Re: Simple Floating-Pt DSP chip question
- From: "Howard Long" <howard@xxxxxxxxxxxxxxxxx>
- Date: Fri, 28 Apr 2006 11:27:06 +0100
Hi Rick
I was reading a report the other day where the
author stated that modern-day floating-point DSP chips
can perform division almost as fast (measured in CPU clocks
cycles) as they can perform multiplication.
Is that true?
(I remember some years ago our pal here Dr. Mike
Rosing--what ever happened to Dr. Mike?--,told me
that division took roughly 30 clock cycles to achieve a
relatively accurate quotient.)
My knowledge is really only with TI's devices, where the TMS320C67xx provide
both a single clock time reciprocal and a single clock time reciprocal of a
square root functions documented in SPRA516, but these are ony accurate to
eight bits of mantissa. In the real world these can be quite useful! I have
a satellite transponder project that makes an attempt to calculate the noise
floor for AGC purposes, and changing the code to use the eight bit
approximations rather than standard 32 bit precision improved speed ten fold
with no perceived degradation of result quality.
Cheers, Howard
.
- References:
- Simple Floating-Pt DSP chip question
- From: Rick Lyons
- Simple Floating-Pt DSP chip question
- Prev by Date: Re: Spectral subtraction
- Next by Date: Re: Optimal sampling for damped sinusoid
- Previous by thread: Re: Simple Floating-Pt DSP chip question
- Next by thread: Re: Simple Floating-Pt DSP chip question
- Index(es):
Relevant Pages
|