Is SHIFT RIGHT (SFTR) implemented properly?

Rolf Strand

unread,

May 22, 2021, 12:25:19 AM5/22/21

to uKenbak-1

I am a newbie here, so maybe this was discussed before.

From the manual:
SHIFT RIGHT 1

Original c7 c6 c5 c4 c3 c2 c1 c0
After c7 c7 c6 c5 c4 c3 c2 c1

Shift right sign extends according to the manual

If I try this on nanoKENBAK-1:

023 377 LOAD A 377

011 SFTR A 1

034 200 STORE A OUTPUT

000 HALT

the output is 177 in stead of 377

I get the same with http://www.neocomputer.org/kenbak/kenbak1-JS.html

Are there any other emulators I can try?

I have been working with this code and reporting issues:

https://github.com/kidmirage/KENBAK-2-5-Build-Files

I have this code that does 16bit math.

Running with some local patches to the IDE it seems to work.

If SFRT is on nanoKENBAK-1 does not sign extend shr1b will need to be modified

to emulate sign extension

; Simple thermostat with Anticipator cycler

; The control loop uses scaled 16 bit math

; The scale point is between the msb and lsb

;

; Psuedo code for the control

; antHeat is 4

; relayState 0 or 1

; The anticipator cycler is a first order lag function.

; It converts the switch differential in to a proportional

; controller based on the proportional Perr.

; With a switch differential of 1F, and antHeat of 4F

; the propotional band is 4F - 1F = 3F

;

; while (1)

; {

; sense = getTemperature();

; Perr = setpt - sense

; ant += (antHeat * relayState - ant) >> 32

; Cerr = Perr - ant

; if (Cerr > 0)

; relayState = 1;

; else if (Cerr <= -1)

; relayState = 0;

; delay(cycleTime)

; }

;

; Hardware:

;

; The Temperature is provided in msb, lsb form from an external circuit

; to the INPUT of the KENBAK-1. It is enabled by an external switch.

; b0 of the OUTPUT register selects: 1 = msb, 0 = lsb

;

; The control relay is connected to b2 of the OUTPUT

; antHeat is orged OUTPUT

; Anticipator heat is 4 degrees.

; This provides a short cut to antHeat * relayState:

; OUTPUT or antHeat*relaState is 4 when relay is on, 0 when off

;

; Application data

org 115

cycDly 1 ; 1 for debug

cycDlyh 0 ; 0 about 8ms per count

setpt 0

setpth 75

sense db

senseh db

Perr db

Perrh db

ant db

anth db

Cerr db

Cerrh db

loopCnt db ; Adjust org so this address is 127

org 128

antHeat db ; antHeat is b2 of OUTPUT

; Application program start

reset org 4 ; begining of program ram

control

; get the current temperature reading from INPUT

set 0,1,OUTPUT ; the input will now be the msb of the sensor

nop ; time to settle

;load B,INPUT

load B,73 ; debug at 73.5

set 0,0,OUTPUT ; the input will now be the lsb of the sensor

nop ; time to settle

;load A,INPUT

load A,128 ; debug at 73.5

; can optimize later to use neg16b and just add setpt

load X,#sense ; point to sense

jmk store16b ; Save sensor reading for debug

; calc Perr = setpt - sense

load X,#setpt ; point to spl

jmk load16b

load X,#sense ; point to sensl

jmk sub16b ; Perr = setpt - sense

load X,#Perr ; point to Perr

jmk store16b ; save Perr

; update the anticipator cycler

; anticipator ant = ant + (HVAC - ant) / 32

load A,0 ; antHeat low is always 0

load B,antHeat ; BA = antHeat, note output all other OUTPUT pins assumed to be 0

load X,#ant ; point to ant

jmk sub16b ; (antHeat - ant)

