Fixes: 153130850
Test: None

Change-Id: I358db66d8ddc75efcff2e2dd9bf4ae4518c1fc01

Two build fixes:
- Use C++17 since it is needed by dependencies in
frameworks/native, and C++17 is used by default.
- Rename header MediaSource.h

Bug: 146373480
Test: mma -j500
Change-Id: I3dacea9645c1e2919888e7f63535d531c32947e9

IMPLEMENTATION_DEFINED is a special format for encoder input frames which is
backed by gralloc.In ChromeOS, IMPLEMENTATION_DEFINED may be backed by
YCBCR_420_888 or RGBX_8888. As for C2AllocationGralloc::map(), any unrecognized
format will be regarded as YCBCR_420_888.

In YUV-backed case, C2AllocationGralloc::map() could map and provide layout
correctly. However in RGB-backed case, map() would fail and the caller will get
an empty layout instance.

ImplDefinedToRGBXMap can provide the layout for RGB-backed
IMPLEMENTATION_DEFINED format case. When the instance is created, it will own
the GraphicBuffer wrapped from input block and lock it, to provide the address,
offset, and rowInc information. The GraphicBuffer will be unlocked and released
under destruction.

Bug: 73059339
Test: android.media.cts.MediaRecorderTest#testSurfaceRecording
Change-Id: Iae8112e9dc8c30d2f0a08ae2a4cc0d7fb9db1c03
(cherry picked from commit 28ae20917491a301ed0122e280ddbb4152151cb2)

Bug: 144474391
Test: CtsMediaTestCases
Change-Id: I0fbb41a8b9f52d10b964b366e58f21798640f40a
(cherry picked from commit 39acc2a8f3a4a9ec10b0f23aada7ce2ad20eb175)

Several minor bugs were fixed caused from library DEP migration.

Removed obsolete codes under cmds/.

Use c++17 instead for building C2VDAComponent_test otherwise there are many
complicated build errors.

In C2V[D|E]ACompIntf_test TestUnsupportedParam, C2ComponentTemporalInfo is
replaced by C2ComponentTimeStretchTuning because the former is removed.

Bug: 145262637
Test: mmm external/v4l2_codec2
Change-Id: Ied744d025ccdda20533e823a3dfd17edc8f12d54

This change ports the decoder portion of the arc codec-tests [1] into
an Android application. The codec-test source is copied with minimal
changes - just renaming some header includes, replacing main() with a
RunDecoderTests function, removing most references to 'arc', and auto
formatting. The new code is the Android activity and the jni glue.

This change also enables presubmit format hooks for java and fixes a
typo in the clang_format hook configuration.

[1] https://chromium.googlesource.com/chromiumos/platform2/+/57c0c80bdb0d671f4fb2293c4c5fa7b5ed9dddf0/arc/codec-test/

Test: existing decoder tests pass when following instructions in readme
Bug: 142423642
Bug: 143584325
Bug: 144681449

Change-Id: I31e02cae5c10bb9936769c75efafe1fa9971bdbb
(cherry picked from commit 46344d0a85479760c5294f21de2efc106ab54048)

V4L2 registers video input/output devices as /dev/videoX.
So, we can use this name pattern instead of /dev/video-decX.

Bug: 140275575
Test: Play YouTube on ARCVM with virtio-vdec

Change-Id: Iacd403c46ef1b1f86c8a1675624a11efe86f6c74
Signed-off-by: Keiichi Watanabe <keiichiw@google.com>
(cherry picked from commit c552e9b97437a15d2311aafb14182b7a7f78bcfd)

C2VEAComponent::IntfImpl assumes a platform supports H264 encoding.
C2VEAComponent::IntfImpl's ctor crashes on a platform that doesn't support.
The ctor runs always on ARC++ startup. This fixes the crash.

