> But, to get back to the real topic, I checked out Intrinsificitienceality's > supercalafragalistic website, and unless I'm misreading something, it's > all 32-bit fixed-point arithmetic - very fast, and also completely useless > for > LL testing large exponents. We need fast double-precision floating-point > (or even fixed-point, but *something* on the order of 64 bits) for that.
3 or 4 32bit muls does a 64bit fixed point mul (depending on which 64 bits you want as a result). 2 32b adds does a 64bit add. if I read it right, this thing has 4x4 = 16 ALUs, each with its own register bank that runs at 2Ghz. *each* of these ALU's has a *pair* of 40 bit multiply-accumulators with a 16 x 32 bit register file. oh. dug into the multiply-accumulators a little deeper. each ALU is only a single 16x16 multiplier, with halfword muxes. its dual 40bit accumulator. big f____ deal. ok, never mind. _________________________________________________________________________ Unsubscribe & list info -- http://www.ndatech.com/mersenne/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
