PICAXE Basic doesn't support ANY double-word (32-bit) arithmetic but it does offer two separate multiplication operators (* and **) which can be combined to form a double-word result. Subroutines for 32-bit Addition and Subtraction are simple enough (if you can't write those then you probably shouldn't be attempting 32-bit maths on a PICAXE) but a request for a 32-bit by 16-bit division routine made an interesting challenge for several "serious" programmers in this thread. I've previously posted a simplified ("31 / 15 bits") version here, which should meet most needs, but this extended version accepts a full 16-bit Divisor and can deliver a 32-bit Result (and 16-bit Remainder). The general aim has been to minimise the memory used, with a reasonably understandable program structure, so details common to both are not repeated here.
PICAXE Basic splits its Multiplication and Division processes into two separate operations (generating either a Result or a Remainder in the case of Division) but this isn't sensible for a (much slower) subroutine because the calculations are almost identical. However, there is some merit in performing individual calculations for "High" and "Low" word outputs (rather as done with PICAXE multiplication) but in this case they must be executed in the correct sequence (High then Low). Below are four routines (or really just one subroutine with four entry points), two for the High or Low word results individually (div3216H and div3216L), one to always use both in sequence (div3216A) and one which checks the input data (div3216C) and uses the appropriate path. The latter returns an Error Code (zero in pass) for invalid data, but as the only possible error is Division by Zero, the calling program could just as easily directly test the Divisor before execution.
Calculation of the High-Result word is trivial (it's just the High-Numerator / Low-Divisor in Basic) and the High-Remainder is automatically zero (because the Divisor is a single word). So the routine is extended slightly to prepare the Numerator for the Low-words routine, where most of the computation takes place. This latter routine shifts the Numerator and Result (in a shared double-word register) repeatedly leftwards, subtracting the Divisor when possible. A single byte variable counts the shifts and uses its top bit as a "Carry/Borrow flag" extension of the Numerator. On return, the High-Numerator is replaced by the Remainder and the Low-Numerator by the Low-Result words.
The "High" and "Low" routines individually need 7 bytes of RAM, but the Low-Numerator would normally be stored before executing the High routine (although it is not required for this calculation) and the High-Result will generally need to be stored until after the Low routine has been executed. So in practice 9 bytes of RAM are required, or 13 bytes if the original Numerator must be retained. The subroutine uses about 80 bytes of Program Memory and executes in a fairly consistent 150ms with a 4MHz clock.
The subroutines are preceded by a short test harness to display the High- and Low-Result words and Remainder. However, for additional testing and demonstration it also calls the following subroutine (showascii) which employs the division routines twice more (so the total execution time is around 500ms at 4MHz) to convert the original (32-bit) result into a decimal number (up to 10 digits) and then display it as a formatted string of ASCII characters. A particular feature is that the final division forces use of the "Low-words-only" routine (since the Result is now one millionth of the original Numerator) to avoid over-writing the High-Result register (being used to store digits which cannot yet be sent for display). Also the dec3ascii routine does NOT use the PICAXE bintoascii, since the customised code is more efficient in memory usage (Program and RAM). It sends the digits with leading zeros suppressed and a comma ',' as a thousands separator. Note that this conversion routine corrupts (over-writes) all the previously calculated values so they may need to be copied/stored first.
PICAXE Basic splits its Multiplication and Division processes into two separate operations (generating either a Result or a Remainder in the case of Division) but this isn't sensible for a (much slower) subroutine because the calculations are almost identical. However, there is some merit in performing individual calculations for "High" and "Low" word outputs (rather as done with PICAXE multiplication) but in this case they must be executed in the correct sequence (High then Low). Below are four routines (or really just one subroutine with four entry points), two for the High or Low word results individually (div3216H and div3216L), one to always use both in sequence (div3216A) and one which checks the input data (div3216C) and uses the appropriate path. The latter returns an Error Code (zero in pass) for invalid data, but as the only possible error is Division by Zero, the calling program could just as easily directly test the Divisor before execution.
Calculation of the High-Result word is trivial (it's just the High-Numerator / Low-Divisor in Basic) and the High-Remainder is automatically zero (because the Divisor is a single word). So the routine is extended slightly to prepare the Numerator for the Low-words routine, where most of the computation takes place. This latter routine shifts the Numerator and Result (in a shared double-word register) repeatedly leftwards, subtracting the Divisor when possible. A single byte variable counts the shifts and uses its top bit as a "Carry/Borrow flag" extension of the Numerator. On return, the High-Numerator is replaced by the Remainder and the Low-Numerator by the Low-Result words.
The "High" and "Low" routines individually need 7 bytes of RAM, but the Low-Numerator would normally be stored before executing the High routine (although it is not required for this calculation) and the High-Result will generally need to be stored until after the Low routine has been executed. So in practice 9 bytes of RAM are required, or 13 bytes if the original Numerator must be retained. The subroutine uses about 80 bytes of Program Memory and executes in a fairly consistent 150ms with a 4MHz clock.
