With this kconfig option set to y SOF does not build on i.MX platforms,
so we disable this optimized conversion.
The warning below is treated as an error.
sof/src/audio/pcm_converter/pcm_converter_hifi3.c: In function ‘pcm_convert_f_to_s16_lin’:
sof/src/audio/pcm_converter/pcm_converter_hifi3.c:681: warning: implicit declaration of function ‘XT_ROUND_S’
Signed-off-by: Paul Olaru <paul.olaru@nxp.com>
There is no such compiler as RF-2017.8 that is in use. The correct
compiler is RG-2017.8 for all i.MX platforms.
Signed-off-by: Paul Olaru <paul.olaru@nxp.com>
Existing code returns every time the API function at index 0, which works
now because we support only a single API function. But, this will break
when patches adding multiple Cadence codec types will be merged.
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
In buffer_size calculation period_bytes should be aligned with dma
before multiplying with period_count. This is an issue for example when
period_bytes could be smaller than the dma alignment size like with 8khz
16 bit mono stream.
Signed-off-by: Jaska Uimonen <jaska.uimonen@intel.com>
Add cmake -DINIT_CONFIG= option that can point at any initial file.
"make clean" does not delete .config any more.
Note reconfiguration does NOT causes recompilation because -imacros
hides the generated .h from CMake's dependency scan. This is not a
regression, that problems exists since -imacros was introduced. At least
it's now possible to "make clean" and rebuild without losing the .config
file.
Fix for #3617
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
A buffer should be prevented from being freed only
if both the sink and the source widgets are active.
In the case of dynamic pipeline loading, a buffer
belonging to a pipeline will need to freed when that
pipeline is no longer active but the other end of the
buffer might still be active. So, change the logic
to check if both the sink and source widget states
are invalid to prevent the buffer from being freed.
Also make sure that the buffer is disconnected from the
active comp before it is freed.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
New implementation does not rely on left shift of signed value and
has no code duplication. Output is the same as original.
found by cppcheck
Signed-off-by: Oleksandr Strelchenko <oleksandr.strelchenko@waves.com>
We need to follow TGL specific HW recommended flow to get/put
PM_RUNTIME_DSP, even for not release_unused clocks requirement.
This will address the PM_GATE IPC failure issue:
https://github.com/thesofproject/sof/issues/3823
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Q_SHIFT_LEFT doesn't check overflow itself. We should do saturation
after left shifts to avoid erroreous value after left shifts.
Signed-off-by: Pin-chih Lin <johnylin@google.com>
When setting parameters some API calls might not be fatal, so there
is no need to fail the apply config phase in this case.
For example, in case of Cadence MP3 codec the only supported PCM
output word size is 16 and 24. Trying to set a PCM word size of 32
will result in a non-fatal error and the word size would be adjusted to
24.
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
The SYSTICK_PERIOD is used for system agent only now, remove the
schedule part from the description.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Reschedule once interrupt is handled by all cores(clients) according to
the earlist task of the list, and enable interrupt on cores that have
tasks to run.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
At interrupt handling & tasks_run(): run each pending task of the core,
e.g. for period tasks:
previously: update task->start based on last_tick(asynchronized), if the
last_tick was delayed, the delay will be accumulated to the subsequent
periods!
new: update task->start based on last task->start, to try to make the
task's running follow the pipeline requirement as more as possible.
Add lock/unlock for domain accessing and updating while keep task can
run simultaneously on multiple cores.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Refine and simplify the domain_set/clear() part:
1. Use UINT64_MAX to denote that the domain is not set.
2. For _domain_set(), set only when needed. e.g. the first task over all
cores being added, or a task asking for earlier timer being added.
The _domain_set() logic is refined as below:
a. For first task, set domain according to the required start, else
b. for one shot task, set domain if it is earlier, else
c. for earlier periodic task, try to make it cadence-aligned with the
existed task, else
d. for later periodic task, simply cover it by the coming interrupt.
3. For _domain_clear() logic, simplify to call it when the last client is
freed.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Do timer_domain refining as below:
Use unified LL_TIMER_SET_OVERHEAD_TICKS=1000 which denote the max
overhead that the timer_domain_set() will take, and remove the
definition from each platforms.
