[CSDb] - User Forums - Shortest code for stable raster timer setup

You are not logged in - nap

CSDb User Forums

Forums > C64 Coding > Shortest code for stable raster timer setup

2020-01-20 16:20

Krill

Registered: Apr 2002
Posts: 2980

Shortest code for stable raster timer setup

While working on my ICC 2019 4K entry (now postponed to ICC 2020, but i hope it'll be worth the wait), i came up with this (14 bytes):

initstabilise   lda $d012
                ldx #10          ; 2
-               dex              ;   (10 * 5) + 4
                bpl -            ; 54
                nop              ; 2
                eor $d012 - $ff,x; 5 = 63
                bne initstabilise; 7 = 70

                [...]; timer setup

The idea is to loop until the same current raster line is read at the very beginning (first cycle) and at the very end (last cycle) of a raster line, implying 0 cycles jitter.

With 63 cycles per line on PAL, the delay between the reads must be 63 cycles (and not 62), reading $d012 at cycle 0 and cycle 63 of a video frame's last line (311), which is one cycle longer due to the vertical retrace.

The downside is that effectively only one line per video frame is attempted, so the loop may take a few frames to terminate, and the worst case is somewhere just beyond 1 second.

The upside is that it always comes out at the same X raster position AND raster line (0), plus it leaves with accu = 0 and X = $ff, which can be economically re-used for further init code.

Now, is there an even shorter approach, or at least a same-size solution without the possibly-long wait drawback?

... 177 posts hidden. Click here to view all posts....

2020-01-30 07:38

Oswald

Registered: Apr 2002
Posts: 5094

in max how many frames would this awesome inc solution would sync up ?

I'd be interested in a solution thats simple AND fast. Not necessarily shortest. Would be a nice addendum to codebase.

What I have currently by Ninja does have a lot of code checking on how it misses the end of a rasterline, a version that fits into a dozen lines would be neater.

2020-01-30 09:08

ChristopherJam

Registered: Aug 2004
Posts: 1409

The INC solution takes at most nine frames, and should average 4.5

Copyfault, there's really no harm in writing to d012, unless you wish to set it to some other value than zero in the init code between the two bookends.

2020-01-30 09:12

Oswald

Registered: Apr 2002
Posts: 5094

wow, then the inc solution is good for everything :)

2020-01-30 20:05

Copyfault

Registered: Dec 2001
Posts: 478

Quote: The INC solution takes at most nine frames, and should average 4.5

Copyfault, there's really no harm in writing to d012, unless you wish to set it to some other value than zero in the init code between the two bookends.

I somehow tend to avoid writing to a reg if there's no real purpose behind.

But back to your INC-based solution: why does it take 9 frames at most? If the upper INC happens to come at some cycle >=9 of Rasterline $ff, it should take longer, more or less comparable to the alternative I presented - or do I miss smth here? AFAIU, both approaches do the same, just the rasterline where the syncing finishes is different (yours at line $100, mine at $101).

Ofcourse, it might also be a wanted side-effect to set $D012=0 if the first Raster-IRQ at line 0 (or $100 resp.) makes sense.

2020-01-31 00:24

Copyfault

Registered: Dec 2001
Posts: 478

Quote: in max how many frames would this awesome inc solution would sync up ?

I'd be interested in a solution thats simple AND fast. Not necessarily shortest. Would be a nice addendum to codebase.

What I have currently by Ninja does have a lot of code checking on how it misses the end of a rasterline, a version that fits into a dozen lines would be neater.

So you basically look for a solution that has the least raster-time demand for syncing, or am I on the wrong path?

Something like this should finish in at most eight rasterlines:

        lda #$08
        sta zp_val
        
        ldx #$fe
loop:   
wait_startline:
        cpx $d012
        bne wait_startline
        inx
        bmi wait_startline
        //at cycle 6..12 of line $ff
        ldy zp_val
waste_cycles:
        dey
        bpl waste_cycles
        
        cpx $d012
        bne loop //leaves at cycle 2 of the first line in which raster is stable ($100..$106)
done:

By debouncing the starting line, we can asure that the no. of cycles at the start of the actual syncing loop lies exactly in the interval [6..12] (and is never different). So the syncing can be done by variance cancelation, which needs one rasterline per correction cycle. As there are seven different possibilites for the variance (6,7,8,9,10,11,12), (up to) seven rasterlines are needed in total (plus the first one for ensuring a "save start").

Maybe this can be done with shorter code, but I think not really faster (unless you really want to do variance halfing which will blow up code size too much for my taste).

2020-01-31 00:38

Oswald

Registered: Apr 2002
Posts: 5094

sorry I did not construct it properly with fast I meant it stabilizes fast, with that I mean max ~0.3 seconds a time span that for us humans doesnt matter :) so 9 frames max will do. however looking at the new version and explanation: your skills at this are truly impressive sir.

2020-01-31 11:03

Rastah Bar
Account closed

Registered: Oct 2012
Posts: 336

What Krill said.

