> Hi, is anyone interested in a quick code challenge?
> I have a quick polling loop on a 16F84 that needs to
> check if either of 2 input pins has gone high.
>
> It starts with the two pin values in w,
> just read from the port and the other 6bits masked to 0.
> The two bits are next to each other in bits 4,5
> and won't change places ever due to hardware.
>
> The last state of the 2 bits is is last_inputs reg.
> The bit positions line up in 4,5 again. The other 6
> bits can be masked to 0 or dont-care.
>
> * absolute quickest way to detect if EITHER pin
> has changed from 0 to 1. Critical.
>
> * quickest ways to exit if no change, and to
> store the new last_inputs value and exit if
> there is any other changes (1s to 0s etc)
>
> Any suggestions? My code is clumsy and first
> checks for any change, 5 cycles including the
> portb read and mask. Then I am left checking to see
> if either pin has gone from 0 to 1 but it's
> very messy since the test I need the fastest
> is the 0 to 1 tests. It's probably one of those real
> simple things that the code geniuses do without
> any effort?? :o)
> -Roman

>
> >It starts with the two pin values in w,
> >just read from the port and the other 6bits masked to 0.
> >The two bits are next to each other in bits 4,5
> >and won't change places ever due to hardware.
> >
> >The last state of the 2 bits is is last_inputs reg.
> >The bit positions line up in 4,5 again. The other 6
> >bits can be masked to 0 or dont-care.
>
> ;; Untested
> ;; 5 cycles if bit5 went 0->1
> ;; 7 cycles if bit4 only went 0->1
> ;; 8 cycles if neither bit changed from 0->1
>   btfss last_inputs, 5
>   btfss PORTB, 5
>   goto Check_bit4
>   goto Bit_Changed
>
> Check_bit4
>   btfss last_inputs, 4
>   btfss PORTB, 4
>   goto Both_Set
>
> Bit_Changed
> ;; At least one bit changed from 0->1
>   movf PORTB, 0
>   movwf  last_inputs
>   .
>   .
>   .
> Both_Set
> ;; Both bits of "last_inputs" were set initially, so exit
>   movf PORTB, 0
>   movwf  last_inputs
>   .
>   .
>   .
>
> There seems to be a more elegant way, but I couldn't get it below 9 cycles
> using 3 other different methods.
>
> --Andrew

>.   btfss last_input, 5
>.   btfss last_input, 4
>.   goto error_path
>;; Either 4 or 5 changed 0->1
>.   movf PORTB, w
>.   movwf last_input
>.
>.   etc...
>error_path
>;; Otherwise, neither changed
>;   do error stuff here
>;

> Date: Fri, 26 Nov 1999 02:41:01 +1100
> From: Roman Black <RemoveMEfastvidTakeThisOuTEZY.NET.AU>
> Reply-To: pic microcontroller discussion list <PICLISTEraseME.....MITVMA.MIT.EDU>
> To: EraseMEPICLISTMITVMA.MIT.EDU
> Subject: Re: [PIC]: bit high code challenge
>
> Hi, is anyone interested in a quick code challenge?
> I have a quick polling loop on a 16F84 that needs to
> check if either of 2 input pins has gone high.
>
> It starts with the two pin values in w,
> just read from the port and the other 6bits masked to 0.
> The two bits are next to each other in bits 4,5
> and won't change places ever due to hardware.
>
> The last state of the 2 bits is is last_inputs reg.
> The bit positions line up in 4,5 again. The other 6
> bits can be masked to 0 or dont-care.
>
> * absolute quickest way to detect if EITHER pin
> has changed from 0 to 1. Critical.
>
> * quickest ways to exit if no change, and to
> store the new last_inputs value and exit if
> there is any other changes (1s to 0s etc)
>
> Any suggestions? My code is clumsy and first
> checks for any change, 5 cycles including the
> portb read and mask. Then I am left checking to see
> if either pin has gone from 0 to 1 but it's
> very messy since the test I need the fastest
> is the 0 to 1 tests. It's probably one of those real
> simple things that the code geniuses do without
> any effort?? :o)
> -Roman
>
> --
> http://www.piclist.com#nomail Going offline? Don't AutoReply us!
> email RemoveMElistservEraseMEEraseMEmitvma.mit.edu with SET PICList DIGEST in the body
>
>

