On Tue, 8 Feb 2005 00:56:24 +0100, Lourens Veen <[EMAIL PROTECTED]> wrote: > On Monday 07 February 2005 21:06, Timothy Miller wrote: > > On Mon, 7 Feb 2005 17:21:39 +0100, Lourens Veen > > > > <[EMAIL PROTECTED]> wrote: > > > On Monday 07 February 2005 13:35, Lourens Veen wrote: > > > > Timothy, is there anything in float25::approx_recp() that seems > > > > problematic to you to implement in hardware? Anything that when removed > > > > would make the hardware much simpler? > > > > > > In particular, that multiplication is now 5 bit (difference) multiplied > > > by 6 bit (invfrac), is that still feasible as a hardcoded bunch of adders > > > or does it require using a built-in multiplier? > > > > I can always synthesize a pipelined multiplier out of random logic. > > I realise that. What I don't know is how expensive things are relative to one > another. Is an extra increment somewhere worth an if-then somewhere else? > That sort of stuff. There is still some wriggle room in that approximate > reciprocal, but I don't know which way to go. any clues would be much > appreciated.
Well, I'm not sure how much logic exactly is used by a 16-bit adder, but say you wanted to multiply that by a 6-bit number. I'd make a 6-stage pipeline, each stage consisting of an adder and a MUX. The 6-bit multiplier would select at each stage whether the shifted multiplicand was added or not. _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
