If you can upgrade to 80186 CPU emulation(a handful of new instructions) you can run the 80186 testsuite (I mentioned it here: Re: Test suite and/or how to find bugs in x86 VM).
You can alternatively take Hottobar's 80386 testsuite(https://github.com/barotto/test386.asm) and rip out all 80186+ instructions.

Finally, UniPCemu supports importing instruction output in the common emulator format for verification(It'll dump information when a mismatch occurs). You'll just have to redirect the output from your emulator to UniPCemu's input(using simple piping) and have UniPCemu start it's verification using the "debuggerin"(unquoted) command line parameter(debuggerout does the opposite, dumping it's own debugger output to stdout instead of reading from stdin). Said read data(line endings automatically converted) from stdin is compared with it's own. If a mismatch occurs, "<VERIFY:MISMATCH>"(without quotes) is logged after any mismatching line not matching UniPCemu's logging.

Of course, it requires that the logging is enabled in it's settings.
Edit: Just added documentation on the command-line parameters to the manual.

Thanks!

The 186 tests looks promising.
I think I will start with that.

Oh my! I am trying to build an emulator too, this post looks useful. Thank you.

I just want to give you a special thanks @superfury for suggesting the 186 tests. It helped me a lot with the final piece of the puzzle.

My emulator is now released here here.

It is still a long way from UniPCemu but I'm quite happy with how it has turned out so far. 😉

Thanks!

Thanks for replying to this thread. I am going to integrate the 186 tests into my WIP emulator as well and use the example from your code for how they are supposed to be run. It's really too bad that the tests don't have a license... If anyone knows the license please let us know. Does anyone know the original source of them before they made their way onto this forum?

I originally found them on the fakex86 project forum before I posted a link to it and the files here.

https://forum.osdev.org/viewtopic.php?f=13&t=23739&start=15

It's the post by Artlav.

Okay I got an answer from Artlav on Twitter. It seems they are GPL 3:
https://twitter.com/theartlav/status/1299816147165081606

The 80186 tests by Artlav were helpful to me, but unfortunately (unless I'm mistaken in my interpretation of them) they make a lot of poor assumptions. For example some of them assume flag results for instructions that set some of their flags to undefined. Others assume exact memory locations when the tests have been modified over time and those locations no longer hold. I still have CPU bugs and I am still looking for good 8086 CPU tests that can be loaded in place of a BIOS or run completely independently. To see how I'm using the Artlav CPU tests, you can find the tests in my repository here:
https://github.com/davecom/DK86PC/tree/master/CPUTests

Does anyone have any alternative 8086 CPU tests?

Like I said, try the 80386 testsuite and strip out the 80286+ stuff. Just real mode and strip out the 32-bit instructions and protected mode stuff.

Otherwise, if you can create command line output in the common log format, you can compare to UniPCemu at least, which should be able to find more bugs where present(using piping between the programs).

The current commit of UniPCemu has a little 32-bit instruction(RCR) bugfix which isn't in a released version yet, but that shouldn't be a problem with the 16-bit/8-bit instructions, which should be working properly in UniPCemu afaik.
Perhaps try looking at Bochs handling of flags calculation for math instructions (add, sub etc.)? That was the way I finally managed to get it working in UniPCemu. Look at flags.c in UniPCemu's source code for that. Although it's not using lazy flags (which don't seem to cause big improvements in interpreting emulation anyways. They don't even show up in profiling results).

Yea, the 186 tests got me to a point where FreeDOS runs (seemingly flawless). So does most games that I have tested.
But I can still not boot any version of MS-DOS or Windows 3.0 yet. So there are obviously still issues with my emulator.

My current approach is to do what @superfury suggested. Logging input/output of all instructions and then compare it to another emulator.

UniPCemu is an interesting choice if instruction logging is already supported.

superfury wrote on 2021-01-31, 11:57:

Like I said, try the 80386 testsuite and strip out the 80286+ stuff. Just real mode and strip out the 32-bit instructions and protected mode stuff.

Unfortunately, it's not so trivial to do that. The creator of the 80386 test suite looked into the issue:
https://github.com/barotto/test386.asm/issues/12

phix wrote on 2021-02-01, 09:13:

Yea, the 186 tests got me to a point where FreeDOS runs (seemingly flawless). So does most games that I have tested.
But I can still not boot any version of MS-DOS or Windows 3.0 yet. So there are obviously still issues with my emulator.

