[alsa-devel] [PATCH] Bugfix: Fix resampling when client and slave both use format float

[Maarten Baert]

I found a bug in libasound. When both the client and the slave (in my
case, the JACK plugin) try to use the float format, but the sample rate
or channel count does not match, libasound *should* insert linear
conversion plugins to convert from float to linear, then resample/remap
channels, and then convert back to float (because apparently the
resamplers and channel remapper don't support floating point, only
'linear' i.e. integers). Currently this doesn't work,
snd_pcm_plug_change_format doesn't know what to do and simply returns
EINVAL. As a result, snd_pcm_hw_params fails even though the HW params
were perfectly valid (it indicates that both the float format and any
sample rate are supported).

In my test, this broke audio for WINE (and any other application that
tries to use float, such as aplay with the right settings) when I wanted
to use the JACK plugin (which only supports the float format).

This patch fixes this bug by doing a conversion to S16 and back when
resampling or remapping is needed. And while I was at it, I also removed
a check that had no effect because the exact same check is already being
done at the start of the function.

I still think it's a bit silly that libasound requires integers for
resampling, because both libsamplerate and libspeex use float internally
for resampling. So now ALSA is literally doing a
float-to-s16-to-float-to-s16-to-float conversion. But changing that
would have been a lot harder.

Maarten Baert
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fix-float-resampling.patch
Type: text/x-patch
Size: 864 bytes
Desc: not available
URL: <http://mailman.alsa-project.org/pipermail/alsa-devel/attachments/20140221/a3927af3/attachment.bin>