It do the following:
The SurfaceTexture.updateTexImage() calls are moved from the video renderers into MediaCodecVideoDecoder, and the destructor of the texture frames will signal MediaCodecVideoDecoder that the frame has returned. This CL also removes the SurfaceTexture from the native handle and only exposes the texture matrix instead, because only the video source should access the SurfaceTexture.
It moves the responsibility of calculating the decode time to Java.
Patchset2 Refactor MediaCodecVideoDecoder to drop frames if a texture is not released.
R=magjed@webrtc.org
Review URL: https://codereview.webrtc.org/1440343002 .
Cr-Commit-Position: refs/heads/master@{#10706}
This makes it clearer this code not meant to be used as an API.
I could not find any use of this in downstream code.
BUG=webrtc:5095
TESTED=git cl try -c --bot=android_compile_rel --bot=linux_compile_rel --bot=win_compile_rel --bot=mac_compile_rel --bot=ios_rel --bot=linux_gn_rel --bot=win_x64_gn_rel --bot=mac_x64_gn_rel --bot=android_gn_rel -m tryserver.webrtc
R=stefan@webrtc.orgTBR=magjed@webrtc.org
Review URL: https://codereview.webrtc.org/1440873005 .
Cr-Commit-Position: refs/heads/master@{#10699}
Not fully tested yet. Verified in test loopback application
with fake VP9 codec factory.
Assume that encoder generates bitstream in non flexible mode with
one temporal and one spatial layers.
R=magjed@webrtc.org
Review URL: https://codereview.webrtc.org/1451953002 .
Cr-Commit-Position: refs/heads/master@{#10695}
To avoid breaking downstream, the "interface" directories were copied
into a new "common_video/include" dir. The old headers got pragma
warnings added about deprecation (a very short deprecation since I plan
to remove them as soon downstream is updated).
The header guards are also identical to avoid mixing them up in the transition.
BUG=webrtc:5095
TESTED=Passing compile-trybots with --clobber flag:
git cl try --clobber --bot=win_compile_rel --bot=linux_compile_rel --bot=android_compile_rel --bot=mac_compile_rel --bot=ios_rel --bot=linux_gn_rel --bot=win_x64_gn_rel --bot=mac_x64_gn_rel --bot=android_gn_rel -m tryserver.webrtc
Review URL: https://codereview.webrtc.org/1418913006
Cr-Commit-Position: refs/heads/master@{#10659}
The original purpose with keeping one pending frame in SurfaceViewRenderer was to reduce latency for the first rendered frame when we are waiting for the Surface to be created. However, it is very dangerous to hold a pending frame indefinitely when used with a SurfaceTexture, because the SurfaceTexture only has one frame and thus holding a frame in the renderer will freeze everything and typically cause timeout crashes.
Review URL: https://codereview.webrtc.org/1435413006
Cr-Commit-Position: refs/heads/master@{#10638}
Reason for revert:
Causes fallback to SW decoder if a renderer is put in the background.
Original issue's description:
> Patchset 1 is a pure
> revert of "Revert of "Android MediaCodecVideoDecoder: Manage lifetime of texture frames" https://codereview.webrtc.org/1378033003/
>
> Following patchsets move the responsibility of calculating the decode time to Java.
>
> TESTED= Apprtc loopback using H264 and VP8 on N5, N6, N7, S5
>
> Committed: https://crrev.com/9cb8982e64f08d3d630bf7c3d2bcc78c10db88e2
> Cr-Commit-Position: refs/heads/master@{#10597}
TBR=magjed@webrtc.org,glaznev@webrtc.org
NOPRESUBMIT=true
NOTREECHECKS=true
Review URL: https://codereview.webrtc.org/1441363002 .
Cr-Commit-Position: refs/heads/master@{#10637}
This CL attempts to annotate accesses on >16 API levels using as
small scopes as possible. The TargetApi notations mean "yes, I know
I'm accessing a higher API and I take responsibility for gating the
call on Android API level". The Encoder/Decoder classes are annotated
on the whole class, but they're only accessed through JNI; we should
annotate on method level otherwise and preferably on private methods.
This patch also fixes some compiler-level deprecation warnings (i.e.
-Xlint:deprecation), but probably not all of them.
BUG=webrtc:5063
R=henrika@webrtc.org, kjellander@webrtc.org, magjed@webrtc.org
Review URL: https://codereview.webrtc.org/1412673008 .
Cr-Commit-Position: refs/heads/master@{#10624}
This reverts commit 12f680214e28dc5f0a13ac8afc0d1445f89e67e6.
Original cl in https://codereview.webrtc.org/1396073003/
Prepare MediaCodecVideoEncoder for surface textures.
This refactors MediaVideoEncoder to prepare for adding support to encode from textures. The C++ layer does not have any functional changes.
