Don't have access to TAS-stuff at the moment so I'll just try to post from memory. In the no-input, mid-jump frames press R twice before moving L at 7*93 (don't remember hundreds value) and then another frame of L at 7*12. This results in reaching the 2-2B exit on frame 7*36 instead of 7*37.
Removing input on frames 6994, 6995 solves the 3-3 timer problem, woo. Seems that getting hit as the 2-3 boss dies (so $75=4) makes the hit on the 3-3 boss avoidable, and I have no idea why. I haven't looked into whether $75=0 could be in any way beneficial (or detrimental) for level 4 though.
Maybe it will have some use in the future. Without diving in deep and likely just repeating things you've already tried, I'll have to agree that the early levels should be well optimized at this point. Though I still might spend some time looking for ways to use the $75 thing. Edit: Found the mess of a script I made a couple years ago to look for places where $75=4 might usefully affect stuffLanguage: luacnt=0 function JugFinder() Interrupt = AND(memory.readbyte(0x4C),0x40) if (Interrupt > 0) then cnt=cnt+1 else cnt = 0 end if (cnt > 1) and (memory.readbyte(0x1FC) == 0x87) and (memory.readbyte(0x1F3) == 0xD8) then emu.pause() gui.text(1,226,cnt) end end
Running this over the current pacifist WIP input doesn't seem to point to any obvious places where $75 looks useful, but since I never bothered to find out why it affects some projectiles and not others, I don't really know if some subtle change could change that. MESHUGGAH: Can you think of any tossed/thrown/shot/spit objects where a 2 or 3 pixel change might matter? Another edit: It pains me to bring this up, but has anybody considered that to be strictly pacifist, the demon's head should not be killed?Language: luaNGtest = savestate.create() local function check75(RAM) if memory.readbyte(RAM) > 0x10 then --excludes normal enemy spawns from the check eX=RAM+0x60;eXf=RAM+0x58 --offset for enemy X position eY=RAM+0x80;eYf=RAM+0x78 --offset for enemy Y position ram=RAM;blah=1 end end while true do savestate.save(NGtest) emu.frameadvance() memory.register(0x400, 8, check75) if blah == 1 then eXa=memory.readbyte(eX)+(memory.readbyte(eXf)/256) eYa=memory.readbyte(eY)+(memory.readbyte(eYf)/256) savestate.load(NGtest) memory.writebyte(0x75, 4) --simulates the glitch emu.frameadvance() eXb=memory.readbyte(eX)+(memory.readbyte(eXf)/256) eYb=memory.readbyte(eY)+(memory.readbyte(eYf)/256) if eXa ~= eXb or eYa ~= eYb then frame=emu.framecount();xdif=eXa-eXb;ydif=eYa-eYb print(string.format("%05d, %02X, %4.2f, %4.2f",frame,ram,xdif,ydif)) --when glitch has an effect, output frame count, column for enemy data, positional discrepancy end savestate.load(NGtest) emu.frameadvance() end blah=0 end
Projectiles that would be nice to manipulate2-2 first screen bullet slot 4 - to avoid interrupt or get a half pixel to left 3-1 sprite limit route which depends on last bullet (edit: probably slot 7) shot after a slot gets free 3-2 bullet slot 5 - to avoid interrupt or get at least 1.5 pixel to right (spawn delay misery)
Sprite limit route is letting spawn the bullet guy and respawning a bird between the two guy on the platforms near the end, so Ryu would follow only 2 bullets (closer than he would follow 3 bullets) but in the end the bullet hits him so it becomes much slower. edit: regarding the demon's head, I vote for the pacifist spirit.
It was very long ago, before I started reversing that function. I also tried to predict it, but gave up after seeing that it would need to include interrupts that you don't know how many will happen and when.
Possible mprovements and their effect to 5-2 final knife bitch:
* feos' RTA trick at 5-1 start saves 5 frames for 1 HP. This way the knife bitch fires 13 frames earlier, which is really bad.
Frame 26788
* the same 5-1 boost and avoiding the ballerinas (jumping roundhouse kickers) at 5-2 for 13 frames the knife bitch still throws bad knife
Frame 26825
* avoiding only the ballerinas let's us jump over the knife bitch, but this way we need to manipulate bird 6 to be slower so we can boost over slot7 bird, this way we finish stage 5 with 1 health (3dmg = avoiding 2dmg ballerinas and 1dmg bat at 5-3)
Frame 26833
So currently I'm sticking to the last one.
Other improvements pointed out by Scumtron, 5-3 final walljumps were not perfect, so 2 frames can be saved at each place (3 times), saving 6 frames. And subtimer will be need to manipulate to get our sweet 3 frames back. Stay tuned :)
edit: and another one spotted by Scumtron is better jumping at 5-3 1st screen between tysons to save 1 frame.Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
Stack corruption: interesting, I've tried to recreate the situation on other levels without success (picking up a "bonus 1000 points" powerup and position differently to update score graphically while reaching the interrupt zone since killing the bird doesn't affects this).
It looks like a lag frame, however if you append 1 frame (to reach the ladder), it will strike again at 13314
edit: probably useless fact but it's 16*60 difference between 13314 and 12354moozooh: MESHUGGAH is continuing work on a pacifist run, not an improvement to the published run. Personally, I wasn't sold on a pacifist run being submission-worthy until feos made some test runs and made clear that, done right, it would take full application of every little thing known about the game—including some stuff that just isn't relevant when you can kill everything. As for any%, thanks to a couple recent finds from MESHUGGAH and feos, I've resumed my sporadic efforts of improving it and it's currently 65 frames faster (should be a few more by later today).
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
