Microchip provides the following code as a 16 x 16 unsigned multiply in
several 18F datasheets (I got it from the 18F1220/1320 datasheet):

; RES1:RES0 = ARG1L * ARG2L

MOVF   ARG1L, W
MULWF   ARG2L
MOVFF   PRODH, RES1
MOVFF   PRODL, RES0

; RES3:RES2 = ARG1H * ARG2H

MOVF   ARG1H, W
MULWF   ARG2H
MOVFF   PRODH, RES3
MOVFF   PRODL, RES2

; RES3:RES2:RES1 += ARG1L * ARG2H

MOVF   ARG1L, W
MULWF   ARG2H
MOVF   PRODL, W
ADDWF   RES1, F
MOVF   PRODH, W
ADDWFC   RES2, F
CLRF   WREG
ADDWFC   RES3, F

; RES2:RES2:RES1 += ARG1H * ARG2L

MOVF   ARG1H, W
MULWF   ARG2L
MOVF   PRODL, W
ADDWF   RES1, F
MOVF   PRODH, W
ADDWFC   RES2, F
CLRF   WREG
ADDWFC   RES3, F

This code takes 28 code words and 28 cycles (contrary to the timing in the
datasheet which claims 24 code words/24 cycles (I think they counted the
MOVFFs as 1/1 instread of 2/2).

I figured that this could be improved upon and this is what I came up with:

; RES1:RES0 = ARG1L * ARG2L

movf   ARG1L,F
mulwf   ARG2L
movff   PRODL,RES0
movff   PRODH,res1

; RES2:RES1 += ARG1L * ARG2H

mulwf   ARG2H
movf   PRODL,W
addwf   RES1,F
clrf   WREG
addwfc   PRODH,W
movwf   RES2

; NOTE: It is OK not to carry into RES3 here
; because the maximum value in RES2:RES1:RES0
; would be: FEFF01, which doesn't carry into
; RES3

; RES3:RES2 += ARG1H * ARG2H

movf   ARG1H,W
mulwf   ARG2H
movf   PRODL,W
addwf   RES2,F
clrf   WREG
addwfc   PRODH,W
movwf   RES3

; RES3:RES2:RES1 += ARG1H * ARG2L

movf   ARG1H,W
mulwf   ARG2L,W
movf   PRODL,W
addwf   RES1,F
clrf   WREG
addwfc   RES2,F
addwfc   RES3,F

This is 26 code words / 26 cycles, or 2/2 better than the datasheet routine.

Can anybody improve on this??

There is a similar situation with the 16 x 16 signed multiplication routine.

Bob Ammerman
RAm Systems

--
http://www.piclist.com#nomail Going offline? Don't AutoReply us!
email listserv@mitvma.mit.edu with SET PICList DIGEST in the body