On 2023-11-16 4:10 PM, Jaroslav Kysela wrote:
On 16. 11. 23 12:22, Cezary Rojewski wrote:
From: Jaroslav Kysela <perex@xxxxxxxx>
Improve granularity of format selection for S32/U32 formats by adding
constants representing 20, 24 and MAX most significant bit >
The MAX means the maximum number of significant bits which can
the physical format hold. For 32-bit formats, MAX is related
to 32 bits. For 8-bit formats, MAX is related to 8 bits etc.
The drivers may use snd_pcm_hw_constraint_subformats with
a simple format -> subformats table.
I am afraid, the above sentence is no more correct and the current code
does not follow my original idea. It's a bit step back to the initial
code. But I admit that from the API POV, it's workable now (with the
added refine mechanism for one format).
I'll update the commit message, indeed no longer up-to date. Will also
try to explain the thought process which is currently missing, as you
noted below.
As noted several times, this is not my preferred implementation (I would
keep only the constraint function which will be called by drivers on
demand in the ALSA PCM core code). The latest proposed simplification
may be applied in the ASoC core (store S32_LE subformat mask in
snd_soc_pcm_runtime and install this intersected constraint for FE PCM -
user space). Something like ASoC core does for the msbits constraint.
If nobody else thinks that it's a good direction, please, add a note to
the comment that this implementation (extend snd_pcm_hardware structure)
is a compromise for the ASoC code with details.
I'd like to avoid complicating ASoC runtime structures as storing
hw_params is not among their current task list, at least not currently.
But I'm open for any improvements in the future.
+ if (f == SNDRV_PCM_FORMAT_S32_LE)
Missing mask check: (f == SNDRV_PCM_FORMAT_S32_LE && *subformats)
Otherwise the MSBITS_MAX won't be set for S32_LE by default.
Ack.
+ m.bits[0] |= *subformats;
Jaroslav
