speex_resample_float() does not work with speex compiled with
--enable-fixed-point, because speex expects its float input
to be normalized to ±32768 instead of the more usual ±1.
It is possible to fix speex_resample_float(), as demonstrated at
http://lists.freedesktop.org/archives/pulseaudio-discuss/2014-May/020617.html
However, a better idea is to avoid using the speex-float resampler and
the associated s16 <-> float conversions that speex will immediately undo
internally if it is known that speex has been compiled with FIXED_POINT.
So, transparently change speex-float-* to speex-fixed-* in that case.
Signed-off-by: Alexander E. Patrakov <patrakov@gmail.com>
Reported-by: Fahad Arslan <fahad_arslan@mentor.com>
Cc: Damir Jelić <poljarinho@gmail.com>
Cc: Peter Meerwald <pmeerw@pmeerw.net>
FIXED_POINT detection is based on code by Peter Meerwald.
State can be used by remap function implementations to
speed up the remapping, e.g. by precomputing things or
even by generating specialized code for a specific channel
remapping task
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
Initialization of the remap structure now happens in one place
Rename calc_map_table() to setup_remap(), copy sample format and
channel specs; the remap structure is initialized when we know the
work sample format of the resampler
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
pa_init_remap_func() only sets the appropriate remapping function, it
does not initialize the pa_remap struct
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
Reintroduces a cleaned-up version of commit 30ce3a14e5 which
was reverted by 1ce71cbd82; for more information see
http://thread.gmane.org/gmane.comp.audio.pulseaudio.general/17479/focus=17487
The patch intends to reduce computational load when resampling AND remapping. The PA
resampler performs the following steps:
sample format conversion -> remapping -> resampling -> sample format conversion
In case the number of output channels is higher than the number of input channels, the
resampler has to be run more often than necessary. E.g. in case of mono to 4-channel remapping,
the resampler runs on 4 channels separately.
To ímprove this, the PA resampler pipeline is made adaptive:
if out-channels <= in-channels:
sample format conversion -> remapping -> resampling -> sample format conversion
if out-channels > in-channels:
sample format conversion -> resampling -> remapping -> sample format conversion
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
the patch changes the interface of the (internal) fit_buf() function:
fit_buf() manages the memblock of the buf chunk, it reallocates the memblock
if the requested number of bytes ('len') if larger than the memblock's size ('size')
and optionally preserves 'copy' bytes
the code should be in line with the comment now
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
I think this makes the code a bit nicer to read and write. This also
reduces the chances of off-by-one errors when checking the bounds of
sample rate values.
I think this makes the code a bit nicer to read and write. This also
reduces the chances of off-by-one errors when checking the bounds of
the sample format value.
buf in struct ffmpeg_data is reset() initially and freed, but never
actually used
when a new block is allocated ffmpeg_data->buf[c].length is used
(which is always 0) to compute the new block size
so, drop buf
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
leftover_buf points to the output buffer of a stage containing leftover
data; similar for leftover_buf_size and have_leftover
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
some resampler implementations (e.g. libsamplerate and ffmpeg) do not consume
the entire input buffer; the impl_resample() function now has a return value
returning the number of frames in the input buffer not processed
these frames must be saved in appropriate buffer and presented together with
new input data
also change the parameter names from in_samples, out_samples to in_n_frames,
out_n_frames, respectively (n_frames = samples / channels)
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
This patch fixes a small mistake where we actually log that we are
reverting to the auto resampler if we can't use the 'copy' resampler but
never do the revert.
This would lead to a crash if the user chooses the 'copy' resampler and
then tries to play something that needs to be resampled.
With very low input sample rates the memory pool max block size may
not be big enough, in which case we should return the size of one
frame. Returning zero caused crashing.
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=68616
This adds asserts to check if the implementation has an update rate
function defined for the unlikely event that some implementation forgets
to assign a update rate function we can simply bail.
It is expected from the resampling implementations to have such a
function even if the state of the resampler is completely reset.
This patch fixes this assertion:
Assertion 'r->i_ss.rate >= r->o_ss.rate' failed at ../../src/pulsecore/resampler.c:1744, function peaks_init(). Aborting.
The pa_resampler struct contains many implementation specific
structures. These create overhead and don't belong there anyways.
This patch moves the implementation specific structures out of the
pa_resampler structure.
This consumes less power, has low (no?) perceivable difference, and
allows the default configuration to work out of the box on low-end
systems (such as netbooks).
The patch intends to reduce computational load when resampling AND remapping. The PA
resampler performs the following steps:
sample format conversion -> remapping -> resampling -> sample format conversion
In case the number of output channels is higher than the number of input channels, the
resampler has to be run more often than necessary. E.g. in case of mono to 4-channel remapping,
the resampler runs on 4 channels separately.
To ímprove this, the PA resampler pipeline is made adaptive:
if out-channels <= in-channels:
sample format conversion -> remapping -> resampling -> sample format conversion
if out-channels > in-channels:
sample format conversion -> resampling -> remapping -> sample format conversion
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
Remixing one channel map to another is (except for special cases) done
via a linear mapping between channels, whose corresponding matrix is
computed by calc_map_table(). The k-th row in this matrix corresponds to
the coefficients of the linear combination of the input channels that
result in the k-th output channel. In order to avoid clipping of samples
we require that the sum of these coefficients is (at most) 1. This
commit ensures this.
