NMI TOD clock alarm interrupt

[Whammo]

This is a number of applications to exploit the TOD clock.
The alarm, as it is a NMI, calls QUIT, so it could be used as a watchdog timer.
The alarm NMI now only gets cleared by the NMI routine as per Kernal.
Now ptod with colons!
Leverage the power of your c64 as a clock.

: demit dup emit ;

: bcdin key demit $30 -
>r 1 key demit dup
$d <> if $30 - >r 1+ key demit
     then emit
1- r> swap 0<> if r> $10 * or then ;

: ampm ." am or pm? "
key demit $41 <> if $80 or then cr ;

: hour ." 1-12? " bcdin ampm $dd0b c! ;
: min ." 0-59 " bcdin $dd0a c! ;
: sec ." 0-59 " bcdin $dd09 c! ;
: 10th ." 0-9 " bcdin $dd08 c! ;

: rtod $dd0b dup c@ dup $7f and >r
        $80 and swap
        1- dup c@ >r
        1- dup c@ >r
        1-     c@ >r
        r> r> r> r> ;

: ptod  hex rtod          \ prints TOD
3 0 do dup 10 u< if '0' emit then
. $9d emit ':' emit loop
. decimal $80 and if   ." pm"
                else ." am"
                then cr ;

: settod  \ sets TOD
hour min sec 10th ptod ;

: setalarm \ sets alarm, enables interr
$80 $dd0f c! \ alarm bit
settod $84 $dd0d c! \ nmi enable
$2a1 c@ 4 or $2a1 c! \ rs232 enables
0 $dd0f c! ;  \ tod bit

: clalarm 4 $dd0d c!
$2a1 c@ $fb and $2a1 c! ; \ disable nmi, rs232 enables

: elapsed 0 0 0 0 \ used for benchmarki
          $dd0b c!
          $dd0a c!
          $dd09 c!
          $dd08 c! ;

 : time source >in @ /string
dup >in +! elapsed evaluate
ptod ;
\ time $c000 dump

: clock page
begin $13 emit ptod key?
until key drop ;

[polluks]

Cool! There was a hardware bug at midnight IIRC...

[Whammo]

There is no correction for the defect that makes the TOD clocks follow:

11:59:59.9 AM
with
12:00:00.0 PM

instead of AM, although 0.1 seconds later, the clocks' PM will be correct. A
similar flaw exists for midnight.

[Whammo]

An easy way to show this working is:

include tod
settod
setalarm
: go ptod recurse ;
go

It will break the GO loop accordingly.

[Whammo]

You'll notice when warp mode in VICE is active, unlike the jiffy clock, TOD isn't sped up.

[ekipan]

I rewrote your code to help myself understand it. It uses Johan's nice 00. word. Seems to work, though I know almost nothing about C64 hardware. The only reason >b isn't called >bcd is so am can fit on one line :P

decimal \ marker -tod- : vv -tod- v ;

\ time of day ( dsec sec min hour )
: >b  10 /mod 4 lshift or ; \ bcd
: am  >r 0 -rot r> >b swap >b rot >b ;
: pm  am $80 or ;
\ e.g. 4 30 57 pm -- 0 $57 $30 $84

\ print:
: 00.  s>d <# # # #> type ;
: a/p  $80 and if 'p' else 'a' then ;
: tod.  dup >r $1f and  base @ >r hex
 00. ':' emit 00. ':' emit 00. drop
 r> base !  r> a/p emit ." m " ;

\ fetch/store tod registers:
: c@+  dup c@ swap 1+ ;
: c!-  tuck c! 1- ;
: tod@  $dd08 c@+ c@+ c@+ c@ ;
: tod!  $dd0b c!- c!- c!- c! ;

\ alarm interrupt & register mode:
: alset!  if $84 else 4 then $dd0d c! ;
: almode! if $80 else 0 then $dd0f c! ;
: alarm!  1 almode! tod! 0 almode! ;

\ user interface:
: setalarm  1 alset! alarm! ;
: clalarm  0 alset! ;
: ptod  tod@ tod. ;
: lcsr  0 do $9d emit loop ;
: poll  key? if key else 0 then ;
: clock begin ptod 11 lcsr poll until ;
\ e.g. 1 23 45 pm tod! page clock