Bug: None
Test: No crash happens in libv4l2_codec2.so on ARC++ startup
Change-Id: If8c312b5583cfe9582d1c73317fcd901fa018898
(cherry picked from commit c559c23c692220894db1972919fa53818c12838e)

gralloc stack might need the info that an allocated buffer will be consumed
by hardware video encoder. Specifying BufferUsage::VIDEO_ENCODER in Android
results in GRALLOC_USAGE_HW_VIDEO_ENCODER. This sets VIDEO_ENCODER mask for
a destination buffer on converting in C2VEAFormatConverter. Without
C2VEAFormatConverter, the input buffer given by C2VEAComponent client is
passed as-is. That said, it is another part that has a responsibility to
set the proper usage mask.

Bug: 144135251
Test: ARC++ encoder on kukui
Change-Id: I5cf2d462677ba14643fc564d99e2a8a64145cbce
(cherry picked from commit 7e4a6634db0d12a86eaf1a5ead29206076046fc4)

[automerger skipped] DO NOT MERGE - qt-qpr1-dev-plus-aosp-without-vendor@5915889 into stage-aosp-master am: b214db3d
am: b260c87a -s ours
am skip reason: subject contains skip directive

Change-Id: I5abf7c9aa196768301f93b1a6901c24341b03c60

am: b214db3d

Change-Id: I9551f23150868fa262f4f70fba3fe45ea5e21fb4

Bug: 142003500
Change-Id: Ia0e2379efe718832962021332c5e0f232e15a1aa

Most ChromeOS devices are able to create 8 concurrent VDAs for ARC video
clients. This is set as a constant in GAVDA to reject more than 8 clients
by INSUFFICIENT_RESOURCES.

However, for new device Kukui, the max concurrent VDA number is 4. Since
we cannot change the constant nor obtain this value dynamically. The
alternative way is in C2VDAComponent side, we treat initialization error
PLATFORM_FAILURE as INSUFFICIENT_RESOURCES. In fact, Codec2.0 SW codecs
also adapt this method.

Bug: 143194688
Test: android.media.cts.MediaCodecCapabilitiesTest#testGetMaxSupportedInstances
Change-Id: Ia25c9b27976bb71dd1d7166eece12e88fbfb4659
(cherry picked from commit 78981bbe6b64597aaf67d2fc67083715af073bf2)

In current C2VEAFormatConverter design, it always does format conversion and
copy for each frame even if the source and target format are identical.

This CL we added the scheme to avoid unnecessary copy while detecting the
source format is already the same as target.

Bug: 73059339
Bug: 118544836
Test: CtsMediaTestCases android.media.cts.MediaRecorderTest on Soraka
Test: nexus7 camera App on Soraka
Change-Id: I3ab1387cbc7508777d44f0f4319f5e95cd2da17d
(cherry picked from commit 2a1d153e160b6ccb22b7249899161f9cb5cc9097)

The maximal supported AVC level is adjusted up to Level5.1 which is the same as
legacy ArcVideoEncoder.

ProfileLevelSetter is refined to be able to set lower level, e.g. Level1.0. This
is the potential issue in current Codec2.0 config update procedure. Because config
setting is one by one, the order of config setting may affect the final setting if
there are dependent configurations.

Bug: 140407694
Test: android.media.cts.MediaRecorderTest#testProfileAvcBaselineLevel1
Change-Id: Iea9bfc4d8b6fd7a8bd8ffa3a8f262ff24aa86518
(cherry picked from commit ca6e942d0926cb9331a7f81ae6bda5d92dc29d73)

In VEA API, UseBitstreamBuffer() doesn't require to correlate to Encode(). That
is, in C2VEAComponent's aspect, output buffers are not guaranteed to be FIFO.

Instead, we should apply a unique buffer index for each output buffer and
assign it to the corresponding BitstreamBufferReady callback. It could help
C2VEAComponent to identify which output buffer is returned.

Bug: 73059339
Test: pass arcvideoencoder native e2e test on Kevin.
Test: pass CTS android.media.cts.EncodeDecodeTest on Kevin.
Change-Id: I0c7a7768f9266c3c908dfe4506cbffa47a3459a4
(cherry picked from commit 9bb2cd5745bee36c877b67bc6ad16656cf6eb574)

When component flush is called, component will initialize VEA again and
callback RequireBitstreamBuffers is expected. However, if component is stopped
(released) in the middle of them, RequireBitstreamBuffers callback is no longer
cared. And in fact previous code would cause CHECK error since the component
state is no longer CONFIGURING.