- Moves ResetEncoder to always work on the codec thread
- Adds use of ThreadChecker.
- Change Java MediaEncoder.Init to return true or false and introduce method getInputBuffers.
- Add simple unit test for Java MediaCodecVideoEncoder.
The pure revert of the revert is in patchset 1.
Patchset 2, moves getting the input buffer to before storing pending timestamps etc to fix b/24984012.
BUG=webrtc:4993 b/24984012
Review URL: https://codereview.webrtc.org/1406203002
Cr-Commit-Position: refs/heads/master@{#10622}
This make sure that the texture copy is syncronized.
To reproduce the problem I:
Reverted "Revert of "Android MediaCodecVideoDecoder: Manage lifetime of texture frames" https://codereview.webrtc.org/1378033003/"
commit 543b6ca30a43eeb069c699291460ce6bacc7959d.
Reverted "Enable SurfaceViewRenderer for AppRTCDemo"
commit 7076729c57c27aa813760d2038be02c36f4d7649.
and ran ApprtDemo in loopback and changed the orientation a couple of times.
TBR=glaznev@webrtc.org
Review URL: https://codereview.webrtc.org/1437823002
Cr-Commit-Position: refs/heads/master@{#10598}
revert of "Revert of "Android MediaCodecVideoDecoder: Manage lifetime of texture frames" https://codereview.webrtc.org/1378033003/
Following patchsets move the responsibility of calculating the decode time to Java.
TESTED= Apprtc loopback using H264 and VP8 on N5, N6, N7, S5
Review URL: https://codereview.webrtc.org/1422963003
Cr-Commit-Position: refs/heads/master@{#10597}
Add resource name to log outputs to distinguish local renderer from remote renderer.
This Cl also adds some thread checks and factors out a small helper function makeBlack().
Review URL: https://codereview.webrtc.org/1420203003
Cr-Commit-Position: refs/heads/master@{#10596}
This change VideoCapturerAndroid to attempt 3 times with a period of 300ms to open the camera if it fails.
This is so that if another application have it already opened, it would have more time to release it.
BUG=b/25190234
Review URL: https://codereview.webrtc.org/1422023007
Cr-Commit-Position: refs/heads/master@{#10559}
If SurfaceViewRenderer can't keep up with the stream of incoming frames it has to drop frames. Currently, new frames are dropped until the old pending frame is rendered. This CL drops the old pending frame instead.
Review URL: https://codereview.webrtc.org/1417063005
Cr-Commit-Position: refs/heads/master@{#10558}
Also distinguish between camera failures and failures due to that buffers has not been returned.
Adds unit tests for making sure CameraEventHandler.onError is triggered if frames are not returned.
BUG=b/25514149
Review URL: https://codereview.webrtc.org/1415013006
Cr-Commit-Position: refs/heads/master@{#10555}
JNI already has jstring<->UTF8 string conversion, so using that should
save ~1mb off android binaries (ICU is *large*), probably around
300-400k after compression.
BUG=
Review URL: https://codereview.webrtc.org/1430023005
Cr-Commit-Position: refs/heads/master@{#10545}
MediaCodec.stop() call may hang in some rear cases. To avoid
application hang this call need to be done on separate thread and
possible error reported back to application.
Application may elect to continue executing and use another codec
instance for encoding/decoding or stop the call and exit.
BUG=b/24339249
R=magjed@webrtc.org
Review URL: https://codereview.webrtc.org/1425143005 .
Cr-Commit-Position: refs/heads/master@{#10467}
Plus log first few decoder frames in and out events.
BUG=b/25287910
Review URL: https://codereview.webrtc.org/1423843005
Cr-Commit-Position: refs/heads/master@{#10439}
I replaced quitSafely() with a CountDownLatch. The reason for not using ThreadUtils.invokeUninterruptibly() is that I want to stop accepting frames asap, and invokeUninterruptibly() would still accept frames during the waiting time.
BUG=webrtc:4742
Review URL: https://codereview.webrtc.org/1418223002
Cr-Commit-Position: refs/heads/master@{#10393}
The purpose with this change is to support older API levels by replacing EGL14 (API lvl 17) with EGL10 (API lvl 1). The main purpose is to lower API lvl requirement for SurfaceViewRenderer from API lvl 17 to API lvl 15. Also, camera texture capture will work on API lvl < 17 (and texture encode/decode in MediaCodec, but we don't use MediaCodec below API lvl 18?).
GLSurfaceView/VideoRendererGui is already using EGL10.
EGL 1.1 - 1.4 added new functionality, but won't affect performance. We don't need the functionality, so there should be no reason to not use EGL 1.0.
I have profiled AppRTCDemo with Qualcomm Trepn Profiler on a Nexus 5 and Nexus 6 and couldn't see any difference.
