IGT0 cycle fuckery

[5:28 PM] Dabomstew: i think IGT has to overflow on the frame that the game starts loading a map
[5:28 PM] Dabomstew: which in this case is the first map off load
[5:32 PM] entrpntr: that would explain why it starts failing on random floors
[5:33 PM] gifvex: and always fails on same random floors with same save
[5:33 PM] Dabomstew: the game disables the lcd every time it loads a map, which throws off the interrupt timing cycle
[5:38 PM] Dabomstew: here's my theory
1) delayframes x20 are called before LoadMap is started (this number might be different for loading maps after the first, but it's also irrelevant)
2) after the 20th vblank interrupt is done, it returns to the function and starts with LoadMap
3) LoadMap calls DisableLCD soon enough that it's still on the same vblank, so cycles are still offset by the IGT overflow code
4) Because of the cycle offset, we exit DisableLCD's wait loop 4 cycles later [or earlier or w/e it is]
5) One last vblank interrupt is triggered IMMEDIATELY when the game re-enables vblank interrupt at the end of DIsableLCD, despite lcd being off. This is also offset by 4 cycles
6) GG

[5:41 PM] Dabomstew: state 2 enters DisableLCD at 2388CF2, exits waitloop at 023906CE and triggers vblank at 023906F4
state 1 enters DisableLCD at 02388D16 (+24h), exits waitloop at 023906D2 (+04h) and triggers vblank at 023906F8 (+04h)
[5:42 PM] Dabomstew: i havent investigated whether we get fucked if the phantom vblank is the one to overflow the IGT
[5:42 PM] Dabomstew: going to check that now
[5:44 PM] Dabomstew: state2 enablelcd 023CA7E2 nextvblank 023D2840
state2 modified to overflow in the phantom vblank: enablelcd 023CA806 (+24h) vblank 023D2864 (+24h)
[5:44 PM] Dabomstew: so yeah 1/30 to get fucked
[6:00 PM] Dabomstew: there are 2 "frames" where the overflow fucks you

1st frame overflow (IGT0 = 37f):
> overflow frames only - 4 cycle offset
> overflow frames and seconds - 8 cycle offset

2nd frame overflow (IGT0 = 36f):
> overflow frames only - 24 cycle offset
> overflow frames and seconds - 48? cycle offset

[6:00 PM] Dabomstew: not actual ingame frames necessarily, but calls to vblank and subsequently the IGT incrementer
[6:01 PM] Dabomstew: the second is far more drastic than the first, one would think
[6:01 PM] Dabomstew: but either has the potential to be devastating
[6:03 PM] Dabomstew: none of that is guaranteed at all, but I would imagine the first one is more likely to have a more similar outcome to the "non-fucked" situation, yes
[6:03 PM] Dabomstew: the only part I know for certain is the number of cycles things are offset by
[6:03 PM] Dabomstew: and why it happens