Adding "hip-amd-precompile" build option. If enabled, AMD kernels
will be precompiled at build time. Also "hip-hipcc-arch" build option
(corresponding to --offload-arch hipcc option) is added
so that user can specify target GPU arch instead of auto-detection by hipcc
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8923>
In a previous version of the clock time conversion code, scheduled playback used
to be started (from 0) when transitioning to PLAYING and stopped when
transitioning PLAYING->PAUSED. This worked fine when converting running times
using the internal clock. However, now the decklink clock will produce values
that are monotonically increasing and do not reset to 0 at the same moments as
running time anymore. This means that the clock adjustments could attempt to
convert a small running time based on a large clock time e.g. after pausing
for many hours. As the adjustment code is a simple linear interpolation based on
the current clock times (large) using the provided value (small), the small
differences in the rate could result in very large differences in the
output time.
Fix by instead using both internal and external clock times based on the values
that gst_clock_get_calibration() will return. By doing so, small changes in the
rate calculations between the internal and external clock times will not result
in potentially large differences in the output internal time from
gst_clock_unadjust_with_calibration().
Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/4197
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9267>
Prevent race condition where gst_play_sink_do_reconfigure() could be called
from a pad probe while stream synchronizer pads are being released during
GST_STATE_CHANGE_PAUSED_TO_READY transition.
The race occurred when:
1. State change starts releasing stream synchronizer pads
2. Pads are unblocked earlier in the state change, allowing events to flow
3. A streaming thread triggers sinkpad_blocked_cb -> gst_play_sink_do_reconfigure
4. Reconfiguration tries to use already-released pad pointers
5. New pad creation fails with assertion in gst_pad_iterate_internal_links
The fix adds GST_PLAY_SINK_LOCK around the pad cleanup to ensure atomic
cleanup and prevent concurrent access during state transitions.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9233>
After gst-validate-1.0 notices the first expected error on the bus, it will
queue the completion of that action while the pipeline is still processing the
error and getting to the basesrc for the second error.
Mark the second error as 'sometimes=true', as it's not critical to the test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9118>
Actions are no longer always called directly from the main loop
so we can't depend on G_SOURCE_REMOVE to remove it. It's also
possible while returning up the call chain after running stop
to try to re-add it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9118>
Colorimetry bt2100-pq is a variant of colorspace V4L2_COLORSPACE_BT2020
where the transfer function is SMPTE 2084 which is used by HDR content.
Colorimetry 1:4:5:3 is a full-range variant of colorspace
V4L2_COLORSPACE_470_SYSTEM_BG
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9166>
Android 10 (API 29) added support for isHardwareAccelerated() to
MediaCodecInfo to detect whether a particular MediaCodec is backed by
hardware or not. We can now use that to ensure that the video hw-codec
is PRIMARY+1 on Android, since using a software codec for video is
simply not feasible most of the time.
If we're not able to detect isHardwareAccelerated(), perhaps because
the Android API version is too old, we try to use the codec name as
a fallback.
Also rank PRIMARY+1 the c2.android c2.exynos and c2.amlogic audio
codecs alongside OMX.google, because they are known-good.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9225>
Most of the messages can be printed with INFO threshold since they are
only printed on plugin registration.
Fix printing of codec caps, since GST_PTR_FORMAT truncates the output
in almost every case that I saw.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9225>
Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/4317
This patch fixes a regression in
a35bf1e3847351703542869755b6b9e536b9d2fd.
The new calls to tracers in the affected change were being done while
holding the pad object lock, which wasn't the case in the old ones,
leading to the latency tracer deadlocking in gst_object_get_parent().
The dependency on the pad lock for those calls was accidental. This
patch removes it by temporarily unlocking during the affected calls,
not unlike how it's done when calling a probe callback or the
send_event() function of the downstream element.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8925>
When there are gaps between segments in the DASH manifest (e.g. it's a
recording of a live stream during which the camera was turned on and off
a couple times), seeking to a timestamp that belongs into a gap leads to
repeat_index calculation yielding a nonsensical negative number (because
ts < segment->start).
In such event set the playback to continue at the first repetition of
the segment that follows after the gap.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8620>
A user-supplied window handle (external_view) becomes the superView
of internal_view, which is closed with [view removeFromSuperview].
This fails silently if external_view = NULL (no handle supplied).
Call [win_internal_id close] in this case. Fixes#4432.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9049>
The copy of the exact same stream-start event prevents the multiqueue's sink
event function from being called because it is already stored on both pads at
link time
The text streams are no longer considered sparse by the multiqueue, so
interleave calculation is broken and makes us consume a lot of ram and we can
end up killed by the kernel because of this
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8412>
We never implemented correctly the selection of memories in a buffer given the
plane.
This patch uses gst_buffer_find_memory() for that. The offset is checked via the
video meta either in the input and output buffers, or the default offset given
the format and size.
This patch also requests the video meta option for the output buffers.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9219>
This patch adds the possibility to attach video metas to allocated buffers if
API users request that option.
The main reason to add this meta is for the future usage of
gst_buffer_find_memory(), where the requested offset determines the memory
index.
In the case of multi-memory Vulkan images the offset of every component is
larger than the default offset determined by the format and frame size, because
of the buffer minimum size and memory alignment determined by the used driver.
Then, the offset is set in the buffer video meta.
In the case of single memory Vulkan images, the default offset is set in the
video meta.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/9219>