Don't add any overhead if no needed(e.g. if the start > current +
overhead). This will help to schedule the task ASAP, and make the
catching up (of the previous delay) possible.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
rename the last_tick to next_tick, which means the one already set
to the timer, for more easy to understand.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
So far we only used topologies with rate set
to 48 000Hz. This might change in the future.
So use rate from parameters passed from kernel.
While here, align sof_ipc_dai_esai_params struct
with the one from kernel.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
One-touch "make -C installer rsync" combines fast incremental build,
staging and deploy in one command.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
The tools build is independent from the firmware build. The next step is
to invoke it from here if needed.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
This patch fixes the error that happens with the hard-coded PGA
names Dmic0 and Dmic1 in pipelines pipe-eq-iir-volume-capture.m4
and pipe-eq-iir-volume-capture-16khz.m4. Pipelines for beamformer
TDFB have the same hard-coded PGA names and are fixed too.
If impacted pipelines are used for any other purpose than DMIC endpoints
the topology graph gets messed up. With this change the caller macro,
e.g. intel-generic-dmic.m4 needs to set the PGA_NAME macro when the
respective pipeline is instantiated. If it is not set the default
name via N_PGA() macro is used.
The macro intel-generic-dmic-kwd.m4 is updated to set the Dmic0
PGA name as well as the earlier missed PIPELINE_FILTER1 and
PIPELINE_FILTER2 definitions for passing filter coefficients.
Fixes: #3378
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
Lets use the same version of Xtensa toolchain as Intel in order
to avoid compilation problems.
It is difficult to support multiple toolchain versions because specific
headers needs to be updated, so lets have the same version supported
between IMX and Intel.
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Comment is incorrect. There are multiple cases when a config is pushed
by userspace that will result in no callback.
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
The main topology directory contains a number of topologies those
are not used directly by Linux kernel machine drivers. Also there
are variant topologies built those are modified versions of the
main topology. The topologies for testing and devolopment are moved to
directory development. The topologies those may be used by end users
to enhance the audios are moved or built to dsp_enhancements directory.
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
Cppcheck is complaining that we are running a 32bit int off the end.
This shouldn't happen here in any arch. Looking at the math and
comments, it looks like idx should be a 64 type or cast to a larger
container before the bit shift. Therefore I am enforcing a strict type
on it.
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
Cppcheck check doesn't understand the relationship between trace section
start and end so it is giving a false positive because it thinks they
are two separate objects, not two pointers to the same region.
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
The current version of the code effectively relies on the compiler to
optimize out the loop, this both trusts the compiler, annoys
cppcheck which also not being entirely clear to the reader. Solution is
to use the compile switch to remove all relevant code.
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
libc doesn't return a type, so it just no-ops silently on null, lets to
the same having this error makes walk back patterns more annoying
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
The proposed calculation doesn't take negative or zero input-x into
consideration, and we also don't need such use cases. Just return 0
for those input-x.
Signed-off-by: Pin-chih Lin <johnylin@google.com>
Used ABS(x) instead of x for comparing to ONE_OVERT_SQRT2(~0.707)
because x could be a negative value and here we want to transfrom x
from range (-1, -0.5];[0.5, 1) to (-1, -0.707];[0.707, 1)
And revised code line 233~237 to be easier to understand.
Signed-off-by: Pin-chih Lin <johnylin@google.com>
This patch replaces use of pipe-volume-capture macro or
pipe-low-latency-capture to pipe-volume-switch-capture in nocodec
topologies for APL, BDW, CHT, CNL, ICL, and TGL based platforms. It
allows to test of volume component mute switch control in SSP loopback
that is used by default in nocodec topologies. The testing of mute
switch feature is simplest to do with loopback topologies. Nocodec
topologies are not in mainstream usage so this is the safest option to
enable testing.
The mute switch add should be possible to all capture topologies but
it can be done later once it is confirmed it is safe to do (avoid
accidental muted audio or problems with UCM).
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
Those two enums should not be mixed, because the first one is
only internal enum for FW while the second one is part of ABI.
Signed-off-by: Karol Trzcinski <karolx.trzcinski@linux.intel.com>
There is a bug in upstream cppcheck where dereferencing a pointer does
not result in a type check in sizeof as far as cppcheck is concerned.
This results in a false divide by sizeof(pointer) error.
See https://trac.cppcheck.net/ticket/10179
Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
Paul Olaru