Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Brock Lee wrote:
I finished making a tas two player game of Metal Slug and I'm trying to convert it to AVI but it slows down a lot. I know sometimes the game itself does, but the videos does it in so many spots (the bosses are the worst). Is there any way to fix this?
Use emulator built-in dump (if you can get that to work). That gets the speed correct (the emulation slows down if it can't dump fast enough).
For maximum quality, use some lossless video codec like Camstudio codec.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Brock Lee wrote:
I have the video of that mission on FBA and it goes until it originally froze. Is there any way to continue off that movie, or is my only option to make a new movie from that point and later try to slap the two together?
If FBA works anything like other emulators, load the movie in readonly mode, play it to the end and switch to readwrite mode. That should allow you to continue from that point (make a savestate there?)
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
jlun2 wrote:
I looked at the page of your WIP, and assuming the time is half of the actual time based on your encodes, it's 1:08:45.10 by the time you end your WIP.
Well, it is not quite half, but should be close. Dolphin timing is screwed, it would take dump with AVsync hack or some memory watching (of course, having to find the address first) to figure out the true length.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Reportedly Kaizo Mario World 3 will crash on anything accurate-for-sound after defeating the final boss .
What would happen if defeated fast enough not to crash the sound processor I don't know/recall, but the fastest method known is not even close.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Mr. Pwnage wrote:
Your finish time is enough to conclude that Supaplex runs at 70fps (with a move taking 8 frames), while RnD is stuck at 50, so all the times I set get tagged with a 40% penalty.
Quite close. 125875/1796 fps [70.08(63028953229398663697104677060133) fps] to be exact.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
STBM wrote:
Well, I actually have to restart this run since I found a way to improve my 1-1... So I was thinking about taking the latest version of Dolphin (3.5-367) since the emulation became more accurate.
I would like to know if it was ok for the encoders or if they would have to do all the work again - in which case I would stick with the previous version of Dolphin.
Well, the AVsync changes would have to be ported to the newest version (presumably wouldn't be hard). Otherwise it would be the same for encoders.
Also, this game is fscking nightmare to encode. Sound shuttering (which isn't that difficult to fix actually), and codec poison on scale far beyond anything I have ever seen before[1][2][3]...
Oh, and given the time it takes to do another run and the nastyness of encoding, I would leave this submission alone (letting it be published)...
[1] Back years ago, Flygon used to give Da Aztex (a PC Demo) as example of bad codec poison. This just totally blows that away...
[2] Also, I tried to encode Alien Carnage HD with some funky upscaling and blending. That turned out to be so fscking poisonous I quickly aborted the encode... But even that is downright codec-friendly compared to this.
[3] I wondered before Rgamma encoded this run if this would be first run with 2GB+ encodes... Turned out that the proposed encodes are all 4GB+...
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
dleary wrote:
I am new to the TAS scene, but I am confused about this movie and the rules.
Consider HappyLee's SMB1 run, which is a 1-frame improvement over klmz's previous run.
This movie has 19 frames of blank input at the end of the fm2 file.
The movie it replaces, by klmz, has 22 frames of blank input at the end of the fm2 file.
The TAS rules here http://tasvideos.org/MovieRules.html#MovieMustBeComplete say
"It should end with the last input. Don't leave any blank input at the end of the movie."
By this measure, klmz's movie could be better... it takes 1 frame longer to beat the game, but the last input is a couple of frames earlier than HappyLee's last input.
How the endpoint is judged is bit fuzzy. For SMB1, the time to touch the final axe is counted (and that is the traditional point to cut off input with that game).
Some other types seen with other games:
- Cut off as soon as ending will be reached.
- Cut off as soon as ending will be reached fastest possible.
- Cut off when control is finally lost.
And I'm sure there are some other types as well...
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
GuidMorrow wrote:
Where do I find a setting that does a piece check after finishing the download? This way I don't have my movie interrupted partway and have to go back and check manually.
I used CFV for checking downloads against torrents. But there probably are GUI-based ones too.
GuidMorrow wrote:
It seems to be aggravated more when small movies use large (>= 256 KiB) block sizes. Movies should dynamically allocate block size based on the length of the movie, frame size and the encoding used (standard vs. 10-bit 4:4:4).
The encoding shouldn't affect it, as primary/10bit444 is often approximately the same size. And yeah, the block size should be chosen based on the file size.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
GoldFibre wrote:
I'm trying to write a script that repeatedly loads a savestate in lsnes and tries different inputs. The savestate seems to save in an array, but I can't get it to load back properly.
