Software Vault: The Gold Collection/Software Vault - The Gold Collection (American Databankers) (1993).ISO/cdr47/pctuto.zip/DISK1-B.EXE/lha/MISHMASH.DOC

Text File | 1990-08-06 | 31KB | 842 lines

The PC Assembler Tutor - Copyright (C) 1990  Chuck Nelson
______________________


                             MISHMASH


This document contains several assembler programs. It has no page breaks and
no footnotes so you can cut the programs directly out of the text with a word
processor.


                             BLOCK MOVE

The first subroutine does a block move from one place in memory to another. It
is designed so the sorce block and the target block can be overlapping. It
first calculates the total address of the sorce block and the target block. If
the sorce block is below the target block the move starts at the top of the
source block and moves down. If the source block is above the target block the
move starts at the bottom of the source block and moves up. This makes sure
that overlapping data will not be clobbered.

This calculates the full 20 bit address. It was designed for BASIC; BASIC
sometimes requires the full 20 bit address. For many languages, all you need
to do is look at the offset addresses since segments cannot overlap. This is
NOT true of something called the "HUGE" mode, where you need to calculate the
full 20 bit address.


+++++++++++++++++++++ << START OF PROGRAM >> ++++++++++++++++++++++

include /pushregs.mac
_TEXT SEGMENT PUBLIC 'CODE'
        ASSUME cs:_TEXT
        PUBLIC BlockMove
; - - - - - - - - - -
; BlockMove ( from.seg, from.off, to.seg, to.off, byte.count)
; for BASIC
; MOVSW is from DS:[SI] to ES:[DI]

        FROM_SEG_ADDRESS        EQU     [bp+14]
        FROM_OFFSET_ADDRESS     EQU     [bp+12]
        TO_SEG_ADDRESS          EQU     [bp+10]
        TO_OFFSET_ADDRESS       EQU     [bp+8]
        BYTE_COUNT_ADDRESS      EQU     [bp+6]
; - - - - - - - - - -
BlockMove proc far
        push    bp
        mov     bp, sp
        PUSHREGS  ax, bx, cx, dx, si, di, es, ds

        ; AX:BX is the total FROM address
        ; DX:DI is the total TO address
        ; (FROM address > TO address) ->  upwards
        ; (FROM address < TO address) ->  downwards

        ; calculate 20 bit total address
        sub     ax, ax          ; zero AX
        mov     si, FROM_SEG_ADDRESS
        mov     bx, [si]        ; from_seg to BX
        sub     dx, dx          ; zero DX
        mov     si, TO_SEG_ADDRESS
        mov     di, [si]        ; to_seg to DI

        mov     cx, 4           ; shift 4 bytes
shift_loop:
        shl     bx, 1
        rcl     ax, 1           ; carry from BX -> AX
        shl     di, 1
        rcl     dx, 1           ; carry from DI -> DX
        loop    shift_loop

        ; AX:BX and DX:DI now contain the total address of the
        ; segment start. Now add the offsets.
        mov     si, FROM_OFFSET_ADDRESS
        add     bx, [si]
        adc     ax, 0
        mov     si, TO_OFFSET_ADDRESS
        add     di, [si]
        adc     dx, 0

        ; AX:BX and DX:DI are now the total addresses of the first
        ; byte to be moved. First compare AX and DX and go to the
        ; appropriate routine depending on which address is higher.
        ; If AX and DX are the same, then compare BX and DI and go
        ; to the appropriate routine. If BX = DI, the block is being
        ; moved onto itself, so just exit (there is no work to be done).

        cmp     ax, dx
        ja      bottom_to_top   ; FROM is higher
        jb      top_to_bottom   ; TO is higher
        cmp     bx, di
        ja      bottom_to_top   ; FROM is higher
        jb      top_to_bottom   ; TO is higher
        jmp     exit

bottom_to_top:
        mov     si, TO_SEG_ADDRESS
        mov     es, [si]        ; to_seg to ES
        mov     si, TO_OFFSET_ADDRESS
        mov     di, [si]        ; to_offset to DI
        mov     si, BYTE_COUNT_ADDRESS
        mov     cx, [si]        ; byte count to CX
        mov     si, FROM_SEG_ADDRESS
        mov     ax, [si]        ; temporary storage for new DS
        mov     si, FROM_OFFSET_ADDRESS
        mov     si, [si]        ; from_offset to SI
        mov     ds, ax          ; now move from_seg to DS
        sub     bx, bx          ; clear BX
        shr     cx, 1           ; divide by 2, remainder in CF
        rcl     bx, 1           ; move CF to low bit of BX
        cld                     ; clear DF (go up)
        rep     movsw           ; the block move (count in CX)
        and     bx, bx          ; one extra byte?
        jz      exit
        movsb                   ; move one last byte
        jmp     exit

