Shadow glitches

[arakis]

@LevYas wrote in #3:

There are some glitches with shadows though

But this could be because of OpenGL ver 3.3. Even the freshest Nvidia drivers did not bring me 4.

Originally posted by @LevYas in #3 (comment)

[arakis]

This is a known issue and is regardless of the OpenGL version. There's a bug in the shadow calculation. Some brainstorming:
There could be a problem in the source of

• https://github.com/AximoGames/AxEngine/blob/master/Assets/Shaders/common/lib.frag.glsl#L29
• https://github.com/AximoGames/AxEngine/blob/master/Assets/Shaders/common/lib.frag.glsl#L98
• Maybe wrong Near/Far-Plane calculation?

I've spent lots of time in shadow debugging in the past. Any help would be great.

Nut sure, but as I remind this occurs only in DirectionLightning (shadow Map). Try PointLight (shadow Cube) here:
https://github.com/AximoGames/AxEngine/blob/master/Demo/DemoApplication.cs#L189

[arakis]

Just a notice: in Marbles ( https://github.com/AximoGames/AxMarbles ) and in this Sample ( https://www.aximo.games/samples/shadows/ ) only PointLight is used and there are no shadow glitches. So, it seems it happens only with DirectionalLight. Any help in fixing the shadow bug would be great.

[LevYas]

I'm afraid I will not be able to help. I've just started to learn OpenGL and OpenTK as a hobby and came across your engine as a real-word usage example :)
BTW, I'm very disappointed there is no TDD practice for shaders. I think they're ideal for unit testing - a specific set of arguments (coordinates, uniforms, ...), a specific set of output (color for pixel, ...), but I failed to find any testing framework for this, only 1.5 proof of concepts several years old and that's all.
I have no idea how people making shaders like this without a decent debugger and tests https://www.shadertoy.com/view/Ms2SD1

[arakis]

There are no isolated shader tests, but some render tests: https://github.com/AximoGames/AxEngine/tree/master/Tests/RenderTests

They work in the following way:

• Generate an image of a scene. The images are stored in the repository ( https://github.com/AximoGames/AxEngine/tree/master/Tests/Assets/TestOutputs/LightTests )
• On every "unit" test (it's more a integration test) all scenes are rendered and compared with the stored image. If there's a notable difference, the test is failing.

Generally, even the render tests are a life and time saver :-)

[kbitGit]

So I tested a bit and found the following:

• The Reason, that the OpenGL Version is 3.3, is because OpenTK default initializes the Context with Version 3.3
  You can change the Version by Adding APIVersion = new Version(4, 1) in the NativeWindowSettings in the RenderWindow.cs
• The Bug presists in Version 4.1 with Intel
  
  and Nvidia.
  

Both were tested with the Context set to Version 4.1

[kbitGit]

I disabled the moving Light and looked with Renderdoc to check how the shadowmaps are and noticed, that the directional light Shadowmap ist just white.

And the Problem seems to be in DeferredRendering Part of the Engine, since the only shadows I see in Renderdoc are in the DeferredRenderPipeline

If the Renderdoc Capture is interessting, I can upload it to dropbox or something similar, since Github doesn't allow for files bigger than 10MB.

[arakis]

Interesting. But if the shadow map (used for direction light shadows) is "empty", there should not be any shadow? Did you checked the shadow map "Index" (=Slice)?

[kbitGit]

Yes, with index 1 there is a shadow map.
Is it normal, that there is an white image as output in the first render pass?

And the Issue with the shadow is caused by the shadow map

So the next step now is to find out why the shadow map has this problem.

[arakis]

Greate work! Just some ideas:

• Near-Plane is crossing the sphere?
• Shadows are renderend with normal cull mode = back, like normal rendering.

I know for shadows, cull mode = front shuld be used. But with that cull mode, the shadows seems to be less accurate. You can test it here: https://github.com/AximoGames/AxEngine/blob/master/Render/Objects/SimpleVertexObject.cs#L266

[kbitGit]

It was the near plane intersecting with the sphere, that caused the issue.

I changed in the LightObject.cs the nearplane to 0.1f

But the shadow of rotating directional still disappears and reappears.
I thing there might be a problem, that the shadow map doesnt catch the complete map.

[arakis]

amazing! thx very much for tracking down the problem. Regarding the missing shadows that are far away: I have no idea how to solve this in a perfect manner. One idea is to move the directional shadow light with the player. But even in this case, the map needs to be bigger: More memory, more computing time or shadow have more artifacts. So, what do you think is a good/acceptable way to render shadows that are far away?

[kbitGit]

Sadly no :/

[arakis]

Did you found out why the shadows are not correctly rendered if cull mode is changed?

[kbitGit]

I did not check that, but I can look later

[kbitGit]

Hm it's really weird. This is the shadows when I enable Cullmode Front:

I used the following code to enable it in SimpleVertexObject.cs :

GraphicsDevice.Default.CullFaceMode = CullFaceMode.Front;
//GL.CullFace(CullFaceMode.Front);
//GL.Disable(EnableCap.CullFace);
mat.Vao.Draw();
//GL.Enable(EnableCap.CullFace);
//GL.CullFace(CullFaceMode.Back);}
GraphicsDevice.Default.CullFaceMode = CullFaceMode.Back;

The weird part, is that the shadowmaps seem to be fine
Slice 0

Slice 1

I have currently no clue, where the problem can be.