Minutes of the GEnie Forth RoundTable Guest Conference with special guest: Robert L. Smith co-Topics: Floored Division and Floating Point Date: 10/12/89 Time: 22:36EDT Contents Copyright(c) GEnie Forth RoundTable. Entire document may be freely copied and distributed as long as source credit is maintained. Attendees: [R.SMITH44] [GARY-S] [[Kevin] APPERT] [[Jack Brown] J.BROWN11] [[Dennis] D.RUFFER] [[Jax] FIGCHAPTERS] [THE.MUMMY] [M.GARLOCK] Minutes: is here. any last questions Bob ? <[Jack Brown] J.BROWN11> Hi Everyone <[Kevin] APPERT> hi, Jack <[Jack Brown] J.BROWN11> Hi Kevin <[Kevin] APPERT> I guess there are at least 43 other R.Smiths on Genis <[Jack Brown] J.BROWN11> And a lot of Jack Brown's too HELLO ALL 2 minutes to conference is here. <[Jack Brown] J.BROWN11> Hi Fig Chapters <[Kevin] APPERT> Hi, Jax <[Jax] FIGCHAPTERS> Howdy Mr. Brown! Hows the weather in BC? <[Jax] FIGCHAPTERS> Hi Kev. <[Jack Brown] J.BROWN11> Sunny today, rain yesterday <[Jax] FIGCHAPTERS> Hi RS! 1 minute to conference <[Jack Brown] J.BROWN11> I forget the rules Gary... <[Jax] FIGCHAPTERS> Have *I* got question for Guest !?!?!? is here. hi all - ready to go FORMAL !!! is here. <[Jax] FIGCHAPTERS> The Mummy ??? hello is here. <[Jax] FIGCHAPTERS> Does it program in Forth? Before we begin tonight's conference I need to remind ... everyone this is a formal conference. .. .. .. That means NO interlacing. Request a turn to speak by typing ... !! or /RAI . I will keep a list and call on whoever ... is next. PLEASE follow these simple rules : ... PLEASE make your first response ok (so we know YOU know). ... Then, keep your lines short, ending incomplete thoughts ... with three dots (...) and ending your question/statement with ga . thank you.

The GEnie Forth RoundTable is very pleased to welcome ... Robert L. Smith, PhD, Research Specialist at Lockheed Palo Alto. ... Bob received his BS in Physics from Cal Tech and his MS and PhD ... in EE from Stanford. He was one of the original members ... of the Forth Implementation Team, and a member of the IEEE ... Floating Point Committee. Bob became a member of the Forth ... Forth Standards Team (and its secretary) in 1982, and was partly ... responsible for the "floored division" that became part of ... the 83 Standard. He wrote many articles for 'FD', mostly about ... the Standard; and was a member of X3/J14 ANS Forth Technical ... Committee, and is a current member of FIG's Board of Directors. Bob's co-topics this evening are floating point and floored ... division. It is my pleasure to welcome tonight's guest, ... Robert Smith .. .. .. .. .. ga Bob OK THANK YOU. for floored division, it helps to focus on the modulus or remainder rather than the quotient. Most users.. use noly only positive argusments, so floored or non-floored give the same results. For almost all cases that I know of, if you have at least a negative numerator, you propbably.. should use floord division. For Floating Point: (1) should Forth have it at all? and (2) .. If so, should it be in the Standard? And (3) IEEE flting point? That is probably enough from here, so GA with comments. <[Jax] FIGCHAPTERS> Bob, thanx for coming, you and Wil Baden keep us believing in the Standard! ... <[Jax] FIGCHAPTERS> I'm a FLOORED man at gut level, but whatever :-) ... <[Jax] FIGCHAPTERS> My question is a little more general ... <[Jax] FIGCHAPTERS> REading BASIS 9 ... <[Jax] FIGCHAPTERS> I can't help but see it through my Project Manager glasses . .. <[Jax] FIGCHAPTERS> I write specs at work ... <[Jax] FIGCHAPTERS> I see in BASIS 9 ... <[Jax] FIGCHAPTERS> A document somewhat less that 1/4 finished ... <[Jax] FIGCHAPTERS> If you guys & gals don't get some help soon ... <[Jax] FIGCHAPTERS> How long is it going to take to finish?! Don't answer ... <[Jax] FIGCHAPTERS> if it violates the Vth Amendement! :-) I can't answer. For one thing, I am no longer a member of the.. Committee. I agree that there is still a fair amount of work left. <[Jack Brown] J.BROWN11> you say that no one needs... <[Jack Brown] J.BROWN11> negative arguments yet... <[Jack Brown] J.BROWN11> one of my first problems with negative arguments... <[Jack Brown] J.BROWN11> was with a graphics package... <[Jack Brown] J.BROWN11> when converting to floored division... <[Jack Brown] J.BROWN11> and I know others have had the same problem... <[Jack Brown] J.BROWN11> so the comment that -ve arguments can be ignored <[Jack Brown] J.BROWN11> is false... <[Jack Brown] J.BROWN11> any commnet... Jack: I think what I said was that most of the time we use positive args. Clearly there are times when we need negative args. There are a number of clearly defined cases where we need floored div. If you can come up with a clean explaination, in detail.. of why you need rounding toward zero, I and some others whould be interested. Jack did you have follow-up ? ga <[Jack Brown] J.BROWN11> Yes... <[Jack Brown] J.BROWN11> will I guess that I would really like to ... <[Jack Brown] J.BROWN11> see provision for both types of division in the standard... <[Jack Brown] J.BROWN11> and I think your and Wil Baden's <[Jack Brown] J.BROWN11> proposals along these lines may... <[Jack Brown] J.BROWN11> provide a solution... thank you. I hope that my proposal gets a positibe response from the.. committee. It is not 100 % everything we need, but.. goes a long way. Bob Berkey would like to see.. a signed double number divided by a singele, as an additional function. <[Jax] FIGCHAPTERS> Bob , an argument in favor of neg flor which I have not heard mentioned ... <[Jax] FIGCHAPTERS> runs as follows, if I am correct: <[Jax] FIGCHAPTERS> The new RISC chips, including SC32 RISC/STACK chip ... <[Jax] FIGCHAPTERS> actually implement neg floor more efficiently than zero floor ... <[Jax] FIGCHAPTERS> since with no divide instructions, ... <[Jax] FIGCHAPTERS> successive add till 0> ... <[Jax] FIGCHAPTERS> is the fastest alg. for a divide. <[Jax] FIGCHAPTERS> correct? I am not 100% sure of that. The fastest technique depends on just exactly.. how a given chip works. Sometimes a technique of non-restoring.. division is better. I have usually found it quite tricky.. to do for a full signed range. It is certainly easier.. to code by taking care of the signs separately. I hope thats.. enough for now. <[Dennis] D.RUFFER> I don't understand what is the objection to the current... <[Dennis] D.RUFFER> way the TC has addressed the issue... <[Dennis] D.RUFFER> by having the vendors specify what the results are... <[Dennis] D.RUFFER> and providing a way to do the other method... <[Dennis] D.RUFFER> I would think we alread have enough... <[Dennis] D.RUFFER> is it REALLY neccessary to make everyone... <[Dennis] D.RUFFER> supply both algorithms? The current TC method is highly non-transportable. I can't combine.