Re: [openMSX-devel] Timing of OUT

[Maarten t. H. <ma...@tr...>]

On Tuesday 31 December 2002 11:21, Wouter Vermaelen wrote:

> > The alternative is to make sure all calls occur in timestamp order. T=
his
> > means the Z80 is not allowed to perform I/O past the next sync point,
> > even if the instruction that performs the I/O starts before the next =
sync
> > point. Any ideas how this can be implemented? It would really help me=
 in
> > trying to get the VDP code as clear as possible.
>
> Unfortunately this is not so easy to solve. The CPU emulation must be a=
ble
> to stop and resume in the middle of an instruction. I have a few ideas =
to
> implement this. Even without a noticeable performance hit I hope. Howev=
er
> it requires a major rewrite of the current CPU code and thus also a lot=
 of
> time.

Resuming in the middle of an instruction is one option. It is also possib=
le to=20
abort the instruction and when the Z80 is scheduled again, restart it ins=
tead=20
of resuming it. That might simpify the state kept in the Z80 class.

Aborting an instruction means that before you do I/O (both I/O mapped and=
=20
memory mapped, both read and write) you check the I/O timestamp is lower =
than=20
the end timestamp of the Z80 scheduled time. If this is not the case, any=
=20
processing done for this instruction is discarded and control is returned=
 to=20
the scheduler. Since most (all?) instructions only change the Z80 state a=
fter=20
I/O, I hope this won't be too hard.

> How urgent do you need this? I could implement some temporally hacks
> if you like (for example perform IO at the same time as the start of th=
e
> instruction).

The workaround from your other mail is good enough for me to continue=20
debugging the VDP code. However, we want near-perfect emulation, so the i=
ssue=20
should be solved one day. Since it is not a simple issue, I don't mind if=
 it=20
takes some time.

Bye,
=09=09Maarten