Hi, I'm no developer, but I remember that os2museum.com always has some interesting texts to read..

https://www.os2museum.com/wp/undocumented-808 … opcodes-part-i/
https://www.os2museum.com/wp/undocumented-8086-opcodes/

From what I remember, MS-DOS/PC-DOS did things often differently.

The early versions (v2?) of MS-DOS tried to read system files in a single run, whereas later versions couldn't do that anymore because the system was too large to fit a cluster.
The Japanese versions (DOS/V) did some pretty weird stuff, too.

From what I remember, MS-DOS (incl. DOS622) also implemented a CALL5 interface to ease porting of CP/M-80 programs.

Modern DOSes and DOSBox do not emulate this aspect of MS-DOS.

https://www.os2museum.com/wp/a-word-on-the-call-5-spell/

https://www.os2museum.com/wp/the-a20-gate-it-wasnt-wordstar/

https://www.os2museum.com/wp/who-needs-the-ad … paround-anyway/

https://www.os2museum.com/wp/wordstar-needs-a … ess-wraparound/

Anyways, you can also try DCP or a real OS like PC-MOS/386, too.
Older versions of PC-MOS/386 (v1 to ~v3) do run fine on an XT.

https://m.youtube.com/channel/UCM_rzw6WcXibc7 … ry=PC-MOS%2F386

Version 5 became open source, even.

PC-MOS/386 released under GPL

Wow, that's a pretty cool project, I also always wanted to code some kind of emulator ... if you want to test the (graphics) compatibility of your emulator someday, try some of these releases 😀

http://www.pouet.net/prod.php?which=65371
http://www.pouet.net/party.php?which=204&when=1996
http://www.pouet.net/topic.php?which=11077&page=1#c530904

And if you add adlib support, I've coded a small 8086 compatible adlib player for my projects ..

👍 Yeah, that 8088MPH in particular will blow your mind...it's meant to be run only in the original IBM hardware, so no easy task for emulators.

Earlier thread about UniPCemu + 8088MPH:
UniPCemu cycle accurate 8088 implementation

mr.cat wrote on 2021-03-16, 18:13:

👍 Yeah, that 8088MPH in particular will blow your mind...it's meant to be run only in the original IBM hardware, so no easy task for emulators.

seems that 86box finally made it ...
https://www.youtube.com/watch?v=CWDRSx1LYTU

TheMechanist wrote on 2021-03-16, 17:54:

Wow, that's a pretty cool project, I also always wanted to code some kind of emulator ... if you want to test the (graphics) com […]
Show full quote

Wow, that's a pretty cool project, I also always wanted to code some kind of emulator ... if you want to test the (graphics) compatibility of your emulator someday, try some of these releases 😀

http://www.pouet.net/prod.php?which=65371
http://www.pouet.net/party.php?which=204&when=1996
http://www.pouet.net/topic.php?which=11077&page=1#c530904

And if you add adlib support, I've coded a small 8086 compatible adlib player for my projects ..

+1

Speaking of CGA.. What about adding that 640x400 mode?
The Olivetti/AT&T graphics mode is a superset of CGA.
Adding it wouldn't break anything, would it? 😀
Would be interesting for Windows 3.0, for example.
Or many of the Turbo Pascal programs.

Olivetti/Logabax/AT&T/Toshiba 640x400 hi-res graphics mode

Just to share how I ended up doing in the end.

I created a hardware validator using a real 8088 and then use that in conjecture with the 186 tests @superfury mentioned.

It has been a very helpful tool and I found a lot of bugs with it. I hope it can be useful to some one else as well. 😁

Congrats, seems, that you reached your goal and running windows 3.0 is a great milestone!

Now, looking at the final (?) product - would you say, that it's an accurate emulation of a XT - in terms of speed, compatibility and so on?

unfortunately VirtualXT doesn't build on Ubuntu 18.04 (latest Zig from github) ..

1"Semantic Analysis [1615/2537] 
2./build.zig:221:23:error: no member named 'defineCMacroRaw' in struct 'std.build.LibExeObjStep'
3            pirepheral.defineCMacroRaw("PI8088");
4                      ^
5./build.zig:150:22: error: container 'builtin' has no member called 'zig_version'
6    const a = builtin.zig_version.major;
7                     ^
8./build.zig:101:12: error: no member named 'defineCMacroRaw' in struct 'std.build.LibExeObjStep'
9        lib.defineCMacroRaw("VXT_CPU_286");"