Prior to this commit tests/remix-test.c gives 52 of 132 matrices that
violate this property. For example:
'front-left,front-right,front-center,lfe' -> 'front-left,front-right'
prior this commit after this commit
I00 I01 I02 I03 I00 I01 I02 I03
+------------------------ +------------------------
O00 | 0.750 0.000 0.375 0.375 O00 | 0.533 0.000 0.267 0.200
O01 | 0.000 0.750 0.375 0.375 O01 | 0.000 0.533 0.267 0.200
Building the matrix is done in several steps. However, only insufficient
measures are taken in order to preserve a row-sum of 1.0 (or leaves it
at 0.0) after each step. The current patch adds a post-processing step
in order check for each row whether the sum exceeds 1.0 and, if
necessary, normalizes this row. This allows for further simplifactions:
- The insufficient normalizations after some steps are removed. Gains
are adapted to (partially) resemble the old matrices.
- Handling unconnected input channls becomes a lot simpler.
- Separate the cases with PA_RESAMPLER_NO_REMAP or PA_RESAMPLER_NO_REMIX
set and remove redundant if-conditions.
- Fix C90 compiler warning due to mixing code and variable declaration.
- Do not repeatedly count number of left, right and center channels in
the input channel map.
The logic of calc_map_table() remains unaltered.
Problem: s16 to s32 conversion is performed as s16->float->s32 (via work
format float) for resamplers TRIVIAL, COPY, PEAKS.
Precision and efficiency suffers: e.g. 0x9fff results in 0x9ffe4001 (instead
of 0x9fff0000) and there are two sample format conversions instead of one
conversion.
Solution: If input or output format is s16, then choose the work format
to be s16 as well.
If remapping is to be performed, we could stick to work format float32ne for
precision reseans. This is debateable.
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
To reproduce, add resampler-method = ffmpeg in daemon.conf
then start PA, and load module-loopback
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xb2f1db40 (LWP 23047)]
0x00000000 in ?? ()
(gdb) bt
0 0x00000000 in ?? ()
1 0xb7c463cb in pa_resampler_set_input_rate (r=0x80e9438, rate=44011) at pulsecore/resampler.c:365
2 0xb7c6321d in pa_sink_input_process_msg (o=0x80e87a0, code=3, userdata=0xabeb, offset=0, chunk=0x0)
at pulsecore/sink-input.c:1833
3 0xb7e9840b in sink_input_process_msg_cb (obj=0x80e87a0, code=3, data=0xabeb, offset=0, chunk=0x0)
at modules/module-loopback.c:538
4 0xb7c2709b in pa_asyncmsgq_dispatch (object=0x80e87a0, code=3, userdata=0xabeb, offset=0, memchunk=0xb2f1d17c)
at pulsecore/asyncmsgq.c:322
5 0xb7c4c6e3 in asyncmsgq_read_work (i=0x80dd580) at pulsecore/rtpoll.c:564
6 0xb7c4b34a in pa_rtpoll_run (p=0x80fb7e0, wait_op=true) at pulsecore/rtpoll.c:238
7 0xb7dd90af in thread_func (userdata=0x80afe88) at modules/alsa/alsa-sink.c:1785
8 0xb7bf3291 in internal_thread_func (userdata=0x8095d08) at pulsecore/thread-posix.c:83
9 0xb7ab9d4c in start_thread (arg=0xb2f1db40) at pthread_create.c:308
10 0xb79f3ace in clone () at ../sysdeps/unix/sysv/linux/i386/clone.S:130
This fixes at least one crash that has been observed. The
multiplication in trivial_resample() overflowed when
resampling from 96 kHz to 48 kHz, causing an assertion
error:
Assertion 'o_index * fz < pa_memblock_get_length(output->memblock)' failed at pulsecore/resampler.c:1521, function trivial_resample(). Aborting.
Without the assertion, the memcpy() after the assertion
would have overwritten some random heap memory.
libsamplerate_resample() assumed that src_process() would
always consume the whole input buffer. That was an invalid
assumption leading to crashes.
This patch adds a leftover memchunk for storing any
non-consumed input. When pa_resampler_run() is called next
time, the leftover is prepended to the new input.
Changes in v3:
- Make the calculations in pa_resampler_result() and
pa_resampler_max_block_size() more readable and more
correct.
- Rework the leftover storing: instead of using a dedicated
buffer for it, store it in the beginning of remap_buf.
This can avoid some memory copying. (The idea was
suggested by Wang Xingchao.)
- Use a generic save_leftover() function instead of doing
the leftover copying in the resampler implementation.
- Use the leftover logic also with the speex and ffmpeg
resamplers.
[ed: dropped the speex bit since the API guarantees that
it will consume everything -- Arun]
Changes in v2:
- If add_leftover() is called with zero-length input while
the leftover length is non-zero, we don't try to acquire
the input memblock.
- Instead of taking a reference to the original input in
libsamplerate_resample(), we copy the leftover data to a
new memblock. This is done, because otherwise, if the
input is one of the internal buffers, the data can get
overwritten before reading it in add_leftover().
- Store add_leftover_buf size in bytes instead of samples
(more convenient, but less consistent with other code).
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=47156
