4 Sep
2016
4 Sep
'16
9:39 p.m.
Just thought I'd share with you all one of my pet projects for the past few
years. I have written an HP-35 and HP-45 simulator for the PDP-8, which
uses the original microcode from these calculators. Thus, it is just as
accurate as the original calculators, I would believe.
Not really sure why I decided to do this, but I suppose I enjoy both HP
calculators and PDP-8 programming. I guess it's nice to have a scientific
calculator at one's disposal too while sitting in front of a PDP-8, too. It
also shows that something like this could've existed back in the early
1970s if one were so inclined to do it. Perhaps it's also a bit of homage
to the Expensive Desk Calculator for the TX-0 and PDP-1.
Anyways, you can grab the code from here:
https://github.com/drovak/hpcalcpdp8
The HP-45 support is new as of the past couple of days; I went ahead and
through in a help function to explain the keys. Toggle between 4XXX and
0XXX on the switch register to toggle between HP-45 and HP-35 modes,
respectively. Feel free to give feedback; I've tested it fairly extensively
in SimH with no major issues other than no error indication on the HP-35. I
have a debug mode which prints each register every instruction cycle, but I
haven't used it too much in recent times; I went ahead and IFDEF'ed it as
it also reads from the switch register.
OS/8 support will come soon, as that shouldn't be challenging. You need 8k
to run this, though I also uploaded my original hp35pdp8.pal code which
runs in 4k. It lacks the help function (and obviously the HP-45 simulator),
but otherwise, works pretty much the same.
Big thanks to Eric Smith for his 'nonpareil' work which inspired much of
the simulator.
Kyle
4 Sep
4 Sep
10:43 p.m.
On Sep 4, 2016, at 21:39, Kyle Owen <kylevowen at
gmail.com> wrote:
Just thought I'd share with you all one of my pet projects for the past few
years. I have written an HP-35 and HP-45 simulator for the PDP-8, which
uses the original microcode from these calculators. Thus, it is just as
accurate as the original calculators, I would believe.
That is silly, fantastic, absurd, and wonderful. All at the same time.
--
Mark J. Blair, NF6X <nf6x at nf6x.net>
http://www.nf6x.net/
5 Sep
5 Sep
1:14 a.m.
On Sun, Sep 04, 2016 at 10:43:19PM -0700, Mark J. Blair wrote:
Agreed :) Period correct programming.
/P
On Sep 4, 2016, at 21:39, Kyle Owen <kylevowen
at gmail.com> wrote:
Just thought I'd share with you all one of my pet projects for the past few
years. I have written an HP-35 and HP-45 simulator for the PDP-8, which
uses the original microcode from these calculators. Thus, it is just as
accurate as the original calculators, I would believe.
That is silly, fantastic, absurd, and wonderful. All at the same time.
6 Sep
6 Sep
11:10 p.m.
I updated the project to include optional OS/8 support. I won't say I've
tested it extensively, but it does seem to be working as expected in SimH,
anyways. I updated the README to reflect the additions. The directory
structure was also updated to something more sane.
The keen observer will note that I also changed up some of the debugging
features which are likely to not be useful to anyone other than the
author...and even then, I only used the features a few times when getting
the thing running initially. But, it's there if you need it, all
configurable through the switch register as detailed in the source.
Glad some folks got a kick out of it enough to try it out! Feel free to
suggest improvements where you see fit. I was thinking about adding support
to read keystrokes from a file for macro programmability...but that might
be too absurd even for this project. Maybe the HP-41C simulator is next...
:)
Kyle
7 Sep
7 Sep
4:27 a.m.
On Wed, Sep 7, 2016 at 12:10 AM, Kyle Owen <kylevowen at gmail.com> wrote:
Maybe the HP-41C simulator is next...
That will definitely be entertaining.
It will need three PDP-8 fields just for the 41C ROMs (12Kx10), and
400 words for the basic 41C RAM (16 "status" registers, 64
data/program registers, all of 56 bits each).
I think the 41CX might not fit, because it would need six fields for
the 41CX ROMs, and 2320 words for the RAM (total 464x56). The
simulator code and OS/8 would have to fit in less than 1.5 fields.
8:15 a.m.
On Sep 7, 2016 6:27 AM, "Eric Smith" <spacewar at gmail.com> wrote:
On Wed, Sep 7, 2016 at 12:10 AM, Kyle Owen <kylevowen at gmail.com> wrote:
I could imagine OS/8 would come in handy here, as we could swap fields out,
maybe between several fields of ROM instructions, as kind of a cache
approach. There would be a ROM file that's required for use.
Hmm...but the biggest issue I see is the complex display, don't you think?
Maybe if we use a VC8E option...which would be fun to support for the
original simulator, too!
Kyle
Maybe the HP-41C simulator is next...
That will definitely be entertaining.
It will need three PDP-8 fields just for the 41C ROMs (12Kx10), and
400 words for the basic 41C RAM (16 "status" registers, 64
data/program registers, all of 56 bits each).
