Minutes of the FIGGY BAR RT Conference.

Date: 07/12/90   Time: 22:32EDT
   
     The question of the night (in preperation for next weeks guest conference)
     'Have you used the SC32 or SBC32 ?'. Other items discussed include NASA,
     COSMIC,Kitts Peak VAX Forth,Johns-Hopkins,signed carry,+LOOP,LOOP,DIVIDE,
     inverted borrow,STICKY BIT,FOR/NEXT

Attendees:
[GARY-S]
[[Kevin] APPERT]
[[Robert] R.BERKEY]
[[Hail, Denver] JAX]

Minutes:
<[Kevin] APPERT> is here
<GARY-S> did you see my notice just now ?
<[Kevin] APPERT> didn't see anything on the door
<GARY-S> I just posted it - to wit, have you used Silicon Composer's SBC32
<[Kevin] APPERT> nope.  I know Jax had the hots for it a while back
<GARY-S> George Nicol is next weeks guest
<[Kevin] APPERT> is this the Silicon Composer's (APL) chip, or Harris?
<GARY-S> APL
<GARY-S> You 'just' missed George as a matter of fact
<GARY-S> We have a BUSY August working
<[Kevin] APPERT> I've heard great things about the APL chip.  I hope the design 
has reached maturity
<GARY-S> It is my understanding SC and J-H have done some nice clean-up
<R.BERKEY> is here.
<[Kevin] APPERT> I would guess mostly at the Johns-Hopkins end.   I don't think 
Silicon composers has the resoures for didling on the silicon level.
<[Robert] R.BERKEY> What new from Scott Squires?
<GARY-S> Have not heard from Scott in  l-o-n-g time
<GARY-S> Apparently the diddling was mostly done at the foundry
<GARY-S> (mostek ?)
<[Kevin] APPERT> I don't think it's Mostek
<[Kevin] APPERT> maybe we can ask George...
<GARY-S> sure - that would be a good question, plus the evolution since John 
hayes and Marty Fraeman were guests
<[Kevin] APPERT> is he still producing onsites (x-ray litho) or did he buy a 
batch?
<GARY-S> I really don't know, kevin
<[Kevin] APPERT> I bet you could call them up and get their design for media 
cost.  you paid for it, after all
<GARY-S> How did I pay for it ?
<[Kevin] APPERT> tax $
<GARY-S> I pay for the street in front of my house too, but I bet I'd play hell 
helping myself to a chunk of it
<[Kevin] APPERT> you can get Kitt Peak Vax Forth for the cost of sending Kitt 
Peak a blank tape
<[Kevin] APPERT> (let me know if anybody wants to do that, by the way)...
<[Kevin] APPERT> only at NASA do they want you to pay twice...
<[Kevin] APPERT> once with tax dollars, and then again with lisence fees and so 
on...
<[Kevin] APPERT> if you want to use the technology you have already paid for.
<GARY-S> Do you guys receive COSMIC tapes from NASA ?
<[Kevin] APPERT> we use COSMIC NASTRAN...
<[Kevin] APPERT> the most expensive 'public domain' software in history.
<GARY-S> Is anything worth while in it ?
<[Robert] R.BERKEY> Any features missing you'd like to see in the chips?
<[Robert] R.BERKEY> I know I would like a signed carry bit.
<GARY-S> Robert, which chip ? the SC32, RTX ? Both ?
<[Robert] R.BERKEY> If someone were to get a copy, what would they have?
<[Kevin] APPERT> yes, when it's a customer requirement that finite element 
models be delivered in NASTRAN format, it's worthwhile (customer=NASA)
<[Kevin] APPERT> I'd like them all to be cheaper, and available in 
CLASS-S/Rad-hard etc.
<[Robert] R.BERKEY> What's CLASS-S?
<[Kevin] APPERT> space qualified
<[Robert] R.BERKEY> Gary, a general question--any of the chips.
<GARY-S> signed carry isn't a given ?
<[Robert] R.BERKEY> For one, it's what +LOOP needs.  I think it's the XOR of the
<[Robert] R.BERKEY> sign of the subtrahend and the carry.  The carry bit is 
clearly an
<[Robert] R.BERKEY> unsigned bit.
