New discovery on Z80 I/O block instructions

[ZjoyKiLer]

I've discovered a new undocumented behavior of the inir/indr/otir/otdr instructions:

As in cpir/cpdr/ldir/lddr, when the repeat condition is met, the CPU generates an extra M-cycle of 5 T-states to decrement PC. This M-cycle performs additional flag changes (discovered and cracked by David Banks (AKA hoglet)), but until now it was not known that MEMPTR is also modified, being set, as in all other block instructions during this M-cycle, to PCi + 1 (PCi = value of PC at the start of the instruction, before it is incremented).

After adding this behavior to my Z80 core, z80memptr.tap from Patrik Rak's Zilog Z80 Test Suite v1.2 fails in the following 2 steps:

Code: Select all

102 INIR->NOP'            FAILED
CRC:F3B1BE2F   Expected:0A537B63
103 INDR->NOP'            FAILED
CRC:F3B1BE2F   Expected:0A537B63

Which matches the results obtained on real hardware by Richard Chandler, who kindly tried the test at my request on one of his machines with NEC CPU:




Of course, this "new" behavior has been tested on Visual Z80 Remix.

I've opened a pull request to fix Patrik's test suite: https://github.com/raxoft/z80test/pull/3

[R-Tape]

I look forward to someone writing a game that tests this in the similar way to MEMPTR snake.

[Morkin]

Every emulator author after reading this:

[TheMartian]

Hi!

Does this happen in all Z80 models, or just NEC? I'm curious.

I'll try later to execute on my machines. Any suggestion or request?

[Lethargeek]

Does this happen in all Z80 models, or just NEC? I'm curious.

couldn't be all, not every one failed Patrik's test before afair, otherwise too many people would complain

[jamesh]

I not sure what kind of CPU I have, there is big "Z" with Zilog under it (original Zilog's?), but it fails those two tests. Issue 6A board.

[ZjoyKiLer]

Hi!

Does this happen in all Z80 models, or just NEC? I'm curious.

I'll try later to execute on my machines. Any suggestion or request?

Confirmed on Zilog and NEC. I assume that ST CMOS models also have this behaviour, but it is not confirmed.

USSR and MME models could behave differently.

[ZjoyKiLer]

couldn't be all, not every one failed Patrik's test before afair, otherwise too many people would complain

Actually, I don't know of anyone who specifically tried z80memptr.tap on a real Zilog or NEC on Sinclair hardware until yesterday. People usually try z80full.tap and z80ccf.tap. I have a Russian friend who tried z80memptr.tap on USSR Z80 clones, and failed several steps. He used a Profi, a Russian Speccy clone, which tends to produce CRC errors in the I/O instructions because the values read from the I/O ports are not the expected ones...

[TheMartian]

Looks like I've got a round of testing to do. I thought that z80full.tap already covered the MEMPTR.

Live and learn...

[Weiv]

Every emulator author after reading this:

Not me) I'm glad to update my emulator everytime when new info about emulation accuracy comes. I know that there is no perfection in emulation but we can get as close to it as we can

[Lethargeek]

Actually, I don't know of anyone who specifically tried z80memptr.tap on a real Zilog or NEC until yesterday on Sinclair hardware.

i dont remember old discussions details now, but do you mean even Patrik himself never tried it on a real hw

People usually try z80full.tap and z80ccf.tap. I have a Russian friend who tried z80memptr.tap on USSR Z80 clones, and failed several steps.

if it was T34Bx1 then it's supposed to have slightly different memptr behaviour afair

[ZjoyKiLer]

i dont remember old discussions details now, but do you mean even Patrik himself never tried it on a real hw

yep... I was there when Patrik was working on the new releases of his test suite. We knew about new discoveries about the CCF/SCF instructions and the flags of block instructions, which motivated the update. We didn't check all the tapes of the new version on real hardware, sorry.

if it was T34Bx1 then it's supposed to have slightly different memptr behaviour afair

Yes, a T34VM1, and MME models too, 80-CPU, U880...

[Lethargeek]

yep... I was there when Patrik was working on the new releases of his test suite. We knew about new discoveries about the CCF/SCF instructions and the flags of block instructions, which motivated the update. We didn't check all the tapes of the new version on real hardware, sorry.

owww... trust no one

[Patrik Rak]

i dont remember old discussions details now, but do you mean even Patrik himself never tried it on a real hw

There was so many people testing the interrupted flags on the real hardware, using both mine and Ped7g's tests, that it never occurred to anyone the memptr behavior was not tested. Apparently, many people think that z80full covers it all. Or perhaps it was tested on hardware which supplied wrong IN value, so we dismissed it as that. I think it comes as a surprise to everyone involved that no one noticed before.

Well, FWIW, the test suite was updated to 1.2a with the updated CRCs now. Apologies to everyone and thanks to ZjoyKiLer for finding this out.

[Zoran]

Not me) I'm glad to update my emulator everytime when new info about emulation accuracy comes. I know that there is no perfection in emulation but we can get as close to it as we can

I totally agree.
Many thanks to ZJoyKiller for investigation, to people who tested on hardware, to Patrik for his testing suite (version 1.2 of memptr test, imperfect as it was, actually revealed this).
Now after other people did the hard work, implementing it was easy.

[ZjoyKiLer]

BTW, many thanks to @TheMartian, who is testing all the tapes from Patrik's Zilog Z80 Test Suite on many of her machines, so that finally everything is 100% tested once and for all.

Some of the results can be checked here.