I think the 41CX might not fit, because it would need six fields for
the 41CX ROMs, and 2320 words for the RAM (total 464x56). The
simulator code and OS/8 would have to fit in less than 1.5 fields.
5:18 p.m.
On Wed, Sep 7, 2016 at 9:15 AM, Kyle Owen <kylevowen at gmail.com> wrote:
I could imagine OS/8 would come in handy here, as we
could swap fields out,
maybe between several fields of ROM instructions, as kind of a cache
approach. There would be a ROM file that's required for use.
I thought about swapping, but the microcode jumps around so much that
it would be insanely slow.
8:53 a.m.
I'd vote for the HP-1xC line myself. You can get 5 different calculators
(financial, 3 x scientific and programmer) for nearly the effort of the
first one.
The HP-16C (http://www.hpmuseum.org/hp16.htm) would be especially helpful
as it can easily be converted to calculate with a 12 bit with carry width
and octal operation with just a few keystrokes. Even does 1's
complement/18 bits if you happen to have a PDP-1 lying around.
On Tue, Sep 6, 2016 at 11:10 PM, Kyle Owen <kylevowen at gmail.com> wrote:
I updated the project to include optional OS/8
support. I won't say I've
tested it extensively, but it does seem to be working as expected in SimH,
anyways. I updated the README to reflect the additions. The directory
structure was also updated to something more sane.
The keen observer will note that I also changed up some of the debugging
features which are likely to not be useful to anyone other than the
author...and even then, I only used the features a few times when getting
the thing running initially. But, it's there if you need it, all
configurable through the switch register as detailed in the source.
Glad some folks got a kick out of it enough to try it out! Feel free to
suggest improvements where you see fit. I was thinking about adding support
to read keystrokes from a file for macro programmability...but that might
be too absurd even for this project. Maybe the HP-41C simulator is next...
:)
Kyle
1:29 p.m.
There is a nice HP-16C emulator for Windows - and Android.
http://www.wrpn.emmet-gray.com/
Also has Java version
(and sources)
Keven Miller
----- Original Message -----
From: "Marc Howard" <cramcram at gmail.com>
To: "General Discussion: On-Topic and Off-Topic Posts"
<cctalk at classiccmp.org>
Sent: Wed 07 Sep 2016 09:53 AM
Subject: Re: HP-35/45 Simulator for PDP-8
I'd vote for the HP-1xC line myself. You can get
5 different calculators
(financial, 3 x scientific and programmer) for nearly the effort of the
first one.
The HP-16C (http://www.hpmuseum.org/hp16.htm) would be especially helpful
as it can easily be converted to calculate with a 12 bit with carry width
and octal operation with just a few keystrokes. Even does 1's
complement/18 bits if you happen to have a PDP-1 lying around.
8 Sep
8 Sep
6:30 a.m.
On Wed, 7 Sep 2016, Kyle Owen wrote:
I updated the project to include optional OS/8
support. I won't say I've
tested it extensively, but it does seem to be working as expected in SimH,
anyways. I updated the README to reflect the additions.
...
I tried it out: I dowloaded the hp35only.pal-source, because it needs only
4K of memory, kermited it to my PDP8/E and succeeded assembling with PAL8
without modifying the PALBERT-code.
After starting it worked well. :-)
As I'm always looking for 4K-applications for our older machines,
I downloaded the .BN to our classic-8 and started it. It gave
"strange" output. :-(
A look in the source showd some BSWs, an operation that doesn't exist
on pre PDP8/E machines. I replaced the BSWs with "JMS BSWEMU", a short
subroutine to perform the byte-swap and now the hp35 runs on the
classic-8 with 4K of memory too. I also had to replace the MQLs with CLAs.
Next step is to modify the output, to run it with a teletype...
Klemens
--
klemens krause
Stuttgarter KompetenzZentrum fyr Minimal- & Retrocomputing.
http://computermuseum.informatik.uni-stuttgart.de
3:53 p.m.
On Thu, Sep 8, 2016 at 8:30 AM, Klemens Krause <
krause at informatik.uni-stuttgart.de> wrote:
I tried it out: I dowloaded the hp35only.pal-source, because it needs only
4K of memory, kermited it to my PDP8/E and succeeded assembling with PAL8
without modifying the PALBERT-code.
After starting it worked well. :-)
Awesome, glad to hear!
As I'm always looking for 4K-applications for our older machines,
I downloaded the .BN to our classic-8 and started it.
It gave
"strange" output. :-(
A look in the source showd some BSWs, an operation that doesn't exist
on pre PDP8/E machines. I replaced the BSWs with "JMS BSWEMU", a short
subroutine to perform the byte-swap and now the hp35 runs on the
classic-8 with 4K of memory too. I also had to replace the MQLs with CLAs.
Next step is to modify the output, to run it with a teletype...
You've inspired me to try my hand at converting the current version into a
4K HP-35 only version with a macro switch. Teletype output is already in
the works too; I have some ideas there.
How does the following compare to your BSWEMU, by the way? This ensures
that the link bit remains untouched, which may or may not be important in
every case of BSW in my application.