<[Robert] R.BERKEY> The original Hayes design I thought was that way.
<GARY-S> Obviously, I don't use the chip, but I'm amazed signed carry is not 
present..
<GARY-S> it was even available on the 8-bit general uP's
<[Robert] R.BERKEY> What I'm thinking of has the same properties as overflow, 
bit occurs
<[Robert] R.BERKEY> at the transition between -1 and 0.
<JAX> is here.
<GARY-S> I understand carry
<[Robert] R.BERKEY> For some reason I keep running into it--over and over.
<[Robert] R.BERKEY> BASE LOOP /
<[Robert] R.BERKEY> And it's not in the Intel 80x8x family.
<[Robert] R.BERKEY> Which is why +LOOP implementations use the overflow bit.
<[Hail, Denver] JAX> You shoulda seen the hail! Flooded my office, busted my 
hatchback and did $1500 worth of body damage to my Mazda!
<[Kevin] APPERT> are you insured?  when was this?
<[Hail, Denver] JAX> On the 80196, the carry bit is inverted for borrow.
<GARY-S> Saw a friend's Mazda after a trip through South Dakota hail, looked 
like the test base at a ball-peen hammer factory
<[Hail, Denver] JAX> Exactly, gars. It was yesterday, Kev, read this morning's 
papers!
<[Kevin] APPERT> BBC Newsreal didn't mention it
<[Hail, Denver] JAX> Hail the size of baseballs. The biggest I saw was golfball 
size.
<GARY-S> Jax - did you ever play with SC32/SBC32, George Nicol is next week's 
guest
<[Robert] R.BERKEY> JAX do you see the borrow, or inverted carry as useful?
<[Robert] R.BERKEY> I know this is something of confusion on 80x8x.
<[Hail, Denver] JAX> I wrote an assembler for Sc32, SCASM.ZOO here on GEnie.
<[Hail, Denver] JAX> John R. Hayes mailed me his (superior) assembler for it.
<[Hail, Denver] JAX> Bob ...
<[Hail, Denver] JAX> I am not the integer math wiz you are ... BUT!!!!!!!!!!!!!!
!!
<[Hail, Denver] JAX> I have checked out what you said about INTEGER MATH ...
<[Hail, Denver] JAX> DIVIDE ...
<[Hail, Denver] JAX> and you are right. 5 -3 / is *indeed -2 with a modulus of 
2.
<[Hail, Denver] JAX> ( um, did I get the modulus right? :-) )
<[Robert] R.BERKEY> Since I haven't run into the inverted borrow, I don't know, 
but I'd
<[Robert] R.BERKEY> probably like it, although I might then want a reversed 
subtract.
<[Hail, Denver] JAX> 80196 has a STICKY BIT ... meaning ...
<[Robert] R.BERKEY> 5 -3 / is  2 rem -1
<[Hail, Denver] JAX> that during a shift a bit went into CARRY and left again ..
<[Hail, Denver] JAX> helps in rounding.
<[Robert] R.BERKEY> Er, -2 rem -1
<[Hail, Denver] JAX> right.
<[Robert] R.BERKEY> STICKY BIT sounds like something they use in floating point 
processors.
<[Hail, Denver] JAX> Same use .../
<[Robert] R.BERKEY> STICKY GUARD and ROUND.
<[Robert] R.BERKEY> What kind of rounding is it for?
<[Hail, Denver] JAX> 80196 has instructions that help in implementing software 
FP
<[Hail, Denver] JAX> 1/2 bit rouding of A/D for instance.
<[Hail, Denver] JAX> if you round 12 bits to eight, for instance ...
<[Hail, Denver] JAX> and the carry bit is set and the sticky bit is set, you 
know that it was >50% remainder.
<[Hail, Denver] JAX> if carry is set, no sticky, it's 50% ...
<[Hail, Denver] JAX> sticky, no carry, 50% > remainder > 0%
<[Hail, Denver] JAX> no carry, no sticky, r == 0%
<[Robert] R.BERKEY> So, you'd never see it in a single shift then?
<[Hail, Denver] JAX> i think you have to reset it yourself if you want to use 
it.
<[Hail, Denver] JAX> with POPF or POPA.
