sound: Implement AFMT_FLOAT support
Needs ReviewPublic
Actions

Authored by christos on Sat, Nov 16, 6:15 PM.

Details

Reviewers

markj
emaste
dev_submerge.ch

Summary

Even though the OSS manual [1] advises against using AFMT_FLOAT, there
are applications that expect the sound driver to support it, and might
not work properly without it.

This patch adds AFMT_FLOAT in sys/soundcard.h and implements AFMT_FLOAT
<-> AFMT_S32_LE conversion functions. As a result, applications can
write/read floats to/from sound(4), but internally, because sound(4)
works with integers, we convert AFMT_FLOAT to AFMT_S32_LE before doing
any processing.

[1] http://manuals.opensound.com/developer/AFMT_FLOAT.html

PR: 281390, 157050, 184380, 264973
Sponsored by: The FreeBSD Foundation
MFC after: 2 days

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 60632
Build 57516: arc lint + arc unit

Event Timeline

christos created this revision.Sat, Nov 16, 6:15 PM

Herald added a subscriber: imp. · View Herald TranscriptSat, Nov 16, 6:15 PM

christos requested review of this revision.Sat, Nov 16, 6:15 PM

Harbormaster completed remote builds in B60632: Diff 146593.Sat, Nov 16, 6:15 PM

christos added a child revision: D47639: beep(1): Use AFMT_FLOAT.Sat, Nov 16, 6:19 PM

Hello,

i can confirm so far that it works fine with emulators/wine.

I did took 2 FMOD games for testing (Way of the Hunter and Roboquest) that required AFMT_FLOAT.

Thank you for your work!

emaste added inline comments.Sat, Nov 16, 9:31 PM

sys/dev/sound/pcm/feeder_format.c
68	I'm a bit worried about a possible performance impact, and we might need to hoist the `fpu_kern_enter` / `fpu_kern_leaave` up a level, but we can profile it and see what the impact actually is.

emaste mentioned this in D47639: beep(1): Use AFMT_FLOAT.Sat, Nov 16, 9:34 PM

agh_riseup.net added a subscriber: agh_riseup.net.Sun, Nov 17, 1:19 AM

Very welcome, thanks for tackling this @christos. Please note that there are sound cards supporting float format, and I intend to experiment with that some time in the future (probably needs bitperfect for now).

sys/dev/sound/pcm/feeder_format.c
68	I'm a bit worried about a possible performance impact, and we might need to hoist the `fpu_kern_enter` / `fpu_kern_leaave` up a level, but we can profile it and see what the impact actually is. +1. Also the way this function is called through a function pointer for every sample seems quite inefficient, at least I don't see how it could be inlined there. Means a restructuring could help all format conversions, not just float. Although I see moving format conversions out to userland as the better option in the middle to long term, anyway.