08980: all sets in konamim2.cpp: Extreme slowdown when loading data from CD

ID	Category [?]	Severity [?]	Reproducibility	Date Submitted	Last Update
08980	Known Issues/To-Do's	Minor	Always	Nov 30, 2024, 20:08	20 days ago

Tester	Hydreigon	View Status	Public	Platform	MAME (Official Binary)
Assigned To	Haze	Resolution	Fixed	OS	Windows 10/11 (64-bit)
Status [?]	Resolved			Driver	konami/konamim2.cpp
Version	0.272	Fixed in Version	0.274GIT	Build	64-bit
		Fixed in Git Commit	355e240	Github Pull Request #

Summary	08980: all sets in konamim2.cpp: Extreme slowdown when loading data from CD
Description	While performance in this driver isn't the best there is, the games tend to slow down to single digit percentages (around 4-6%) when they load data from the CD. This is most noticeable in evilnght's course select screen the moment one shoots a course. This is no regression either; the issue has always been there ever since enough progress has been done to promote some sets to 'working' status.
Steps To Reproduce	Not really much. You can press f11 to display the speed and watch it go to single digit numbers when they load specific content from the CD.
Additional Information	Cuvas mentioned in a reddit post that the reason behind this was "something about what the game does causes MAME to constantly recompile the PowerPC code, which causes the performance to tank." No clue if this statement refers to the overall performance or the specific cause of the huge performance hit.
Github Commit

Flags
Regression Version
Affected Sets / Systems	all sets in konamim2.cpp

Attached Files

No.22513 JimCarlTay Tester Dec 1, 2024, 03:18	Acknowledged. I even notice major slowdown in the start of the attract mode of "Tobe! Polystars." When testing "Evil Night", I decided to check the speed by pressing F11 and due to the recompiler issues mentioned by cuavas, the game speed does not maintain a consistent 100%.
No.22514 cuavas Administrator Dec 1, 2024, 05:49	For some reason, MAME is constantly recompiling code, which is expensive. Unless you actually plan to do major work on the recompiler framework there isn't any point discussing it. Also it isn't a bug per se (just less than ideal behaviour), it isn't a crash or freeze (it's just slow), and it isn't "critical".
No.22742 hap Developer 20 days ago	Haze found the cause, and suspects it's an obsolete hack. ref: https://github.com/mamedev/mame/pull/13143 3dom2.cpp line 1923 // TODO: HACK! #if 1 m_cpu1->set_cache_dirty(); #endif