08530: zombraid: Graphics corruption

ID	Category [?]	Severity [?]	Reproducibility	Date Submitted	Last Update
08530	Graphics	Major	Always	Dec 3, 2022, 12:07	Dec 11, 2022, 14:17

Tester	sgpowers	View Status	Public	Platform	MAME (Official Binary)
Assigned To	Haze	Resolution	Fixed	OS	Linux (64-bit)
Status [?]	Resolved			Driver	seta/seta.cpp
Version	0.250	Fixed in Version	0.251	Build	64-bit
		Fixed in Git Commit	a8c82d8	Github Pull Request #

Summary	08530: zombraid: Graphics corruption
Description	First reported in https://github.com/mamedev/mame/issues/10626 Moved here by request. Regression caused by: https://github.com/mamedev/mame/commit/7cbb49c5c77270898f58ed67238eb54014aa9a0d Ported Notes from GitHub issue (DavidHaywood): " I suspect the problem is the overall display timing is wrong screen.set_vblank_time(ATTOSECONDS_IN_USEC(0)); is completely unrealistic for example" "although changing that makes no difference, so it might be telling us something else"
Steps To Reproduce	Start `zombraid` in 0.250
Additional Information
Github Commit

Flags
Regression Version	0.250
Affected Sets / Systems	zombraid

Attached Files	zomraidcorrupt.png (1,982,909 bytes) Dec 3, 2022, 12:07 Uploaded by sgpowers Example corruption

No.20894 hap Developer Dec 3, 2022, 19:08	Maybe some of the regs don't have effect on video hardware until the start of the frame. Whereas xscroll would have effect on the next scanline. In that case, see my earlier advice on only updating screen when xscroll is written :\|
No.20895 Haze Senior Tester Dec 3, 2022, 20:06 edited on: Dec 3, 2022, 20:12	Only updating the screen when xscroll is written is a hack though, as real hardware draws every scanline based on what it thinks the current values are, so nothing should technically break if you force partial updates, even on every line. Not just drawing every line as a partial scanline update is simply a MAME optimization that ends up masking some other bad assumptions. You'd want proper latching behavior of certain registers at the start/end of frame if some values are only read once per frame. Partial updates here is simply exposing the timing of something else being incorrect, or a value that should be latched not being latched, or maybe even an entire extra sprite RAM buffer present on these later hardware games that isn't on the earlier ones.
No.20897 Haze Senior Tester Dec 3, 2022, 20:27 edited on: Dec 3, 2022, 20:28	I'm not opposed to a hack to disable the partial updates in this case, but it would just be masking a real problem. However, given that 99% of the drivers in MAME mask real problems by only updating the display at a single point in a frame I'm not really sure that would be such a big deal. This is one of the things MAME gets away with, whereas FPGA based solutions, which to my knowledge HAVE to work as proper per-pixel,per-scanline renderers, do not. It would be interesting to see if any research on these hardware types ends up coming from that direction.
No.20907 Mamesick Senior Tester Dec 4, 2022, 07:45 edited on: Dec 4, 2022, 07:45	FWIW, this commits also makes BLANDIA look worse than before, see expecially Athena stage. STRIKE GUNNER STG intro seems also bugged, with a lot of flickering at middle of the screen.
No.20908 Haze Senior Tester Dec 4, 2022, 11:51 edited on: Dec 4, 2022, 11:51	Alright, those are useful datapoints too then. I suspect either the way the chip does buffering in MAME is incorrect, or the later games have some additional external buffering. I may just go with the suggested hack for now, as there really isn't enough evidence to draw better conclusions, but should MAME one day decide to enforce more realistic handling (updating every scanline rather than every frame) by default, the issue will no doubt come back.
No.20938 Haze Senior Tester Dec 7, 2022, 17:24	FWIW even restricting it to xscroll writes isn't enough, the timing is so out on the other games.
No.20942 Haze Senior Tester Dec 10, 2022, 15:29	fixed with a hack (only doing partial updates for caliber50 rather than for everything using the chip) there are definitely underlying timing issues with these games that will need resolving one day.
No.20943 sgpowers Tester Dec 11, 2022, 00:50	Manually applying a8c82d8a86a8af04b81434e283d14cedec2cb792 - looks good to me.
No.20946 Haze Senior Tester Dec 11, 2022, 13:21 edited on: Dec 11, 2022, 13:22	honestly it's a shameful fix, but it just restores the behaviour to what it was before the raster effect support was added, so yeah. MT probably needs a 'worked around with awful hack' resolution ;)