load X,5 ; shift 5 to divide by 32

jmk shr16b ; (antHeat - ant) >> 5

load X,#ant ; point to ant

jmk add16b ; ant += (antHeat - ant) >> 5

jmk store16b ; X still points to ant

; calculate the composite err

; cmperr = Perr - ant

jmk neg16b ; negate anticipator value in BA

load X,#Perr ; point to Perr

jmk add16b ; Cerr = Perr - ant

load X,#Cerr ; point to Cerr

jmk store16b ; store Cerr

; Update the relay state based on composite err

; if Cerr < 0 HVAC = 4 else if Cerr >= 1 HVAC = 0

jmp B,LT,chkOff ; Cerr > 0

rlyon set 2,1,OUTPUT ; Cerr > 0 so bit 2 relay output on

jmp cycleDelay

chkoff

load X,#S_ONE ; point to const scaled one

jmk add16b ; Cerr + 1, if still positive then cerr >-1

jmp B,LT,rlyOff ; Cerr <= -1

jmp cycleDelay

rlyOff set 2,0,OUTPUT ; Cerr <= -1 so bit 2 relay output off

; anticiaptor time constant simulated here

cycleDelay

load A,cycDly ;delay constant for anticipator cycler

load B,cycDlyh ;delay constant for anticipator cycler

load X,#ONE ;point to constant onel = 0x0001

delay jmk sub16b

jmp A,NE,delay

jmp B,NE,delay

load A,loopCnt

add A,1

store A,loopCnt

jmp control

; -------------------- End Appplication ----------------------------------

; -------------------- 16 bit math routines ------------------------------

;

; The math routines are in upper half of RAM

org 0

lsb db ; used for 16bit math indexing

msb db ; used for 16bit math indexing

org 133 ;Skip over the registers

; 16 bit load, store, saveTmp

; on entry X points to value to load

; return msb = B, lsb = A

load16b db

load A,lsb+X

load B,msb+X

jmp (load16b)

store16b db

store A,lsb+X

store B,msb+X

jmp (store16b)

saveTmp db

store A,TMP16

store B,TMP16h

jmp (saveTmp)

; 16 bit math routines

; on entry

; B is msb A is lsb if first operand

; X is the address of the second operand

;

; result msb = B lsb = A

; X is unchanged

; add16b

; sub16b

; neg16b

;

add16b db ; return address

add A,lsb+X

skp 1,0,OCA ; skip if no carry

add B,1 ; adjust for carry

add B,msb+X

jmp (add16b) ; return

sub16b db ; return address

sub A,lsb+X

skp 1,0,OCA ; skip if no carry

sub B,1 ; adjust for carry

sub B,msb+X

jmp (sub16b) ; return

neg16b db

store X,TMPX ; save X

load X,#TMPN16 ; point to tmp

jmk store16b

load A,0 ; load 0 to BA

load B,0

jmk sub16b ; subtract input from 0

load X,TMPX ; restore X

jmp (neg16b) ; return

; 16 bit div 2^n

; call with B = msb, A = lsb, X = places

; return B = msb, A = lsb, X = 0

shr16b db

sloop sft A,R,1 ; shift right lsb

set 7,0,A ; clear lsb b7

skp 0,0,B ; test b0 msb, skip it 0

set 7,1,A ; b0 lsb was 1 so set b7 lsb

sft B,R,1 ; shift right msb, if negative sign extend

sub X,1 ; dec place count

jmp X,NE,sloop ; if not done again

jmp (shr16b) ; return to caller

;Math system variables and constants

org 246

ZERO 0 ; math constant zero 0x0000

S_ONE 0 ; math constant scaled one low 0x0100

ONE 1 ;

ONEH 0 ; math constant one 0x0001

TMP16 db ; temporary 16b value

TMP16h db ;

TMPN16 db ; temp for neg16b

TMPN16h db ;

TMPX db ; place to save X

RAM_END db ; If adding data items make sure this address is < 256

Raw octal dump:

