Minimig (Amiga) core discussion

[Sorgelig]

exe_opcode is missing in process() args, I guess it will break PACK/UNPACK

no, it won't. Quartus will warn and add it while synthesising.

[apolkosnik]

bf_loff_dir was also gone, pull request submitted.

exe_opcode is missing in process() args, I guess it will break PACK/UNPACK

no, it won't. Quartus will warn and add it while synthesising.

[xzarian]

I'm having the same issue, can't boot with fastram > 8m. (68020,wb3.1) Before I think it was set to 24m, but I guess that that options gone now? I haven't used this core in months.

[Sorgelig]

I'm having the same issue, can't boot with fastram > 8m. (68020,wb3.1) Before I think it was set to 24m, but I guess that that options gone now? I haven't used this core in months.

currently i have no idea why.

[GoingDown]

I'm having the same issue, can't boot with fastram > 8m. (68020,wb3.1) Before I think it was set to 24m, but I guess that that options gone now? I haven't used this core in months.

Do you use latest "official" core version which is provided by update script? Because I had similar issue with that version, but later test versions provided by Sorgelig via manual download in this thread works perfectly.

[xzarian]

I'm having the same issue, can't boot with fastram > 8m. (68020,wb3.1) Before I think it was set to 24m, but I guess that that options gone now? I haven't used this core in months.

Do you use latest "official" core version which is provided by update script? Because I had similar issue with that version, but later test versions provided by Sorgelig via manual download in this thread works perfectly.

Yes, I was using latest official core, but it now works with the test version Sorgelig posted.

[Milongero]

Hello,

i just got back from vacation and did an update. the Minimig core hasn't changed in a week.
So I tested again different settings to get Lotus 2 and III to run.
Nothing worked until I triggered a core reset in the Minimig OSD. After that both games ran.
For me, the core is now running wonderfully well.
Greetings Lutz

[apolkosnik]

Destination fix for BFFFFO, BFEXT*, MULU.L DIVUL.L

In BFFFFO, BFEXT*, MULU.L DIVUL.L according to the documentation, should only allow Dx as the destination.
https://github.com/MiSTer-devel/Minimig ... ik:patch-3

After going through the Toni's cputester output, I've noticed that some instructions were writing to address registers rather than data registers. I'd like to get a second set of eyes on this before I create a pull request. Thanks!

[JimDrew]

I have a pretty thorough test suite for the 020+ instructions, and I focused on the all of the bitfield based instructions. I will take a look at the Minimig core.

SOME bitfield instructions do have extended modes that do use address registers, ie:

Code: Select all

TEST_BFINS:
   lea   Me(pc),a0
   moveq   #1,d0
   moveq   #0,d1
   moveq   #0,d2
   bfins   d0,(a0){d1:d2}
   nop
   rts

Me   dc.l $55667788
   dc.l 0,0,0,0,0


Code: Select all

TEST_BFCHG:
   lea   Me(pc),a0
   moveq   #0,d2
   moveq   #0,d3
   bfchg   (a0){d2:d3}
   nop
   rts
   
Me   dc.l $55667788
   dc.l 0,0,0,0,0


While other instructions use a data register, ie:

Code: Select all

TEST_BFFFO:
      bfffo   OurStatus+1{0:8},d0
      nop
      rts
      
OurStatus   dc.w   256


I have a slew of these test snippets that I wrote originally for testing FUSION-PC's CPU core, and more recently used to help debug the Vampire and Replay CPU cores.

[Futaura]

Wasn't sure where best to report this, but @kolla directed me here... Attached is a test case showing that the MULU.L instruction is broken - it can incorrectly give different results when the parameters are transposed, depending on the numbers involved:

FAIL: MULU.L $a26bd7e0,$7667c08f = $37459620 ($37459620)
PASS: MULU.L $7667c08f,$a26bd7e0 = $b7459620 ($b7459620)

Both the above should obviously yield the same result, as they do on a real CPU. Given the difference between the two results above is $80000000, I'm guessing it is probably an issue related to signedness somewhere.

[apolkosnik]

Jim, Thank you, that sounds great.
The main issue that I'm trying to address here is when the second word of the previously noted instructions starts with MSB set, which was making the result appear in the address register rather than data register. It's a corner case that could only happen with erroneously generated code.

I have a pretty thorough test suite for the 020+ instructions, and I focused on the all of the bitfield based instructions. I will take a look at the Minimig core.
...
I have a slew of these test snippets that I wrote originally for testing FUSION-PC's CPU core, and more recently used to help debug the Vampire and Replay CPU cores.

