02960: speglsht: Regression in Super Eagle Shot, looping broken attract mode

ID	Category [?]	Severity [?]	Reproducibility	Date Submitted	Last Update
02960	Graphics	Critical (emulation)	Always	Feb 24, 2009, 14:13	Feb 19, 2010, 13:27

Tester	incog	View Status	Public	Platform	MAME (Official Binary)
Assigned To	Mamesick	Resolution	Fixed	OS	Windows XP/Vista 32-bit
Status [?]	Resolved			Driver	seta/speglsht.cpp
Version	0.129	Fixed in Version	0.136u4	Build	Normal
		Fixed in Git Commit		Github Pull Request #

Summary	02960: speglsht: Regression in Super Eagle Shot, looping broken attract mode
Description	Looping broken attract mode, shows checksum error and shows an error on inserting a coin. Works in 0.128, Broken in 0.129
Steps To Reproduce
Additional Information
Github Commit

Flags
Regression Version	0.128u4
Affected Sets / Systems	speglsht

Attached Files	speglsht.diff (1,054 bytes) Feb 7, 2010, 14:25 [Show Content]
Attached Files	real_fix.diff (1,580 bytes) Feb 8, 2010, 17:05 [Show Content]

No.03817 Phil Bennett Developer Feb 24, 2009, 18:56	Reducing the MDRV_QUANTUM_TIME frequency fixes it. This suggests to me that there's a missing CPU sync mechanism.
No.05647 M.A.S.H. Senior Tester Feb 7, 2010, 14:25	I fixed the regression in Super Eagle Shot. INTERRUPT_GEN( irq4_gen ) was changed in drivers\speglsht.c from cpunum_set_input_line(machine, 1, R3000_IRQ4, ASSERT_LINE); to cpu_set_input_line(device, R3000_IRQ4, ASSERT_LINE); The same changes was also made for srmp5 and this game has the same problems since MAME 0.129u2, in this version it was fixed. I used the fixes also for Super Eagle Shot and it works! Use attached DIFF file.
No.05648 Haze Senior Tester Feb 7, 2010, 14:35	Both those lines should do the same thing. the R3000 is CPU1. irq4_gen is called by CPU1 (the R3000) so I don't understand why the behavior is different.
No.05653 Haze Senior Tester Feb 7, 2010, 18:33 edited on: Feb 7, 2010, 18:36	actually, I see what's happened. Your description doesn't match your DIFF! The 'fix' you've posted allows it to clear the IRQ line, so that it isn't always ASSERTED. This is very different to the one line change you suggest is responsible in the report. Your fix is hacky tho, really, there should probably be an IRQ ack, or the line should be cleared after a certain period, or possibly when the interrupt is taken, automatically.
No.05663 Mamesick Senior Tester Feb 8, 2010, 17:07	Fixed. Easier than I expected. I'll submit for next release. Note that same mechanism applies to srmp5.c too, where I removed the hack and implemented correct IRQ ack.