Bug: 73059339
Test: CtsMediaTestCases android.media.cts.MediaCodecTest#testReleaseAfterFlush
Change-Id: Iafac39b8a4a0da0fac0dd899c304ba1d0f35660f
(cherry picked from commit 2feac90f0f51c5f2539f43529912aeb4ddb29cc6)

Previously, C2VEAComponent regarded the workaround CSD-holder C2Work and put
the CSD info extracted by the first output bitstream into it. However, there
are several emerging issues of this workaround which increase the complexity of
our implementation (see b/140975410).

We decided NOT to support the workaround and in practice VEA is not able to
output CSD (SPS/PPS stream) prior to getting any input. C2VEAComponent will
just return the workaround C2Work, and CSD info will be put into the
corresponding C2Work of the first input buffer.

Bug: 140975410
Bug: 73059339
Test: CtsMediaTestCases android.media.cts.VideoEncoderTest#testOtherH264Flex720p
Change-Id: I94922ef189a39ea75d3ffcba729e46c8e149cf6a
(cherry picked from commit 8e0e00d0dea48d105b89864f72d0e5d95a06685f)

When the component is in error state, it is not guaranteed VDA is still
responsible. Rather than sending VDA->Reset() request and waiting on
callback, the component will release VDA adaptor immediately on stop.

Furthermore, after this change we might not guarantee there is no
remaining task after onStop() is executed, which will change the state to
UNINITIALIZED immediately. We modified EXPECT_RUNNING_OR_RETURN_ON_ERROR()
to just return if tasks are executed while component state is
UNINITIALIZED, instead of assertion.

Bug: 139457871
Bug: 140357179
Test: pass android.security.cts.StagefrightTest#testStagefright_bug_21443020 on Kevin
Test: pass CtsMediaTestCases on Kevin

Change-Id: I3eef02f66619cfea8a72bed40b5ac83c282ede10
(cherry picked from commit f7d8029f9f18a0f8be138fed41add63417d5d5da)

The previous implementation has a bug that we assume on start process
VEAAdaptor is already created. In fact, if the component has been
through start --> stop --> start, then VEAAdaptor is nullptr on the
second start process. This CL fixed the above issue.

Bug: 73059339
Test: CtsMediaTestCases android.media.cts.MediaCodecTest#testReconfigureWithoutSurface
Change-Id: I328c9241938d6fa81753f11af14f239b6a24a183
(cherry picked from commit c4146fbc682bcd70250838f96c1555cce1107135)

C2VEAFormatConverter is designed to convert C2VEAComponent input frames into
required pixel format (NV12) and coded size. It will allocate additional
graphic blocks for format conversion.

Bug: 73059339
Test: arcvideoencoder native e2e test on Soraka and Kevin
Change-Id: If2e8198563a2eafebe3cc56c473755c2ccf89988
(cherry picked from commit 8a2903d1aa56102b22e5c48e0607c0d126a40802)

In current C2 framework, CSD info would be reported in different manners
whether encoder is in input surface mode and byte-buffer mode.

Byte-buffer mode:
A work with no input buffer will be queued prior to other works to get CSD
info only (CSD-holder work). When the first bitstream buffer is encoded by
VEA, the component needs to extract the CSD info and put into CSD-holder work
as a parameter.

Surface mode:
No CSD-holder work. When the first bitstream buffer is encoded by VEA, the
component needs to extract the CSD info and put into the same work
corresponding to this bitstream as a parameter.

Bug: 73059339
Test: android.media.cts.EncodeDecodeTest#testEncodeDecodeVideoFromBufferToBufferQCIF
Test: android.media.cts.EncodeDecodeTest#testEncodeDecodeVideoFromSurfaceToSurfaceQCIF
Change-Id: I13ede4c3b725e27cd0eb66b07925fdf93b5bafb4
(cherry picked from commit 0922077bcc55af276402129b23110de9abfee5f7)

In the past we check most of the function tasks to be executed while component
state is running (not UNINITIALIZED) and assert if violated. However, in some
cases we might not guarantee there is no remaining task after onStop() is
executed, which will change the state to UNINITIALIZED immediately.

