8x16 bit signed math in assembler

[mess]

can someone point me to a good tutorial for doing math on a pic?

I want to multiply a 8bit unsigned variable with a 16bit signed variable...

thanks,

Michaël

[stryd_one]

piclist Source Code Library :)

I love that list!

[mess]

Hi Stryd,

I couldn't find an example there and most of the code over there

takes no advantage of the hardware multiplier

will this solution work?

it first takes the absolute value of the 16bit signed value,

does the 16x8 multiply and makes it signed again

if someone knows of a better solution let me know

(note that the code below is inline assembler called from an interrupt routine)

the C equivalent:

out[HL] = out[HL] *cutoff /256

...
;out = out*cutoff/256 (8*16 multiply, keep 2 MS bytes)
; calculate the absolute value of out
	movff	_out_h, _IRQ_TMP2
	movff	_out_l, _IRQ_TMP1
	btfss	_IRQ_TMP2, 7	; if sign bit is not set goto get_abs_pos
	bra		_get_abs_pos
_get_abs_neg
	comf	_IRQ_TMP2, F
	comf	_IRQ_TMP1, F
	incf	_IRQ_TMP1, F
	skpnz
	incf	_IRQ_TMP2, F
_get_abs_pos

; multiply lsb of abs(out) with cutoff
	banksel	_cutoff
	movf	_cutoff,W, B
	mulwf	_IRQ_TMP1
	movff	_PRODH, _IRQ_TMP1
; multiply msb of abs(out) with cutoff
	mulwf	_IRQ_TMP2
	movf	_PRODL, W
	addwf	_IRQ_TMP1, F
	skpnc
	incf	_PRODH, F
	movff	_PRODH, _IRQ_TMP2
; if out was negative, convert to negative value
	banksel _out_h
	btfss	_IRQ_TMP2, 7,B	; if sign bit is not set goto mult_end
	bra		_mult_res_pos
_mult_result_neg
	comf	_IRQ_TMP2, F
	comf	_IRQ_TMP1, F
	incf	_IRQ_TMP1, F
	skpnz
	incf	_IRQ_TMP2, F
_mult_result_pos
	movff	_IRQ_TMP1, _out_l
	movff	_IRQ_TMP2, _out_h
...

[stryd_one]

:)

[stryd_one]

Did it work? Seems OK here...

I've been thinking about making a new sdcclib which uses the hardware mul' wherever possible, whatya think? Could save a lot of time and effort in optimising... I need some divide routines which return a remainder too....

That said, it's 1030am and I am still awake and cutting code, my head could be rolling down the street for all I know ;D

[mess]

I haven't tried it yet, I've been rewriting the core sound stuff in asm

because most of the code was inline asm anyway

right now I'm studying for my exams so it will take some time

before I can verify this code snippet

maybe it's better to use the 16x16 signed multiplication from the datasheet

and take out the stuff that you don't need (compensation for sign bit etc)

an optimised version of the sdcc lib would be nice

but there are too many things on my todo list already  :)