32bit assembler math

Javalin

Hiya

Sure i've seen a link to some working code for 32bit assembler math, multiplication & division.

Can't see it in the links.

Can somebody point me a link?

James

Javalin

Looks like this might do it - from the propeller wiki:

                mov   r,#0
loop
                shr   y,#1   wc
        if_c    add   r,x    wc
                shl   x,#1
        if_nc   tjnz  y,#loop

Assuming the the input is X·* Y

James

Javalin

In fact that works - bar the typo.

                mov   r,#0
loop
                shr   y,#1   wc
        if_c    add   r,x    wc
                shl   x,#1
        if_nc   tjnz  y,#loop

should be:

                mov   x,#10
                mov   y,#10
                mov   r,#0
loop
                shr   y,#1   wc
        if_c    add   r,x    wc
                shl   [b]r[/b],#1
        if_nc   tjnz  y,#loop
 
                ' result left in R

A worked example·looks like:

-> R = 10 * 10
Loop    Y             R (dec)       X (dec)
1       5 ( %101)     10 (%1010)    10 (%1010)
2       2 ( %10)      30 (%11110)   10
3       1 ( %1)       40 (%101000)  10
4       0 ( %0)       100(%1100100) 10
END

Sorted!

James

lonesock

If you make y the lesser of the 2 values, it will need fewer passes to get to the final 1 bit. Here's my code, sorry that the variable names are different. ('lo' is the output...so named in case I want to add a 'hi' for 64 bit results.)

{===== Peasant Multiplication =====}
' Stops on 0, works as long as the
' product fits in 32 bits. (So this
' is good for e.g. 24-bit x 5-bit.)
' All values treated as unsigned.
peasant_multiply
              ' put the smallest op in op1
              mov       lo, op1
              max       op1, op2
              min       op2, lo
              ' clear the accumulator
              mov       lo,#0
              ' do the multiplication
:loop         shr       op1,#1 wc,wz            ' move the least significant bit of op1 into C         
        if_c  add       lo,op2                  ' if the bit was a 1, add op2 to the accumulator
              shl       op2,#1                  ' adjust op2 up by a power of 2 
        if_nz jmp       #:loop                  ' continue if op1 is non-zero
              ' done...return the result in lo
peasant_multiply_ret
              ret

Jonathan

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
lonesock
Piranha are people too.

lonesock

Oops, forgot to add, if you go to [noparse][[/noparse]Help]=>[noparse][[/noparse]Propeller Help] in the Propeller Tool, there is a section called "Math Resources"=>"Math Examples". Also, I believe DeSilva's PASM tutorial has some examples.

edit: and Chip came up with this uber-fast sqrt:

''      'mask' is a register containing $40000000 which gets used directly and rotated all the way back to its original value
        mov     root,#0         'reset root
 :loop  or      root,mask       'set trial bit
        cmpsub  input,root  wc  'subtract root from input if fits
        sumnc   root,mask       'cancel trial bit, set root bit if fit
        shr     root,#1         'shift root down
        ror     mask,#2     wc  'shift mask down (wraps on last iteration)
 if_nc  jmp     #:loop          'loop until mask restored

Jonathan

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
lonesock
Piranha are people too.

Post Edited (lonesock) : 4/15/2010 3:49:40 PM GMT

Javalin

Hiya lonesock,

Cool - thanks for that. Yours works - mine didn't!

Do you have a 32bit divide too?

Thanks,

James

Javalin

ah, found the bit in the spin intepreter (and in another thread here)·:

asmdivide32bit          mov     t1,#0                   ' no idea
                        mov     t2,#32                  ' multiply/divide
                        
:mdiv                   shr     y,#1            wc,wz   ' divide
                        rcr     t1,#1
            if_nz       djnz    t2,#:mdiv
:mdiv2                  cmpsub  x,t1            wc
                        rcl     y,#1
                        shr     t1,#1
                        djnz    t2,#:mdiv2
asmdivide32bit_ret      ret

put values in X and Y, and quotiant result is left in Y

James