Talk:GB/C Tests

From Emulation General Wiki
Revision as of 17:32, 4 July 2018 by Syboxez (talk | contribs)
Jump to navigation Jump to search

Hardware and SameBoy results

Does someone mind adding these results to the detailed table? The first 3 are real hardware.

http://www30.zippyshare.com/v/thAaNQfJ/file.html

Testing also needs to be done with other emulators to separate DMG mode and CGB mode, along with separating DMG_Sound and CGB_Sound tests in the first (simple) table. Right now some emulators seem "more accurate" than even real hardware due to this. --Syboxez (talk) 15:39, 5 March 2018 (EST)

I'll add the hardware tests leftmost to the table. I don't know how the other results are categorized (emu release date? test date?), so I don't know where to put in the SameBoy results. FosterHaven (talk) 16:13, 5 March 2018 (EST)
"Some emulators seem more accurate than even real hardware" probably just means the grade system is flawed. Perhaps make two separate grades for DMG and CGB accuracy? Omit the CGB Sound tests from the DMG grade, omit the DMG sound test from CGB grade, etc. I could make a new type of cell along the lines of "Skipped" if emulators have certain modes for different models and the test is for model accuracy. FosterHaven (talk) 16:41, 5 March 2018 (EST)
Added BGB and Gambatte tests as well and added version numbers: http://www62.zippyshare.com/v/LZKQaSJj/file.html
All tests were performed today (5 March 2018). I don't think we need emu release date or test date if we have version numbers (the tests haven't changed). I also don't think we should omit CGB_Sound and DMG_Sound tests, but we should consider that if the emulator fails in the same way as real hardware, then it counts as a pass (with a note making that obvious, perhaps a green fail?). I modified the gambatte source code to force CGB mode, and it already had a force DMG option. --Syboxez (talk) 16:58, 5 March 2018 (EST)
Also BizHawk is redundant, since it is just Gambatte r344 (Source: http://tasvideos.org/Bizhawk/GB.html) --Syboxez (talk) 17:17, 5 March 2018 (EST)
I think the tests for all emulators need to be redone to include the difference between DMG mode and CGB mode (if that emulator distinguished from DMG and CGB). --Syboxez (talk) 17:20, 5 March 2018 (EST)
So I guess new results are added to the right then? What do we do with TASVideos' results in the meantime? Their page says it was last modified a year ago. Perhaps retest the emulators as they are now or test the recommended emulators at the Game Boy page? FosterHaven (talk) 17:40, 5 March 2018 (EST)
Yes. New results are to the right (real hw and SameBoy are unchanged). I think we should scrap TASVideos' results entirely and redo the tests to account for CGB vs DMG.
Also worth mentioning these tests: http://gbdev.gg8.se/wiki/articles/Test_ROMs
I don't have an MGB, so I can't test on that. Their results seem to match mine (DMG_Sound had a black screen on non-DMG hw and emus in CGB mode, so I ran them individually) --Syboxez (talk) 17:49, 5 March 2018 (EST)

For those that would like to repeat the Gambatte CGB tests, download the source code, and in libgambatte/src/mem/cartridge.cpp, change the line "cgb = header[0x0143] >> 7 & (1 ^ forceDmg);" to "cgb = true;", then compile it. --Syboxez (talk) 17:10, 5 March 2018 (EST)

I did that, but in CGB mode for example Mooneye testsuite doesn't even show any graphics. I suppose that there must be some easy fix, any help? (Also, it's good to set it to "cgb = (1 ^ forceDmg);", so the force DMG option still works ;) --Sczther (talk) 10:05, 31 March 2018 (EDT)
I'd recommend using the libretro core which already has the option to force different hardware. --Syboxez (talk) 11:18, 31 March 2018 (EDT)

Should we even bother keeping the top (simple) chart? It is duplicate information as the more detailed one right below it and the accuracy percentages are misleading. --Syboxez (talk) 19:05, 5 March 2018 (EST)

I removed it. If someone feels like it is useful information, then reverse the change and try to separate DMG and CGB into two seperate tables (with AGB going along with CGB). We should judge each emulator based on what hardware revision it is trying to emulate. The PS1 only needs one chart since there is only one model to emulate, where all revisions of the original hardware pass all tests (not including PS2 and/or any official emulation). I also think we should change the color for expected failures (ie DMG failing CGB tests and vice versa) to a unique color (blue) with a color legend. --Syboxez (talk) 20:56, 5 March 2018 (EST)

MAME 0.193 (Ubuntu PPA) results

I've tested the linked blargg ROMs with my copy of MAME 0.193 and the results are pretty close to higan's.

CGB Sound
1-7 passed
8-10: 01
11: 04
12: 02
CPU Instr
1-11 passed
DMG Sound
1-12 passed (mame said the driver's sound was imperfect. perhaps they were talking about CGB?)
Halt Bug
failed
Instr Timing
pass
Interrupt_time
failed
Mem Timing 2
1-3 passed
OAM
1 ok
2 2
3 ok
4 3
5 2
6 ok
7 1
8 2

Some of these tests are not the same, like DMG Sound (the table shows results for DMG Sound 2). Was it just not renamed? Maybe each group in the table could link to the proper test? FosterHaven (talk) 17:40, 5 March 2018 (EST)

Judging by these results, I'm guessing all tests were done in DMG mode. Does MAME have CGB support? I didn't include the halt_bug test because it wasn't on the TASVideos page, although it might be worth including anyway. Eventually we should add other tests besides blargg's, but that can wait until this page is near-final (mainly because I don't feel like doing that many tests myself :P). Also I think the DMG_Sound test is V2, but I cannot confirm. Maybe you can ask blargg? --Syboxez (talk) 19:33, 5 March 2018 (EST)
Your guess is correct, the results were from the gameboy driver, which is DMG. Their CGB driver is called gbcolor, but I haven't tested it though. FosterHaven (talk) 20:14, 5 March 2018 (EST)
As for the claimed sound imperfections, tests aren't always indicative of accuracy, passing or failing. Tests only show one very specific function of the system. Perhaps other tests would show different results (Mooneye, Gambatte, etc.) I'd also be interested in looking at the accuracy of FPGA cores (like kevtris's Game Boy and Game Boy Color cores in the NT Mini), although I don't know if that's the scope of this wiki. Maybe we should have a page dedicated to FPGA emulation and options relating to that. --Syboxez (talk) 20:37, 5 March 2018 (EST)

Removing old and obsolete emulators

Hello, I do think that many of these are currently used by completely no-one and are just wasting space, for example vba-rr(vba-m and bizhawk cover the uses), DMGboy and nintemulator. Others, like VBA, no$gb, TGB Dual, VGB and gearboy are mostly undeveloped, or inferior to other options. It would be beneficial to add used emulators that offer new features, as those are important to have judged, for example GBE+ of HhugBoy.


New SameBoy tests

Sameboy v0.11 now passes the DMG register tests for CGB Sound. I haven't done anymore tests due to lack of time, but it's worth redoing them. -Syboxez

Just ran the tests. I believe this now makes SameBoy the most accurate GameBoy emulator due to it passing all the OAM tests. Should we recommend it above all the others?