BSWSUB, 0
DCA BSWTMP /SAVE VALUE
SZL /IS LINK ZERO?
TAD (0100) /NO, ADD 100
DCA BSWLNK /SAVE LINK
TAD BSWTMP /GET VALUE TO BSW
CLL /CLEAR LINK
AND (7700) /KEEP TOP HALF
RAL /ROTATE IT LEFT
TAD BSWLNK /ADD LINK
DCA BSWLNK /SAVE IT
TAD BSWTMP /GET VALUE
AND (0077) /KEEP BOTTOM HALF
TAD BSWLNK /GET REST OF IT
RTL /ROTATE INTO POSITION
RTL
RTL
JMP I BSWSUB /RETURN WITH ORIGINAL LINK AND BSW
BSWTMP, 0
BSWLNK, 0
Essentially, I shift the top six bits once to the left into a now-emptied
link bit, add in the bottom six bits and the previous link bit to the left
of the bottom six bits, then shift 6 times to the left; the old link is now
in the link position, and the two halves of the AC are swapped.
I'm sure I've seen some code before that does this, but I can't seem to
find any now that I'm looking for it. Maybe there's a shorter way. I think
this takes 23 words if you include the (0100), (7700), and (0077), which
may or may not also be used elsewhere in the first page where I put this
subroutine.
Thanks,
Kyle
5:07 p.m.
From: Kyle Owen: Thursday, September 08, 2016 3:53 PM
How does the following compare to your BSWEMU, by the
way? This ensures
that the link bit remains untouched, which may or may not be important in
every case of BSW in my application.
I'm sure I've seen some code before that does
this, but I can't seem to
find any now that I'm looking for it. Maybe there's a shorter way. I think
this takes 23 words if you include the (0100), (7700), and (0077), which
may or may not also be used elsewhere in the first page where I put this
subroutine.
Here's my slightly optimized version, for what it's worth:
1 *400
2 /
3 / BSW emulation
4 /
5 00400 0000 bsw, .-.
6 00401 3216 dca saveac / Save AC
7 00402 7630 szl cla / Link set?
8 00403 7140 cll cma / Yes, remember it
9 00404 3217 dca savel
10 00405 1216 tad saveac / Get 0 xxx xxx yyy yyy
11 00406 0220 and c7700
12 00407 1216 tad saveac / x xx xxx0 yyy yyy
13 00410 7006 rtl
14 00411 7006 rtl
15 00412 7006 rtl / 0 yyy yyy xxx xxx
16 00413 2217 isz savel / Was link set?
17 00414 7020 cml / Yes, restore it
18 00415 5600 jmp i bsw / return
19 00416 0000 saveac, .-.
20 00417 0000 savel, .-.
21 00420 7700 c7700, 7700
22 $
There is a fair chance that each of saveac, savel, and c7700 can be reused
elsewhere.
Some assemblers flag the "(" construct when used on page 0, BTW. It's more
correct
to use "[" there (or to avoid both).
I'm a "family of 8" guy, so I avoid using BSW to begin with.
Vince
5:27 p.m.
From: "Vincent Slyngstad" <v.slyngstad at frontier.com>
Sent: Thursday, September 08, 2016 5:07 PM
7 00402 7630 szl cla / Link set?
8 00403 7140 cll cma / Yes, remember it
That should probably be:
7 00402 7620 snl cla / Link set?
8 00403 7140 cll cma / No, remember it
Needs more testing :-/.
Vince
5:43 p.m.
On Sep 8, 2016 8:28 PM, "Vincent Slyngstad" <v.slyngstad at frontier.com>
wrote:
From: "Vincent Slyngstad" <v.slyngstad at frontier.com>
Sent: Thursday, September 08, 2016 5:07 PM
Probably can make it "snl" instead of "snl cla" since you have a
"dca"
before it, but it doesn't save code space or time. Just an observation.
Kyle
7 00402 7630 szl cla / Link
set?
8 00403 7140 cll cma / Yes, remember it
That should probably be:
7 00402 7620 snl cla / Link set?
8 00403 7140 cll cma / No, remember it
Needs more testing :-/.
5:57 p.m.
From: Kyle Owen: Thursday, September 08, 2016 5:43 PM
On Sep 8, 2016 8:28 PM, "Vincent Slyngstad"
wrote:
Your message crossed my offlist message about the 0 in line 10 needing
to be an "L", and the useless cll in "cll cma". I still like the cla
in "snl
cla",
even though it doesn't do anything. But the cll on the next line seems
misleading. All of which is probably more on this subject than other folks
care about :-).
Vince
That should probably be:
7 00402 7620 snl cla / Link set?
8 00403 7140 cll cma / No, remember it
Needs more testing :-/.
Probably can make it "snl" instead of "snl cla" since you have a
"dca"
before it, but it doesn't save code space or time. Just an observation.
9 Sep
9 Sep
8:07 a.m.
On Thu, 8 Sep 2016, Vincent Slyngstad wrote:
From: Kyle Owen: Thursday, September 08, 2016 5:43 PM
> On Sep 8, 2016 8:28 PM, "Vincent Slyngstad" wrote:
>> That should probably be:
>> 7 00402 7620 snl cla / Link set?
