For fixed conversion ratios (44.1->48kHz), the speex resampler is
overkill. Using pre-defined tables and a polyphase implementation the
CPU load can be reduced big time, or for the same load the audio quality
can be increased.
I uploaded the latest code of the library we've been using internal
Hello,
> The PulseAudio integration is still very preliminary and not ready for merge
> upstream, the detection of SSE at compile time isn't done, the filters will be
> updated later with better ones, but I thought I would ask for early feedback
> from the list.
it seems SSE3 is required
src/Mak
> it seems SSE3 is required
Indeed.
> src/Makefile.am:
> +libpulsecommon_@PA_MAJORMINOR@_la_CFLAGS += -D__SSSE3__ -msse -msse2 -
> msse3 -march=core2 -mfpmath=sse
>
> doesn't -msse3 imply -msse2/-msse?
I honestly have no idea. The code was originally developed with ICC for
atom, not sure what g
Le 06/01/2012 19:59, Pierre-Louis Bossart a écrit :
> For fixed conversion ratios (44.1->48kHz), the speex resampler is
> overkill. Using pre-defined tables and a polyphase implementation the
> CPU load can be reduced big time, or for the same load the audio quality
> can be increased.
>
> I uploa
> I see in the third commit all those large tables. Are they computer
> generated? If so, could a python/ruby/shell/whatever script be included
> to generate those tables again?
They are generated by a Matlab script. I don't think there's any secret
sauce preventing us from releasing that script i
On Fri, 2012-01-06 at 21:09 -0600, Pierre-Louis Bossart wrote:
> > it seems SSE3 is required
>
> Indeed.
>
> > src/Makefile.am:
> > +libpulsecommon_@PA_MAJORMINOR@_la_CFLAGS += -D__SSSE3__ -msse -msse2 -
> > msse3 -march=core2 -mfpmath=sse
> >
> > doesn't -msse3 imply -msse2/-msse?
>
> I honest
> Hopefully the other Intel folks will have a better solution, but in the
> mean time, the right solution to me seems to be to drop -march, put all
> those CFLAGS in configure.ac, try to compile a segment of code with it,
> and see if it all goes okay. So the actual enabling will only be done
> if
On Wed, 2012-01-11 at 11:44 -0600, Pierre-Louis Bossart wrote:
> > Hopefully the other Intel folks will have a better solution, but in the
> > mean time, the right solution to me seems to be to drop -march, put all
> > those CFLAGS in configure.ac, try to compile a segment of code with it,
> > and