Improve auto lod by comparing every attribute

[fire]

Fixes for some of the issues in #57416

• a struct to a float indexing change.
• always use 8 bones / 8 weights.
• Add a ui for changing lod indices

[fire]

@akien-mga do you think a pr that only increases the minimum to 256 triangle and or exposes the setting would be acceptable for 4.0?

[akien-mga]

I don't know, that depends on the scope of the actual change and whether it's tested to fix reported critical issues.

[clayjohn]

@akien-mga do you think a pr that only increases the minimum to 256 triangle and or exposes the setting would be acceptable for 4.0?

IMO this is definitely 4.1 material. LOD changes are support difficult for us to test as the surface area for testing is huge.

[jcostello]

@akien-mga do you think a pr that only increases the minimum to 256 triangle and or exposes the setting would be acceptable for 4.0?

IMO this is definitely 4.1 material. LOD changes are support difficult for us to test as the surface area for testing is huge.

Can this break projects between 4.0 and 4.1?

[fire]

@clayjohn the only alternative that I can think of is set autolod to off and mark as experimental?

[realkotob]

This fixes critical issues which is why I think it should go into 4.0, otherwise practically every user importing animated models will have broken results.

The other alternative as mentioned is to have autolod be off by default and marked as experiment, but in that case then this should also be merged anyway.

Testing this is not as complicated as it seems, just need to open a sizeable project (like the TPS demo) and reset the generated autolods and see if the game's assets look ok.

[fire]

Out of band conversation. We agreed to have no change.

[clayjohn]

I tested this on the TPS demo a few weeks ago and found that this resulted in almost no LODs being generated. Additionally the base LOD had way more vertices. The net effect was that the scene consistently had way more geometry than needed, even if LOD bias was set to always use the worst quality

[fire]

Can you provide some test case?

[clayjohn]

Can you provide some test case?

All I did is re-import the main meshes in the 4.0 branch of the TPS-demo https://github.com/godotengine/tps-demo. You can reproduce my finding with any one of the meshes in the project (i.e. just reimport the robot if you want to save time)

[fire]

@clayjohn

<param index="3" name="minimum_index_count" type="int" />
<param index="4" name="maximum_mesh_error" type="float" />

Thoughts on extracting this pr to add hard overrides to the lod import with minimum_index_count and maximum_mesh_error.

[clayjohn]

@clayjohn

<param index="3" name="minimum_index_count" type="int" />
<param index="4" name="maximum_mesh_error" type="float" />

Thoughts on extracting this pr to add hard overrides to the lod import with minimum_index_count and maximum_mesh_error.

Sounds like it could be helpful. But we definitely need to investigate a bit more. I have a feeling we should be able to produce better meshes with the default settings.

[fire]

I tried a few of the test cases and after the AABB test fixes they're not failing anymore.

Can reopen if there's a need.