twin16: ghost sprites in miaj

[jotego]

commit: 5ae3cb8

As the main character moves around, he leaves ghosts behind. This can be seen in scene 1, where the character on the ground should not be visible. The one climbing up the stairs is the right one.

swcore twin16 ver/miaj
jtsim -s 1

[skutis]

Maybe the sprite table is not emptied correctly?

[jotego]

Should it be emptied even if the sprite pre processor is not used?

[skutis]

On real hardware the parsed sprite table is emptied, it depends on how you have implemented the parsed sprite table scanning.

[jotego]

Currently, it is only cleared when the pre-processor is used. I see that we need to clear it each time then.

[jotego]

A couple of notes from a conversation with skutis:

• objbufinit starts right at blanking. The 1-line delay was not correct
• the value used to clear the buffer is 8'h00, not MAME's 8'hff
• the CPUs cannot write to the sprite RAM while the DMA is active

The last point would require modifying jttwin16_share so oram_we is not asserted during the DMA phase. A signal should be carried from jt00778x all the wait to jttwin16_share

[jotego]

Skutis looked into this a bit more. If the CPU tries to access the object RAM while DMA is active, it will get locked because DTACKn does not get asserted for oram access if DMA is busy.

@rp-jt modify twin16_main:

jtframe_68kdtack_cen #(.W(5),.RECOVERY(1)) u_dtack(
// ...
    .bus_cs     ( bus_cs    ),
    .bus_busy   ( bus_busy  ),
    .bus_legit  ( bus_legit )
//...
);

• include oram_cs as part of bus_cs
• when oram_cs&dma_bsy, bus_busy should also be asserted
• make bus_legit=oram_cs

This should fix the boot-up test