[libre-riscv-dev] [Bug 44] IEEE754 FPU inverse-sqrt

[bugzilla-daemon at libre-riscv.org]

http://bugs.libre-riscv.org/show_bug.cgi?id=44

--- Comment #24 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #23)
> (In reply to Luke Kenneth Casson Leighton from comment #22)
> > CSA 3-2:
> > -------
> > 
> > SUM = A XOR B XOR C
> > CRY = NAND(NAND (A,B), NAND(B,C), NAND(A,C))
> > 
> > CSA 4-2:
> > -------
> > 
> > S0, Cout = CSA32(A,B,C)
> > SUM, CRY = CSA32(S0, D, Cin)
> > 
> > that err seems to be it, where normally there would be 5 inputs
> > and 3 outputs (Cin ripples through to Cout).
> > 
> > so err it's not CSA 4-2 at all, it's CSA 5-3 (oink?)
> > 
> > CSA 4-2 would presumably be much simpler?
> 
> it's csa 4-2. notice that there isn't a path from cin to cout.

Am assuming it is for use in chaining.

> I personally prefer csa 3-2 since it takes half as much logic but adds more
> than half as many inputs for the same number of outputs, so it's more
> efficient.

The Tanako paper specifies an algorithm
that requires both.

A ripple carry adder is also mentioned.
That will be the next one to look up.

-- 
You are receiving this mail because:
You are on the CC list for the bug.