top_to_bottom:
        mov     si, TO_SEG_ADDRESS
        mov     es, [si]        ; to_seg to ES
        mov     si, TO_OFFSET_ADDRESS
        mov     di, [si]        ; to_offset to DI
        mov     si, BYTE_COUNT_ADDRESS
        mov     cx, [si]        ; byte count to CX
        mov     si, FROM_SEG_ADDRESS
        mov     ax, [si]        ; temporary storage for new DS
        mov     si, FROM_OFFSET_ADDRESS
        mov     si, [si]        ; from_offset to SI
        mov     ds, ax          ; now move from_seg to DS
        add     si, cx          ; move to top of block
        sub     si, 2           ; we were 1 word too far
        add     di, cx          ; move to top of block
        sub     di, 2           ; we were 1 word too far
        sub     bx, bx          ; clear BX
        shr     cx, 1           ; divide by 2, remainder in CF
        rcl     bx, 1           ; move CF to low bit of BX
        std                     ; set DF (go down)
        rep     movsw           ; the block move (count in CX)
        and     bx, bx          ; one extra byte?
        jz      exit
        inc     si              ; top byte of word
        inc     di              ; top byte of word
        movsb                   ; move one last byte

exit:
        POPREGS  ax, bx, cx, dx, si, di, es, ds
        mov      sp, bp
        pop     bp
        ret     (10)

BlockMove endp
; - - - - - - - - - -
_TEXT ENDS
END

++++++++++++++++++++++ << END OF PROGRAM >> +++++++++++++++++++++++




                    MULTIPLICATION AND DIVISION


The following are routines for multiple word multiplication and division. They
are the core routines. There must be an intermediate routine which prepares
the information correctly for the core routine and then calls the core
routine. Among other things, these intermediate routines must:

     1) deal with signed numbers. They must convert any negative numbers into
     positive numbers and keep track of the signs. Then they must alter the
     signs of the results if necessary.

     2) make copies of numbers for the core routine when the core routine will
     destroy or alter the number during the calculation.

     3) make decisions about valid results for the multiplication routines. If
     we multiply two numbers of length N words, then the result can be N + N
     words long. What do you want to do if the result is over N words long? It
     is your decision.

     4) transfer the results back to the programs if necessary.

These are the things we did in chapter 16, and they are necessary here as
well. In all the following routines you need to pay attention to the lengths.
Some lengths are in BYTES and some lengths are in WORDS. Make sure you know
which is which.




                         BLOCK MULTIPLICATION

The first multiplication program uses block multiplication. This is simply the
multiple word multiplication that you did in chapters 13 and 16. This time,
instead of multiplying  n X 1  words, we will be multiplying  n X n  words.
The most important thing that this routine does is minimize its work. If n =
100 words, then it is possible for the routine to do 10,000 multiplications.
This takes a lot of time. If we have two 100 word numbers but the first one is
127,911 and the other one is 4,926,948,187,062 the first number has
significant information in two words and the second number has significant
information in three words. We only need to multiply  3 X 2 = 6 words instead
of 10,000 words. As you can see, this will cut the time by a factor of over
1000. This routine requires that the result be distinct from either the
multiplicand or multiplier and be  n + n  (2n) words long.

First we clear the area for the result. The next section finds the highest
non-zero word of both the multiplicand and multiplier. If either is 0 the
result is 0, so we exit (the result is cleared and is 0). After that comes the
multiplication proper. We multiply the complete multiplicand by one multiplier
word, then cycle to the next multiplier word and so on. We add each DX:AX pair
to the temporary result and propagate any carry that results from the
addition. The result cannot be larger than N + N words, so we will never
propagate past the result area. This is as fast as you can multiply numbers on
the 8086.


