This commit introduces IOSGLMemory which is a GLMemory that falls back to
GstAppleCoreVideoMemory for CPU access. This is a temporary solution until
IOSurface gets exposed as a public framework on iOS and so we can use
IOSurfaceMemory on both MacOS and iOS.
https://bugzilla.gnome.org/show_bug.cgi?id=769210
Similar to vtdec_hw, this commit adds a vtenc_h264_hw element that fails
caps negotiation unless a hardware encoder could actually be acquired.
This is useful in situations where a fallback to a software encoder
other than the vtenc_h264 software encoder is desired (e.g. to x264enc).
https://bugzilla.gnome.org/show_bug.cgi?id=767104
When doing GLMemory avfvideosrc negotiates UYVY. This change allows avfvideosrc
! tee name=t ! ... ! glimagesink t. ! ... ! gldownload ! vtenc_h264 ! ...
to do GLMemory and 0-copy with the encoder (with the CV meta).
Otherwise, gst_vtenc_negotiate_profile_and_level will double-release as
it checks for profile_level != NULL. This caused crashes when the
vtenc instance is stopped and then restarted.
https://bugzilla.gnome.org/show_bug.cgi?id=757935
Even when we fail to encode frame, we should still enqueue it so
it could be passed into handle_frame (with output_buffer == NULL).
Otherwise, we risk GstVideoEncoder's queue of frames growing unbounded.
Note: We're slightly changing the renegotiation code to accommodate for
frames without output buffers, but this commit takes no ownership over
the way negotiation is being done.
https://bugzilla.gnome.org/show_bug.cgi?id=750669
VTCompressionSessionEncodeFrame retains the CVPixelBuffer during
encoding, and will release it as soon as it can (e.g. before it even
calls our callback). This means we can safely release input buffer
at this point, possibly allowing the system to reuse it sooner.
https://bugzilla.gnome.org/show_bug.cgi?id=750671
It is incorrect to modify the frame properties after passing them, since
VTCompressionSessionEncodeFrame takes reference and we have no control
over when it's being used.
In fact, the code can be simplified. We just preallocate the frame
properties for keyframe requests, and pass NULL otherwise.
https://bugzilla.gnome.org/show_bug.cgi?id=748467
When doing texture sharing we don't need to call CVPixelBufferLockBaseAddress to
map the buffer in CPU. This cuts about 10% relative cpu time from a vtdec !
glimagesink pipeline.
... and hope that everything will be fine. This shouldn't really happen but
previously happened during shutdown. It should be fixed in videoencoder now,
but better be on the safe side here.
The property is in kbit/s and we store it in bit/s, so just multiply and
divide by 1000. No need to put a factor of 8 in there.
kVTCompressionPropertyKey_AverageBitRate is also in bit/s according to
its documentation.
The object lock only protects the session, as we modify
the session from other threads when the bitrate property
is changed. Don't hold it much longer than for session
related things.
And we need to release the video decoder stream lock before
enqueueing a frames. It might wait for our callback to dequeue
a frame from another thread, which will then take the stream
lock too and deadlock.
It is not required on OSX apparently and was only added in 10.9.6 there.
Calculating the correct level from the configuration is not trivial, so let's
just not set a level at all here.
iOS has special stride requirements that we don't know yet, so copy
input buffers into buffers allocated by iOS for now.
Later we should check the stride and probably provide a buffer pool for these
buffers so upstream can directly write in there.