The IDE assembler is consistent with itself

Not sure if it matches nanoKENBAK-1

000 000 000 004 102 200 200 123 111 002 200 200 023 200 223 167

367 214 223 165 367 205 223 167 367 245 223 171 367 214 023 000

124 200 223 173 367 245 223 005 367 301 223 173 367 232 367 214

367 260 223 171 367 232 223 175 367 214 145 100 122 200 347 112

223 367 367 232 145 110 347 112 022 200 024 163 124 164 223 370

367 245 043 120 143 120 024 177 003 001 034 177 347 004 000 000

000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000

000 000 000 001 000 000 113 000 000 000 000 000 000 000 000 000

000 000 000 000 000 000 026 000 126 001 357 205 000 036 000 136

001 357 214 000 034 372 134 373 357 223 000 006 000 212 201 103

001 106 001 357 232 000 016 000 212 201 113 001 116 001 357 245

000 234 376 223 374 367 214 023 000 123 000 367 245 224 376 357

260 000 011 072 000 202 001 172 000 051 213 001 243 302 357 301

000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000

000 000 000 000 000 000 000 000 001 000 000 000 000 000 000 000

Rolf Strand

unread,

May 22, 2021, 1:55:59 AM5/22/21

to uKenbak-1

The emulator here sign extends on SFTR:

http://www.historicsimulations.com/kenbak1.html

Rolf Strand

unread,

May 22, 2021, 3:42:31 AM5/22/21

to uKenbak-1

Above control converted to another assembler: (with some typos fixed)

# Simple thermostat with Anticipator cycler

# The control loop uses scaled 16 bit math

# The scale point is between the msb and lsb

#

# Psuedo code for the control

# antHeat is 4

# relayState 0 or 1

# The anticipator cycler is a first order lag function.

# It converts the switch differential in to a proportional

# controler based on the proportional Perr.

# With a switch differential of 1F, and antHeat of 4F

# the proportional band is 4F - 1F = 3F

#

# while (1)

# {

# sense = getTemperature()#

# Perr = setpt - sense

# ant += (antHeat * relayState - ant) >> 32

# Cerr = Perr - ant

# if (Cerr > 0)

# relayState = 1#

# else if (Cerr <= -1)

# relayState = 0#

# delay(cycleTime)

# }

#

# Hardware:

#

# The Temperature is provided in msb, lsb form from an external circuit

# to the INPUT of the KENBAK-1. It is enabled by an external switch.

# b0 of the OUTPUT register selects: 1 = msb, 0 = lsb

#

# The control relay is connected to b2 of the OUTPUT

# antHeat is orged OUTPUT

# Anticipator heat is 4 degrees.

# This provides a short cut to antHeat * relayState:

# OUTPUT or antHeat*relaState is 4 when relay is on, 0 when off

#

# Application data

org 115

cycDly db 1 # 1 for debug

cycDlyh db 0 # 0 about 8ms per count

setpt db 0

setpth db 75

sense db 0

senseh db 0

Perr db 0

Perrh db 0

ant db 0

anth db 0

Cerr db 0

Cerrh db 0

loopCnt db 0 # Adjust org so this address is 127

A equ 0

B equ 1

X equ 2

PC equ 3

OUTPUT equ 128

OCA equ 129

OCB equ 130

OCX equ 131

INPUT equ 255

org 128

antHeat db 0 # antHeat is b2 of OUTPUT

# Application program start

reset org 4 # beginning of program ram

control

# get the current temperature reading from INPUT

set1 OUTPUT,0 # the input will now be the msb of the sensor

nop # time to settle

# ldb INPUT

ldb #73 # debug at 73.5

set0 OUTPUT,0 # the input will now be the lsb of the sensor

nop # time to settle

# lda INPUT

lda #128 # debug at 73.5

# can optimize later to use neg16b and just add setpt

ldx #sense # point to sense

call store16b # Save sensor reading for debug

# calc Perr = setpt - sense

ldx #setpt # point to spl

call load16b

ldx #sense # point to sensl

call sub16b # Perr = setpt - sense

ldx #Perr # point to Perr

call store16b # save Perr

# update the anticipator cycler

# anticipator ant = ant + (antHeat - ant) / 32

lda #0 # antHeat low is always 0

ldb antHeat # BA = antHeat, note output all other OUTPUT pins assumed to be 0

ldx #ant # point to ant

call sub16b # (antHeat - ant)