Instead of assertion, we modified to just return if tasks are executed while
component state is UNINITIALIZED. It makes sense because the component is
already released everything. And it could also avoid unnecessary process
crashes.

Bug: 73059339
Test: android.media.cts.MediaCodecTest#testAbruptStop passed and no crash
Change-Id: Ib5b68ed51f3ba137b981edc24519dd4ba7144532
(cherry picked from commit 23cccc9f965581241573bad6a9f840a8363bb0d7)

Add usage flag BufferUsage::VIDEO_DECODER for video decoder while allocating
graphic buffers. Note that this flag is defined in IAllocator but not in
legacy GRALLOC_USAGE.

Bug: 139859772
Test: check usage bit is propogated to C2AllocatorGralloc
Test: pass GtsExoPlayerTestCases on Soraka
Change-Id: Ief673eff25df34026c66c8941f6b54c9e49c8b11
(cherry picked from commit 89006a485724cbef83b6e220026047639763cdd0)

V4L2VDA assumes a buffer with V4L2_BUF_FLAG_LAST doesn't contain a decoded frame. However, the fact
is a buffer with V4L2_BUF_FLAG_LAST might contain a decoded frame (not always though). This CL
removes the wrong assumption.

Bug: 131122595
Test: ARC++ decoder E2E test (i.e. arc_video_decoder_e2e_test) with pure Android solution
Change-Id: Iaf2d425b019a3d0438552c5ea3d2fb1852043777
(cherry picked from commit fe275e331fb9afab7d04fa01881ff6b802df32b9)

V4L2VDA calls NotifyEndOfBitstreamBuffer() when BitstreamBufferRef is destructed.
BitstreamBufferRef is destructed when a file descriptor info is registered in |input_record|.
This is obviously wrong if we use a file descriptor for bitstream buffer because we don't copy the
content to memory owned by v4l2 driver. Therefore, the bitstream buffer can be overwritten in a
client side for a new bitstream buffer before a v4l2 driver will consume the buffer. The bitstream
buffer shall not be overwritten until a v4l2 driver will consume.

This CL delays NotifyEndOfBitstreamBuffer() by keeping the reference of BitstreamBufferRef until
DQBUF is complete on OUTPUT queue, after which a driver should not refer the buffer defenitely.

Bug: 131122595
Test: arcvideodecoder native e2e test on eve

Change-Id: I5b357f9395acbad53417c843800465138d92d619
(cherry picked from commit 85e5d31bcbd6eb7d5df75355ef664e4bbb3d070a)

Note: the original code is under video_types.(h|cc) in Chromium

Bug: 73059339
Test: compile image
Change-Id: I3a5a71d2a9ea076e6e6807a4bb5342f4c8279624
(cherry picked from commit 24ce720487f1c92229d50e0018631f7eef7a928a)

The previous check of flushed input/output buffer by frame index is not
correct for output. isFlushedState() is introduced to provide a better
approach by checking if the component state is under two situations:
1) flushed by stop, and 2) flushed by flush.

Bug: 73059339
Test: pass arcvideoencoder native e2e test on Kevin and Soraka
Change-Id: Ifb1c2d5ef6098bf95b94b5a4b2d8491b856991dd
(cherry picked from commit 2940ca2475746235f32fbf92dc6b9758e5ab683c)

For C2Work, the output worklet should be assigned with the output bitstream
which is encoded from the corresponding input frame. In VEA's aspect, the
callback of bitstream ready is paired with timestamp, which is obtained from
the corresponding input frame. That is, the component should use timestamp to
match the output bitstream with the pending work.

Bug: 73059339
Test: arcvideoencoder native e2e test on Soraka and Kevin

Change-Id: I235a3d8aa2f1e33de762a25a1e8ce2414a18e629
(cherry picked from commit c541e2c0d99424ced9294b4f7a95b1753e1b7e79)

The number of allocated buffers by a driver by VIDIOC_REQBUFS should not be smaller than requested
unless the driver runs out of memory. This CL correct the exisiting check for CAPTURE queue and add
the check for OUTPUT queue.

