The rounding done when converting to smaller sample widths fails for
saturated positive samples; in this case, the sample value overflows and
becomes negative. We are better off without rounding, even if we then
have quantization noise (of at most half the resolution of the least
significant bit).
ALSA bug#3360; Debian #437827; GNOME #436192; LP #116990; Mandriva #33908
The entries in the current Versions file don't work as expected.
Since the first ALSA_0.9 has already snd_* global definitions, all
the rest are simply ignored. Some symbols (e.g. snd_hw_params_*)
indeed work because they have explicit symver definitions, but
ALSA_0.9.6 and later are all superfluous.
This patch clean up these useless entries. Also, the patch restricts
the matching patterns for _snd_* and __snd_* entries to reduce the
unneeded export symbols.
Rename the local functions to snd1_* so that they won't be exported
out of alsa-lib.
Some functions are still kept because aserver requires them. Sigh.
softvol can be also a pass-thru when the given control already exists
as a hardware control, and the name isn't set properly because of
slave creation. This patch fixes it.
The PCM name isn't set properly on empty and asym plugins due to its call
of snd_pcm_open_slave(). Now a new function snd_pcm_open_named_slave()
is created and make snd_pcm_open_slave() an inline function calling the
new one with name=NULL.
The mmap of multi plugin seems broken (for a long time!) due to its
creation of local buffer via snd_pcm_mmap(). Since the multi plugin
just needs to shadow the mmap buffer of each slave, it now has
mmap_shadow=1 and its own mmap/unmap method to do shadowing.
PCM direct plugins didn't update the timestamp properly.
Now it always starts the slave PCM with MMAP tstamp_mode so that the
timestamp will be being updated. When a client is set up as MMAP
tstamp_mode as well, simply copy this slave timestamp. Otherwise
status callback calculates the current timestamp as usual.
The rate plugin has ugly hacks in poll_descriptors callback to adjust
avail_min when partial read/write occurs. This causes often unexpected
problems like XRUNs, especially with two-period cases.
Let's remove that beast, it's rather harmful than useful.
Signed-off-by: Stas Sergeev <stsp@aknet.ru>
When calculating the value range of a control, the variables cannot be
initialized with zero because this would prevent the minimum from having
a value above zero or the maximum from having a value below zero.
Remove obsoleted sequencer instrument layer from alsa-lib.
The old symbols are compiled in as default as dummy functions
(unless --disable-old-symbols is given to configure) so that
the old binaries can still work more or less.
Putting assert in the public macros isn't good idea at all.
Let's get rid of them.
Also, clean up snd*_alloca() functions to use a helper macro
instead of copy and paste.
Export helper functions to convert dB level and range.
snd_tlv_*dB*() are to convert dB level or range directly from TLV data.
snd_ctl_*dB*() are to get dB level or range from a control element.
Activating assert() in the default error message handler isn't always
good for producitve systems. Make this optional and enable only when
a special configure option is given (i.e. for explicit debugging).
This patch causes snd_pcm_dmix_close() to up a semaphore after downing it
if it is unable to discard it. It prevents some deadlock that I am
getting when a couple of applications interact and one of them closes the
device and later re-opens it.
From: Mike Gorse <mgorse@mgorse.dhs.org>
