Quoting Copyfault
The EOR#$FF beams us into yet another interval, but the sub will know work consistently: either we are before the critical cycle and the cmp#... gives a clear carry (effectively substrating one more than needed), or the carry is set, exactly when the [SBC raster] will be done one line later than needed. Either case, we end up with "one digit off" in the substraction, which only affects the interval of the resulting values.
I'll leave it for now... hopefully :)=)
Just had the idea to substitute the SEC by a cmp#val, and to continuously update val to hold the lo-byte value of the critical cycle in a line. While the overhead for this feels far to high, it even doesn't work, because the carry is set as long as the timer value is above the compare-value, but we'd need it the other way round. Maybe someone has a better idea...I give it a last shot. When this mini-dejittering is used "inline", it probably means we know the rasterline we're in and that the code is not in a subroutine but rather "thrown in" some rastercode. This'd make the overhead I had in mind when tracing that val-value more or less obsolete. The cmp#val *can* be used when the rasterline is know beforehand; we only have to invert the timer-value:
lda $dd04 eor #$ff cmp #<(($4cc7-57-line*63)^$ff) sbc $d012 and #$3f
The EOR#$FF beams us into yet another interval, but the sub will know work consistently: either we are before the critical cycle and the cmp#... gives a clear carry (effectively substrating one more than needed), or the carry is set, exactly when the [SBC raster] will be done one line later than needed. Either case, we end up with "one digit off" in the substraction, which only affects the interval of the resulting values.
I'll leave it for now... hopefully :)=)