+++++++++++++++++++++ << START OF PROGRAM >> +++++++++++++++++++++++

; block multiplication using standard 8086 multiplication
; block_multiply ( length , multiplicand, multiplier, temp_result )

; length is the number of WORDS
; length is a number, but the others are addresses. The temp_result
; space must be (2 X length), and must be distinct from the other
; varibles since it will be overwritten by the routine. This is
; a far routine for C, and after setting up BP, we have:
;
;      TEMP_RESULT_ADDRESS     EQU   [bp + 12]
;      MULTIPLIER_ADDRESS      EQU   [bp + 10]
;      MULTIPLICAND_ADDRESS    EQU   [bp + 8]
;      DATA_LENGTH             EQU   [bp + 6]


INCLUDE \pushregs.mac
; - - - - - - - - - - - - - - - - - - - -
DATASTUFF SEGMENT PUBLIC 'DATA'
multiplicand_top_address      dw   ?
multiplier_top_address        dw   ?
temp_bottom_address           dw   ?
current_multiplier_address    dw   ?
DATASTUFF  ENDS
; - - - - - - - - - - - - - - - - - - - -
CODESTUFF SEGMENT PUBLIC 'CODE'

     PUBLIC  block_multiply
     ASSUME CS:CODESTUFF, DS:DATASTUFF

     TEMP_RESULT_ADDRESS     EQU   [bp + 12]
     MULTIPLIER_ADDRESS      EQU   [bp + 10]
     MULTIPLICAND_ADDRESS    EQU   [bp + 8]
     DATA_LENGTH             EQU   [bp + 6]

; - - - - - - - - - -
block_multiply proc far

     push  bp
     mov   bp, sp
     pushf               ; save DF value
     PUSHREGS ax, bx, cx, dx, si, di, es
     push  ds            ; es = ds
     pop   es

     ; clear temp_result
     mov   di, TEMP_RESULT_ADDRESS
     mov   cx, DATA_LENGTH
     shl   cx, 1        ; 2 X LENGTH is buffer length
     mov   ax, 0        ; zero for clearing
     cld                ; upwards
     rep   stosw        ; store ax

     ; find the highest multiplicand word which is non-zero
     mov   di, MULTIPLICAND_ADDRESS
     mov   dx, DATA_LENGTH
     mov   cx, dx       ; cx = length in words
     mov   bx, dx
     dec   bx           ; first word is at offset 0
     shl   bx, 1        ; bx = top word
     add   di, bx       ; di = address of top word
     std                ; downwards
                        ; ax is still 0
     repe  scasw        ; continue as long as es:[di] is 0
     jne   first_top_found    ; found non-zero word
     jmp   exit_mult    ; multiplicand is 0 so result is 0

first_top_found:
     add   di, 2                         ; we went 2 too far
     mov   multiplicand_top_address, di  ; address of top non-zero word

     ; no registers have been modified except di and cx
     ; use the same ax, bx and dx values as before for multiplier.

     ; find the highest non-zero multiplier word
     mov   di, MULTIPLIER_ADDRESS
     add   di, bx                 ; di = address of top word
     mov   cx, dx                 ; cx = length in words
                                  ; ax is still 0
     repe  scasw                  ; continue as long as es:[di] is 0
     jne   second_top_found       ; found non-zero word
     jmp   exit_mult              ; multiplier is 0 so result is 0

second_top_found:
     add   di, 2                         ; we went 2 too far
     mov   multiplier_top_address, di    ; address of top non-zero word

     ; the multiplication  ********************
     mov   ax, TEMP_RESULT_ADDRESS
     mov   temp_bottom_address, ax       ; start at bottom
     mov   si, MULTIPLIER_ADDRESS
     mov   current_multiplier_address, si  ; save address

outer_multiplication_loop:
     ; set up the registers
     mov   cx, [si]                ; move current multiplier to cx
     mov   di, MULTIPLICAND_ADDRESS
     mov   bx, temp_bottom_address

inner_multiplication_loop:
     mov   ax, cx                  ; multiplier word to ax
     mul   WORD PTR [di]           ; multiplicand - result in DX:AX
     add   [bx], ax                ; low word of multiplication
     adc   [bx+2], dx              ; high word of multiplication
     jnc   no_more_carry           ; extra work if CF=1
     mov   si, 4
     ; keep propagating the carry till CF = 0
propagate_carry:
     add   WORD PTR [bx+si], 1
     jnc   no_more_carry
     add   si, 2                   ; next word
     jmp   propagate_carry

no_more_carry:
     add   bx, 2         ; next word of temp result
     add   di, 2         ; next word of multiplicand
     cmp   di, multiplicand_top_address  ; finished?
     ja    next_multiplier_word
     jmp   inner_multiplication_loop

next_multiplier_word:
     mov   si, current_multiplier_address
     add   si, 2
     cmp   si, multiplier_top_address
     ja    exit_mult                ; end of multiplication

     mov   current_multiplier_address, si  ; save address
     add   temp_bottom_address, 2   ; increment for next start
     jmp   outer_multiplication_loop