>> 8 00403 7140 cll cma / No, remember it
>>
>> Needs more testing :-/.
Yes, I think, there is a bug in your code:
SNL CLA /CLA IS REDUNDANT, AS AC is 0 here
CLL CMA /CLL MAKES NO SENSE BECAUSE LINK IS NOT SET, IF THE SNL
/DOES NOT SKIP
CLL /IS NEEDED AFTER THE SNL, OTHERWAYS THE MSB WOULD BE
/INVERTED AFTER THE BSW-ROUTINE.
I tried both versions with the emulator, and both gave identical
results. So I tested another "BSW"-replacement, simply:
BSWEMU, 0000
RTR
RTR
RTR
JMP I BSWEMU
This also worked. ;-) I didn't look in the code, but evidently the
BSWs are only used, to get the higher 6 bits of a word.
And now another question came up. What does a hp-calculator answer,
if you ask fot tan(90)?
Answer:
hp35_emu: 90 t gives 9.9999999999 E99
hp35_real: didn't find the power supply :-(
hp9100: 90 tan gives 9.9999999999 E99
hp9810: 90 tan gives 9.9999999999 E98 (yes, E98!)
hp9820: 90 tan gives "NOTE 01"
Here are the three version of BSW:
/ K. KRAUSE K. OWEN V. SLYNGSTEDT
BSWEMU, NOP /BSWSUB, 0 /BSW, 0
DCA AC / DCA BSWTMP / DCA SAVEAC
RAL / SZL / SNL
DCA LINK / TAD C0100 / CMA
TAD AC / DCA BSWLINK / CLL
RTR / TAD BSWTMP / DCA SAVEL
RTR / CLL / TAD SAVEAC
RTR / AND C7700 / AND C7700
AND M0077 / RAL / TAD SAVEAC
DCA BH / TAD BSWLNK / RTL
TAD AC / DCA BSWLNK / RTL
AND M0077 / TAD BSWTMP / RTL
RTL / AND C0077 / ISZ SAVEL
RTL / TAD BSWLNK / CML
RTL / RTL / JMP I BSW
TAD BH / RTL /SAVEAC,0
CLL RAL / RTL /SAVEL, 0
TAD LINK / JMP I BSWSUB /C7700, 7700
RAR / BSWTMP
JMP I BSWEMU / BSWLNK
AC, 0000 / C7700
LINK, 0000 / C0077
BH, 0000 / C0100
M0077, 0077
/
I added literals an temporary locations to the code, because one cannot
be sure, that a special literal, as for example 0100 or 0077 exists in
the other code.
Klemens
--
klemens krause
Stuttgarter KompetenzZentrum fyr Minimal- & Retrocomputing.
http://computermuseum.informatik.uni-stuttgart.de
9:42 a.m.
On Fri, Sep 9, 2016 at 10:07 AM, Klemens Krause <
krause at informatik.uni-stuttgart.de> wrote:
I tried both versions with the emulator, and both gave identical
results. So I tested another "BSW"-replacement, simply:
BSWEMU, 0000
RTR
RTR
RTR
JMP I BSWEMU
This also worked. ;-) I didn't look in the code, but evidently the
BSWs are only used, to get the higher 6 bits of a word.
I'm pleasantly surprised this worked, as I see several cases in my code
where I'm using BSW to shift 6 times to the left, which isn't the same as
shifting 6 times to the right since there's that pesky link bit, unless I'm
missing something obvious here. Hmm. I'm updating the code to selectively
use three RTRs or three RTLs instead of BSW with the single field option
enabled.
And now another question came up. What does a hp-calculator answer,
if you ask fot tan(90)?
Answer:
hp35_emu: 90 t gives 9.9999999999 E99
hp35_real: didn't find the power supply :-(
hp9100: 90 tan gives 9.9999999999 E99
hp9810: 90 tan gives 9.9999999999 E98 (yes, E98!)
hp9820: 90 tan gives "NOTE 01"
Huh. E98...that's pretty odd. Wonder what the algorithm looks like for the
9810. The real HP-35 gives E99 like the emulator.
I'll push the changes soon; please give it a try on your 4K machine and let
me know how it does!
Thanks,
Kyle
11:13 a.m.
Just pushed the changes; please give it a try! Printing terminal support is
next, then maybe I'll tackle the HP-45 error display.
Thanks!
Kyle
8 Sep
8 Sep
5:38 p.m.
On Sep 8, 2016 8:07 PM, "Vincent Slyngstad" <v.slyngstad at frontier.com>
wrote:
Here's my slightly optimized version, for what it's worth:
Nice work. Definitely shorter than mine. I was just working on an
optimization that used ISZ, but you beat me to it!
Some assemblers flag the "(" construct when used on page 0, BTW. It's
more correct
to use "[" there (or to avoid both).
Yup, I tend to use '(' when I'm writing code that could go anywhere, and
'[' for special cases, like when I'm setting up the "JMS I [BSWEMU]"
or the
like.
