Post by thread:switch (case) for the PIC

Does anyone have a simpler switch (multiple case) than this below? I would like to decide where to go based on W contents, but not for 256 cases, only a sparse number of them. Look at the code, table, and give your comments, please. It uses 2 temporary storage for compared byte and loop counter. I call it putting in W the value to be compared, the destination case returns at the end. casech equ 0ch tmp equ 0dh ; switch movwf casech clrf tmp swloop movf tmp,w addwf tmp,w ;2*n gets in a 'retlw' instruction call swtable ;andlw 0ffh ;(optional) return if found NULL ;btfsc _z ;return subwf casech,w btfsc _z goto swjmp incf tmp,f goto swloop swjmp movf tmp,w addwf tmp,w addlw 1 ;2*n+1 targets a 'goto' instruction ;goto swtable ;(optional if table don't follows) swtable addwf PCL,f retlw case_a goto rot_a retlw case_b goto rot_b ... ... retlw case_n goto rot_n retlw 0 ;NULL marks end (optional) ; rot_a ... ... return ;returns to main (switch caller) best regards, Rildo Pragana

>Does anyone have a simpler switch (multiple case) <snip> The easiest I know of is something like,... ; enter with value in W. ANDLW 0x07 ; constrain to within jump table. ADDWF PC,F ; add to PC GOTO CASE_0 GOTO CASE_1 GOTO CASE_2 GOTO CASE_3 GOTO CASE_4 GOTO CASE_5 GOTO CASE_6 GOTO CASE_7 ; ; never get's here ; Ray Gardiner, Shepparton, Victoria, Australia, spam_OUTrayTakeThisOuTnetspace.net.au

Andrew Warren has given an optimal implementation of a switch statement. However, for its interest, here is a modestly tidied-up version of the original loop. casech equ 0ch tmp equ 0dh ; switch movwf casech clrf tmp ; swloop call swtable incf tmp,f ; subwf casech,w btfsc _z goto swtable ; incf tmp,f goto swloop ; ; swtable movf tmp,w addwf PCL,f ; retlw case_a goto rot_a retlw case_b goto rot_b ... ___Bob

> Andrew Warren has given an optimal implementation of a > switch statement. However, for its interest, here is > a modestly tidied-up version of the original loop. [table-driven switch statement deleted] Cute, but your approach uses 13 cycles per case, at a savings of at most one byte per. On that basis I would have to regard Andy's as being better all around. That does not mean it cannot be improved upon, though. :-) On a non-5x/non-508 part, you may use the addlw instruction to improve things a bit. For example, [hopefully I'm doing this right]... suppose you need to distinguish Y or y, N or n, and 0-9. Switcher: addlw 256-':' ; One character past '9' btfss C goto DigCheck addlw 256+':'-'Y' btfsc Z goto Yep addlw 256+'Y'-'y' btfsc Z goto Yep addlw 'y'-'N' btfsc Z goto Nope addlw 256+'N'-'n' btfsc Z goto Nope goto Reject DigCheck: addlw ':'-'0' btfss C goto Reject goto GotDigit ; W is digit value 0..9 The use of 'addlw' rather than 'xorlw' allows the program to check for ranges, and to perform a binary search. The only downside is that the code can get more complicated (even if it's faster/shorter) than with the 'xorlw' approach.

>> Andrew Warren has given an optimal implementation of a >> switch statement. However, for its interest, here is >> a modestly tidied-up version of the original loop. > >[table-driven switch statement deleted] > >Cute, but your approach uses 13 cycles per case, at a savings of at most >one byte per. On that basis I would have to regard Andy's as being better >all around. Cute?? My intention wasn't to produce a time-optimised version of a sparse table lookup. Rildo Pragana put up an implementation that used array indexing and a loop, and I wanted to show Rildo a variation of this technique. Looking at Rildo's code you see he keeps an array index, and then calculates an array offset on each of the two occasions he accesses the array. This is suggestive of a high level programmer who has 'hand compiled' the code. Most low level programmers would take the approach of just stepping straight through the array. I wanted to show this. {Quote hidden}>That does not mean it cannot be improved upon, though. :-) >On a non-5x/non-508 part, you may use the addlw instruction to improve >things a bit. For example, [hopefully I'm doing this right]... suppose you >need to distinguish Y or y, N or n, and 0-9. > >[switch statement deleted] > >The use of 'addlw' rather than 'xorlw' allows the program to check for ranges, >and to perform a binary search. The only downside is that the code can get >more complicated (even if it's faster/shorter) than with the 'xorlw' approach. The problem specification was to search a list of arbitrary values. To redefine the problem specification to include value ranges and sorted values, and to then declare a solution to this modified problem an 'improvement' is kinda cute. ___Bob