Specifically, this CL:
* Update EglBase to use EGL10 instead of EGL14.
* Update imports from EGL14 to EGL10 in a lot of files (plus changing import order in some cases).
* Update VideoCapturerAndroid to always support texture capture.
Review URL: https://codereview.webrtc.org/1396013004
Cr-Commit-Position: refs/heads/master@{#10378}
Add events to track when camera is requested to open,
when first camera frame is available and when camera is
closed.
BUG=b/24271359
R=perkj@webrtc.org
Review URL: https://codereview.webrtc.org/1398793005 .
Cr-Commit-Position: refs/heads/master@{#10306}
This reverts commit 90754174d98d6b71fd4aaed897bd54980f7e59c4.
Revert "Fix use of scaler in MediaCodecVideoEncoder"
This reverts commit ec93628e75fdb81f23635b39b5f3da846bcefd21.
R=magjed@webrtc.orgTBR=glaznev@webrtc.org
BUG=webrtc:4993 b/24984012
Review URL: https://codereview.webrtc.org/1407263002 .
Cr-Commit-Position: refs/heads/master@{#10300}
The code that depends on the reverted CL is disabled but not removed. NativeHandleImpl is reverted to the previous implementation, and the new implementation is renamed to NativeTextureHandleImpl. Texture capture can not be used anymore, because it will crash in peerconnection_jni.cc.
Reason for revert:
Increased HW decoder latency and crashes related to that. Also suspected cause of video tearing.
Original issue's description:
> This CL should be the last one in a series to finally
> unblock camera texture capture.
>
> The SurfaceTexture.updateTexImage() calls are moved from
> the video renderers into MediaCodecVideoDecoder, and the
> destructor of the texture frames will signal
> MediaCodecVideoDecoder that the frame has returned. This
> CL also removes the SurfaceTexture from the native handle
> and only exposes the texture matrix instead, because only
> the video source should access the SurfaceTexture.
>
> BUG=webrtc:4993
> R=glaznev@webrtc.org, perkj@webrtc.org
>
> Committed: https://crrev.com/91b348c7029d843e06868ed12b728a809c53176c
> Cr-Commit-Position: refs/heads/master@{#10203}
TBR=glaznev
BUG=webrtc:4993
Review URL: https://codereview.webrtc.org/1394103005
Cr-Commit-Position: refs/heads/master@{#10288}
This CL should reduce the number of timeouts for dequeueInputBuffer() which results in the log "MediaCodecVideo: dequeueInputBuffer error" followed by software fallback for VP8/VP9 and codec restart for H264.
A timeout always happen for dequeueInputBuffer() when frames_received_ > frames_decoded_ + num_input_buffers. The following code tries to drain the decoder before enqueuing more input buffers:
// Try to drain the decoder and wait until output is not too
// much behind the input.
if (frames_received_ > frames_decoded_ + max_pending_frames_) {
ALOGV("Received: %d. Decoded: %d. Wait for output...",
frames_received_, frames_decoded_);
if (!DeliverPendingOutputs(jni, kMediaCodecTimeoutMs,
true /* dropFrames */)) {
ALOGE << "DeliverPendingOutputs error";
return ProcessHWErrorOnCodecThread();
}
if (frames_received_ > frames_decoded_ + max_pending_frames_) {
ALOGE << "Output buffer dequeue timeout";
return ProcessHWErrorOnCodecThread();
}
...
}
However, for H264, |max_pending_frames_| can currently be larger than the number of input buffers so that the code above is never executed. This CL limits |max_pending_frames_| to the number of input buffers.
TBR=glaznev
BUG=b/24867188,b/24864151
Review URL: https://codereview.webrtc.org/1394303005
Cr-Commit-Position: refs/heads/master@{#10273}
SurfaceViewHelper requires EGL14 that was added in API level 17. Since the SurfaceViewHelper is only neeed when we capture to textures, this cl change back to not use it when we are capturing to byte buffers.
Also, thread.quitsafely was added in level 18. Instead a new ThreadUtil method has been added for this.
BUG=b/24782220
TEST = run
ninja -C out/Debug libjingle_peerconnection_android_unittest && CHECKOUT_SOURCE_ROOT=`pwd` build/android/adb_install_apk.py --debug out/Debug/apks/libjingle_peerconnection_android_unittest.apk && ./third_party/android_tools/sdk/platform-tools/adb shell am instrument -w -e class org.webrtc.VideoCapturerAndroidTest org.webrtc.test/android.test.InstrumentationTestRunner on a device running Android 4.1 (I tried Nexus 7, the first version)
Review URL: https://codereview.webrtc.org/1401023003
Cr-Commit-Position: refs/heads/master@{#10265}
