Refactoring done to more easily and cleanly add SIMD optimizations and
to remove `FullyConnectedLayer` from the RNN VAD api.
Minor improvements (readability, API):
- `FullyConnectedLayer` gets the ActivationFunction enum and not
a function view anymore
- SSE2 optimization moved into `FullyConnectedLayer::ComputeOutputSse2`
- layer name added for improved logs
Bug: webrtc:10480
Change-Id: Ida4903a67655e19ef0464f378c433c1f6e96dca7
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195444
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32766}
This reverts commit 0496a4121188a26013dca007bf6e9a7ab6d961b6.
Reason for revert: Causes unexpected changes in perf tests.
Original change's description:
> Add class InterArrivalDelta to goog_cc
>
> This cl copies modules/remote_bitrate_estimator/inter_arrival.x to inter_arrival.h and interrival_delta.cc in goog_cc in the first patchset.
> In the following- this class is modified to use webrtc::Timestamp and webrtc::Timedelta in order to avoid having to use 24 bit time repressentation.
>
> Bug: none
> Change-Id: I9befe6e3e283cf7e21efa974ae33e8a83e26cbe6
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/194004
> Commit-Queue: Per Kjellander <perkj@webrtc.org>
> Reviewed-by: Christoffer Rodbro <crodbro@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#32733}
TBR=perkj@webrtc.org,crodbro@webrtc.org
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: none
Change-Id: I725b246f6ec0c293cb3ada39b1a65a14ef9a001e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/196343
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Christoffer Rodbro <crodbro@webrtc.org>
Commit-Queue: Christoffer Rodbro <crodbro@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32765}
Changes:
1) Scoped class
This is a special class for GLib based objects which we need to manually
delete with different functions. Wrapping these objects into Scoped class
will destroy them automatically when they go out of scope.
2) Window sharing support
Unlike screen sharing, with window sharing we are required to obtain more
information from the PipeWire stream, like video crop metadata, which we
use to properly set size of our buffer.
3) Support for DmaBuf and MemFd buffer types
As of now, we expected the PipeWire stream will provide only plain data
which we just need to copy to our buffer. We now add support for new
buffer types, which are often preferred for better effeciency.
4) Minor bugfixes:
a) Additionally accept PipeWire streams using alpha channels (BGRA, RGBA)
b) Add lock over PipeWire loop to prevent potential issues until we fully
intialize everything we need
c) When obtaining buffers, make sure we work with the latest one
Bug: chromium:682122
Change-Id: I64638d5dcbe18e7280550dca0b01b17c511ac98a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/194100
Commit-Queue: Jamie Walch <jamiewalch@chromium.org>
Reviewed-by: Jamie Walch <jamiewalch@chromium.org>
Cr-Commit-Position: refs/heads/master@{#32763}
This reverts commit 6e7167456b5eba36c7985d6a74f1d191958d4e0f.
Reason for revert: Unexpected perf change
Original change's description:
> Adds experimental libvpx VP9 speed settings.
>
> Using the field trial WebRTC-VP9-PerformanceFlags, this CL allows you to
> configure the libvpx VP9 encoder with a list of flags to affect the
> quality vs speed tradeoff. This CL adds support for:
>
> * Speed (effort), for the temporal base layer frames
> * Speed for higher (non-base) layer frames
> * De-blocking (as part of the loopfilter) enabled for:
> 0 = all frames
> 1 = all but frames from the highest temporal layer
> 2 = no frames
>
> Each entry in the list has a threshold in min number of pixels needed
> for settings in the entry to apply.
>
> Example: Two spatial layers (180p, 360p) with three temporal
> layers are configured. Field trial "WebRTC-VP9-PerformanceFlags" set to:
> "min_pixel_count:0|129600,base_layer_speed:5|8,high_layer_speed:7|8,deblock_mode:1|2"
> This translates to:
> S0:
> - TL0: Speed 5, deblocked
> - TL1: Speed 8, deblocked
> - TL2: Speed 8, not deblocked
> S1:
> - TL0: Speed 7, not deblocked
> - TL1: Speed 8, not deblocked
> - TL2: Speed 8, not deblocked
>
> Bug: webrtc:11551
> Change-Id: Ieef6816d3e0831ff53348ecc4a90260e2ef10422
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/188461
> Reviewed-by: Michael Horowitz <mhoro@webrtc.org>
> Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
> Commit-Queue: Erik Språng <sprang@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#32749}
TBR=sprang@webrtc.org,ssilkin@webrtc.org,mhoro@webrtc.org
Change-Id: If910963441ac1a0e002aac7066791c7cc7764a1a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: webrtc:11551
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/196344
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32762}
I was not able to reproduce chromium:1146676 locally, so the change in merge.cc is a speculative fix.
