Gngeo behaves differently from real hardware when scaling sprites vertically

[city41]

Hi Damien,

Would you say this Gngeo repo is currently the best place for bug fixes/features?

I've been exploring vertically scaling sprites, and I found Gngeo behaves a little differently from real hardware. Details here:

https://github.com/city41/neo-geo-scaling

If I can spare some free cycles, I might take a stab at fixing this.

[dciabrin]

Hey city41

Thanks for the nice detailed report in the link.

I'm not in touch with Mathieu Peponas, the original author of GnGeo, however I do try to get fixes as isolated as possible in my fork's branches.

So if anything differs from what the real hardware is doing, I think it's fair to fix it in this fork.

What I would do as well is trying to replicate the difference on a more recent emulator like Mame. In the past I already spotted differences between the two (e.g. IRQ handling).

If the problem also reproduces on Mame, we might want to dig further and propose a fix for both emulators.

[city41]

MAME and Gngeo both render the same way (I added the MAME screenshot to the README at the scaling repo).

But since I am using a NeoSD to test on real hardware, it is possible the NeoSD causes VRAM to have different values than it otherwise would. It's possible MAME and Gngeo are reproducing what a real game would do, I'm not sure how to eliminate the NeoSD as a variable.

[city41]

I configured NeoSD to boot straight to the game, so possibly that is an accurate simulation? If so, then the differences between real hardware and MAME/Gngeo remain.

[city41]

neodev said on the TerraOnion discord

Neosd doesn't do any scaling
But you can't trust on emulator accuracy
To run a completely unmodified boot, you need to enable arcade mode (cartridge mode in aes), then take the sd card out
It will boot without ing-game trigger, and it's 100% original cart accurate

I did that, and I still see the differences between emulators and real hardware, so I am thinking this is a true difference.

[city41]

Started a MAME discussion here

[city41]

I think this boils down to my error. I was not padding the C ROMs. The NeoSD will load tiny C ROMs and they work. But I am guessing the tiny C ROMs cause strangeness when scaling.

After padding the C ROMs to a more expected size (I did 2mb), the black boxes on hardware went away. Now hardware, MAME and Gngeo all agree with each other.

[city41]

I have found two ways gngeo scales sprites different from MAME and real hardware. I captured them here

https://github.com/city41/gngeoVsmame

Here is how gngeo scales the test sprites

Versus MAME

I have switched to MAME for my dev purposes. But I still like gngeo and would like to use it. I feel like gngeo is a better chance to build great dev/debug tools such as a vram viewer. As time allows I plan to keep getting to know gngeo's code base and see if I can fix these issues.

In general I find MAME is very accurate. I have yet to find a case where it differs from real hardware.