<[Hail, Denver] JAX> There's also a VT flag, oVerflow Trap.
<[Hail, Denver] JAX> You reset it, if there is an overflow, it sticks set.
<[Robert] R.BERKEY> I think a line got lost--sticky seems to be an imprecise 
indicator.
<[Hail, Denver] JAX> Helps in long calcs where all you want to know is if 
overflow *ever* occurred.
<[Hail, Denver] JAX> So Bob, what is most efficient way to implement DO LOOP on 
such a 16-banger?
<[Robert] R.BERKEY> LOOP is never a problem.  It's +LOOP .
<[Hail, Denver] JAX> why is LOOP never a prob?
<[Hail, Denver] JAX> And tell me about +LOOP
<[Robert] R.BERKEY> because the carry bit detects end-of-loop.
<[Robert] R.BERKEY> +LOOP is a signed increment, but carry is only detecting 
unsigned
<[Robert] R.BERKEY> increments.
<[Hail, Denver] JAX> if you set up your base indices to be +8000
<[Robert] R.BERKEY> The usual approach is to use the overflow bit.
<[Hail, Denver] JAX> then carry detects a "" 1 +LOOP "" correctly, you say?
<[Hail, Denver] JAX> which is what LOOP means, right?
<[Hail, Denver] JAX> so V flag for detecting rolls past 0x8000?
<[Robert] R.BERKEY> If that's not available I forget off of the top of my head, 
but
<[Robert] R.BERKEY> the code is in the 8080 F83.
<[Hail, Denver] JAX> Never again! (8080) :-)
<[Robert] R.BERKEY> The carry implementation is why the text for NEXT is 
incorrect in BASIS.
<[Hail, Denver] JAX> woof?
<GARY-S> hmmm
<[Hail, Denver] JAX> (reaching for BASIS
<[Robert] R.BERKEY> They imply you must test for a result.
<[Robert] R.BERKEY> A result of zero, which is usually less efficient, as 
adding an increment
<[Robert] R.BERKEY> of one will have already done the test.
<[Robert] R.BERKEY> Or subtracting a one depending on how you implement it.
<[Hail, Denver] JAX> "If the current (innermost) loop index is zero, discard 
the loop control parameters and execute the words following NEXT. Otherwi
se, decrement the index and execute the words between FOR and NEXT.
<[Robert] R.BERKEY> Why is ANS insisting on  I  working with NEXT .
<[Hail, Denver] JAX> Maybe the mistake ks in not specifying that NEXT counts 
down?
<[Robert] R.BERKEY> Yeah, I guess I could ignore that text and not be overly 
worried.
<[Hail, Denver] JAX> Wait, the text for FOR specifies.
<[Hail, Denver] JAX> " ... Begin a loop which terminates based on control 
paramters. The loop index begins at  u  and terminates based on the limit
0   .
<[Hail, Denver] JAX> Clear enough.
<[Hail, Denver] JAX> What's the prob? Easy to implement.   I   is a compiler 
word, anyway.
<[Hail, Denver] JAX> On *most* modern micros ...
<[Robert] R.BERKEY> The problem is implementing both FOR NEXT and DO LOOP on 
one system.
<[Hail, Denver] JAX> FOR ... NEXT makes *perfect* sense.
<[Robert] R.BERKEY> But those two I's are not necessarily the same.
<[Hail, Denver] JAX> I thought that at first, bob, but no biggus dealus.
<[Robert] R.BERKEY> If the compiler is used, then you must play some games.
<[Robert] R.BERKEY> It's doable, but since programmers don't like I with FOR 
NEXT anyway,
<[Robert] R.BERKEY> and other FOR NEXT programmers have already come up with -I 
 , I don't
<[Robert] R.BERKEY> understand why ANS rejected my proposal to not require I .
<[Hail, Denver] JAX> : I STATE @ NOT ABORT" Compile only" DUP FOR-FLAG? IF 
FOR-I ELSE DO-FLAG? IF DO-I ELSE ABORT" Loop Control Error" THEN THEN ; IMM
EDIATE
<[Robert] R.BERKEY> I thought maybe you were there and knew the discussion.
