So that we can reuse optimized versions in unoptimized noise
functions.
Do allocation a little different so that we can align everything
from the start.
Loop though coefficients, when all are 0, clear the destination.
When only one coefficient is used, simply copy with volume.
Otherwise run the complete convolution.
Don't count the sum for unused matrix coefficients. When doing pairing,
we set the mask to ~0 and so we would otherwise count too much.
Apply normalization to the active matrix coefficients, not whatever the
loop variables end up being at the end of the loops.
This fixes several integer overflow problems in the POD parser, as well
as fixing a returns-twice warning from GCC and integer truncation
problems in SPA_FLAG_CLEAR and SPA_ROUND_DOWN_N. The integer overflows
can result in a tiny POD being treated as a huge one, causing
out-of-bounds reads.
A flush command is not supposed to stop playback but just clear the
current state. Normally, to avoid complications, an application will
Pause, Flush and optionally Start to do things smoothly without
interfering with the process loop, but things should not crash if that's
not the case.
Fixes#2726
spa_pod_parser_can_collect and spa_pod_parser_getv were inconsistent
with each other. When parsing crafted POD, this could cause a type
confusion or stack overflow.
HFP 1.8 states in 4.35.1:
Due to the small inconsistency between the GSM standard [2]) and the current
Headset specification ([3]), the HF shall also accept the “=” symbol,
in place of “:”, as a valid separator for this unsolicited result code.
iOS seems to use "=", so accept that as well.
See #2463
Both HFP and HSP require an AT command from HF/HS to AG to
be terminated by CR, not LF. (HFP 1.8, 4.34.1; HSP 1.2, 4.8.1)
See #2463
Fixes: 0b2d3730b6 ("bluez5: Add HFP HF support")
`strsep()` is simpler to use than `strtok()` or `strtok_r()`
while being thread-safe and reentrant.
Although it is an extension, it is available in both glibc and musl.
A Suspend should clear all the negotiated state and start a new
negotiation in Start. Use a flag to control this.
This avoids recalculation of state for each pause/play state change.
See #2701
When we are already past the size of the buffer, don't bother calling
the overflow callback anymore, the buffer is already corrupted.
Otherwise it would be possible to have the overflow callback fail the
first time around, some data will be skipped, and then the next
overflow callback would succeed, giving the impression that all is
fine.
Add a unit test for this.
Add a resampler option to prefill the resampler with 0. This then
results in the resampler always outputing and consuming the same
amount of data instead of a short buffer in the beginning.
When rfcomm_hfp_ag() returns false, an "ERROR" reply is sent.
When testing if the SLC is configured, 2 "ERROR" replies are sent, which
should not be done.
For backward compatibility with old Wireplumber releases, support the
old api.bluez5.a2dp.sink/source names, and use them in object events
instead of the media.sink/source names.
Previously, if "add_listener" was called on the monitor device, then it
did not necessarily emit events about all devices because it called
`enum_devices()` which does not emit events about already existing
devices. So the very first listener would get all existing devices,
but subsequent ones would not get events about the existence of devices
that have already been seen by the monitor. Fix that by simply emitting
events about all existing devices if the current listener is not the first.
libcamera's CameraManager runs the event handlers on its own thread,
thus synchronization is needed to ensure correct functionality.
Implement that by collecting hotplug events into a mutex protected queue
and signalling the main loop where they can be processed and the
"object_info" event can be safely emitted.