I'm a "family of 8" guy, so I avoid using BSW to begin with.
I should probably get in the habit of avoiding it...but it's so darn
convenient! And faster. And I'm all about efficiency! Truth be told, if I
ever get a pre-8/E computer, I'd likely be changing my programming habits.
Kyle
13 Sep
13 Sep
1:38 p.m.
Vince,
This struck me as a pretty clever bit of code. My first cut used 23 words
including 2 masks and 3 temporaries.
This is the working version of Vince's original. It uses 18 words
including 1 mask and 2 temporaries.
I have done some edge testing and it appears to work.
30 / THIS VERSION BY VINCE SLYNGSTAD
31 00210 0000 BSWI, .-. /ENTRY POINT
32 00211 3175 DCA SAVEAC
33 00212 7420 SNL /REMEMBER LINK STATE
34 00213 7040 CMA /AS A -1 FOR ISZ TEST
35 00214 3176 DCA SAVEL
36 00215 7100 CLL /NEED THE LINK CLEARED
37 00216 1175 TAD SAVEAC /GET 0 XXX XXX YYY YYY
38 00217 0177 AND C7700 / 0 XXX XXX 000 000
39 00220 1175 TAD SAVEAC / X XXX XX0 YYY YYY
40 00221 7006 RTL / X XXX 0YY YYY YXX
41 00222 7006 RTL / X X0Y YYY YYX XXX
42 00223 7006 RTL / 0 YYY YYY XXX XXX
43 00224 2176 ISZ SAVEL /WAS LINK SET?
44 00225 7020 CML /YES, RESTORE LINK
45 00226 5610 JMP I BSWI /RETURN
46 $
As Klemens Krause points out you do need the CLL (line 36) somewhere before
the second TAD SAVEAC or it will flip
what was the original leftmost bit (AC 0).
I have thought about this a bunch and have come up with an improved
version. It also uses 18 words including 2 constants
two temporaries. This is one less instruction executed.
32 00210 0000 BSWI, .-. /ENTRY POINT
33 00211 3174 DCA SAVEAC
34 00212 7430 SZL /REMEMBER LINK STATE
35 00213 1177 TAD C0100 /PRE ROTATE LINK POSITION
36 00214 3175 DCA SAVEL
37 00215 7100 CLL /NEED THE LINK CLEARED
38 00216 1174 TAD SAVEAC /GET 0 XXX XXX YYY YYY
39 00217 0176 AND C7700 / 0 XXX XXX 000 000
40 00220 1174 TAD SAVEAC / X XXX XX0 YYY YYY
41 00221 1175 TAD SAVEL / X XXX XXL YYY YYY
42 00222 7006 RTL / X XXX LYY YYY YXX
43 00223 7006 RTL / X XLY YYY YYX XXX
44 00224 7006 RTL / L YYY YYY XXX XXX
45 00225 5610 JMP I BSWI /RETURN
Not bad but I realized even more was possible.
31 00210 0000 BSWI, .-. /ENTRY POINT
32 00211 3174 DCA SAVEAC
33 00212 7430 SZL /REMEMBER LINK STATE
34 00213 1176 TAD C0100 /PRE ROTATE LINK POSITION
35 00214 7100 CLL /NEED THE LINK CLEARED
36 00215 1174 TAD SAVEAC /GET 0 XXX XXX YYY YYY
37 00216 0175 AND C7700 / 0 XXX XXX 000 000
38 00217 1174 TAD SAVEAC / X XXX XX0 YYY YYY
39 00220 7006 RTL / X XXX LYY YYY YXX
40 00221 7006 RTL / X XLY YYY YYX XXX
41 00222 7006 RTL / L YYY YYY XXX XXX
42 00223 5610 JMP I BSWI /RETURN
I realized I could add in the link with the first TAD and the result is the
same so was able to eliminate the DCA SAVEL and
the corresponding TAD SAVEL and since SAVEL was not used anymore I got rid
of that as well. This is 15 words long which
includes the two constants and the one temporary.
I think the only way you could do this faster would be to use a whole field
as a lookup table but that uses a lot more space.
48 00227 0000 BSWT1, .-.
49 00230 3175 DCA SAVEAC
50 00231 6271 CDF 070 /ASSUME LOOKUP TABLE IS ON PAGE 7
51 00232 1575 TAD I SAVEAC
52 00233 6201 CDF 000 /RESTORE DATA FIELD
53 00234 5627 JMP I BSWT1 /RETURN
Something like that anyway.
Thanks for an interesting bit of optimization!
Doug
On Thu, Sep 8, 2016 at 6:07 PM, Vincent Slyngstad <v.slyngstad at frontier.com>
wrote:
From: Kyle Owen: Thursday, September 08, 2016 3:53 PM
--
Doug Ingraham
PDP-8 SN 1175
How does the following compare to your BSWEMU, by
the way? This ensures
that the link bit remains untouched, which may or may not be important in
every case of BSW in my application.