Bug: None
Test: ARC e2e decoder test
Change-Id: I95fe5564979b77594ebbfa83815037755e85c38c
(cherry picked from commit 7bbeeecb65916f75bd089e8f118b392494db0da6)

Bug: 136716638
Test: m -j32

Change-Id: Ie3a652c5ecd6d7b597efb6fcfa090f942d5dd4a1
(cherry picked from commit 0d5ae9523d04984d5205b845b2733cf58f681aef)

1. C2VEAAdaptorProxy should be guarded by #ifdef V4L2_CODEC2_ARC
2. Thread-safety attributes are not able to compile by bertha. I
   haven't found the root cause yet so removed them first.

Bug: 73059339
Test: build bertha_x86_64-eng on pi-arcvm-dev
Change-Id: Ib0ea84f4c24072faffd68a8bd4e9d1cee66b1e2b
(cherry picked from commit c8fab05fe388259b3b148400b573f685a5bab870)

Please refer to go/codec2-v4l2-encoder for design details.

Note: there is a hack to pretend YV12 by I420 of input buffers since currently
Chromium VEA only accepts I420 format. YV12 will be used for input byte-buffer
mode, which is adopted by arcvideoencoder native test. By doing so, the encoded
stream will have UV-swapped while playing.

Bug: 73059339
Test: pass arcvideoencoder native e2e test.

Change-Id: Ibfcb19f5a397ccbc934f3800443d19dafb8e42f8
(cherry picked from commit 13e650b2031da1cb6c8948391896dd8fcf7161a1)

Previously we want to use timestamp as the indicator for component to find the
corresponding work on notifying from VEA. However, it would be not robust
enough since timestamp could be repetitive and wrap-back. Index (frame index)
should be the better indicator due to its uniqueness and is always increased
monotonically.

Bug: 73059339
Test: mmm external/v4l2_codec2
Change-Id: I23ac7f1830f2405163604e25f7be92b98fabd949
(cherry picked from commit 155167ddd100311e1aa2b15649b21cb0580bee22)

We are adding video encode accelerator (VEA) related codes into this library.
So we would like to rename it to *_accel.

Bug: 73059339
Test: build image
Change-Id: I66c574b66d4fdf9014db304b0d97c19c248718c1
(cherry picked from commit 7ba9f723f767e78b04b330f66901beeadff4915d)

C2VDAComponent has got offsets and strides info by using C2ConstGraphicBlock::map(). The function
actually maps an owned buffer in memory by calling lockYCbCr with SW flag. We computed offsets from
the based addresses. The computed offsets are not always correct and it's not guaranteed a buffer is
mappable always either. The correct way is to execute lockYCbCr() without SW flag. However,
C2ConstGraphicBlock::map() doesn't have the way. This CL changes C2VDAComponent to directly execute
lockYCbCr() to native_handle in C2GraphicBlock and compute offsets and strides from android_ycbcr.

Bug: 137161142
Test: ExoPlayerDemo app on kevin
Change-Id: Ia8a49ef0960770bc35ba7fc05476e4b86c266202
(cherry picked from commit 9ba0c64347b9bdd75acfb8c2db0857ac37f822e5)

C2VDAComponent allocates a graphic buffer by gralloc. The size is requested by C2VDAAdaptor. gralloc
might adjust stride. C2VDAComponent needs to inform the adjusted size to C2VDAAdaptor by
C2VDAAdaptor::AssignPictureBuffers(). C2VDAComponent calls GraphicBuffer::lockYCbCr with usage=0 to
an output buffer. Note that lockYCbCr doesn't map the buffer with usage=0.

Bug: chromium:979115
Test: Regular playback and Secure playback with ExoPlayerDemo app

Change-Id: Ia3f2dbdc561114dc734eb862ac800aa0ca202f1b
(cherry picked from commit 0df5811519ed485d62d8c54a6e2f7ce4268edc64)