     ; end of the multiplication  *************

exit_mult:
     POPREGS ax, bx, cx, dx, si, di, es
     popf              ; restore DF value
     mov   sp, bp
     pop   bp
     ret               ; a C return, so don't pop arguments.

block_multiply  endp
; - - - - - - - - - -
CODESTUFF ENDS
END

++++++++++++++++++++++ << END OF PROGRAM >> +++++++++++++++++++++++



If you understand all of this you can go on. The next one is even more
difficult.



                     BINARY MULTIPLICTION

This is how the 8086 does multiplication internally. It is a series of shifts
and additions. We can do the same thing with base 10 numbers.


               24763 X 275

                    24763  ---
                    24763   |
                    24763   5
                    24763   |
                    24763  ---
                   247630  ___
                   247630   |
                   247630   |
                   247630   7
                   247630   |
                   247630   |
                   247630  ---
                  2476300  ---
                  2476300   2

                6,809,825

In the base 10 system this is tedious. In the base 2 system this works well.
You either do NO addition or you do 1 addition. We start at the bottom and add
(either once or not at all), then shift the whole number left one bit. We
repeat this cycle till we are finished with the whole multiplier. Once again,
the pivotal operation is finding the highest non-zero word before starting.
This is about 5 times slower than the first method. The only reason that it is
here is to prepare you for the binary division routine.

We need to reserve an extra word above the multiplicand. If the multiplicand
is 6 words long, we need 7 words for the multiplicand. The 6th word will shift
into that 7th word 1 bit at a time. At the end of our 16 bit cycle, all words
will have shifted up one word.

As the multiplication progresses, the bottom words of the multiplicand will be
0 so we don't bother to add these 0 words.

We load the multiplier into DX one word at a time. We then check this word one
bit at a time. If the bit is 1 we add, if the bit is 0 we do nothing. We shift
the multiplicand left 1 bit each time, whether we add or not.



+++++++++++++++++++++ << START OF PROGRAM >> ++++++++++++++++++++++

; binary multiplication using shifts and addition
; binary_multiply ( length , multiplicand, multiplier, temp_result )

; length is the number of WORDS
; length is a number, but the others are addresses. The temp_result
; space and the multiplicand space must be ((2 X length)+1) WORDS,
; and must be distinct from the calling variables since they will be
; overwritten by the routine. This is a far routine for C, and after
; setting up BP, we have:

;      TEMP_RESULT_ADDRESS     EQU   [bp + 12]
;      MULTIPLIER_ADDRESS      EQU   [bp + 10]
;      MULTIPLICAND_ADDRESS    EQU   [bp + 8]
;      DATA_LENGTH             EQU   [bp + 6]


include \pushregs.mac
; - - - - - - - - - - - - - - - - - - - -
DATASTUFF SEGMENT PUBLIC 'DATA'
multiplicand_length          dw     ?
multiplier_length            dw     ?
lowest_non_zero_word         dw     ?
DATASTUFF ENDS
; - - - - - - - - - - - - - - - - - - - -
CODESTUFF SEGMENT PUBLIC 'CODE'

     PUBLIC binary_multiply
     ASSUME  cs:CODESTUFF, ds:DATASTUFF

     TEMP_RESULT_ADDRESS     EQU   [bp + 12]
     MULTIPLIER_ADDRESS      EQU   [bp + 10]
     MULTIPLICAND_ADDRESS    EQU   [bp + 8]
     DATA_LENGTH             EQU   [bp + 6]

; - - - - - - - - - -
binary_multiply proc far

     push  bp
     mov   bp, sp
     pushf              ; save DF value
     PUSHREGS ax, bx, cx, dx, si, di, es
     push  ds           ; es = ds
     pop   es