ldx #5 # shift 5 to divide by 32

call shr16b # (antHeat - ant) >> 5

ldx #ant # point to ant

call add16b # ant += (antHeat - ant) >> 5

call store16b # X still points to ant

# calculate the composite err

# cmperr = Perr - ant

call neg16b # negate anticipator value in BA

ldx #Perr # point to Perr

call add16b # Cerr = Perr - ant

ldx #Cerr # point to Cerr

call store16b # store Cerr

# Update the relay state based on composite err

# if Cerr < 0 HVAC = 4 else if Cerr >= 1 HVAC = 0

jblt chkOff # Cerr > 0

rlyon set1 OUTPUT,2 # Cerr > 0 so bit 2 relay output on

jmp cycleDelay

chkoff

ldx #S_ONE # point to const scaled one

call add16b # Cerr + 1, if still positive then cerr >-1

jblt rlyOff # Cerr <= -1

jmp cycleDelay

rlyOff set0 OUTPUT,2 # Cerr <= -1 so bit 2 relay output off

# anticiaptor time constant simulated here

cycleDelay

lda cycDly #delay constant for anticipator cycler

ldb cycDlyh #delay constant for anticipator cycler

ldx #ONE #point to constant onel = 0x0001

delay call sub16b

jane delay

jbne delay

lda loopCnt

adda #1

sta loopCnt

jmp control

# -------------------- End Appplication ----------------------------------

# -------------------- 16 bit math routines ------------------------------

#

# The math routines are in upper half of RAM

org 0

lsb db 0 # used for 16bit math indexing

msb db 0 # used for 16bit math indexing

org 133 #Skip over the registers

# 16 bit load, store, saveTmp

# on entry X points to value to load

# return msb = B, lsb = A

load16b db 0

lda lsb,x

ldb msb,x

jmp (load16b)

store16b db 0

sta lsb,X

stb msb,X

jmp (store16b)

saveTmp db 0

sta TMP16

stb TMP16h

jmp (saveTmp)

# 16 bit math routines

# on entry

# B is msb A is lsb if first operand

# X is the address of the second operand

#

# result msb = B lsb = A

# X is unchanged

# add16b

# sub16b

# neg16b

#

add16b db 0 # return address

adda lsb,X

skp0 OCA,1 # skip if no carry

addb #1 # adjust for carry

addb msb,X

jmp (add16b) # return

sub16b db 0 # return address

suba lsb,X

skp0 OCA,1 # skip if no carry

subb #1 # adjust for carry

subb msb,X

jmp (sub16b) # return

neg16b db 0

stx TMPX # save X

ldx #TMPN16 # point to tmp

call store16b

lda #0 # load 0 to BA

ldb #0

call sub16b # subtract input from 0

ldx TMPX # restore X

jmp (neg16b) # return

# 16 bit div 2^n

# call with B = msb, A = lsb, X = places

# return B = msb, A = lsb, X = 0

shr16b db 0

sloop shra 1 # shift right lsb

set0 A,7 # clear lsb b7

skp0 B,0 # test b0 msb, skip it 0

set1 A,7 # b0 lsb was 1 so set b7 lsb

shrb 1 # shift right msb, if negative sign extend

subx #1 # dec place count

jxne sloop # if not done again

jmp (shr16b) # return to caller

#Math system variables and constants

org 246

ZERO db 0 # math constant zero 0x0000

S_ONE db 0 # math constant scaled one low 0x0100

ONE db 1 #

ONEH db 0 # math constant one 0x0001

TMP16 db 0 # temporary 16b value

TMP16h db 0 #

TMPN16 db 0 # temp for neg16b

TMPN16h db 0 #

TMPX db 0 # place to save X

RAM_END db 0 # If adding data items make sure this address is < 256

Mark Wilson

unread,

May 22, 2021, 4:14:32 AM5/22/21

to uKenbak-1

It does look like I missed that subtlety, I will investigate fixing it.

