() and EncodeInternal and created default implementations of both
variants of EncodeInternal(), each calling the other. At least one of
the variants must be implemented in a subclass or we'll run out of stack
and explode. Would be nice if we could catch that before runtime. :/
The new interface to EncodeInternal() is protected, since it should
never be called from the outside.
Was unable to mark the old EncodeInternal() as RTC_DEPRECATED, since the
default implementaion of the new variant needs to call it to work around
old implementations. The old Encode() variant is deprecated, at least.
Added a test for backwards compatibility in audio_encoder_unittest.cc.
For the added test I broke out MockEncodeHelper from
audio_encoder_copy_red_unittest.cc and renamed it MockAudioEncoderHelper.
Review URL: https://codereview.webrtc.org/1725143003
Cr-Commit-Position: refs/heads/master@{#11823}
The array is reset in Init() but not the indexer. This makes the start point undefined after Init() for re-initializing an AudioLoop. This can be fixed.
BUG=
Review URL: https://codereview.webrtc.org/1727353002
Cr-Commit-Position: refs/heads/master@{#11739}
With this change the following tests have been successfully
passing in the iOS Simulator for iPhone 5 and iOS 9:
* audio_decoder_unittests
* common_video_unittests
* modules_tests
* rtc_api_objc_tests
* rtc_pc_unittests
* system_wrappers_unittests
* voice_engine_unittests
The modules_unittests and common_audio_unittests are
handled in https://codereview.webrtc.org/1698033002/
BUG=webrtc:4755
NOTRY=True
Review URL: https://codereview.webrtc.org/1694353003
Cr-Commit-Position: refs/heads/master@{#11646}
Both were related to very large jumps in RTP timestamps.
BUG=webrtc:5488
Review URL: https://codereview.webrtc.org/1685103002
Cr-Commit-Position: refs/heads/master@{#11569}
In some rare occations (very low energy signal), a shift value happened
to be negative. This is now fixed by using the WEBRTC_SPL_SHIFT_W32,
which in essence checks the sign of the number of shifts and performs a
right or left shift accordingly.
The fix reverts to how the code was written in old NetEq; see
4d363ae305/webrtc/modules/audio_coding/neteq/normal.c (165).
BUG=webrtc:5490
Review URL: https://codereview.webrtc.org/1675293002
Cr-Commit-Position: refs/heads/master@{#11546}
This pulls in several fixes and gets Visual Studio 2015 support.
The new repo is located at https://github.com/gflags/gflags
which is mirrored in Chrome infrastructure at
https://chromium.googlesource.com/external/github.com/gflags/gflags
New configuration headers were generated according to README.webrtc
on Windows and Linux. I verified the Linux generated ones are working
on Mac. The generating headers on Mac are identical with only a minor
difference (an __unused attribute) that doesn't effect the build.
BUG=webrtc:5185
NOTRY=True
NOPRESUBMIT=True
TESTED=Successfully ran:
out/Release/video_quality_measurement --input_filename=resources/foreman_cif.yuv --width=352 --height=288
to verify flags are still being parsed properly.
I also ran the compile trybots and the baremetal bots
(since they run tests that have gflags flags).
Review URL: https://codereview.webrtc.org/1679263002
Cr-Commit-Position: refs/heads/master@{#11539}
If a StatisticsCalculator::PeriodicUmaAverage object was created and
then deleted without any samples being logged, the destructor would call
the Metric() method, which calculated sum_/counter_. However, with no
samples logged, counter_ is 0.
This was found and verified using UBSan tests; see the bug for more info.
BUG=webrtc:5490
R=ivoc@webrtc.org
Review URL: https://codereview.webrtc.org/1678773003
Cr-Commit-Position: refs/heads/master@{#11534}
This is needed because the target is defined in webrtc/common.gyp
and its current location crosses package boundaries when generating
projects for some build systems.
NOTRY=True
Review URL: https://codereview.webrtc.org/1665603003
Cr-Commit-Position: refs/heads/master@{#11496}
When the file was rewound, the remaining audio read was inserted at
the start of the destination array, not where the first reading
attempt ended.
R=ivoc@webrtc.org
Review URL: https://codereview.webrtc.org/1612053002
Cr-Commit-Position: refs/heads/master@{#11343}
Several unittests were disabled on android, this CL will reenable them. One of
the tests was accidentally disabled on all platforms, and now no longer gives a
bitexact result.
BUG=webrtc:3343,webrtc:5349
Review URL: https://codereview.webrtc.org/1532903002
Cr-Commit-Position: refs/heads/master@{#11323}
There's no need for this class to have a vtable since there exists only a single implementation (per platform). It's also not good for performance.
BUG=
R=pbos@webrtc.org
Review URL: https://codereview.webrtc.org/1601743004 .
Cr-Commit-Position: refs/heads/master@{#11306}
