For component 'inactive, but ready' status, naming 'SETUP' may
be more understandable than 'STOPPED', and 'SETUP' is consistent
with ALSA.
new() params() start()
none ======> init =======> setup ======> running
<====== <======= <======
free() reset() stop()
If its sink is host, we need set it to 16bit(hard coded ATM);
if its sink is dai, we need set it to 32bit SSP format;
otherwise, for internal component, we uses STREAM_FORMAT_S32_LE.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We don't copy when host_avail is 0(usually stream finished), at the
same time, we need wait until the dma copy is finished before we
can start next pulling copy.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
1. only pause at running stage;
2. resetting buffer pointers and local_elem for stopping, to let next
starting begin from original position;
3. adding processes for COMP_CMD_AVAIL_UPDATE.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We need to clear those pointers at new(), and undo at reset()
those initialization what we did in set_params().
For new(), rzalloc() will help set host_data members to 0s and
NULLs.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We need to wait until the last bytes/period is finished in dai,
before we can notify host side about that, otherwise, host side
will trigger stop too early, and we will miss rendering them.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Here add R/W pointers check for playback, make sure the next
copy size is not bigger than the host avail bytes.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
This read pointer indicate how much we have read, we can use it
to inform host side to refill the free periods.
But as it doesn't really mean that we have rendered all of them
(there may be several periods in the pipeline buffer), and for
the last bytes/period, host side will trigger stop immediately
once it got this notification.
To avoid lost for last several periods issue, we postpone this
notification to the finish of last bytes/period copy in dai,
which will be add in the following patches.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We use host_pos_blks to store the latest read pointer, and
report *host_pos to host side, here rename host_pos_blks to
host_pos_read to make it understandable.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
To implement host side buffer R/W pointer check, we need add
buffer update procedures at first, here we introduce host_avail
and host_free to indicate the buffer usuage status.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
It used IPC_INTEL_STR_PAUSE plus IPC_INTEL_STR_RESET to stop a
stream, but sometimes we application only intend to stop the
stream(e.g. speaker-test with large buffer size( > .wav file
size)), which means we only reset the buffer pointers but don't
need set params again(for reset, we need set params again).
Here add this stream operation IPC_INTEL_STR_STOP to handle
this case.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We need the application buffer pointer(write pointer for playback)
update from the host side, to make the ALSA pcm buffer pointer
checking possible.
Here we introduce component command COMP_CMD_AVAIL_UPDATE for the
update event notification.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
The presentation_posn used for dai rendered bytes readback to
host side should be uint64_t type, according to the driver side
definition.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We uses pull mode for running pipeline, that means, we
start each copy in pipeline_schedule_copy().
So here remove this superfluous one period copy.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
It reset dai_pos at prepare stage before, which will
make it wrong when the prepare is called more than
one time.
Here move the reset to MMAP_POS command handling,
which should happen at stream allocating stage only.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We may meet src buffer bound, sink buffer bound, and
splitting may not finish at one time DMA copy, that
is, next_size < split_remaining, ...
This patch added to handle all of those cases, and
prepare for the buffer read/write pointer checking
in future.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
The Right channel was assigned with Left channel by mistake
in vol_s16_to_s32 and vol_s32_to_s32, here correct it.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
It will treat full buffer as empty one at current
comp_update_buffer(), as w_ptr is equal to r_ptr,
this will make us missing those datas or at least
the buffer->avail/free are not correct.
Here spit comp_update_buffer() into consuming one
and producing one, for consuming, 'r_ptr == w_ptr'
update means buffer is becoming empty, and for
producing, 'w_ptr == r_ptr' means buffer is becoming
full.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
The element frame_size in dai_config struct is actually
sample size for each channel, e.g. 24 for S24_4LE, so
rename it to sample_size to avoid confusion.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
As the normal static pipeline can work fine after the 1.33->1ms
buffer and 16->32 bits internal buffer transforming finished,
let's change back to use normal static pipeline by default.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
As we have changed internal buffer to 32 bits, to
avoid clamping, we need to change accumulator to
64 bits.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
As we are using 32 bits data in internal buffer, we need
24bit<==>32bit volume scaler for host<==>processing<==>dai
converting.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
When the host buffer is not 1ms aligned and splitting happens,
the increment is not period size, this wrong pointer will make
pcm data missed/clipped and sound noisy.
Changing to use the actual size last dma copy consumed,
local_elem->size, to fix the issue.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Adding stream_format to indicate the stream format that the
dai buffer is using, which will be used for volume copy
functions mapping.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Make the mixer compute two channels per loop iteration. This allows for
easier vectorisation. The loop is also more efficient if we update or
source/sink pointers after mixing has completed.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
The conversion should up/down convert only if source/sink is host or dai.
Volume should be 32 bit at all other times
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Use period frames instead of period size in bytes. This makes the pipeline
configuration less ambiguous and also allows for host, dai and internal frame
sizes to be defined.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Host buffers are not guaranteed to be continuous physical pages or a factor
of the DSP period size. Add support for non continuous host buffers of any
size.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
The host buffer may not be a multiple of host PHY page size, so provide
so pass this value as part of our host buffer init.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Add support to immediately reload the DMA during a DMA transfer completion
callback. This allows components to respond quickly to any physical
DMA copy updates i.e. period split over two non continuous physical host
pages.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
add 16bit<==>24bit volume copy function and mapping, for
24 bits ssp output/input.
here also optimize all other volume copy functions.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
switch the BCLK generation from shim ssp clock divider to
using SSCR0.SCR, as the divider may lead to jitter.
clear and remove M/N divider part code.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
for DMA controller, we need mask dsp local memory before
setting it to DMAC registers, otherwise, DMAC will locate
to wrong memory address.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We need clear/reset buffer content to 0s each time
creating a pipeline buffer, to avoid random noise
when using it.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
In order to use native XTENSA compiler the core target needs to
be overwite with the good platform. This patch is adding an optional
option in order to be able to overwrite the core selection.
Example:
- Set Xtensa export PATH=$PATH:<PATH>/xtensa/XtDevTools/install/tools/RD-2012.5-linux/XtensaTools
./configure --with-arch=xtensa --with-platform=baytrail --with-dsp-core=CHT_audio_hifiep --with-root-dir=<PATH>/xtensa/XtDevTools/install/tools/RD-2012.5-linux/XtensaTools --host=xtensa CC=xt-xcc OBJCOPY=xt-objcopy OBJDUMP=xt-objdump
make
make bin
Signed-off-by: Sebastien Guiriec <sebastien.guiriec@intel.com>
In order to set the Root directory of the DSP compiler we should
use --with-root-dir instead of --with-tool-dir
Signed-off-by: Sebastien Guiriec <sebastien.guiriec@intel.com>
On the linux driver side, it need get version infomation
from fw, including:
type -- Reef
version -- major.minor
build number
last commit id -- gID
So here we use the package string, which come from git
version, which can provide all those fw version infos.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
We are using 6 states, 2 of them are substates:
init, idle, running, paused,
draining(sub state when transferring from running to idle)
pausing(substate when transferring from running to paused).
Don't response new command when in substates, only stay in
substates for limited timeout. e.g. 2ms for draing and 1ms
for pausing.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Keyon Jie