UAE CPU core based 680x0 tester/validator

[Toni Wilen]

It is not latest but only recent changes are interrupt timing test changes which are still work in progress.

[dirkwhoffmann]

Hi Toni, quick question: I've read that extra hardware is required to do bus error testing. What kind of hardware would I need?

[Toni Wilen]

Originally Posted by dirkwhoffmann

Hi Toni, quick question: I've read that extra hardware is required to do bus error testing. What kind of hardware would I need?

I don't think you can get that hardware, it was repurposed Individual computers 68000/010 "A1200 accelerator" test board. (I got it because I for example did software for ACA500 and ACA500+). It isn't that complex, it has jumpers to select few different address spaces and one jumper to select if read or write access generates bus error when accessing selected address space.

It is almost useless feature anyway, none of 68000/010 based Amigas can generate bus errors. Only reason I did it because it was possible and UAE core is used by Hatari Atari ST emulator and Atari ST uses bus errors. And I don't think there are any Atari ST programs that expect bus error side-effects. Like long access only doing low or high word part of instruction or condition codes being only partially updated etc. Behavior is similar to address errors.

[dirkwhoffmann]

Thanks for the details answer! I was just curious about what kind of hardware was used. I already expected some kind of comparator triggering a bus error when the address on the address bus (partially) matches the selected one.

[dirkwhoffmann]

Yesterday I tried to create 68010 address error tests with the following options:

feature_target_src_ea=0x37fff1,0x7111
feature_target_dst_ea=
feature_undefined_ccr=1
mode=all

It works if cpu equals 68000, but if cpu=68000 is replaced by cpu=68010, zero tests are generated. Am I doing something wrong?

[Toni Wilen]

I don't know. There has been lots of changes since last "stable" version. (Current version is not address/bus error tested, it most likely is broken)

Does same happen with AEDST?

Next version has nice quality of life update: no more dozens of tiny files but single big file (with configurable max before it gets split) that is "streamed" during test run (memory requirements are not increased).

And of course 68000 IPL test but that won't get published too quickly
(EDIT: IPL test needs serial port TX to RX jumper and cycle-accurate serial port emulation + interrupt emulation. No external hardware needed anymore)

[dirkwhoffmann]

AEDST is working. Only AESRC seems to be broken.

I am looking forward to the next version. Sounds very promising!

[jotd]

I don't understand how to generate all instructions data files. I've enabled

[test=BASIC]
cpu=68000-68010
enabled=1

and most of other blocks, but all I get is Fxxx instructions in 4_xxx folders.

Zoned the config file

Getting it to run on vamos to test musashi core that I updated with MAME updates is another piece of cake

20:38:22.062        lib:WARNING:  [    exec.library]  ? CALL:  558 InitSemaphore( sigSem[a0]=00022c60 ) from PC=014fce -> d0=0 (default)
20:38:22.062        lib:WARNING:  [    exec.library]  ? CALL:  564 ObtainSemaphore( sigSem[a0]=00022c60 ) from PC=014dbc -> d0=0 (default)
20:38:22.062        lib:WARNING:  [    exec.library]  ? CALL:  186 Allocate( freeList[a0]=00022da8, byteSize[d0]=00000014 ) from PC=014f20 -> d0=0 (default)
20:38:22.062        lib:WARNING:  [    exec.library]  ? CALL:  570 ReleaseSemaphore( sigSem[a0]=00022c60 ) from PC=014f5c -> d0=0 (default)

That code seems to prevent the tester from even starting... Damn C startup!

[Toni Wilen]

Originally Posted by jotd

I don't understand how to generate all instructions data files. I've enabled

[test=BASIC]
cpu=68000-68010
enabled=1

and most of other blocks, but all I get is Fxxx instructions in 4_xxx folders.

You have multiple test sets enabled. Make sure BASIC test is the only one with enabled=1 (unless you really want to create multiple sets). Then check the output. Try also verbose=1.

4 = 68040 CPU test.

(Maybe I should create newer binaries someday now that things have stabilized again after IPL tests were finished)

[jotd]

I see, if I enable fpu (fpu=68040) then other tests aren't generated. Only Fxx stuff is generated in all categories. I guess I have to run the program several times to get all data files. Strange.

cputest basic/all

tested all basic stuff on WinUAE. What I would like would be to test all 68040 instructions (FPU or CPU) in one go.

Then fucking vamos won't even start the exe, no help printed nothing.... I never could do anything with that program. Too bad.

[jotd]

Tried harder. It occurs that the new version of vamos works really well and can now run a lot of stuff (lha, propack!)

Unfortunately it doesn't support AllocAbs, it handles the memory like a qemu user would do, using the host memory.

And cputest needs AllocAbs.

Is there a way to get rid of this? Why doesn't need absolute memory? for alignment? In that case, we could allocate more memory and used the aligned part.

[Toni Wilen]

Originally Posted by jotd

Is there a way to get rid of this? Why doesn't need absolute memory? for alignment? In that case, we could allocate more memory and used the aligned part.

It really needs absolute allocation because test memory needs to be exactly in same address as when test was generated. (PC would be different, address registers would be different and more all tests would fail if address space is not same)

(I am not going to work around it by allocating multiple small pieces until whole needed address space is allocated)

[jotd]

I see. That makes perfect sense. It's going to be my challenge to make it run using vamos then. I may have to upgrade the OS parts & mem access of vamos itself to support AllocAbs, not trivial but I think I can handle it.

[jotd]

I'm running cputester in vamos! I had to hack a lot of things to make that work

- first I had to implement AllocAbs that wasn't implemented before
- then I had to hack cputest to ignore mouse presses (mouse aborts test, vamos sees the button as pressed...)

Now I'd request a windows version of "cputester.exe". It's pretty difficult to build it, and the version I have generates data for version 16. I'd like to get the latest version of the tests.

[paraj]

Assume you mean "cputestgen.exe"? I've built a version from latest WinUAE sources and put in the zone.