Posts for TiKevin83


1 2
9 10 11 12 13 14
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Yes both cart revs 1.0 and 1.1 have now synced between interim BizHawk-Gambatte and the GBP console. Here's the resynced v1.0 in 12:16.12. http://tasvideos.org/userfiles/info/61577240561588605 And here are links to the new interim bizhawk which also requires a new prereq install https://ci.appveyor.com/project/zeromus/bizhawk-udexo/build/artifacts https://github.com/TASVideos/BizHawk-Prereqs/releases/tag/2.4.8_1 The full process to console verify on GBP is explained here https://pastebin.com/DXiDT9ZT
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
I've resynced to the new emulator for the 1.0 rev now as well. I should be getting a cart to officially verify it tomorrow. The original 12:08.75 as Mugg said is a 12:12.07 timed with the correct framerate, and 186 frames were added from the GBC BIOS and 48 frames from 4 frames x 12 levels from VBA inaccuracy. Accounting for those puts it at 12:15.99, and the resync came out to 12:16.12. There were ~8 lag frames added in 4-1 which would explain the remaining discrepancy, but there were also 10 frames of delay needed to sync the bonus levels. The time from those 10 frames was probably saved by the LCD off timing problem. So I'm pretty confident a resynced time should be 12:16.12.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
This movie has been resynced to cart rev1.1 which lacks the screen wrap glitch using the interim bizhawk builds for console verification testing: https://youtu.be/_68eQzDKu8k http://tasvideos.org/userfiles/info/61468991527972019 It'll need a lot more polish and a full bizhawk release before attaching it to a movie/new submission if we can justify any true improvements. The run is 12s longer at 12:20.6 due to VBA running 4s too fast at full 60fps, 3s to load the GBC BIOS, 4-4.5s to the lost glitches between cart rev 1.0 and 1.1, and then 48 frames for 4 frames delayed per level x12 levels as part of the VBA inaccuracy.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Natt was around in IRC and walked me through how he set up BizHawk-Gambatte's savestates, and they appear to be fixed properly now
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Ok thanks for pointing me in the right direction. I was able to play back the known good Blue, Yellow, and Crystal Glitchless movies as of the latest commit, but yeah savestating changes appear to have broken, probably with respect to the sound channels.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
I have a fork going which pulls in Sinamas' refactor of Gambatte and other improvements from Gambatte-Speedrun. I got BizHawk's libgambatte building with the changes over the weekend but it's currently booting to only a white screen in BizHawk. If anyone wants to help debug it would be greatly appreciated. https://github.com/TiKevin83/BizHawk
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
If you read back that quote it explains why Gambatte opetates at half the 4194304. The GB is memory bound so Gambatte is only counting memory-bound cycles, which is 1/4 the rate, but the GBC rate is twice the GB rate, so it has to count at half the rate overall.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Is it off by a similar factor to before or is it off by a factor of 2/4? Can you link me to the commit where it was implemented?
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Big congrats merrp. Excellent level of thoroughness for such a TAS, and I love when TASing helps RTA.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
There would need to be some more emulation updates and then a resync with a setup movie that clears save data to a constant state between EMU and Console, so I believe it's beyond the immediate scope of this submission
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
The glitch was found while hunting for the cause of some desyncs, so hopefully gifvex and entr can push the fixes for the desyncs into bizhawk's gambatte before new TASes.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
There are some pretty serious bugs with Gambatte in 2.3.3. Emulation->reboot core is giving me bad sync (whereas double clicking frame 0 to reset the movie will play back in line with console). Also, my export script to generate emu.totalexecutedcycles() on each frame in LUA is broken, presumably because it's being hijacked by the new feature to try to dump the proper movie length in cycles into bk2s.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Thank you for the feedback everyone. I'm glad to hear some appreciated the return to speedrun oriented TAS content and not just ACE payloads. For those expecting a surprise, I imagine we'll try to balance payloads and speedruns at future events. We plan to have an SGDQ TASBot prep thread here again to gather suggestions.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
To clarify the thing with going over estimate, dwango was used to submitting exact run times as estimates, but they round estimates down to the minute now and suggested that for the future we submit estimates of total showtime and not just the TAS playback.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
On Reddit the feedback demonstrated some of Arc's concerns but also why we did what we did for the year. If we had just shown Mario Maker 2 people would've complained about it being too short and not Speedrun related. If we just did the TAS block I organized it would've been criticized for being the same games we always see.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
We don't choose whether games are incentives or what donation goals are set for them. The event went very smooth and I thought we nailed our attribution standards
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Dacicus, negative_seven, and I (TiKevin83) will be teaming up as team "A Team Name" - GLHF everyone!
