I took a look at the startup issues in Thunder Force III, and found that it was crashing on the first vertical blank interrupt, which was set to vector into RAM. I found that TF3 (and TF4 for that matter) check the control registers to see whether they are doing first init or not, and skip a lot of set up in they are not. For TF3's sake, the result was a fatal crash.

Adding this code into the trampoline function fixed TF3:

move.w #0,$A11200	; put the Z80 (and YM2162) in resetmove.w #$8004,$c00004	; VDP R0 = 04 (disable IE0)move.w #$8104,$c00004	; VDP R1 = 04 (disable IE1)move.w #$8B00,$c00004	; VDP R11= 00 (disable IE2)clr.l $A10008		; TF3, at least, uses these registers to decide whether to do a full init!clr.l $a1000c		; zeroing them should force a proper initialization

I also added the TMSS code from the TMSS ROM -- basically checking if the Genesis has security hardware, and zeroing the security register if so, which is what the TMSS BIOS does. But I don't have a machine I can test that on... maybe it's not necessary.

Sadly, the above code patch did not fix TF4's lockup, so there is more to do there. Also, I looked more closely at TF2, and it shows two odd behaviours that the emulator does NOT -- for one, it always thinks it's in Japanese mode (even though the other games do not), and for two, it locks up when the game is started. But these two things happen only on hardware, and I don't have an easy way to test that right now, so I will next need to look at TF4.