Bug: chromium:1146835, chromium:1146676, chromium:1137226
Change-Id: I14472ba5b41e58b2d5f27d9833249c14505af18f
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/194264
Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32759}
The HMM based transparent mode classifier is disabled until an issue
with diverging filters is resolved.
Bug: chromium:1155071
Change-Id: Iee249869f6ece1e48e834b3a4b9249c69a51286c
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/196341
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32756}
This CL brings a large improvement to the RNN VAD CPU performance
by finally using `VectorMath::DotProduct()` for pitch search.
The realtime factor improved from about 390x to 570x for SSE2
(+180x, 45% faster) and to 610x for AVX2 (+235x, 60% faster).
RNN VAD benchmark results:
```
+-----+-------+------+------+
| run | none* | SSE2 | AVX2 |
+-----+-------+------+------+
| 1 | 393x | 572x | 618x |
| 2 | 388x | 568x | 607x |
| 3 | 393x | 564x | 599x |
+-----+-------+------+------+
```
*: baseline, no SIMD used for pitch search, but SSE2 used for the RNN
Results obtained as follows:
1. Force SSE2 in `DISABLED_RnnVadPerformance` for the RNN part in
order to measure the baseline correctly:
```
RnnBasedVad rnn_vad({/*sse2=*/true, /*avx2=*/true, /*neon=*/false});
```
2. Run the test:
```
$ ./out/release/modules_unittests \
--gtest_filter=*RnnVadTest*DISABLED_RnnVadPerformance* \
--gtest_also_run_disabled_tests --logs
```
Bug: webrtc:10480
Change-Id: I89a2bd420265540026944b9c0f1fdd4bfda7f475
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195001
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32755}
Using the field trial WebRTC-VP9-PerformanceFlags, this CL allows you to
configure the libvpx VP9 encoder with a list of flags to affect the
quality vs speed tradeoff. This CL adds support for:
* Speed (effort), for the temporal base layer frames
* Speed for higher (non-base) layer frames
* De-blocking (as part of the loopfilter) enabled for:
0 = all frames
1 = all but frames from the highest temporal layer
2 = no frames
Each entry in the list has a threshold in min number of pixels needed
for settings in the entry to apply.
Example: Two spatial layers (180p, 360p) with three temporal
layers are configured. Field trial "WebRTC-VP9-PerformanceFlags" set to:
"min_pixel_count:0|129600,base_layer_speed:5|8,high_layer_speed:7|8,deblock_mode:1|2"
This translates to:
S0:
- TL0: Speed 5, deblocked
- TL1: Speed 8, deblocked
- TL2: Speed 8, not deblocked
S1:
- TL0: Speed 7, not deblocked
- TL1: Speed 8, not deblocked
- TL2: Speed 8, not deblocked
Bug: webrtc:11551
Change-Id: Ieef6816d3e0831ff53348ecc4a90260e2ef10422
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/188461
Reviewed-by: Michael Horowitz <mhoro@webrtc.org>
Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
Commit-Queue: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32749}
This CL adds a new library for the RNN VAD that provides (optimized)
vector math ops. The scheme is the same of the `VectorMath` class of AEC3
to ensure correct builds across different platforms.
Bug: webrtc:10480
Change-Id: I96bcfbf930ca27388ab5f2d52c022ddb73acf8e6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/194326
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32741}
In preparation for adding AVX2 code, a safe scheme to support
different SIMD optimizations is added.
Safety features:
- AVX2 kill switch to stop using it even if supported by the
architecture
- struct indicating the available CPU features propagated from
AGC2 to each component; in this way
- better control over the unit tests
- no need to propagate individual kill switches but just
set to false features that are turned off
Note that (i) this CL does not change the performance of the RNN VAD
and (ii) no AVX2 optimization is added yet.
Bug: webrtc:10480
Change-Id: I0e61f3311ecd140f38369cf68b6e5954f3dc1f5a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/193140
Reviewed-by: Per Åhgren <peah@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32739}
AudioAttributes::getAllowedCapturePolicy was added in API Level 29.
Update WebRtcAudioTrack to add API Level check before using the API.
Bug: webrtc:12250
Change-Id: Ica6604eb1d7fa736a0e64729a022eefcfb7b3020
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195941
Commit-Queue: Gaurav Vaish <gvaish@chromium.org>
Reviewed-by: Henrik Andreassson <henrika@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32735}
For some time now, calls to EnumerateCapturableWindows could lead to a
deadlock if an application's main thread is waiting on the thread that
is running EnumerateCapturableWindows. This is because calls to
GetWindowText and GetWindowTextLength send a message to the window if
the window is owned by the current process. Since the main thread is
waiting on us, it will never reply to this message and we will hang.
This happens occasionally in Chromium when tearing down the
NativeDesktopMediaList object, e.g. when a user clicks "cancel" on
the capture target picker.
We can avoid this deadlock by checking if the window we are querying
is owned by the current process, and if it is then we must ensure it
is responding to messages before we call a GetWindowText* API.
This change also adds a unit test for this scenario. We create a
window and force it to be unresponsive by creating a deadlock, and
then call GetWindowList and (with the new changes) we should not
hang. Without the new changes to GetWindowListHandler, this test
would hang.
Change-Id: I2523cd735f96fd7ea60708c30cd22e5b525803f0
Bug: chromium:1152841
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195365
Commit-Queue: Austin Orion <auorion@microsoft.com>
Reviewed-by: Jamie Walch <jamiewalch@chromium.org>
Cr-Commit-Position: refs/heads/master@{#32734}
This cl copies modules/remote_bitrate_estimator/inter_arrival.x to inter_arrival.h and interrival_delta.cc in goog_cc in the first patchset.
In the following- this class is modified to use webrtc::Timestamp and webrtc::Timedelta in order to avoid having to use 24 bit time repressentation.
Bug: none
Change-Id: I9befe6e3e283cf7e21efa974ae33e8a83e26cbe6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/194004
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Christoffer Rodbro <crodbro@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32733}
When building WebRTC.framework, building the XCTest test runner is a
problem because it requires Chromium's //base checkout. This workaround
allows to skip that.
No-Presubmit: True
Bug: webrtc:12134
Change-Id: I0d99bd03f27911f46679ee91b0120e7121d1c7d7
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/196081
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Jeremy Leconte <jleconte@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32732}
This CL renames webrtc guava dependencies from
third_party/guava:guava_android_java to
//third_party/android_deps:guava_android_java
This is in preparation for deleting third_party/guava:guava_android_java
BUG=chromium:2560401
No-Presubmit: True
Change-Id: If9227f4ac4d24386896c47eeb38142a76a27a4ea
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195720
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32730}
As requested on bugs.webrtc.org/12096#c2, this CL adds a Chromium
metric OWNERS in order to always have their review when WebRTC's UMA
metrics are updated.
Bug: webrtc:12096
Change-Id: Icd9ab7dda5f7a4ba6ac078f667c1fd39f3314123
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/191702
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32728}
Assume bitrate is evenly distributed between frames. This is wrong for
uneven frame sizes and will underestimate the overhead for simulcast.
However, it will be more correct than the current calculation,
especially for low bitrates when each frame is smaller than one packet.
This is also when overhead matters more since it is a larger fraction
of the total bitrate.
It is also unclear what will happen when using FEC.
Bug: b/166341943
Change-Id: I247b9d0fc7a8ad5daa9b577f55ec16c56efa34c3
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195221
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Jakob Ivarsson <jakobi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32725}
This CL breaks out descriptor specific parts into separate classes. All logic in the newly added classes is just copy pasted from the (previously massive) RtpFrameReferenceFinder with the exception of how frames are being returned, which is now done via return value rather than a callback. Basically, all interesting changes have been made in the RtpFrameReferenceFinder.
Bug: webrtc:12221
Change-Id: I5f958d2fbf4b77ba11c3c6c01d8d0d80e325be60
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/195448
Commit-Queue: Philip Eliasson <philipel@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32717}
