Re: [hatari-devel] test needed on a 4MB STF

[ Thread Index | Date Index | More Archives ]

Le 20/12/2016 à 16:17, Christian Zietz a écrit :
Nicolas Pomarède schrieb:

But if you can take the same photos for the STE setup, then it will be
of use to me too in the near future.

I can do that later...

Do you mean different MMU on STE instead ? I thought the STF ones were
all the same and that it was the STE which had different versions and
"bugs" (like applying bank0 settings to bank1, prenventing to mix
different bank sizes) ?

No, I really mean different MMUs on the *STF*. It's long been known that
IMP MMUs in STF for example don't support 2,5 MiB configurations:
<>. The reason is
that these MMUs also use bank0 settings for both banks, i.e. they don't
allow the banks to differ.

OK, thanks for confirming, I'm not an MMU expert, I know this gave problem to people doing memory extension, but I never pay too much attention to it. Does the STE has the same problem, or all the GST/MCU chips behave correctly with 2 distinct bank settings ?

If they really differ on STF, then I'm interested in the one where you
can have different values for bank0 and bank1 (my test program assumes
they can differ)

Please find my photos here:
<>. On
the STF with 4 MiB RAM and IMP MMU -- as expected -- nothing happens
when I press F5 or F4, as the MMU ignores these settings. There are
however three pictures of my 4MiB STF with "non-IMP" MMU, that are
hopefully helpful.

Great, exactly what I wanted !

Yes, I read it as a base from my code, especially the MAD <-> CAS/RAS
mapping was very interesting. But I found some more "stranger" results
when using 128 K banks, I'd like to check all the cases and write a
small doc about my findings later.

You should know that the memory address lines (MAD9/MAD8) that are not
actively used with a given bank size still do toggle, at least on the
IMP MMU. So, for example if you really have 2048 kiB in a bank but set
the MMU to 512 kiB, MAD9 won't stay constantly high or low. In fact, if
I remember correctly, it follows MAD8 then. This obviously doesn't make
a difference when there really only are 512 kiB in a bank, since MAD9
isn't even connected then.

Exactly, and that's what I can determine with these color patterns :)
On my 1 MB STF, if I set bank1 to 128 K, then I can see that MAD8 (used as R8 on 512 K ram bank, but not used on 128 K bank) is set to the value of C0 (ie MAD0 when CAS values are sent first).

Mail converted by MHonArc 2.6.19+