Broken games or graphical issues report

[davidgfnet]

Hey folks! Great work you are doing here :)
To help you a bit with finding games that have issues or are broken, I ran mgba with your BIOS and the official one and compared them. I run it for 9000 frames and then compare the outputs. There are obviously many differences since each BIOS takes a different amount of cycles and causes the game to drift slightly, but those can be ignored. Only cases where the output is obviously wrong (for instance FIFA 2004 or Doom II), or the game seems to hang/crash (for instance ZooCube, where it reboots to BIOS, or Super Street Fighter which seems to hang somewhere) are interesting for debugging.
Here goes the full report: https://davidgfnet-cultgba-mgba-testing-1c75d903.netlify.app/ (48MB!!!)

I could also classify the games on the SWIs they use automatically if you find it can be useful (might be easy to spot a pattern!), or I could even do automated A/B testing of the different SWI calls (by running both BIOSes and using the official one for some SWIs and yours for others).
Hope it is helpful :)

[skylersaleh]

This is pretty awesome!

When I was fixing bugs in other games, a large chunk of the effort was just figuring out what SWIs were broken. So, if you can automate that, it would be incredibly helpful.

Another, thing that could be really useful to automate is if you could generate a summary for each of the SWIs, how many games use that SWI (both working and non working), and how many broken games use that SWI.

[davidgfnet]

Hey there!

So I can generate something like a list of SWIs that a game uses. (I just run it with the official BIOS so that it works well) This won't cover them all, since I do not play the game (just some dummy inputs to get past most start screens but not a lot, see https://www.davidgf.net/2021/08/15/miniretro/ -> there's videos at the bottom if you dont feel like reading and wanna see action). This should be incomplete but already very much useful for many games. Given there's so many games it's probably good enough if you miss a few SWIs here and there.

With this information, we could already have some sort of idea of what the problematic SWI might be, right? However the "perfect" test would be to run both BIOSes, just using yours for one particular SWI number. Comparing results we could know if the SWI causes trouble or not. The only issue is that we need N*M tests (N being game count and M SWI count). Also might need a bit of work to enable some sort of dual-bios mode in mgba or similar emulator.

So, should I start with the low hanging fruit and get you a SWI list per game?

[skylersaleh]

Just having the SWI list per game would be a great first start. If we have that, we could at least prioritize fixing the SWIs that will give us the most game coverage.

[davidgfnet]

The cpuset/fast routines must have some interesting bug (eyeballing I'd say the bits you check are wrong? According to tonc it's bits 24 and 26 that have a meaning), since games that only use those ones fail with your bios.
Here you go a CSV: https://pastebin.com/86Bw3Zjb

[fleroviux]

Hey. Thanks for your effort here! And sorry for the very late response.
The code should actually use bits 24 and 26 unless there's a bug there that I'm overlooking right now. The comment about bit0/bit1 is about the function table index that the code builds.

[davidgfnet]

Ah you are indeed right! Missread it!
I guess the only thing that might be missing is clearing out bits to ensure that other non-zero bits break that function. (As the TODO says)
Take a look at the csv, theres a couple of games that use these SWIs only and are broken, which indicates theres something going on there. Perhaps they rely on some weird BIOS behaviour (for example they might assume that r0-r2 values after the call follow some pattern).
Glad to help!

[skylersaleh]

Thanks!

I parsed your data to create a histogram showing what the most important SWIs to get correct are.

SWI #	# of games that use this SWI
0xb	778
0x5	551
0xc	524
0x6	409
0x11	394
0x12	361
0x1	235
0x2	162
0xf	128
0x8	114
0xa	106
0xe	91
0x13	79
0x15	78
0x14	63
0x7	31
0x4	21
0x3e	7
0x3d	7
0x3c	7
0x3b	7
0x3a	7
0x39	7
0x38	7
0x37	7
0x36	7
0x35	7
0x34	7
0x33	7
0x32	7
0x31	7
0x30	7
0x2f	7
0x2e	7
0x2d	7
0x2c	7
0x2b	7
0x2a	7
0x29	7
0x28	7
0x27	7
0x26	7
0x25	7
0x24	7
0x23	7
0x22	7
0x21	7
0x20	7
0x1f	7
0x1e	7
0x1d	7
0x1c	7
0x1b	7
0x1a	7
0x16	6
0x18	4
0xd	4
0x17	3
0x9	3
0x10	2

[davidgfnet]

I only recorded SWIs in the [0, 64) range, since i used a 64bit bitmap. It is possible that there are more weird 64+ calls, but AFAIK they do ..nothing? so we should not worry. It's surprising that there are games doing that, perhaps they have bugs or it's an emulator bug too.