I'm sure I've seen some code before that does this, but I can't seem to
find any now that I'm looking for it. Maybe
there's a shorter way. I think
this takes 23 words if you include the (0100), (7700), and (0077), which
may or may not also be used elsewhere in the first page where I put this
subroutine.
Here's my slightly optimized version, for what it's worth:
1 *400
2 /
3 / BSW emulation
4 /
5 00400 0000 bsw, .-.
6 00401 3216 dca saveac / Save AC
7 00402 7630 szl cla / Link set?
8 00403 7140 cll cma / Yes, remember it
9 00404 3217 dca savel
10 00405 1216 tad saveac / Get 0 xxx xxx yyy yyy
11 00406 0220 and c7700
12 00407 1216 tad saveac / x xx xxx0 yyy yyy
13 00410 7006 rtl
14 00411 7006 rtl
15 00412 7006 rtl / 0 yyy yyy xxx xxx
16 00413 2217 isz savel / Was link set?
17 00414 7020 cml / Yes, restore it
18 00415 5600 jmp i bsw / return
19 00416 0000 saveac, .-.
20 00417 0000 savel, .-.
21 00420 7700 c7700, 7700
22 $
There is a fair chance that each of saveac, savel, and c7700 can be reused
elsewhere.
Some assemblers flag the "(" construct when used on page 0, BTW. It's
more correct
to use "[" there (or to avoid both).
I'm a "family of 8" guy, so I avoid using BSW to begin with.
Vince
6 p.m.
From: Doug Ingraham: Tuesday, September 13, 2016 1:38 PM
This struck me as a pretty clever bit of code. My
first cut used 23 words
including 2 masks and 3 temporaries.
Thanks!!
This is the working version of Vince's original.
It uses 18 words
including 1 mask and 2 temporaries.
I have done some edge testing and it appears to work.
30 / THIS VERSION BY VINCE SLYNGSTAD
31 00210 0000 BSWI, .-. /ENTRY POINT
32 00211 3175 DCA SAVEAC
33 00212 7420 SNL /REMEMBER LINK STATE
34 00213 7040 CMA /AS A -1 FOR ISZ TEST
35 00214 3176 DCA SAVEL
36 00215 7100 CLL /NEED THE LINK CLEARED
Ok, I just don't see why one should need to clear link here.
37 00216 1175 TAD SAVEAC /GET 0 XXX
XXX YYY YYY
38 00217 0177 AND C7700 / 0 XXX XXX 000 000
These two lines, the comment would have "L" instead of their first
"0".
39 00220 1175 TAD SAVEAC / X XXX
XX0 YYY YYY
Poof! There goes the prior value of link! And the "0" in the comment
is correct.
40 00221 7006 RTL / X XXX
0YY YYY YXX
41 00222 7006 RTL / X X0Y YYY YYX XXX
42 00223 7006 RTL / 0 YYY YYY XXX XXX
43 00224 2176 ISZ SAVEL /WAS LINK SET?
44 00225 7020 CML /YES, RESTORE LINK
45 00226 5610 JMP I BSWI /RETURN
46 $
As Klemens Krause points out you do need the CLL (line 36) somewhere before
the second TAD SAVEAC or it will flip
what was the original leftmost bit (AC 0).
Someone help me understand this claim.
I have thought about this a bunch and have come up
with an improved
version. It also uses 18 words including 2 constants
two temporaries. This is one less instruction executed.
32 00210 0000 BSWI, .-. /ENTRY POINT
33 00211 3174 DCA SAVEAC
34 00212 7430 SZL /REMEMBER LINK STATE
35 00213 1177 TAD C0100 /PRE ROTATE LINK POSITION
36 00214 3175 DCA SAVEL
37 00215 7100 CLL /NEED THE LINK CLEARED
38 00216 1174 TAD SAVEAC /GET 0 XXX XXX YYY YYY
39 00217 0176 AND C7700 / 0 XXX XXX 000 000
40 00220 1174 TAD SAVEAC / X XXX XX0 YYY YYY
41 00221 1175 TAD SAVEL / X XXX XXL YYY YYY
42 00222 7006 RTL / X XXX LYY YYY YXX
43 00223 7006 RTL / X XLY YYY YYX XXX
44 00224 7006 RTL / L YYY YYY XXX XXX
45 00225 5610 JMP I BSWI /RETURN
Not bad but I realized even more was possible.
31 00210 0000 BSWI, .-. /ENTRY POINT
32 00211 3174 DCA SAVEAC
33 00212 7430 SZL /REMEMBER LINK STATE
34 00213 1176 TAD C0100 /PRE ROTATE LINK POSITION
35 00214 7100 CLL /NEED THE LINK CLEARED
36 00215 1174 TAD SAVEAC /GET 0 XXX XXX YYY YYY
37 00216 0175 AND C7700 / 0 XXX XXX 000 000
38 00217 1174 TAD SAVEAC / X XXX XX0 YYY YYY
39 00220 7006 RTL / X XXX LYY YYY YXX
40 00221 7006 RTL / X XLY YYY YYX XXX
41 00222 7006 RTL / L YYY YYY XXX XXX
42 00223 5610 JMP I BSWI /RETURN
Nice!
