dsum bands

[2:36 AM] entrpntr: so was there a simple explanation how dsum bands work?
[2:36 AM] entrpntr: or in general what causes the slope of dsum to change
[2:36 AM] entrpntr: these patterns I am seeing do not make a ton of sense
[2:37 AM] entrpntr: choice of bootrom inputs keeps things close to the first joypad hit in overworld usually
[2:37 AM] entrpntr: but then they predictably diverge after that
[2:38 AM] entrpntr: substantially too; like for pre-Nerd Paras manips, you only have one bootrom strat that will work based on where you save
[2:40 AM] entrpntr: the latest thing that threw me for a loop was nopal vs nopal(ab) when saving outside of Moon
[2:41 AM] entrpntr: nopal is the only thing that will be in range for the tiles you want Paras on (dsum is ~0 a few tiles before, everything else has diverged to like 210-230 by that point)
[2:41 AM] entrpntr: but by the time you hit Moon entrance, nopal(ab) was at ~20 while nopal was at ~10
[2:41 AM] entrpntr: so it was diverging in one direction prior to Moon entrance, then went the other way
[2:43 AM] entrpntr: also for instance, saving on http://pokeworld.herokuapp.com/rb/1#141,61 vs saving on http://pokeworld.herokuapp.com/rb/1#140,60 causes different rates of change
[2:44 AM] entrpntr: (those 2 tiles IGT0 coordinate with 0 A presses)
[2:45 AM] entrpntr: if you save 0-1 tiles inside of Moon instead, only nopal(ab) will get dsum into a good range
[2:46 AM] entrpntr: I'm guessing there might be a reasonable explanation based on how the Random function works, or how related variables (cycle counter, carry flag) change, but right now I just have data that looks weird
[4:52 PM] ExtraTricky: the bands are caused by the cycle offset of VBLANK
[4:52 PM] ExtraTricky: essentially either it will be 0, 4, 8, or 12 mod 16, and one of those corresponds to the high band (which one depends on where you measure)
[4:53 PM] ExtraTricky: I never was able to find a good way to predict what cycle offset you'd land on
[4:54 PM] ExtraTricky: but essentially whenever vblank turns off is when you can switch bands, depending on when it gets switched on again
[4:55 PM] entrpntr: k that makes sense and basically explains those observations
[4:57 PM] Dabomstew: so a script could check its band, presumably
[4:57 PM] ExtraTricky: yeah easily
[4:57 PM] Dabomstew: if a method that exposes cyclecounter is added to the libgambatte api
[4:57 PM] Dabomstew: not that we had any issues doing that in the past
[4:57 PM] entrpntr: time to add cycle offset pruning to htas