Phantasy Star Online Ver. 2 graphic issue

[mfigueirido]

Platform / OS / Hardware: Windows 11 x64

Flycast version: v2.3.2-221-g4d73cc8e1

Hardware: AMD Ryzen 2600, Radeon RX 570, 16GB RAM

Description of the Issue

Part of the ground shows visual corruption when looked at certain angles.

I'm not expecting this to be fixed as it looks like an edge case related to Z-fighting, specific to my drivers or hardware, but thought it would be good to have it documented.

Save state: Phantasy Star Online Ver. 2 (USA) (En,Ja,Fr,De,Es).state.zip

Debugging Steps Tested

• Happens in standalone and Retroarch.
• Happens in both Vulkan and DirectX11, although it's harder to catch in the latter.
• Native depth interpolation on/off has no effect.
• Changing OIT sorting setting has no effect.
• Saw videos of real hardware and did not notice the issue.

Screenshots

[flyinghead]

It is indeed z-fighting and I can reproduce similar issues in this scene with a Nvidia GTX 1080. Unfortunately there's no easy fix. Related to #1141.

[mfigueirido]

I see. Thanks for checking it out!

[Wunkolo]

Just curious: Does this happen with OpenGL at all too?

[flyinghead]

Yes, same glitch with OpenGL and Vulkan for me.

[mfigueirido]

Yes, this is OpenGL:

[Wunkolo]

Some more Vulkan data-points

Result on my Intel UHD Graphics 750:

Result on my RTX 3090:

All the GPUs mentioned in this thread support D32_SFLOAT_S8_UINT for DEPTH_STENCIL_ATTACHMENT just fine too.

Might have to do with the driver compiling shaders with varying levels of precision for different shaders similar to these issues: https://github.com/grimdoomer/NvidiaPrecisionLossPoc https://icode4.coffee/?p=566

[mfigueirido]

Interesting read... so maybe there's a chance for a fix?

From the second link: "If I could force the optimizer to choose one of the two paths consistently then that should mitigate the issue. Now the solution became clear, instead of calculating the dot product the long way I would use the dp3 intrinsic and a single add. This should optimize the dph function to the point where the shader optimizer wouldn’t be able to perform any further optimizations on it."

[flyinghead]

Not the exact same glitch but I do have intermittent z-fighting on this tex.

[mfigueirido]

Same behaviour: