DxgiOutputDuplicator objects hold a reference to the last frame that
they succesfully captured by maintaining a reference to the
SharedDesktopFrame that was passed as their target. This is done because
the DirectX capture APIs may fail to provide an update if there has been
no (or no substantial) change since the last capture call was made.
However, the higher levels of this capture stack
(DxgiDuplicatorController and ScreenCapturerWinDirectX), were unaware of
this, and assumed that the caller of CaptureFrame is the only one who
may have held a reference to the frame. Thus, when CaptureFrame is
called, the DirectX screen capturer assumes that the oldest frame in its
queue can be safely reused.
In the steady state, where capture is not being switched between
monitors, this is fine as there are no competing DxgiOutputDuplicators
being run and this assumption mostly holds true (or the frame is being
overwritten only when the DxgiOutputDuplicator is also done holding it).
However, when capture is being rapidly switched between multiple targets
(e.g. to show a preview of each of the available monitors), this can
result in a frame being held by one DxgiOutputDuplicator being passed to
another as a valid target and overwritten. In the common case of only a
single monitor this is essentially the same as steady state capture,
where there are no competing DxgiOutputDuplicator. In the other common
case of two monitors being captured, the fact that the
ScreenCaptureFrameQueue has two frames ends up masking this issue. Since
each monitor is captured in the same order, the same frame ends up
getting passed to each DxgiOutputDuplicator, so no data actually ends
up getting overwritten. In the case of 3 monitors, the 1st and 3rd
monitor end up sharing a frame, which when capture fails on one of them
surfaces as the other monitor being duplicately shown.
This change addresses the issue by ensuring that each screen that the
ScreenCapturerWinDirectX *actually attempts* to capture, gets it's own
FrameQueue, and thus essentially brings us back to the "steady state"
case for each monitor. Note that this does increase memory usage of
capturers that are switched between multiple targets by 2 frames/target
used (and actually attempted to be captured).
Alternatives considered:
DxgiOutputDuplicator makes a copy of the frame, rather than holding
a reference
This was rejected because adding an additional copy for every
capture upon getting a new frame, would expensive and could degrade
performance.
Allow the DxgiOutputDuplicators to "fail" when there has been no update
This would result in either a breaking change to the API for consumers
or would require the ScreenCapturerWinDirectX to track these last
captured frames; which would result in essentially the same approach,
but with less abstraction for re-using the frames.
Bug: chromium:1296228
Change-Id: I5442ec40e9f234046010b562b258db63693ccc6b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/256043
Reviewed-by: Mark Foltz <mfoltz@chromium.org>
Commit-Queue: Alexander Cooper <alcooper@chromium.org>
Cr-Commit-Position: refs/heads/main@{#36295}
This class is accessed by Electron for its desktop capture support,
but it breaks with component builds on Windows because the symbols
aren't exported by the dll.
No behavior change at runtime, only modifies the generated .lib
when building as a shared library (static builds are unchanged).
Bug: None
Change-Id: I5dc606846de990c1bf4d375ddbb1c73dfc512762
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/230200
Reviewed-by: Joe Downing <joedow@chromium.org>
Reviewed-by: Niels Moller <nisse@webrtc.org>
Reviewed-by: Jamie Walch <jamiewalch@chromium.org>
Commit-Queue: Jamie Walch <jamiewalch@chromium.org>
Cr-Commit-Position: refs/heads/main@{#34858}
This name change communicates that the recursive critical section
should not be used for new code.
The relevant files are renamed rtc_base/critical_section* ->
rtc_base/deprecated/recursive_critical_section*
Bug: webrtc:11567
Change-Id: I73483a1c5e59c389407a981efbfc2cfe76ccdb43
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/179483
Commit-Queue: Markus Handell <handellm@webrtc.org>
Reviewed-by: Niels Moller <nisse@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Tommi <tommi@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#31754}
This is a trivial change to add more logs in DX capturer components for
debugging purpose.
Bug: chromium:764258
Change-Id: I406127d838a522f0226720434e840c7163b4719d
Reviewed-on: https://webrtc-review.googlesource.com/3541
Commit-Queue: Zijie He <zijiehe@chromium.org>
Reviewed-by: Jamie Walch <jamiewalch@chromium.org>
Cr-Commit-Position: refs/heads/master@{#19960}
In order to eliminate the WebRTC Subtree mirror in Chromium,
WebRTC is moving the content of the src/webrtc directory up
to the src/ directory.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
TBR=tommi@webrtc.org
Bug: chromium:611808
Change-Id: Iac59c5b51b950f174119565bac87955a7994bc38
Reviewed-on: https://webrtc-review.googlesource.com/1560
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Henrik Kjellander <kjellander@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#19845}
