Skip processing individual triangles or vertices of the mesh

Question

In my game I have a billiard table, and a floor below the table. In typical situation the table covers 80% of the screen, so only small amount of the floor is visible. And I render the table first, so invisible part of the floor is discarded by the depth buffer test.

But profiling with Nsight shows me drawing the floor takes significant amount of time, and it is spent in the rasterizer. It seems generating all those pixels of the floor and depth testing them takes a long time.

I'm wondering how to solve this problem and one thing I tried is to divide the floor (a rectangle with two triangles) to multiple rectangles, so some of them could be cancelled in an earlier stage. The problem is, this is still one model / mesh, just with more triangles and I don't know how to start.

What would be a way to achieve this in Monogame + HLSL? Somewhere in vertex shader? Or gemoetry shader (is it supported in monogame?)

Or is there any other way to optimize this?

UPDATE: the whole floor is needed and might be drawn depending on where the camera is, so I can't get rid of parts of it. In typical situations table top covers 80% of the screen because of perspective and looking down at an angle. The game is in 3D.

Answer 1

Break the floor geometry such that invisible pieces are not there at all

Answer 2

You have these options:

• Render the table first, and the floor second and have depth testing on. This will cancel calling the pixel shader on the floor pixels where the table has already been drawn. (though the rasterizer will still do its work) (edit: Oh I see you are already doing that)
• Frustum culling: don't render anything that is not in the view frustum. Usually visibility is determined by checking the object's bounding box against the view frustum.
• Occlusion culling: Don't call the draw if it is occluded. I don't know if occlusion queries are available i XNA though. It is also doable by a software occlusion culling technique, but I doubt it would be fast enough if it is done in C#. This is a broad topic, I won't get into details here.
• Culling primitives in the geometry shader. This is not available in XNA, but in monogame it is when using a DirectX 11 renderer. For each primitive you can test it against the view frustum, and decide if you want to send it to the pixel shader. I don't know if adding this extra shader stage will help you or not though, it's just an idea.