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Requesting to be added to a team, good luck everyone!
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
The RTC isn't too critical because it's a function of the cartridge and we've observed that Crystal cartridges do not all have the same RTC error. MrWint handled this in Gambatte by adding a variable for RTC speed error to the movie file, not sure how you'd want to handle it in GBHawk
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Thanks Alyosha, When you get done with that I'll check all of the testROMs I know of now that they display properly so we can see how close things are to being able to expand beyond Pokemon games with Console Verification.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
All h.264 is YCbCr internally anyway, YUV444 just signals to keep that encode 444 and let YT handle going down to 420. CRF 0 is the same as QP 0 (both lossless, no quantization). CRF and QP do similar things in lossy modes, CRF sets a target QP and the encoder fluctuates around it instead of being stuck to that exact QP.
Post subject: GB/GBC/GBA Encoding
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
I have some suggestions for GB/GBC/GBA encoding based on experience in making this set of 4k encodes for YT from Console capture, with the captures being 1:1 with emulator the same ideas should apply. https://www.youtube.com/watch?v=Oo6HcU0iP4c&list=PLH-rsHQS9LqjDJCGTuLF7YM_FrcsTW9qz Even Integer Scales for Color First and most importantly for quality, I've been sticking to even integer multiples when scaling. YT doesn't have full resolution color, they do 4:2:0, so 15x causes the color to blur at the transitions between original pixels. Then 15x at 4k is 7.5x at 1080 and 5x at 720, so the lower resolutions are also affected. 12x at 4k goes to 6x at 1080 and 4x at 720, so no harm from 4:2:0 at those resolutions. The 360p30 size is also a 2x scale with this setting so it's only impacted by the framerate loss and not color blur. Some might prefer the picture to fit the full screen, IDK I prefer avoiding blur and letterboxing the empty space. Applications for lossless video Now also I was talking with Spikestuff about whether to use lossless or a high quality CRF between 7 and 14. When doing these extreme integer multiple scales of pixel art games, the differences between lossless and lossy aren't as dramatic, and the output size is mostly affected by audio and not video. At only a 4x scale to 720p, my quad core machine did a lossy pass of Yellow NSC at 410fps and lossless at 355fps, with the output size being 14MB for lossy and 50MB for lossless (video only). At a 12x scale for 4k the same machine does 56fps for lossy and 40fps for lossless, 148MB and 191MB. This is for 12 minutes of video: lossless 4k footage of 12x integer scale GBC is only ~1GB/hour. It seems in this weird use case, lossless is not actually substantially worse than lossy for filesize or encode speed. The filesize is still 3x worse at 720 with a 4x scale, but 100MB and 300MB an hour are both very small for video (lossless audio would be larger). This is the ffmpeg line I used to do the upscales for YT along these guidelines. The GB/GBC footage was already letterboxed to 240x160, which worked fine with them both having the same 12x limitation at 4k. ffmpeg -i input.avi -vf scale=2880x1920:flags=neighbor,pad=3840:2160:(ow-iw)/2:(oh-ih)/2 -c:v libx264 -crf 0 -c:a copy -pix_fmt yuv444p output.mkv
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
Here's an updates video of the TAS in 2:44.07 https://youtu.be/0iSRqA4T5wo I played it back without the input viewer once this time so it would play at full speed, then again showing the new inputs. One other thing, I forgot to mention before that the TI-83 appears to have consecutive input lag; there are some places in the speedcode where the "right" key is pressed to reduce lag between two presses of the "(" key.
TiKevin83
He/Him
Ambassador, Experienced Forum User, Moderator, Published Author, Site Developer, Player (120)
Joined: 3/17/2018
Posts: 348
Location: Holland, MI
http://tasvideos.org/userfiles/info/57692685003754860 Updated submission Frame counts comparison: Speedcoding 2370 2154 Reordered a bunch of code and moved some things out of loops that could be done with list math. Levels 1->9 2699 2423 3310 2912 4045 3473 4576 3851 5581 4594 7143 5757 8197 6500 10075 7841 12845 9843 Fixed all the RNG with the new code efficiency, saved 3002 frames in total. I also discovered a previously undocumented use of DelVar, it can be part of a single line if statement. The pastebin with the related code has been updated, line 27 shows the new tech: https://pastebin.com/TMu6pyiB
1 2
9 10 11 12 13 14