[apolkosnik]

Wasn't sure where best to report this, but @kolla directed me here... Attached is a test case showing that the MULU.L instruction is broken - it can incorrectly give different results when the parameters are transposed, depending on the numbers involved:

FAIL: MULU.L $a26bd7e0,$7667c08f = $37459620 ($37459620)
PASS: MULU.L $7667c08f,$a26bd7e0 = $b7459620 ($b7459620)

Both the above should obviously yield the same result, as they do on a real CPU. Given the difference between the two results above is $80000000, I'm guessing it is probably an issue related to signedness somewhere.

It looks like the MSB gets cleared somewhere.

[Sorgelig]

Here is a new test version. It includes all instructions fixes currently available.
The main change is communication protocol between HPS and Minimig.
I've changed it to make it more generic and use common framework for easier maintenance.
ZIP includes both Minimig.rbf and MiSTer binary. You have to replace them both. New MiSTer won't work with old builds of Minimig. Old MiSTer won't work with new Minimig either.
Other cores are unaffected.
If it works well, then it will go to release.

minimig.zip

[breiztiger]

hi

last test core work great with my save configs (A500 stock, A600 stock and A1200 stock or with hdf attach)

i have found a graphics corruption in Pacmania (1988)(Grandslam).adf with aga, it run great in ecs or ocs mode

i think it's not related to last release

[apolkosnik]

Breiztiger,
Would it be possible to document the corruption as an issue on github?

https://github.com/MiSTer-devel/Minimig-AGA_MiSTer

Thanks!

hi

last test core work great with my save configs (A500 stock, A600 stock and A1200 stock or with hdf attach)

i have found a graphics corruption in Pacmania (1988)(Grandslam).adf with aga, it run great in ecs or ocs mode

i think it's not related to last release

[remowilliams]

Latest test core is looking good here so far.

[Sorgelig]

New version with i hope more stable build:

minimig_2.zip

MiSTer binary is the same as in previous build, so i've included it just for completeness.
Please test it.

[rondc]

Sorgelig, doesn't work for me. Get's me on system like menu core.
What is the MiSTer binary attached to Minimig2.zip ?

[rondc]

When running Fusion with Kick 3.1, this is what happens...

Recognizes 82MHz 020

fusinonmister01.png

Just before finishing the finder load, it crashes and reset the Amiga.

fusionmister2.png

This is the crash msg.

fusionmister3.png

The same sequence using a 3.1.4 KickStart

When booting, this is the 1st yellow guru msg:

err01.png

After crashing the Fusion , these are the msgs:

err02.png

And finally the latest yellow guru msg before booting Workbench.

err03.png

ShapeShifter makes an attempt to start but it also reboots the Amiga.

Thanks a lot for your work ! ! !

[optiroc]

New version with i hope more stable build:
minimig_2.zip
MiSTer binary is the same as in previous build, so i've included it just for completeness.
Please test it.

Works like a charm here!

[fille1976]

Thx sorg,gods working now,it was broken for a while,keep up the good work.

[apolkosnik]

Same here, after moving both filkes onto the MicroSD card to replace the existing binaries (scp will not do the trick since: scp: /media/fat//MiSTer: Text file busy ), it works pretty much the same.
Workbench loaded, ran a few demos. No visible difference so far.

New version with i hope more stable build:
minimig_2.zip
MiSTer binary is the same as in previous build, so i've included it just for completeness.
Please test it.

Works like a charm here!

[Sorgelig]

Sorgelig, doesn't work for me. Get's me on system like menu core.
What is the MiSTer binary attached to Minimig2.zip ?

congratulation with not able to read messages.

[apolkosnik]

I think that might be a side effect of one of the latest fixes. Possibly some stack values get corrupted. Lately, I've experienced a crash when running "cputest all abcd.b continue" somewhere before mulu.l, but if I'd start it from the m-instructions, it goes fine.

When running Fusion with Kick 3.1, this is what happens...

Recognizes 82MHz 020
fusinonmister01.png

Just before finishing the finder load, it crashes and reset the Amiga.
fusionmister2.png

[JimDrew]

When running Fusion with Kick 3.1, this is what happens...

Well... other than the Vampire, that's the first time I have seen any FPGA based Amiga emulation even be able to start the Mac with FUSION! I will have to look into see why FUSION is crashing.