[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] the wrong way (or not?) concerning FP



On Wed, Aug 29, 2001 at 09:48:37AM +0200, Juergen Goeritz wrote:
[...]
> I took a look into the links you mailed. Haven't seen much
> about a good division though. I will concentrate on the 64 bit
> version, because 32bit is derivable from this.

I'd rather forget fdiv, for the moment.  Concentrate in Level-1 FP,
that is: fadd, fsub, fmul, f2int, int2f, fiaprx and fsqrtiaprx (IIRC).

	fadd/fsub:
		use my generic adder as the core element and add an
		"input (de)normalizer" to it.

	fmul:
		basically, an unsigned multiplier (like IMU, but
		without all the bells and whistles like MAC and signed
		multiplication) for the fractional part, and a small
		(<= 16-bit) adder for the exponent.

	fiaprx/fsqrtiaprx:
		table lookup (8...16 entries) + some easy bit shuffling

	f2int/int2f:
		bit shuffling

Most units will also need a postprocessor for rounding, range checking
and so on.  Shouldn't be too hard either (unless you try to be 100%
IEEE compliant).

-- 
 Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
 "All I wanna do is have a little fun before I die"
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/