Savestates save into a file (or in case of unsafe rewind, to an object).
exec("save-state foo.lsmv") -- That should save state as soon as possible.
exec("load-state foo.lsmv") -- Load a state.
For unsafe rewinds, there is function movie.unsaferewind and callback on_set_rewind (manual describes more about these).
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Master of Puppets wrote:
Is accuracy of Dolphin anywhere near close to be ready for console verification?
Depends on if game polls even on lag/loading frames and what kind of RNG it uses.
If it polls on lag/load frames, forget about it (load frames are even worse, as those are just plain impossible to emulate correctly).
If it doesn't, but RNG is badly behaved (CPU cycle counts, etc...) then it won't sync.
But otherwise it might actually sync.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Hopper262 wrote:
QEMU is definitely not cycle-accurate, and from what I gather it's not always deterministic either.
The more complicated the system, the less important it is to be cycle-accurate. With PC, cycle-accuracy is only needed for some early games for the original IBM PC (and some very badly made games).
Basically, anything that played properly on much faster hardware than when it was released doesn't need cycle-accuracy.
Determinism is however very important...
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Abahbob wrote:
I recommend atleast testing it because it may change (or fix) the issue, although I know nothing on the subject.
The details of the problem: The video dump is correct, it is the audio dump that is wrong.
This game programs the DSP chip to DMA (Direct Memory Access) the audio in 160 sample chunks. Pretty normal. However, it sometimes resets the audio DMA mid-transfer.
Now, this mid-transfer reset caused previous versions of AVsync hack to lose audio. As this resetting happened quite often, this lead to noticeable desync quickly.
But it gets worse: If one tries to read the missing audio on reset, one gets garbage. And if one inserts zeroes, the sound becomes choppy. The new AVhack does the latter.
In the end, I wrote a program that identifies the missing chunks of the audio track and stretches previous audio over those chunks.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
WST wrote:
Hmm, May the weird behavior of your Sega (sometimes syncing, sometimes not) be related with instability of processing frequency of the real Genesis console? I think, because of that some lag frame may appear/disappear, and it may depend on such factors as power voltage stability (controlled by an 7805 IC) and temperature stability?..
If there is just one oscillator, everything will slow down or speed up in proportion, and since bot syncs to select pulsing, that wouldn't cause the desyncs.
If there are multiple oscillators, all bets are off...
My guess is that it might be related to undefined initial system state (memory and various kinds of internal hardware counters).
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Sir VG wrote:
Would it have been faster to take the alternate route in Stage 8 to avoid the mini-boss? Or would that have taken more time?
It would have taken more time and not avoided the miniboss.
At least levels 6, 8 and 10 have minbosses on split paths. On 6 and 10, the miniboss can be skipped by going through the other path, but not on 8 because it is there too.
Also, that miniboss is so darn annoying I would have loved to skip it. It took far far more time than any other type of miniboss (IIRC, that miniboss type takes like order of magnitude longer than anything else to TAS).
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
I verified that:
- The SHA-1 the torrent file I have matches SHA-1 of the torrent file on the server.
- The SHA-1 of video file I have matches SHA-1 of video file in archive.org collection file metadata.
- The file in the torrent pass hash check against the torrent itself.
All these checks seem OK.
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
SoulCal wrote:
Have you tried both builds on your end to see if there is a difference in the "frame" and "VI" values when you run a game? Maybe it's just RE4 being iffy? I'll try some more hexing in my dtm, but if the numbers don't match up I don't see it syncing. If we can't get a proper encoding emulator, then I don't see my run getting accepted. :(
Well, this seems to be worse than that. Useful first step would be to locate some Dolphin version with sources available that can sync the whole run.
Because if one can get sources for emulator that it syncs on, one can merge AVhack code to it, or in the worst case just hack it to dump enough info to fix the dump...
Emulator Coder, Experienced Forum User, Published Author, Skilled player
(1142)
Joined: 5/1/2010
Posts: 1217
Patashu wrote:
@GhostSonic: If you're capable of coming to Summer Games Done Quick 2013 or Awesome Games Done Quick 2014 and taking your SNES bot with you, you should look into doing it.
Unfortunately, most SNES TASes are done using snes9x 1.43, which is really inaccurate. And even snes9x 1.5x isn't really that accurate (although way ahead of 1.43 in accuracy).
So there probably aren't many SNES TASes that would sync on the real console. :-/