Potentially a breaking change, so I will probably add a flag for backward compatibility...

fcpr...@gmail.com

unread,

May 22, 2021, 10:15:13 AM5/22/21

to uKenbak-1

I looked over my various contributed programs and this will break only one - SerialBlinkenLightsV2.1 (see µKenbak-1: Mindless serial port program). I will fix this to be independent of the shift-right instruction if and when Mark updates the firmware to sign-extend. I'll update that topic with the new version at that time.

The curious part is that I actually wrote this line of code, with exactly this comment:

SHRB 1 # Make random number positive

I clearly made the same assumption without consulting the documentation; why I would have done that is beyond me because I am well aware that right shifts can be either zero-filled or sign-filled.

Nice catch to the OP!

Rolf Strand

unread,

May 22, 2021, 11:08:40 AM5/22/21

to uKenbak-1

I only found it while implementing 16 bit shift right. Needed the detail of the shift operator to implement it, so I had to read the manual.

MICHAEL GARDI

unread,

May 22, 2021, 12:27:28 PM5/22/21

to uKenbak-1

Rolf is being modest. He is very thorough and has been a big help improving my KENBAK-1 IDE ( https://github.com/kidmirage/KENBAK-2-5-Build-Files) with bug reports and useful feature suggestions. Thanks Rolf.

Mark Wilson

unread,

May 22, 2021, 11:19:51 PM5/22/21

to uKenbak-1

Thanks for finding that Rolf.

I have changed the right-shift instruction to preserve the high ("sign") bit.

In working on that, I found there was also a problem with the multi-bit roll left/right instructions so I fixed them too.

For example 004 rolled right 4x (0101) would give 0 rather than 0100

There is no compatibility flag but there is a #define in CPU.cpp (CPU_LEGACY_SHIFT_ROLL) which if defined, restores the "legacy" behaviour.

None of the built-in programs were broken by the change.

I have updated github.

fcpr...@gmail.com

unread,

May 23, 2021, 4:54:20 PM5/23/21

to uKenbak-1

Thanks for the fix(es) Mark.

fcpr...@gmail.com

unread,

May 24, 2021, 10:57:32 AM5/24/21

to uKenbak-1

On Saturday, May 22, 2021 at 11:19:51 PM UTC-4 funnypo...@gmail.com wrote:

None of the built-in programs were broken by the change.

Whoops, it seems that's not the case. The built-in program programCylon (STOP+BTN1) now only toggles bit 0. I didn't disassemble all of it, but I see a 011 in there (SHRA 1).

Mark Wilson

unread,

May 24, 2021, 3:02:25 PM5/24/21

to uKenbak-1

Ah. I seached the assember versions but I think that's just raw bytes and I missed it. I will fix it shortly

Mark Wilson

unread,

May 24, 2021, 3:26:13 PM5/24/21

to uKenbak-1

But really, it was just careless of me. I remember focusing on little test programs to check shift/roll, searching the asm source, running the clocks and Eratosthenes. But I missed the most obvious program?

If I did that in my day job I'd be in trouble.

I apologize. I should be able to fix it tonight, NZ time.

Mark Wilson

unread,

May 25, 2021, 3:24:16 AM5/25/21

to uKenbak-1

OK, Cylon is fixed on github.

Thanks for finding this.

Mark

fcpr...@gmail.com

unread,

May 25, 2021, 8:57:45 AM5/25/21

to uKenbak-1

Great Mark! Re-flashed my µKenbak-1 and nanoKenbak-1 and all is well now.

Vicente González

unread,

May 26, 2021, 9:50:45 AM5/26/21

to fcpr...@gmail.com, uKenbak-1

Thanks Rolf for your valuable work checking and validating the kenbakino firmware,

and more important, thanks Mark for keeping your cool kenbakino up to date.

Vicente

--
You received this message because you are subscribed to the Google Groups "uKenbak-1" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ukenbak-1+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ukenbak-1/f83578bb-7dc2-413e-ba5c-e8031f99489dn%40googlegroups.com.

Reply all

Reply to author

Forward