Re: [linux-audio-dev] Floating-point exceptions

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] Floating-point exceptions
From: Eli Brandt (eli_AT_v.gp.cs.cmu.edu)
Date: pe loka   01 1999 - 20:49:51 EDT


Francois Dechelle wrote:
> DSP algorithms (filters for example) do a lot of floating-point underflow=
> s.
> In theory, masking the corresponding trap bits in the pentium FPU control
> register should do it, and silently flush the underflows to zero.

You're talking about denormals, where there's no implicit leading 1 on
the mantissa? These don't cause underflow exceptions (until they
finally do underflow to zero). They're just ungodly slow.

I am told that Intel hardware implements them through a trap to
software, but this is not an IEEE FP exception -- you can't turn it
off through IEEE. It could just as well be slow microcode. There is
no way I know of to disable denormal-handling and flush to zero,
presumably because that would violate the FP spec.

Some architectures do permit this little violation. Lacking that,
I think you have to resort to adding DC, adding noise, testing for
small values, that sort of thing.

Sometimes I wonder how Linux is running G4 PowerMacs.

-- 
     Eli Brandt  |  eli+@cs.cmu.edu  |  http://www.cs.cmu.edu/~eli/


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:27:12 EST