     ; clear temp buffer
     mov   di, TEMP_RESULT_ADDRESS
     mov   cx, DATA_LENGTH
     shl   cx, 1        ; 2 X LENGTH is buffer length
     mov   ax, 0
     cld                ; upwards
     rep   stosw        ; store ax

     ; find the highest word which is non-zero
     mov   di, MULTIPLICAND_ADDRESS
     mov   dx, DATA_LENGTH
     mov   cx, dx       ; cx = length in words
     mov   bx, dx
     dec   bx
     shl   bx, 1        ; bx = top word
     add   di, bx       ; di = address of top word
     std                ; downwards
                        ; ax is still 0
     repe  scasw
     jne   first_top_found    ; found non-zero word
     jmp   exit_mult    ; multiplicand is 0 so result is 0

first_top_found:
      ; we went 2 too far + 2 for length + 2 extra for bit shift
     add   di, 6
     sub   di, MULTIPLICAND_ADDRESS
     shr   di, 1                     ; divide by 2
     mov   multiplicand_length, di   ; length in WORDS

     ; no registers have been modified except di and cx
     ; use the same ax, bx and dx values as before for multiplier.

     ; find the highest non-zero word
     mov   di, MULTIPLIER_ADDRESS
     add   di, bx         ; di = address of top word
     mov   cx, dx         ; cx = length in words
                          ; ax is still 0
     repe  scasw
     jne   second_top_found    ; found non-zero word
     jmp   exit_mult      ; multiplier is 0 so result is 0

second_top_found:
     ; we went 2 too far + 2 for length
     add   di, 4
     sub   di, MULTIPLIER_ADDRESS
     mov   multiplier_length, di        ; length in BYTES

     ; the multiplication  ********************
     mov   lowest_non_zero_word, 0

multiplicand_loop:
     mov   ax, lowest_non_zero_word      ; # of words shifted
     cmp   ax, multiplier_length         ; length in bytes
     jb    multiply_a_word
     jmp   exit_mult                     ; we are through
     ; ax still has lowest word count
multiply_a_word:
     mov   si, MULTIPLIER_ADDRESS
     add   si, ax                 ; calculate where multiplier is
     mov   dx, [si]               ; this is current multiplier word
     mov   cx, 16                 ; 16 adds and shifts
add_and_shift_loop:
     push  cx
     shr   dx, 1                    ; add if low bit is 1
     jnc   skip_the_addition
     mov   ax, lowest_non_zero_word     ; offset count
     mov   si, MULTIPLICAND_ADDRESS
     add   si, ax
     mov   bx, TEMP_RESULT_ADDRESS
     add   bx, ax
     mov   cx, multiplicand_length       ; length in words
     clc
inner_add_loop:
     mov   ax, [si]
     adc   [bx], ax
     inc   si               ; doesn't affect the carry flag
     inc   si
     inc   bx
     inc   bx
     loop  inner_add_loop
     adc   WORD PTR [bx], 0     ; one last carry is possible

skip_the_addition:
     ; shift one bit to the left
     mov   si, MULTIPLICAND_ADDRESS
     add   si, lowest_non_zero_word
     mov   cx, multiplicand_length         ; length in words
     clc
shift_1_loop:
     rcl   WORD PTR [si], 1
     inc   si                     ; doesn't affect carry flag
     inc   si
     loop  shift_1_loop

     pop   cx
     loop  add_and_shift_loop

     add   lowest_non_zero_word, 2          ; move up one word
     jmp   multiplicand_loop


     ; end of the multiplication  *************

exit_mult:
     POPREGS ax, bx, cx, dx, si, di, es
     popf              ; restore DF value
     mov   sp, bp
     pop   bp
     ret               ; a C return, so don't pop arguments.

binary_multiply  endp
; - - - - - - - - - -
CODESTUFF ENDS
END

++++++++++++++++++++++ << END OF PROGRAM >> +++++++++++++++++++++++




                       BINARY DIVISION


This is by far the hardest to understand. The binary division routine is the
opposite of the multiplication routine. We move the dividend to the remainder
area since it will be modified during the routine. We shift the divisor one
word past the top of the dividend (to make sure that the divisor starts out
larger than the dividend) and then start the shift-subtract cycle. We shift
right 1 bit and then take a look at the two numbers. If the divisor is larger
than the dividend we do nothing and put a 0 bit in the quotient. If the
divisor is smaller, we put a 1 bit in the quotient and subtract the divisor
from the dividend. At the end, what is left of the dividend is our remainder

As usual, we only use only as many words as necessary, both for the numbers
and the individual subtractions.

This is about 5 times slower than the block multiplication. It is possible to
approach the speed of the block multiplication routine by using block division
routine which guesses and then modifies its guess, but it would be almost
impossible to understand what the code does, so I won't show it to you.


+++++++++++++++++++++ << START OF PROGRAM >> ++++++++++++++++++++++

; binary division using shifts and subtraction
; binary_divide ( length , dividend, divisor, quotient, remainder)

; length is the number of WORDS
; length is a number, but the others are addresses. The divisor and
; remainder space will be overwritten one word past the highest non-
; zero word by the subroutine. The remainder space is cleared one word past
; its length. This is a far routine for C, and after setting up BP, we have:

     OUR_DIVIDEND_ADDRESS EQU   [bp + 14]   ; same as remainder address
     REMAINDER_ADDRESS    EQU   [bp + 14]
     QUOTIENT_ADDRESS     EQU   [bp + 12]
     DIVISOR_ADDRESS      EQU   [bp + 10]
     DIVIDEND_ADDRESS     EQU   [bp + 8]
     DATA_LENGTH          EQU   [bp + 6]

include \pushregs.mac
; - - - - - - - - - - - - - - - - - - - -
DATASTUFF SEGMENT PUBLIC 'DATA'

dividend_length               dw    ?
divisor_length                dw    ?
     ; - - - -
top_divisor_address           dw    ?
bottom_divisor_address        dw    ?
top_dividend_address          dw    ?
bottom_dividend_address       dw    ?
current_quotient_address      dw    ?
     ; - - - -
shift_count                   dw    ?
quotient_bit                  dw    ?



DATASTUFF ENDS
; - - - - - - - - - - - - - - - - - - - -
CODESTUFF SEGMENT PUBLIC 'CODE'

     PUBLIC  binary_divide
     ASSUME  cs:CODESTUFF, ds:DATASTUFF

; - - - - - - - - - -
binary_divide proc far

     push bp
     mov  bp, sp
     pushf                   ; save DF value
     PUSHREGS ax, bx, cx, dx, si, di, es
     push ds             ; es = ds
     pop  es

     ; clear quotient
     mov   ax, 0                   ; zero for clearing
     mov   dx, DATA_LENGTH         ; store for later
     mov   cx, dx
     mov   di, QUOTIENT_ADDRESS
     cld                           ; upwards
     rep   stosw
     ; move dividend to remainder area
     mov   si, DIVIDEND_ADDRESS
     mov   di, REMAINDER_ADDRESS   ; our new dividend area
     mov   cx, dx                  ; DATA_LENGTH
     rep   movsw                   ; upwards
     mov   [di], ax                ; extra 0 above dividend space

     ; find the highest divisor word which is non-zero
     ; dx  still has DATA_LENGTH
     mov   bx, dx            ; dx = DATA_LENGTH
     dec   bx
     shl   bx, 1             ; bx = top word (in # of bytes)
     mov   di, DIVISOR_ADDRESS
     mov   bottom_divisor_address, di    ; save for later
     add   di, bx            ; di = address of top word
     mov   cx, dx            ; cx = length in words
     std                     ; downwards
     ; ax is still 0
     repe  scasw             ; look for nonzero
     jne   first_top_found   ; left loop because unequal?
     int   0                 ; divisor is 0 so divide error

first_top_found:
     add   di, 2                         ; we went 2 too far
     mov   top_divisor_address, di       ; store for later
     sub   di, DIVISOR_ADDRESS
     add   di, 2                         ; actual length
     mov   divisor_length, di            ; length in BYTES

     ; no registers have been modified except di and cx
     ; use the same ax, bx and dx values as before for dividend.
     ; find the highest non-zero dividend word
     ; ax is still 0 (from above)
     mov   di, OUR_DIVIDEND_ADDRESS
     mov   bottom_dividend_address, di   ; save for later
     add   di, bx      ; di = address of top word
     mov   cx, dx      ; dx = length in words
     repe  scasw       ; downwards
     jne   second_top_found        ; equal on exit?
     jmp   exit_div    ; dividend = 0 so quotient is 0, remainder is 0

second_top_found:
     ; add 2 for overshoot & 2 for calculating length
     ; top dividend address is just past top of dividend
     add   di, 4
     mov   top_dividend_address, di   ; this is correct
     sub   di, OUR_DIVIDEND_ADDRESS
     mov   dividend_length, di    ; length in BYTES

