help me bug fix 32bit multiplication

[twh/f2]

Hi !

 

I have here a 32bit multiplication routine which is simply adapted from this code from the Codebase64.

 

http://codebase64.org/doku.php?id=base:16bit_multiplication_32-bit_product

 

The following code was adapted by FJC.

Num1	.ds 4
Num2	.ds 4
Num3	.ds 8

​.proc MulLong ; 32 bit multiplication: Num1 by Num2, result in Num3
	lda #0
	sta Num3+4
	sta Num3+5
	sta Num3+6
	sta Num3+7
	ldx #32 ; loop for each bit
multloop
	lsr Num1+3 ; divide multiplier by 2
	lsr Num1+2
	lsr Num1+1
	ror Num1
	bcc rotate
	lda Num3+4 ; get upper half of product and add multiplicand
	clc
	adc Num2
	sta Num3+4
	lda Num3+5
	adc Num2+1
	sta Num3+5
	lda Num3+6
	adc Num2+2
	sta Num3+6
	lda Num3+7
	adc Num2+3
rotate
	ror ; rotate partial product 
	sta Num3+7
	ror Num3+6
	ror Num3+5
	ror Num3+4
	ror Num3+3
	ror Num3+2
	ror Num3+1
	ror Num3
	dex
	bne multloop
	lda Num3
	ldx Num3+1
	rts
.endp

But I'm afraid it has a bug which I can't really fix. The following shows the dilemma: $4 * $16b62 = $5ad88

	mda #$16b62 Num1
	mda #$4 Num2
	jsr MulLong
	mda Num3 Temp	; save result

	jmp *

It's producing: "$1AD88" rather than "$5AD88".

 

However this one works: $16b62 * $4 = $5ad88

	mda #$4 Num1
	mda #$16b62 Num2
	jsr MulLong
	mda Num3 Temp	; save result

FJC: It seems that some upper bits of multiplier/multiplicand aren't handled properly (so result is different depending on which has set or cleared upper bits).

 

Any ideas?! Any help is appreciated!!

 

greetings,

\thomas

Edited February 2, 2017 by twh/f2

[flashjazzcat]

Found it:

	lsr Num1+3 ; divide multiplier by 2
	lsr Num1+2
	lsr Num1+1
	ror Num1

Should surely be:

	lsr Num1+3 ; divide multiplier by 2
	ror Num1+2
	ror Num1+1
	ror Num1

[twh/f2]

i'm not home now... but let's see if this works "in the field"

 

Thank you Jon!!!

[twh/f2]

@home now and yep. it really solves the problem!!!

 

it's a perfectly logical bug fix.. afterwards. I expected something more fundamental to be the problem :-)

[flashjazzcat]

I guess I dumbly invented this bug when hurriedly adapting the code in response to your earlier request. Fortunately the issue doesn't exist in any released software.