Vince
8:02 p.m.
On Tue, Sep 13, 2016 at 7:16 PM, Vincent Slyngstad <v.slyngstad at frontier.com
wrote:
From: Vincent Slyngstad: Tuesday, September 13, 2016
6:00 PM
My first test case would fail without the CLL
BSW=JMS I BSWIND
*0170
C7700, 7700
BSWIND, BSWI
*0200
CLA CLL CML /CLEAR AC AND SET LINK
TAD C7700 / AC <- 7700
BSW /AFTER THIS AC SHOULD BE 0077
HLT /LOOK AT THE PRETTY LIGHTS
BSWI, .-.
I believe this returns 0037 instead of 0077 if the CLL is left out of the
code, but I believe you figured this out.
I've been trying to come up with a way to reorder this but without
success. I am going to have Warren look at it as
he has a twisted mind and sometimes sees things I miss.
--
Doug Ingraham
PDP-8 SN 1175
Poof! There goes the prior value of link! And
the "0" in the comment is
correct.
Never mind. I finally see what Klemens and others were trying to tell me
about the need for CLL.
Vince
14 Sep
14 Sep
6:27 a.m.
On Tue, 13 Sep 2016, Doug Ingraham wrote:
Not bad but I realized even more was possible.
31 00210 0000 BSWI, .-. /ENTRY POINT
32 00211 3174 DCA SAVEAC
33 00212 7430 SZL /REMEMBER LINK STATE
34 00213 1176 TAD C0100 /PRE ROTATE LINK POSITION
35 00214 7100 CLL /NEED THE LINK CLEARED
36 00215 1174 TAD SAVEAC /GET 0 XXX XXX YYY YYY
37 00216 0175 AND C7700 / 0 XXX XXX 000 000
38 00217 1174 TAD SAVEAC / X XXX XX0 YYY YYY
39 00220 7006 RTL / X XXX LYY YYY YXX
40 00221 7006 RTL / X XLY YYY YYX XXX
41 00222 7006 RTL / L YYY YYY XXX XXX
42 00223 5610 JMP I BSWI /RETURN
I realized I could add in the link with the first TAD and the result is the
same so was able to eliminate the DCA SAVEL and
Nice idea, I needed some time to understand whats going on.
Application of the commutativity of addition:
PREROTLINK + XXXXXX + XXXXXX = XXXXXX + XXXXXX + PREROTLINK :-)
I found the next simplification:
BSWI, 0
DCA SAVEAC
RTR / MOVE LINK TO THE PREROTATE POSITION
RTR / ELIMINATE C0100
RTR / LINK IS CLEARED BECAUSE AC WAS 0 AFTER DCA
TAD SAVEAC
AND C7700
TAD SAVEAC
RTL
RTL
RTL
JMP I BSWI
SAVEAC,0
C7700, 7700
Statistically, if 50% link is set, and 50% link is not set both
versions have identical speed, but one constant is eliminated.
Doug Ingraham
PDP-8 SN 1175
Klemens
PDP-8 SN 768
--
klemens krause
Stuttgarter KompetenzZentrum fyr Minimal- & Retrocomputing.
http://computermuseum.informatik.uni-stuttgart.de
8:59 p.m.
On Wed, Sep 14, 2016 at 7:27 AM, Klemens Krause <
krause at informatik.uni-stuttgart.de> wrote:
I found the next simplification:
BSWI, 0
DCA SAVEAC
RTR / MOVE LINK TO THE PREROTATE POSITION
RTR / ELIMINATE C0100
RTR / LINK IS CLEARED BECAUSE AC WAS 0 AFTER DCA
I had been looking at this section as the last piece of clumsy coding.
Your solution is simplicity itself! My ideas were all more complicated and
worse than the original. I actually have the three RTR in a note but it
was with a SNL and a JMP .+4 around them. I didn't stop to think that I
didn't need the test at all. Another idea was to create the constant 0100
so I didn't have to make a constant for it. CLA CLL CML RTR;RTR;RTR. I
don't have to explain why I discarded it. Anyway, Great work! I doubt
this can be made shorter, but I would really like to be proved wrong.
Of course this code is obfuscated to the point that without the comments it
would be difficult to understand what it was supposed to do.
I recently keyboarded the Maindec that is the part 1 test for the EAE. The
guy who coded it appeared to not like doing
DCA XXX
TAD YYY
to take advantage of the fact that the AC is zero. Instead he would do
this to load a
value into the AC
DCA XXX
CMA
AND YYY
This of course does not ever try to touch the Link and so that might have
been the goal although the TAD with Zero AC
should never mess with the Link either. I found reading this code to be
tedious and I want to rewrite the diagnostic in a
more sensible manner. This guy had some other really strange coding
practices.
--
klemens krause
Stuttgarter KompetenzZentrum fyr Minimal- & Retrocomputing.
http://computermuseum.informatik.uni-stuttgart.de
--
Doug Ingraham
PDP-8 SN 1175
13 Sep
13 Sep
12:55 p.m.