Code:
#picaxe 20m2 ; And most others
#no_data
; AllyCat 17/06/12
Symbol NUMERATOR = 2109876543 ; TEST DATA
Symbol DIVISOR = 10 ; Maximum 16 bits (i.e. < 65536)
Symbol numerhi = NUMERATOR / $10000 ; NB: Program Editor (v5.5.1) Maths
Symbol numerlo = NUMERATOR & $0FFFF ; is only valid up to 31 bits
Symbol divsor = DIVISOR & $0FFFF ; (i.e. maximum 2,147,483,647)
symbol topbit = $8000 ; "Weight" of carry bit when left-shifting
symbol reshi = w4 ; Result High word
symbol numhi = w3 ; Numerator High, Returns Remainder (Low)
symbol numlo = w2 ; Numerator Low, Retuns Result Low
symbol divis = w1 ; Divisor, Unchanged on return
symbol pass = b0 ; Loop counter, Carry flag and Error report
divtest:
divis = divsor ; Divisor (maximum 16 bits)
numhi = numerhi ; Numerator High -> Remainder on exit
numlo = numerlo ; Numerator Low -> Result Low
gosub div3216C ; Check data and divide 32 by 16 bits
if pass = 0 then
sertxd ("Cannot divide by zero",cr,lf)
do : loop
endif
sertxd ("Result Hi=",#reshi," Lo=",#numlo)
sertxd (cr,lf,"Remainder=",#numhi) ; Value NOT stored beyond here
sertxd (cr,lf,"Dec.Result=")
numhi = reshi ; Move Result back to Numerator
gosub showascii ; Now Show the Result in ASCII
do : loop ; (~145 bytes to here)
div3216H: ; Calculate High Result word for 32 by 16 bits division
; numhi = Numerator High. Returns with Numerator High ready for "Low Word" calculation
; divis = Divisor; reshi = High Result (High Remainder will be zero)
; numlo = Numerator Low (optional, not required for "High" calculation)
reshi = numhi / divis * divis ; Divisible part of High byte
numhi = numhi - reshi ; Subtract to give numerator for Low calc
reshi = reshi / divis ; High Result (Low not done yet)
done:
return
div3216C: ; Check input data and select entry point for 32 by 16 bits division
; Return immediately with pass = 0 if divisor is zero
pass = 0 ; Pre-set the error code
if divis = 0 then done ; Can't divide by zero
reshi = 0 ; Pre-load zero High Result
if divis > numhi then div3216L ; Jump to use 16-bit Result routine...
; else fall through for full calculation
div3216A: ; Force a full 32 by 16 bit division (regardless of input data)
gosub div3216H ; Calculate High word Result, then ..
; fall through to calculate Low words
div3216L: ; Calculate Low Result and Low Remainder of 32 by 16 bits division
; numhi = Numerator High word -> Remainder on return
; numlo = Numerator Low word -> Result (Low word) on return
; divis = Divisor, unchanged on return. pass = 0 returned if input data error
for pass = 0 to 15 ; Repeat for each bit position
if numhi >= topbit then ; Carry from top bit
pass = pass OR 128 ; Carry flag
endif
numhi = numhi + numhi ; Start shifting numerator left
if numlo >= topbit then ; Carry bit
numhi = numhi + 1 ; Add in the carry
endif
numlo = numlo + numlo ; End of Shifting
if numhi < divis AND pass < 128 then nosub ; Jump if can't subtract
numhi = numhi - divis ; Subtract divisor from (current) numerator
numlo = numlo + 1 ; Add digit into result
pass = pass AND 127 ; Cancel the "subtract" flag
nosub:
next pass
return ; (~80 bytes)
Code:
symbol diblo = b2 ; Low byte of divisor (used by ascii routine)
symbol dibhi = b3 ; High byte of divisor (used by ascii routine)
showascii: ; Send Long Word (32 bits in numhi:numlo) as ASCII decimals
; NB: Registers divis, numlo, numhi, reshi and pass are ALL corrupted on return.
divis = 1000 ; Start by removing the final three digits
gosub div3216C ; Returns Result in reshi, Remainder in numhi
divis = numhi ; Shuffle the words for the next division
numhi = reshi ; Result into Numerator ready for next division
reshi = divis ; Remainder (Final 3 digits to send) now in reshi
divis = 1000 ; Prepare to remove another three digits
gosub div3216L ; Middle 3 digits in numhi, First 4 digits in numlo
pass = numlo / 1000 ; First digit to be sent
numlo = numlo // 1000 ; Next 3 digits
if pass > 0 then
sertxd(#pass) ; Show billions digit
endif
gosub dec3ascii ; Show millions digits
numlo = numhi
gosub dec3ascii ; Show thousands digits
numlo = reshi
gosub dec3ascii ; Show units digits
if pass = 0 then ; Result was zero so send that
sertxd("0")
endif
return ; (~75 bytes)
dec3ascii: ; Send numlo as 3 digits with ',' and leading zeros (when appropriate)
if pass > 0 then ; Flag indicates some digits have been sent
dibhi = numlo /100 // 10 ; Hundreds digit ) Uses less
diblo = numlo / 10 // 10 ; Tens digit )- memory than
numlo = numlo // 10 ; Units digit ) BINTOASCII
sertxd(",",#dibhi,#diblo,#numlo) ; Send comma and leading zeros
else if numlo > 0 then
sertxd (#numlo) ; Send without leading zeros
pass = 1 ; Set the send-zeros flag
endif
return ; (~45 bytes)
Last edited:
