I recently completed a single-segment (RTA) run of the game
Link to video
One thing that should be noted is that you really don't need much luck with the RNG to clear the game this fast. As long as you can ensure Marth is def blessed early on (which I managed with a clock state of 12:00:10 AM, January 2009 on cleared SRAM), Marth will easily be able to solo the game with an 8 mt rapier forge.
Obviously it's more ideal to crit every boss in the game and avoid the C4 armory, but without a powerful means by which to manipulate the RNG, that is just a concession we'll probably have to make.
It's not as though the RNG isn't manipulable at all. The RNG changes as a function of time whenever there's dialogue on the screen, and I believe it changes dramatically faster if you don't set text speed to max. There are certain cutscenes you can skip a little slower than usual to get different RNG, you can change the order in which units attack, attack with different weapons, etc.
I was actually planning to whip up a quick TAS that basically imitates my RTA run and foregoes any deliberate RNG manipulation. Probably wouldn't take more than a day or two to perform.
If anyone wants to help me crack the game's RNG, it is located at addresses 02196E08 to 02196E0B as four one-bit signed values. This might also be of some help. I can track as many initial RNG states for different DS clock times as necessary. It's also possible to burn 1 RN at a time by simply waiting in an armory, which provides plenty of raw data for finding an RNG formula. I just don't know how to extrapolate a formula from even a large sample size of data, perhaps someone with more expertise can help me out on that end.
As a side-note, FE12's RNG is time-based and varies every frame. Only half the game can be warpskipped, and the game contains some very interesting staves, particularly the rescue, thief and again staff. I haven't even been able to determine where the game's RNG addresses lie, but if someone were interested in helping me with that, I would definitely TAS that game in a heartbeat.
I checked and this is an issue of every desmume version I tried, (0.9.4 through 0.9.10, 32 & 64 bit). In fact, I found some additional issues:
-Any in-game save made during a movie is lost once the movie is stopped and a reset is performed
-Deleting a savestate or a movie causes the SRAM to change
-Overwriting a movie causes desyncs
-You can record a movie without the .dsm extension
Among others.
1. Isn’t that normal? Last I checked (long long ago admittedly) you need to browse for the SRAM separately when opening a movie file.
No, you specify the SRAM when recording a movie, but not for opening one.
ALAKTORN wrote:
2. Isn’t that because in-game saves are a single file, and if you overwrite it– well, it’s overwritten, so even if you load different savestates the game save will stay as the overwritten one.
That's not been my experience with any other emulator (notably SNES9x and VBA). As far as I know, loading a savestate should cause the battery to change, otherwise resetting magically negates the effect the savestate had to begin with.
0.9.10
If you record a movie and make in-game saves, the saves don't work
If you play back a movie, it uses the current SRAM state instead of the one used by the movie
0.9.9
If you make an in-game save, loading another savestate with different in-game saves + resetting will cause the original in-game saves to show, which makes no sense.
I'll have a look at some earlier versions, but for the time being, is there a Desmume version that doesn't suffer from weird SRAM issues?
Link to video
From what I understood from the video, you can now go from the floating continent to Kefka's tower with a full party, as well as several illuminas in your inventory, effectively skipping the entire world of ruin.
If you're going to go for such an ambitious goal, you're definitely not going to be able to optimize this to the same extent as your any% run, otherwise it might take several years to complete, and that type of dedication just isn't worth it for a game as obscure as this.
I'm definitely interested in how you will deal with the funds and exp restriction since your only major advantage over dondon's runs are freely rigging crits. Might be a good idea to do a somewhat sloppy test run just to lay the groundwork for knowing what your requirements are since purchases and trades will be exponentially more complicated this time around. There's simply no way to anticipate these things under S-rank settings, let alone 0%.
I'll try to help out however I can, although I'm not around much anymore.
P.S. One thing that would be cool is if you could program in to the 0% patch some kind of condition that caps a unit's tier 1 level once they hit lv 10 (earliest time to user a promo item), and of course cap their tier 2 level instantly. This would avoid level-ups that serve no strategic purpose.
What's the point?
Glitched any% GT code: The TAS goes straight to lower norfair without even the varia suit, ends the game with arbitrary code execution immediately after Golden Torizo.
Glitched any% X-ray: Head to Maridia and X-ray climb until the game glitches into the explosion sequence. Gravity suit would allow you to move faster in Maridia but requires a huge detour + fighting Phantoon.
It goes without saying this run is phenomenal, hypnotic even.
For the whole category debacle, I would like to echo Garisson's statement:
Gametime TASing may lose its luster soon as well with the discovery that the same pause glitch used in this new TAS can be used to manipulate the Gametime to hilariously low numbers. Not only that, but the new arbitrary code discoveries can manipulate those numbers to be whatever the player decides upon. The two low% runs were both originally declined for being superfluous and brought back with the inclusion of the vault (and then published with moons instead????). If you want to be picky with the categories, there's a lot more fluff than just adding an arbitrary code run.
Keeping the 14% categories is debatable, but the in-game categories should definitely go because of the pause glitch.
Heh, been a while since I've visited this topic. I'm only 2 or 3 semesters away from finishing my undergrad, and I've learned quite a bit in the past few years.
Some random observations:
I'm currently taking a C++ course. At first I hated the language since I couldn't even get a handle on inserter and extractor operators (embarrassing, I know :P). I rather like how simple it is to use file streams compared to Java, and being able to overload operators is awesome. Templates are great too. What I don't like are qualifiers like std and all the weird variable types like size_t. Just having a program that doesn't have "using namespace std" already looks too verbose to read smoothly. I also find it annoying how being able to pass by value, reference and const reference adds a layer of complexity to parameters and return types. Can't say I love having to deal with pointers or de-allocating memory from the heap. Regardless, we've been given some great assignments to do, my favorite being a mini-search engine that would rank sentences in a document based on how well it matches a keyword. It was the most fun I've ever had programming and the first time I ever felt "passionate" about a program, losing track of time working on perfecting my creation.
Another interest course I’m taking involves working in a team of 11 people to create a spreadsheet application. Our project is split into 3 iterations, developer, tester and documenter, and our 3 sub-teams rotate roles each iteration. It’s a software engineering course and I absolutely hate the lectures because the content is unbelievably abstract to the point of uselessness without industry experience. The teacher prattles on about software architecture, requirements, domain models, agile development, and it’s just so meaningless without context. On the other hand, it’s a blast to work with a team even if we do often get disorganized. I’ve also learned how to read large programs (12+ classes, each with ~1000 lines of code), which mostly amounts to following function calls and learning to read programs in a non-linear manner.
I find most of my classes get bogged down with extreme levels of abstract detail. In a system hardware course, for example, I found it hard to pay attention to lectures on the MARIE architecture, but then had a great deal of fun mucking around with a MARIE simulator. I’m much more interested in experimenting with code and building useful applications than getting my head lost in trying to understand how a compiler interacts with a linker or how to micro-optimize a portion of code with loop unrolling.
Something I’ve been picking up on lately is that learning how to program is very different from learning a more traditional concept. For instance, you get in the habit of constantly looking at reference material (Ex/ properly using an interface like Observer in java) and reading answers from Stack Overflow as opposed to say, reading a history book and learning a bunch of facts in a linear manner. You get used to drawing diagrams and planning a lot before writing even a single line of code, and setting up all your classes and methods before adding content to them. Hell, lately I’ve been finding that IDEs like Eclipse are so helpful that you can often learn how to use a method just by typing [object name].[autofill] and then letting the mouse hover over the autofilled method to give a description of what it does.
As a last thought, I talked to a friend who works in the industry and he told me that he prefers to hire Comp Sci grads with no experience because older people with multiple master’s degrees don’t even know how to perform basic tasks in Excel, still use Internet Explorer, etc, and that the generation after me are screwed since schools aren’t teaching students how to read & write properly anymore. He encouraged me to look for work even though I haven’t graduated yet, and the fact that I know how to google for answers by itself puts me in the top 75th percentile of the tech workforce. I don’t know if he’s exaggerating or not but it certainly encourages me to go get some first-hand, real world experience of what it’s like to have a career in IT.
Voted no, it's not that technically impressive to manipulate Yoshimitsu's suicide attack to land every round. Plus it's repetitive, even for 2 minutes.
Link to videoWIP (up to Chapter 5)
Somehow I’ve already managed to save over a minute! A more detailed breakdown below:
Prologue: 2666-2609 = 57 frames (0.95 secs)
For some reason the cutscenes skipped faster from SRAM, otherwise the only difference was holding A on character movement.
Chapter 1: 5818-5245 = 573 frames (9.55 secs)
I found a more efficient way to burn RNs with Seth and delayed his level-up to the next chapter (which is actually why I lose time there, relatively speaking).
Chapter 2: 10626-10229 = 397 frames (6.6 secs)
I rescue Ross & Garcia as soon as possible in order to eliminate the “Other Phase” and plan out my movement to reduce the amount of enemy attacks I face (ie more player phase attacks, less enemy phase ones).
Chapter 3: 17124-15095 = 2029 frames (33.8 secs)
Huge strategic overhaul here. I rescue-drop Seth besides the wall so he can kill both nearby ranged attackers. The remaining turns I use a pacifist approach, only killing the boss and avoiding combat with the thief altogether. You’ll notice Seth’s level-up is a lot more tame than usual as well; he only gains hp and skill.
Chapter 4: 25894-22534 = 3360 frames (56 secs)
I land more crits on turn 1 player phase (reduces combat) and have Seth deal with the eastern mogalls instead of Artur (reduces manipulation).
Chapter 5: 30209-26261 = 3948 frames (65.8 secs)
I found a clever way to ninja the chapter with Seth such that the only enemy he'd actually attack is the boss. This actually postpones his level-up so that will result in lost time in C6.
In general you’ll see I’ve been much more thorough testing out alternate strategies and optimizing RN burns & trades, which is why my re-record count is already over 6000. I really want to hit that sub hour benchmark, although it’s starting to look more and more realistic the further I get.
===
Some miscellaneous observations:
Setting text speed -> max saves time in the armory since some of the text prompts are skipped automatically. However, for some reason you can’t skip boss portraits and some cutscenes must be skipped later when this feature is toggled. Due to this, the chapter 3 armory’s actually faster to do the regular way, although this may not be the case for the torch purchase in chapter 6.
Base stat Eirika can OHKO Murray. This is why I allow her to get a no strength level-up in C2.
It's unfortunate that the ending will be so much less cool, then (and we'll have to watch Fomortiis' slow as hell attack animation).
Actually it will still be pretty cool since the fastest way to kill him will probably be to have Cormag land a double pierce crit with a killer lance.
The Purge Sage in chapter 16 is also in a much weirder location in Ephraim route, so we unfortunately won't have access to that either.
The only places where purge are used are the gorgon eggs chapter (no big deal) and the final chapter, which I guess actually matters since you need to kill the bonewalker in front of Lyon to get that crucial 2 turn. There is, however, a way I could do that using Syrene and Cormag since Syrene has enough movement to kill the bonewalker at 1 range from a decent turn 2 vantage point.
---
So I did a quick test to form an educated guess as to how much time will be saved holding A. In a nutshell, normal movement costs 4 frames/square whereas movement holding A only costs 2. Thus, the time saved per chapter is equal to the amount of movement in said chapter.
I took C2 and C15 as samples (short chapter and long chapter, respectively), and extrapolated that C2 has exactly 56 frames of movement, whereas C15 has roughly 220. As a crude guess, if half the chapters in the game are short and half are long, this totals to 56(10) + 220(10) = 2760 frames saved, roughly a minute.
This is actually kind of disappointing because I was expecting more time to be saved. Then again, maybe C15 was a poor test chapter since it's beaten in 3 turns. Regardless, I'm kind of wondering if it's even worth whatever controversy starting from SRAM will cause if the savings are so minimal. Then again, holding A does make the run at least appear to be much faster paced, even if it doesn't affect what eats up the bulk of the time in such runs, which is the animation time for actually killing enemies and the idle time between their movements.
I think starting from SRAM to get the faster unit movement will cause the run to be rejected. It's sad that you won't be able to get sub-hour because of it, but that's the way it is.
But what would be the reasoning behind it? Originally I was bothered by the idea of using SRAM since I thought it would change the beginning RNG seed and therefore make the run impossible to reproduce without using the exact same SRAM. But this isn't the case, as stated in my previous post. Any SRAM that's cleared the game will yield the same starting RNG, so it's easy to reproduce a run like this from many possible SRAM states. Not only that, there's two published TASes for the game (yours and mine) that could easily be used as verification movies, which automatically rules out any possibility of cheating.
Are there any major known improvements in chapters 1-8, or will you be able to cannibalize those from the existing run?
Well, one definite improvement is moving the cursor without holding B on movement that normally causes the camera to scroll back to the character before moving. It's not too significant; if I had to guess it would save like a minute over the whole run.
e: I'm also curious to see how Cormag ends up--is he better enough than Vanessa that he's worth using over her despite not being able to triangle attack Fomortiis?
Actually I forgot about that. In chapters 1-8 you don't need Vanessa to fight at all, which actually saves time since you don't have to invest 28 levels into her over the course of the run (1 -> 20 then 1 -> 10) since Cormag can instantly be converted to a godly unit in the phantom ship.
In the RTA run I basically have Cormag solo the phantom ship and feed him 6 manipulated level-ups, which is easy considering you control the enemies halfway through turn 2. At level 15/1 his stats were comparable to Seth, aside from lackluster luck and res, which really don't matter in a TAS. I could even manage with fewer levels since pierce crits truly are broken. The only issue I had with him was getting javelin hits to land against multiple fighters in C15, but that's trivial to manipulate in a TAS.
All to say, he easily replaces Vanessa on Ephraim route, and does so without having to suffer through a period of earlygame suckage.
Link to videoLink to video
These are the new real-time WRs for both routes. Notice that Ephraim route is 3 minutes faster, although in theory it should be roughly 4 minutes.
I don't want to get into the specifics unless anyone is curious, but being able to 3 turn the Phantom Ship using the torch glitch makes C11 Eph about 1 minute faster than its Eir route counterpart, which is what ultimately makes Eph route decisively faster.