<[Hail, Denver] JAX> ( forgot the TRUE before the second ABORT" )
<[Hail, Denver] JAX> ( forgot the DUP before the DO-FLAG? )
<[Robert] R.BERKEY> You still must save and restore the state of FOR-FLAG?
<[Hail, Denver] JAX> NO!
<[Hail, Denver] JAX> just don't use FFFF for every darn compiler control flag!
<[Robert] R.BERKEY> You going to allow nested loops?
<[Robert] R.BERKEY> FOR  I  DO   I LOOP I NEXT
<[Hail, Denver] JAX> You generate a DO or FOR , use two *different* compiler 
security flags than the BRANCH compiler security flag.
<[Hail, Denver] JAX> FOR I DO I LOOP I NEXT would work just peachy.
<[Robert] R.BERKEY> That requires another stack to hold the compiler security, 
as
<[Hail, Denver] JAX> NOPE!
<[Robert] R.BERKEY> ANS allows programs to manage the parameter stack to some 
extent.
<[Hail, Denver] JAX> Look, every time you generate a DO ... LOOP, you have 
tossed a compiler security flag on the stack (in many implementations).
<[Robert] R.BERKEY> Maybe I'm missing something.  I'll take another look at 
your comments.
<[Hail, Denver] JAX> I'm just saying, use these security flags a bit more 
cleverly.
<[Hail, Denver] JAX> FFFF constant BRANCH-FLAG BFFF constant DO-FLAG 7FFF 
constant FOR-FLAG.
<[Hail, Denver] JAX> then your normal vanilla L&P style compiler security 
*also* insures the correct   I   gets compiled.
<[Hail, Denver] JAX> You might argue that this weakens compiler security  ...
<[Hail, Denver] JAX> But *sheesh*, how "secure" is a -1 flag for a compiler 
security flag, esp. when used by *every* control construct!?!?!?! :-)
<[Hail, Denver] JAX> I mean, in some systems,   IF ... LOOP will compile!
<[Hail, Denver] JAX> ((((( BOOM )))))
<[Robert] R.BERKEY> I just noticed that FOR NEXT has "sys" on the stack.
<[Hail, Denver] JAX> "buf of courz, m'sieu"
<[Robert] R.BERKEY> That allows implementations some more options there, while
<[Robert] R.BERKEY> disallowing branches out of FOR NEXT loops.
<[Robert] R.BERKEY> DO LOOP is  "dest" so that means that there is only one item
<[Hail, Denver] JAX> Well, that's a poor description of DO ... LOOP then.
<[Robert] R.BERKEY> on the stack.
<[Hail, Denver] JAX> There oughta be   SYS   there.
<[Robert] R.BERKEY> Which means that DO LOOP security must go elsewhere.
<[Hail, Denver] JAX> I'll try to remember to mention that at Vancouver.
<[Robert] R.BERKEY> But if you make DO "sys" then you can't program
<[Robert] R.BERKEY> DO   WHILE   LOOP   ELSE   THEN    ( add an UNLOOP 
somewhere )
<[Hail, Denver] JAX> the 11th commandment: ( after "thou shalt not pick ...) :
<[Hail, Denver] JAX> THOU SHALT NOT UNLOOP!
<[Hail, Denver] JAX>  :-)
<[Robert] R.BERKEY> Huh? What wrong with UNLOOP ?
<[Robert] R.BERKEY> Why does X3.J14 want I to be defined with FOR NEXT ?
<[Robert] R.BERKEY> That another one of the points.
<[Robert] R.BERKEY> JAX, a question I asked earlier--any features you'd like to
<[Robert] R.BERKEY> see on the chips George Nichols will be discussing soon?
<[Hail, Denver] JAX> excuse, I was discussing Gogol with my son!
<[Robert] R.BERKEY> But not googol?
<[Hail, Denver] JAX> My thoughts on SC32, etc. ....
<[Hail, Denver] JAX> will be contained in my Sept. article for ESP magazine ...
<[Hail, Denver] JAX> presuming they publish that article!
<[Hail, Denver] JAX>  :-)
<[Hail, Denver] JAX> What's up, GARS ... bwana mighty quiet tonite.
<GARY-S> didn't care to interrupt an interesting exchange
<GARY-S> Any last thoughts
======= End of Steno notes ======