\ benching: e.g. ' slow timed
: slow  $a000 dump n n n ;
: timed  0 0 0 0 tod! execute ptod cr ;
: in> source >in @ /string dup >in +! ;
: time  in> ['] evaluate timed refill ;

Some history of this comment, check edits for full details:

• First edit reversed tod!'s input order, was actually broken before. Also inlined the Constants into tod@ and tod!
• Then added a missing drop to ptod.
• Removed outdated wording and added this changelog.
• Extracted timed and rest, changed keyin to chrin
• Removed chrin bcdin settod added am pm tod. alset! almode! alarm!
• Deleted incorrect chrin from below. Changed c@@/c!! into c@+/c!-.
• rest into in>, as in Whammo's original code.

Here's the removed input words in case they might be useful elsewhere:

: keyin  key dup emit ;
: bcdin  chrin '0' - chrin '0' -
 dup 9 u> if drop exit then
 swap 4 lshift or ;

[Whammo]

You are fortunate. Poking around in such things is perhaps less fun in Assembly, and certainly less fun with BASIC. 👍

[ekipan]

Speaking of assembly, here's a couple more niceties. A more general time and chrin instead of keyin, with perks like editing and a cursor.

Edit 1: though for some reason timed always prints 00:00:00am, no matter how long the executed word takes. Not sure what's going on there.
Edit 2: time $a000 dump n n n gave four seconds. Guess one dump wasn't slow enough :P
Edit 3: moved the code to the main post.

[ekipan]

I hope it's not a bother if I post more pickings at this.

I wrote am and pm words that convert hour minute second into bcd and rearrange them into the format tod! and tod. expect (also extracted tod. from ptod). That way you can just reuse the forth interpreter and delete the number input words.

Edit: Code moved into main post.

The alarm doesn't seem to do anything, unless I'm supposed to write more code that responds to it somehow, not sure.

1 23 50 pm setalarm
ok
1 23 45 pm tod! clock
ok:23:52pm

[Whammo]

Not a bother at all. In fact, it's why I bother to post my own musings.

[Whammo]

Alarm is a matter of pointing the NMI vector to one's own assembly code.

[Whammo]

restore_handler in interpreter.asm is mostly a copy of the kernal's NMI handler.
This is easily durexForthified and and appended with the appropriate code to redirect it to your favorite word when appropriate and to the kernal's routines or quit when it is appropriate.

restore_handler
    pha             ; save a
    txa             ; copy x
    pha             ; save x
    tya             ; copy y
    pha             ; save y
    lda	#$7f        ; disable all CIA 2 interrupts
    sta	$dd0d       ;
    ldy	$dd0d       ; save CIA 2 interrupt control register for kernal_nmi
    bpl brk_handler ; CIA 2 is not the NMI source if the most significant bit is not set.

kernal_nmi
    jmp $fe72       ; all CIA 2 NMI's fall through to the Kernals' RS-232 routines

brk_handler         ; all non-CIA NMI (RESTORE key) and brk instructions- via IRQ vector end up here.
    pla             ; drop y -the return stack will be reset by QUIT anyway
    pla             ; pull x
    tax             ; restore parameter stack pointer for QUIT
    jmp QUIT        ; already under sei from NMI stub in Kernal or from IRQ to brk_handler

such as this, as yet, untested:
Proof that labels are cool.
Bit 2 needs to be set in $2a1 ENABL or rs232 activity will clear the CIA alarm NMI enable bit.

code rsthandler 
pha, txa, pha, tya, pha, 
$7f lda,# $ddod dup sta, ldy,
0 @: bpl,
tya, 4 and,# 1 @: bne,
$fe72 jmp, 0 @@
pla, pla, tax, ' quit jmp,
1 @@  ' mycode jsr, \ mind yr, $fe72 requires it.
$fe72 jmp,  \ let the Kernal deal with the stack
;code

We have the NMI code. Now we want to signal an alarm without causing issues with the console, or another running program.
If we use the opportunity of the NMI to change the IRQ vector to our signal, say, flashing the border color on the IRQ and playing a bell sound every 60 jiffies we will presumably also require yet another NMI routine to expect RESTORE in order to clear the alarm by resetting the IRQ vector.