08969: area51, maxforce: R3000 high CPU load

ID	Category [?]	Severity [?]	Reproducibility	Date Submitted	Last Update
08969	Gameplay	Major	Always	Nov 17, 2024, 03:31	13 days ago

Tester	peterson1965	View Status	Public	Platform	MAME (Official Binary)
Assigned To		Resolution	Open	OS	Windows 10/11 (64-bit)
Status [?]	Confirmed			Driver	atari/jaguar.cpp
Version	0.271	Fixed in Version		Build	64-bit
		Fixed in Git Commit		Github Pull Request #

Summary	08969: area51, maxforce: R3000 high CPU load
Description	AREA 51 R3000 AND MAXIMUM FORCE SINCE 0.247 AND UP. 0.246 IS FINE AREA 51 WORKS PERFECT ON CLONES AND AREA51/MAXIMUM FORCE DUO .256 mame area51 -bench 90 Average speed: 1296.88% (89 seconds) mame maxforce -bench 90 Average speed: 957.33% (89 seconds) .271 mame area51 -bench 90 Average speed: 224.21% (89 seconds) mame maxforce -bench 90 Average speed: 214.23% (89 seconds)
Steps To Reproduce
Additional Information
Github Commit

Flags
Regression Version	0.257
Affected Sets / Systems	area51, maxforce

Attached Files

No.22465 Robbbert Moderator Nov 21, 2024, 11:40	My computer is much slower than yours it seems. >mame area51 -bench 90 Average speed: 104.75% (89 seconds) >mame maxforce -bench 90 Average speed: 105.50% (89 seconds) Sadly, it's the nature of MAME to get slower and slower... and slower. This is why so many other emulators are eating its lunch. But perhaps there really is a problem. We'll see if anything gets done.
No.22496 VasiliyFamiliya Tester Nov 25, 2024, 08:33	I think it may have some connection with a speed saltations to the opposite side - for example, in the Total Vice recentry. Or, for exemple, few days ago I found that Gunblade NY (i.e Sega Model 2B-CRX) runs faster than on previous official releases - around 50~70%. Did an official release builders have any update of GCC or any other devtoolkit elements?
No.23129 hap Developer 13 days ago edited on: 13 days ago	FWIW the commit that caused the slowdown: https://github.com/mamedev/mame/commit/4af341e49bacaa899ed95345f11874fee8dd726d It looks like cojagr3k_main_speedup_r doesn't work anymore since it reads from cache instead of from the addressmap? The driver needs a different way to detect the idle loop.