> The noise source is pretty expensive, particularly for GR_GAUSSIAN.
> Also, you're using the generic fir filter kernel (straight C++).  On
> x86 and x86-64 we've got hand-coded assembler for taking advantage of
> SSE and 3DNow SIMD instructions.

The noise source is very expensive.  We should find some alternative
implementation which is faster.

> I don't know if there's a Solaris equivalent to "oprofile" (profiling
> using the h/w performance counters), but the first step to making it
> faster would be to determine where it's slow.

As a former UltraSPARC designer, I can give the following advice --

- Floating point is much better than integer math
- VIS (Visual Instruction Set) is a big help.  VIS is the SPARC
equivalent of MMX and SSE.  Actually, VIS came first, so you could argue
that MMX is a clone of VIS :)

You also need to keep in mind that you can't get blood from a stone.
The UltraSPARC II was a great processor for its day, but it's about 9
years old now.

Matt


_______________________________________________
Discuss-gnuradio mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/discuss-gnuradio

Reply via email to