audio: what encoding kernel accepts?

[Tetsuya Isaki <isaki%pastel-flower.jp@localhost>]

Hi,

I'd like to know our audio specification.  i.e.,
what encoding the kernel accepts?

This table shows the status of current implementation, not current
documentation.

		audioplay(1)	audio(4)	AUDIO_GETENC(*1)
------------------------------------------------------------
s8		accept		accept		not support
s16{be,le}	accept		accept		supported(*2)
s24{be,le}	accept		reject		not support
s32{be,le}	accept		accept		not support
u8		accept		accept		not support
u16{be,le}	?		?		supported by emulation
u24{be,le}	?		reject?		not support
u32{be,le}	?		?		not support
mulaw		accept		accept(ifdef)	supported by emulation

  note:
  *1 The mismatch of AUDIO_GETENC is described in PR/52585.

  *2 On LE, s16le is supported natively and s16be is by emulation.
     On BE, s16be is supported natively and s16le is by emulation.
     Even if the hardware does not support slinear16 natively.

  memo:
  o I confirmed that audioplay(1) accepts u8.wav, s{16,24,32}le.wav,
    s8.au, s{16,24,32}be.au, and mulaw.au.
    But I don't know about u{16,24,32}.

  o I confirmed that audio(4) rejects s24{le,be} by audioplay(1).

Is this result expected (= well-planned)?

I don't say that "audio(4) does not accpet 24bit, let's add it."
Before rev1.26, auconv.c supported 8/16bit only.
After rev1.27, auconv.c supports {8,16,24/24,24/32,32} but mostly
broken (see PR/52586) but no one noticed it because most of them
was dead code.

It's better to clarify the specification before implementing.

I'm not an audio man, so I don't know details but,
Will audio(4) support ulinear{16,24,32} or slinear24 ? i.e.,
is it worth to pay the maintainance cost to keep supporting it?

Thanks,
---
Tetsuya Isaki <isaki%pastel-flower.jp@localhost / isaki%NetBSD.org@localhost>