Here is another method (13 bytes, stabilizes in less than a frame). When entering from Basic, timer A of CIA#1 is running. That can be used to check if the last cycle of an RMW instruction falls on the first "BA low, AEC high" cycle of a badline, as follows:

sync: lda $dc04
      sec
      sta ZP    ;RMW instruction
      sbc $dc04
      cmp #51   
      bne sync:

If and only if the last cycle of STA ZP is executed on the first "BA low, AEC high" cycle of a badline will there be exactly 51 cycles between LDA $DC04 and SBC $DC04 and the routine will exit on the last cycle of a badline.

2020-01-31 23:50

Copyfault

Registered: Dec 2001
Posts: 478

Quoting Copyfault

[...]
But back to your INC-based solution: why does it take 9 frames at most?[...]

This kept me awake for quite some time now. Think I have an explanation for it - finally!

If I do the calculations correctly (read: set up my surrounding framework including those frame-counters right;)), the lda-based method takes at most 7 frames. Uh, why is it now 7, even less than those 9 frames maximum for the inc-based approach?

The answer lies in the respective entry points of the delay loops. Taking a look at the INC-method, we see that it starts with

waitline:
   inc $d012
   bne waitline
   ...

If this first waiting loop has finished, the delay part begins (that we decided upon to be filled with init code f.e.). To simplify things, let's hold the case of starting this code in the middle of line=$ff (it would instantly come true whilst being off more than 9 cycles from the start of that line) back for a moment. How many cycles are over when leaving the waiting loop? It's 4 cycles iff $d012=$ff on the fourth R-cycle of the INC, but it amounts to 12 cycles iff $d012=$ff happens one cycle later! So this gives a variance of 12-4=8.

Exactly this variance is what we need to get rid of to have a stable raster. The INC-&LDA-loops presented in this thread cancel one cycle of variance per frame. For the INC-approach, this means we need 8 frames for the worst case (i.e. 12 cycles off).
Now we still have that "bad case" I had ignored for the sake of simplification. In fact, it does not do too much harm: in case the loop really starts mid of the testing line ($ff in the INC-approach), the first delay loop run will go fail. As the loop construction ensures 71 cycles between each $d012-checks at the start of each delay loop, with ggT(71,63)=1 (coprimeness) plus the fact that one run of the waiting loop is 9, the next start of the delay loop will be at a cycle c of type c = 9*k + 71 = 9*(k+1) - 1 #= -1 (mod 9) [mind that 9 is a factor of 63=7*9, thus skipping a multiple of 9 will get you to the exact same cycle position of any other line (or the same in the next frame); that -1(mod9) ensures that the position is changed!]
This means, from the second run of the delay loop onwards till the end, we step through the cycles of the first nine cycles of the line.
So back to counting the no. of frames that is needed at most: this "bad case" adds one to this frame count. So the INC-approach has a max frame count of 9.

Looking at the LDA-based method, we have a waiting loop like this:

waitline:
   lda $d012
   bne waitline
   ...

This part is finished 2-8 cycles after the beginning of line=$00. Following the above arguments, this approach needs at most 6+1(for the "bad case")=7 frames. Interesting fact is that the waiting loop here also needs a factor of 63 (=7*9), i.e. 7 cycles for one run. So here we have that c-formula like this: c = 7*k + 71 = 7*(k+10) + 1 #= 1 (mod 7). Thus we deal with a 7-cycle window in this case.
One other thing to mention is that with that lda, there's no chance to check explicitly for a unique rasterline (or you use compare opcodes, but it'll take more bytes!!!). The fact that line=$000 consists only of 62 cycles and the construction of the delay loop ensure that the check if this line will always fail. This is no real problem either, as we hit line=$100 once per frame, so the overall approach will come to an end!

Maybe someone is interested enough to read this, maybe this was all clear to you. Anyway, I felt the urge to write it down now that I finally understood it (I think).

2020-02-01 00:06

Copyfault

Registered: Dec 2001
Posts: 478

Quote: sorry I did not construct it properly with fast I meant it stabilizes fast, with that I mean max ~0.3 seconds a time span that for us humans doesnt matter :) so 9 frames max will do. however looking at the new version and explanation: your skills at this are truly impressive sir.

Ah come on, I'm just too fond of playing aroung with things that seem to keep certain mathematical mysteries inside;) Does not really help to get things *done*
To the opposite: I'd say you are the one to adore here! Will never ever reach that level of coding that you simply own, Oswald! I mean it:)

But thanks for your kind words. Gives me the positive feeling that there are people like you out there that care about explanations'n'stuff!

2020-02-01 11:54

Rastah Bar
Account closed

Registered: Oct 2012
Posts: 336

I find this problem surprisingly hard to understand. I think I get most of what you are saying, but aren't you neglecting the presence of badlines? The number of cycles available to the CPU is less on badlines and can even vary because of RMW instructions in the init code. So it seems there may be cases where neither of the approaches (INC, LAX) locks. Or am I mistaken?