     ; if dividend length < divisor length, we are done
     cmp   di, divisor_length
     jae   shift_divisor
     jmp   exit_div

shift_divisor:
     ; figure out shift count.
     ; change divisor length from bytes to words
     ; di is still dividend length
     mov   ax, di                  ; dividend_length
     mov   dx, divisor_length
     sub   ax, dx                  ; amount of shift
     add   bottom_divisor_address, ax   ; current bottom
     add   bottom_dividend_address, ax  ; current bottom
     add   ax, 2                   ; 2 extra bytes for shift
     mov   shift_count, ax         ; save shift count
     shr   dx, 1                   ; divisor length - BYTES to WORDS
     mov   cx, dx                  ; cx is amount of data to shift
     inc   dx                      ; one word extra for shift
     mov   divisor_length, dx      ; new divisor_length  (WORDS)
     ; prepare pointers for the shift
     mov   si, top_divisor_address
     mov   di, si                  ; destination pointer
     add   di, ax                  ; add the shift
     mov   top_divisor_address, di ; new top of divisor
     rep   movsw                   ; downwards
     ; zero bottom of divisor
     mov   ax, 0
     mov   cx, shift_count
     shr   cx, 1                   ; shift count in words
     rep   stosw

     ; set up quotient info
     mov   ax, QUOTIENT_ADDRESS
     add   ax, shift_count
     sub   ax, 2                   ; address of top word
     mov   current_quotient_address, ax
     mov   quotient_bit, 0001h     ; bit to rotate

     ; ***** the division  *****************

division_loop:
     cmp   shift_count, 0    ; if 0, we are done
     ja    do_shift_16
     jmp   exit_div

do_shift_16:
     ; ++++++++++ SHIFT AND SUBTRACT LOOP ++++++
     mov   cx, 16
shift_16_loop:
     push  cx                ; save counter

     ; +++++++++  SHIFT ++++++++++
     ; shift divisor one bit to the right
     ror   quotient_bit, 1
     mov   si, top_divisor_address
     mov   cx, divisor_length         ; length in words
     clc                              ; clear CF
shift_1_loop:
     rcr   WORD PTR [si], 1
     dec   si                ; doesn't affect carry flag
     dec   si
     loop shift_1_loop

     ; +++++++++ CHECK FOR SKIP SUBTRACTION +++++++
     ; skip subtraction if dividend < divisor
     mov   di, top_divisor_address
     mov   si, top_dividend_address
     mov   cx, divisor_length
     std                     ; decrement pointers
     repe  cmpsw             ; cmp dividend, divisor
     jb    skip_subtraction  ; dividend < divisor

     ; +++++++++++++++ SUBTRACTION ++++++++++++++++
     ; OR 1 into quotient
     mov   si, current_quotient_address
     mov   dx, quotient_bit
     or    [si], dx

     mov   si, bottom_divisor_address
     mov   di, bottom_dividend_address
     mov   cx, divisor_length      ; words
     clc                           ; clear CF
subtraction_loop:
     mov   dx, [si]
     sbb   [di], dx
     inc   si
     inc   si
     inc   di
     inc   di
     loop  subtraction_loop

     ; dividend >= divisor, so we have no final borrow

     ; +++++++++++ AFTER SUBTRACTION ++++++++++++++++
skip_subtraction:
     pop   cx
     loop  shift_16_loop

     ; reset the pointers and counters for the outer loop
     sub   shift_count, 2
     sub   top_divisor_address, 2
     sub   top_dividend_address, 2
     sub   bottom_divisor_address, 2
     sub   bottom_dividend_address, 2
     sub   current_quotient_address, 2
     jmp   division_loop

     ; end of the division  *************

exit_div:
     POPREGS ax, bx, cx, dx, si, di, es
     popf              ; restore DF value
     mov  sp, bp
     pop  bp
     ret            ; a C return, so don't pop arguments.

binary_divide  endp
; - - - - - - - - - -
CODESTUFF ENDS
END

++++++++++++++++++++++ << END OF PROGRAM >> +++++++++++++++++++++++