>
> (don't mind the "." in front of each line... my email client seems not to
> recognize tabs or spaces in front of text)
>
> Hmm, how about this.  6 cycles regardless of which bit changed.  7 cycles if
> neither changed 0->1.
>
> .   movf PORTB, w
> .   andlw 0x30  ; mask out 4,5
> .   xorwf last_input, f
> .   btfss last_input, 5
> .   btfss last_input, 4
> .   goto error_path
> ;; Either 4 or 5 changed 0->1
> .   movf PORTB, w
> .   movwf last_input
> .
> .   etc...
> error_path
> ;; Otherwise, neither changed
> ;   do error stuff here
> ;

> Drew Vassallo wrote:
> >
> > (don't mind the "." in front of each line... my email client seems not to
> > recognize tabs or spaces in front of text)
> >
> > Hmm, how about this.  6 cycles regardless of which bit changed.  7 cycles if
> > neither changed 0->1.
> >
> > .   movf PORTB, w
> > .   andlw 0x30  ; mask out 4,5
> > .   xorwf last_input, f
> > .   btfss last_input, 5
> > .   btfss last_input, 4
> > .   goto error_path
> > ;; Either 4 or 5 changed 0->1
> > .   movf PORTB, w
> > .   movwf last_input
> > .
> > .   etc...
> > error_path
> > ;; Otherwise, neither changed
> > ;   do error stuff here
> > ;
>
>
> Hi Andrew (and thanks everyone else re this challenge!).
>
> But now I'm getting confused. Doesn't the xorwf
> produce a 1 for every bit that has changed, and a 0
> for every unchanged bit??
>
> So this is just a test for change?? How does it
> just detect a 0-1 change but not a 1-0 change?
>
> I already had a fast routine to detect any change:
>
> subwf last_input,w
> skpz
>
> so did I miss something? :o)
>
> I forgot to mention before too that after it detects
> a 0-1 going input I need to know which input it was
> to get to that routine asap. :o)

> Another way to approach this problem is with state machines. As I see it you
> have these conditions which are of interest:
>
>     bit5    bit4
> SA:  low     low
> SB:  low    high
> SC: high     low
> SD: high    high
>
> Then the low to high transitions may be checked slightly faster:
>
> while in state SA:
>
>     movf   PORTB,W
>     andlw  MASK
>     skpnz
>   ...
>
> However, this assumes your in a tight polling loop. If you're really
> multitasking, then you have the overhead of determining which state to switch
> to.
> Just a thought.

>
> At 12:43 AM 2/22/01 +1100, Roman Black wrote:
> >Thanks to everyone who offered ideas for detecting
> >2 inputs going 0-1 in the fastest time. I learned a
> >lot and it was a good help.
> >
> >Basically now I'm torn between using int on change
> >feature, which has advantages but sucks when you
> >need to rely on it to detect two different inputs.
>
> Are you using TMR0 during the time when you need to monitor for those pin
> changes?  If not, you can use the T0CK input as another interrupt
> input.  You simply change TMR0 clock input to external, set the polarity
> for the edge that you want (rising or falling), pre-load TMR0 with 0xFF,
> clear T0IF, and set T0IE.  Next correct edge on pin T0CK causes an
> interrupt.  That, coupled with the INT pin gives you 2 distinct interrupt
> inputs.
>
> Another choice is to add one of those SOT 23-5 single gate logic packages -
> use a XOR gate feeding the INT pin.  You could fit that chip on the other
> side of the board under the PIC - no extra board space required.  Downside
> is that you need 3 PIC pins - the 2 inputs (paralleled with the XOR gate
> inputs) plus the INT pin.
>
> INT on change feature is not reliable if you are accessing (reading or
> writing) port B in any fashion while you want to monitor for pin
> changes.  If you can arrange your pin assignments so that you NEVER read or
> write port B while looking for those pin changes, INT on port change is usable.
>
> dwayne