13 Aug
2010
13 Aug
'10
7:56 p.m.
On Fri, 13 Aug 2010, torbenh wrote:
arch is already using gcc-4.5
other distros will probably follow during this year.
True... though I keep seeing bug reports about, "doesn't
work when I compile with gcc 4.5" :-)
the vectorizer can generate pretty much optimal code,
if your a bit
careful, and use the restrict keyword.
since the most optimal thing is to weave the mixing code into the actual
processing a library doesnt make sense imo.
Hmmm... makes sense... though we're not all Torben's. :-)
I was also thinking that another benefit of a library like
this would be:
* efficient, optional clipping
* block partitioned routines that efficiently utilize
the cache[1]
* dithering routines
* routines to interleave and de-interleave audio streams
* (maybe) format conversions (e.g. float<->int)
-gabriel
[1] OK, I'm just sort of talking out of my arse on this
one. I doubt that most DSP code is even threatening
to bust out of the cache. But, for instance, if you
set a high buffer size (4096) and mix 16 chans,
that maxes out the cache on my own CPU (2048k).