McGuire@solbourne.com (Jim McGuire) (06/11/91)
In the "Technology" section, EE Times, June 3, page 29 there is an article discussing a floating point coprocessor design done at IBM's Watson research center. Some quotes: "The Hybrid Number Execution Unit uses a time-honored computational aid - logarithms - to amplify the performance of basic operations" "The key to the hybrid computational unit is a new approach to representing logarithms that closely parallels the IEEE P754 32-bit floating-point format." "The result is a conversion method simple enough to cast in hardware..." "Computer simulations indicate that multiplication, division, squaring, and square root can all be performed by the coprocessor in 10 nanoseconds." After all the discussion we've had lately over floating point, I expected some discussion of this. So does anyone have comments? -- Jim McGuire, Solbourne Computer Speaking for mcguire@solbourne.com myself only!
dana@hardy.hdw.csd.harris.com (Dan Aksel) (06/11/91)
In article <1991Jun10.171824.10980@solbourne.com> McGuire@solbourne.com (Jim McGuire) writes: > "The key to the hybrid computational unit > is a new approach to representing logarithms > that closely parallels the IEEE P754 32-bit > floating-point format." > > "The result is a conversion method simple enough > to cast in hardware..." I've never heard of this. The technique is to take the log of a IEEE P754 number as it comes in from memory? Then mulitplies become adds, divisions become subtraction, exponentiate becomes multiplication... AND ADDITION AND SUBTRACTION BECOME A MESS ! I assume there is some quick translation to get a log of a number? I'd really like to hear more about this one. Anyone out there with more details, I'd appreciate a followup post. Thank you. --- Dan Aksel
kenton@abyss.zk3.dec.com (Jeff Kenton OSG/UEG) (06/12/91)
In article <1991Jun10.171824.10980@solbourne.com> McGuire@solbourne.com (Jim McGuire) writes: > "The key to the hybrid computational unit > is a new approach to representing logarithms > that closely parallels the IEEE P754 32-bit > floating-point format." > > "The result is a conversion method simple enough > to cast in hardware..." This reminds me of the early Wang calculators, which used logs internally. -- ----------------------------------------------------------------------------- == jeff kenton Consulting at kenton@decvax.dec.com == == (617) 894-4508 (603) 881-0011 == -----------------------------------------------------------------------------