ASoC: bcm: Use power-of-2 bclk_ratios

[pelwell]

The soundcard drivers originally used snd_pcm_format_physical_width, but a later commit changed that to snd_pcm_format_width because the in-memory sample storage width should not be a factor in determining the bclk_ratio. However, the physical width rounds the sample bits up to the nearest power of 2, which makes it easier to find integer clock divisors.

Restore the old behaviour, but with an implementation that makes it clear what is going on.

See: #6104

[HiassofT]

Thanks a lot, this looks fine to me.

I don't have any of the soundcards that caused noticable issues here but I could verify that this PR works fine with my Cirrus Logic audio card (rpi-cirrus driver):

Without this PR the driver requested 2 slots of 24 bit, but the arizona driver choose to use the next higher bclk rate that was a multiple of the sample rate so it worked fine. As the interface is running in I2S mode with the RPi as a clock consumer the slight misconfiguration on the RPi side didn't matter.

dmesg with dynamic debug of snd_soc_arizona enabled showed:

wm5102-codec wm5102-codec: AIF1: Configuring for 2 24 bit TDM slots
wm5102-codec wm5102-codec: AIF1: BCLK 2822400Hz LRCLK 44100Hz

With this PR the driver now correctly requests 2 slots of 32bit which matches the actually supported bclk rate and the configuration is fine both on RPi and codec side

wm5102-codec wm5102-codec: AIF1: Configuring for 2 32 bit TDM slots
wm5102-codec wm5102-codec: AIF1: BCLK 2822400Hz LRCLK 44100Hz

[pelwell]

Thanks for reporting this.