Now because of this, I believe that it's now possible to TAS Sacred Stones in under an hour, which warrants an updated TAS. To get sub hour requires saving 8 mins 51 secs, 4 of which are already accounted for through the route change. The other 4 would come from starting the run from a savestate (probably just use the published TAS as a verification movie), as this allows units to move twice as fast when holding A. I can easily see this shaving more than 4 minutes, and combining this with better strats & optimization, I should realistically be able to meet my goal.
So, a couple basic questions:
1) Is anyone bothered by using a savestate to increase unit movement speed? It won't affect the RNG at all since the RNG seed is always 8,56,21 after a reset.
2) Should Eirika and Ephraim route be separate categories, or just go with whichever happens to be fastest?
3) Is there any actual interest in seeing this improvement or should I not even bother? I know these FE runs tend to have very niche audiences and all.
At any rate, if I do work on this, progress will be very gradual since I'm pretty busy these days. I certainly won't be concocting a run in ~1 month like I did with my last Sacred Stones TAS (the currently published one). Probably more like a few months to a year or so.
I don't care whether the game gets technically finished or not, from the encode it clearly appears that Cheetahmen dies on the final boss and restarts at stage 3, then the encode abruptly cuts off.
Link to video
I could understand if Cheetahmen dies and you reach a credits screen or something, but that's clearly not the case.
As for the whole initial state of hardware controversy, I actually don't mind it and agree with this post:
However, if it can happen on an actual cart, via power cycling, I don't care what initial register state is possible vs which was used. We'd know its possible to get a certain outcome from certain hardware, so we have to emulate that as best we can. Maybe the IC is in some glitched metastable state. Emulating that faithfully isnt even sensible. But emulate it we must, as best we can, and if that means by arbitrarily selecting startup values that match known behaviours (as we do for RAM) then so be it.
On the other hand, I fear this sets a bad precedent if every movie on the site could theoretically be obsoleted by those clever enough to find the best initial hardware state for their TAS, which increases the amount of technical knowledge required to make a TAS. We should be looking to decrease the learning curve for new TASers, not increase it, make it as accessible as possible.
I found a Path of Radiance TAS WIP by accident on NicoVideo. It was so good I had to upload it to youtube (see below).
Link to videoNicovideo link- Click where it says “mylist/38311285” in the video description for the full playlist
Braster’s user page- The guy who uploaded the video. It says he “started a broadcast” a bunch of times in his description, could mean he streams his FE9 TASing.
The run is on the Japanese version and on maniac mode difficulty. In addition to the usual path drawing RNG manipulation, I noticed the RNG was also manipulated by pressing Y, which goes to the chosen character’s menu (where you see their portrait, stats, etc). In fact, this was used to manipulate BEXP level-ups in the base, which is very encouraging. The run also uses the 255% crit glitch, a glitch where slim weapons & thunder tomes get maxed out crit if you make a forge with reduced crit. This makes Mia extremely valuable, as vantage + 100% crit = carnage.
Unfortunately the WIP ends on chapter 10 and before Marcia is used, it really leaves me wanting more. My motivation to TAS 9 and 10 is now in overdrive now that I know what the final result looks like. I could seriously see either becoming a TAS of the year candidate.
I definitely want to get into contact with the maker of the run since clearly that person has cracked the game’s RNG and at the very least knows the memory addresses that affect the RNG. From what I could tell the run looks very optimized already, so maybe I’ll just wait for him to finish it and encourage him to submit here.
I don't want to get into the specifics unless anyone is curious, but being able to 3 turn the Phantom Ship using the torch glitch makes C11 Eph about 1 minute faster than its Eir route counterpart, which is what ultimately makes Eph route decisively faster.
Now because of this, I believe that it's now possible to TAS Sacred Stones in under an hour, which warrants an updated TAS. To get sub hour requires saving 8 mins 51 secs, 4 of which are already accounted for through the route change. The other 4 would come from starting the run from a savestate (probably just use the published TAS as a verification movie), as this allows units to move twice as fast when holding A. I can easily see this shaving more than 4 minutes, and combining this with better strats & optimization, I should realistically be able to meet my goal.
So, a couple basic questions:
1) Is anyone bothered by using a savestate to increase unit movement speed? It won't affect the RNG at all since the RNG seed is always 8,56,21 after a reset.
2) Should Eirika and Ephraim route be separate categories, or just go with whichever happens to be fastest?
3) Is there any actual interest in seeing this improvement or should I not even bother? I know these FE runs tend to have very niche audiences and all.
At any rate, if I do work on this, progress will be very gradual since I'm pretty busy these days. I certainly won't be concocting a run in ~1 month like I did with my last Sacred Stones TAS (the currently published one). Probably more like a few months to a year or so.