On Wed, 7 Sep 2016, Kyle Owen wrote:
Glad some folks got a kick out of it enough to try it
out! Feel free to
suggest improvements where you see fit. I was thinking about adding support
to read keystrokes from a file for macro programmability...but that might
be too absurd even for this project. Maybe the HP-41C simulator is next...
:)
I tried the new version of your program with option
STRT8=1
First PAL8 complained about "IFNZER". PAL8 wants "IFNZRO"!
After changing this, assambly was successfull, and I could start the
program on the straight-8. All seemed to be ok, with the exception,
that sine (keyboard I) and roll (keyboard L) do not work with your
new version. These keys are just ignored.
Same behaviour on my 8/E, no difference to the straight-8.
Another issue: your BSWSUB has still the bug, that
DCA BSWTMP
SNL
CMA
DCA BSWLNK
CLL / THIS MUST BE!
the CLL is missing. Without this, the bit 0 on entry of this routine
is inverted, when the link is set on entry.
Remember, the link is not a carry!
Next question: my hp35 has 36 keys. The 36th key is not accessible
for normal users. It's hidden below the "ENTER" key. What does this
key do? If I remember right, it's a kind of debug-key. I believe it
shifts the whole register including exponent and signs digit by digit
to the left.
Klemens
--
klemens krause
Stuttgarter KompetenzZentrum fyr Minimal- & Retrocomputing.
http://computermuseum.informatik.uni-stuttgart.de
1:05 p.m.
On Tue, Sep 13, 2016 at 1:55 PM, Klemens Krause
<krause at informatik.uni-stuttgart.de> wrote:
Next question: my hp35 has 36 keys. The 36th key is
not accessible
for normal users. It's hidden below the "ENTER" key. What does this
key do? If I remember right, it's a kind of debug-key. I believe it
shifts the whole register including exponent and signs digit by digit
to the left.
The keyboard dispatch is done using a "goto keys" instruction, which
replaces the low eight bits of the program counter with a hardware
keycode.
On the HP-35, the "right half of ENTER" switch doesn't do anything
useful. The location it jumps to is just an arbitrary instruction in
the code.
On the HP-45, the "right half of ENTER" key was carefully planned to
be useful to enter the undocumented stopwatch mode, by a STO
right-ENTER key sequence. If you put a shim under the right half of
the enter key, then STO ENTER will do it. Otherwise you can get it as
a phantom key by pressing STO, then a simultaneous press of CHS, 7,
and 8. It takes some practice to get that reliably by hand; some
people used the "penny trick" to press the three keys together.
4:30 p.m.
On Tue, Sep 13, 2016 at 2:55 PM, Klemens Krause <
krause at informatik.uni-stuttgart.de> wrote:
First PAL8 complained about "IFNZER". PAL8 wants "IFNZRO"!
After changing this, assambly was successfull, and I could start the
program on the straight-8. All seemed to be ok, with the exception,
that sine (keyboard I) and roll (keyboard L) do not work with your
new version. These keys are just ignored.
Same behaviour on my 8/E, no difference to the straight-8.
Huh, palbart understands both IFNZER and IFNZRO, so I'll fix it to use the
latter.
The keys were updated for the HP-35 and HP-45 to be more intuitive for the
commonly-used functions, 'S' and 'R' are now sine and roll, which were
formerly store and recall. Store and recall are now '<' and '>',
respectively. Look in the code for the key lookup table; I have an ASCII
table of keys arranged in the HP-35/45 keyboard pattern. Both the HP-35
only and HP-35/45 versions use the same keyboard code.
Another issue: your BSWSUB has still the bug, that
DCA BSWTMP
SNL
CMA
DCA BSWLNK
CLL / THIS MUST BE!
the CLL is missing. Without this, the bit 0 on entry of this routine
is inverted, when the link is set on entry.
Remember, the link is not a carry!
Hmm. Sounds like I need some more complicated test vectors! Either that or
the link is never set prior to entry, which I doubt. I'll look at adding
Doug's new code to the mix. Nice work, Doug!
Next question: my hp35 has 36 keys. The 36th key is not accessible
for normal users. It's hidden below the
"ENTER" key. What does this
key do? If I remember right, it's a kind of debug-key. I believe it
shifts the whole register including exponent and signs digit by digit
to the left.
Sounds like Eric has answered your question, but no, this simulator doesn't
support the extra button without changing one line of code; I can add that
as an option if interested for the HP-45 portion, but I'd think the
stopwatch mode serves zero purpose on this simulator, as it would never run
anywhere close to the accurate time anyways.
Kyle
3020
days inactive
3030
days old
27 comments
9 participants
participants (9)
-
cramcram＠gmail.com -
dpi＠dustyoldcomputers.com -
kevenm＠3kranger.com -
krause＠informatik.uni-stuttgart.de -
kylevowen＠gmail.com -
nf6x＠nf6x.net -
pontus＠Update.UU.SE -
spacewar＠gmail.com -
v.slyngstad＠frontier.com