Need help mapping UV coordinates on cube voxel faces with tilemap texture

This is a READ-ONLY snapshot of the Godot Community Forums when TwistedTwigleg was the owner, which is hosted for archival reasons only. The current Godot Community Forums can be found here.

Some links may point to external websites, including the current Godot Community Forums. These websites are not associated with TwistedTwigleg.

Need help mapping UV coordinates on cube voxel faces with tilemap texture

MirceaKitsune

I've found myself in a bit of a pickle again and am going to need ideas. I reached the point in my voxel terrain engine where I'm working on adding textures. If I could texture each face independently this would be an easy one, but I read that for every material the mesh needs to be split which creates extra draw calls and reduces the performance I tried hard to maximize. So I decided to go for mixing most categories into a single tilemap and using UV coordinates to map an image to each face. Only problem is, I can't figure out how to get the coordinates right. To explain with a code snippet of the function in cause below, the full file and project can also be found on Github:

https://github.com/MirceaKitsune/godot_cubedot/blob/main/scripts/lib/voxel_mesh.gd#L39

func get_tris():
	var c: PackedVector3Array
	var u: PackedVector2Array
	var invert = dir == 1 or dir == 2 or dir == 5
	if dir == 0 or dir == 1:
		c = [pos + Vector3(0, -size.x, -size.y), pos + Vector3(0, -size.x, +size.y), pos + Vector3(0, +size.x, -size.y), pos + Vector3(0, +size.x, +size.y)]
		u = [Vector2(c[0].y, c[0].z), Vector2(c[1].y, c[1].z), Vector2(c[2].y, c[2].z), Vector2(c[3].y, c[3].z)]
	elif dir == 2 or dir == 3:
		c = [pos + Vector3(-size.x, 0, -size.y), pos + Vector3(-size.x, 0, +size.y), pos + Vector3(+size.x, 0, -size.y), pos + Vector3(+size.x, 0, +size.y)]
		u = [Vector2(c[0].x, c[0].z), Vector2(c[1].x, c[1].z), Vector2(c[2].x, c[2].z), Vector2(c[3].x, c[3].z)]
	elif dir == 4 or dir == 5:
		c = [pos + Vector3(-size.x, -size.y, 0), pos + Vector3(-size.x, +size.y, 0), pos + Vector3(+size.x, -size.y, 0), pos + Vector3(+size.x, +size.y, 0)]
		u = [Vector2(c[0].x, c[0].y), Vector2(c[1].x, c[1].y), Vector2(c[2].x, c[2].y), Vector2(c[3].x, c[3].y)]

The gist of that code is simple: Both c and u are an array containing 4 entries each representing a corner, one for spatial corners (4 x Vector3) the other for UV coordinates (4 x Vector2)... dir is the axis in which the face is pointing, this is important for 3D positioning and less relevant here, just to explain why I have those cases separating the X Y Z axes. Currently that code takes the corners in the direction the face spans across and treats them as UV coordinates. I already tested it and get good results with a single texture, some drift but if that was the only issue I'd correct it with a fixed offset.

With this test texture I have 2 x 2 tiles, so I'd need to zoom the texture in twice then show a 0.5 x 0.5 section per face. If I multiply u[0] / u[1] / u[2] / u[3] with 0.5 I should be getting the right scale, however I also need to offset the position accordingly. With trial and error I might be able to find an offset that works, yet I have an even bigger problem: After the scale syncs I need to choose which tile to show, I can't have one tile on one voxel then the other on the voxel next to it like a checkerboard.

The obvious solution would seem to manually set the UV corners per face and drop the face's position offset, but there's another problem there: There's also the a face merging algorithm I posted about earlier, meaning any surface may be a stretched rectangle representing multiple other surfaces with the same image. To make matters worse I also have a variable voxel size and already decided to go default at a resolution of 0.5m rather than the 1m most Minecraft inspired engines use; This shouldn't be a big issue as I want half-voxels to cut into the texture which may remain scaled to 1m by default... my real problem is the varying position and scale offsets caused by face mergers causing larger faces that need to repeat the same texture giving the appearance the smaller faces are there.

In that function I know the center position and scale of the face as well as its corners in 3D space. How do you suggest I use them to correctly put the UV coordinates at a position of choice to display a specific section of my texture? Will think on it till tomorrow and see what answers I can come up with myself.

MirceaKitsune

Made some progress but running into extremely weird issues. I generate the top-left corner and the surface syncs across faces, but the offset is off so the whole thing renders in-between cubes at various resolutions. I'm now writing a global offset applied to the position in both the X and Y axis which varies with resolution.

	# rp is position, works as long as long as offset is okay
	# rs is scale, works as long as there's sync between continuous surfaces
	var rp = 1 + (res * Data.settings.resolution)
	var rs = 1 + (res * Data.settings.resolution)

	var c: PackedVector3Array
	var u_start: Vector2
	if dir == 0 or dir == 1:
		c = [pos + Vector3(0, -size.x, -size.y), pos + Vector3(0, -size.x, +size.y), pos + Vector3(0, +size.x, -size.y), pos + Vector3(0, +size.x, +size.y)]
		u_start = Vector2(rp - fmod(pos.y - size.x, rs), rp - fmod(pos.z - size.y, rs))
	elif dir == 2 or dir == 3:
		c = [pos + Vector3(-size.x, 0, -size.y), pos + Vector3(-size.x, 0, +size.y), pos + Vector3(+size.x, 0, -size.y), pos + Vector3(+size.x, 0, +size.y)]
		u_start = Vector2(rp - fmod(pos.x - size.x, rs), rp - fmod(pos.z - size.y, rs))
	elif dir == 4 or dir == 5:
		c = [pos + Vector3(-size.x, -size.y, 0), pos + Vector3(-size.x, +size.y, 0), pos + Vector3(+size.x, -size.y, 0), pos + Vector3(+size.x, +size.y, 0)]
		u_start = Vector2(rp - fmod(pos.x - size.x, rs), rp - fmod(pos.y - size.y, rs))

	var u: PackedVector2Array
	u = [
		Vector2(u_start.x, u_start.y),
		Vector2(u_start.x, u_start.y + size.y * 2),
		Vector2(u_start.x + size.x * 2, u_start.y),
		Vector2(u_start.x + size.x * 2, u_start.y + size.y * 2)
	]

The inexplicable thing is that when I go below a step size of 0.5 the pattern inexplicably changes by reversing at a different scale:

# @ res = 0.125, ofs = 0.4375
# @ res = 0.25, ofs = 0.375
# @ res = 0.5, ofs = 0.25
# @ res = 1, ofs = 0.5
# @ res = 2, ofs = 1

Why isn't the correct offset for 0.25 = 0.125, or for 0.125 = 0.0625? What could be the pattern that makes it jump from 1 to 0.5 to 0.25 to 0.375 to 0.4375? The trend for the required offset seems to be -0.5 -0.25 +0.125 +0.0625 which quite confuses me since everything scales the same way.

MirceaKitsune

I ultimately gave up and went for splitting each texture / material its own mesh. Not because the math was too hard (which it was) but because this optimization was incompatible with the face optimization system.

It took me a day to realize that a stretched rectangle cannot repeat the same section of a texture: If you set its UV coordinates to cover just that section it will focus on it, but the pixels stretch with the face as the whole square is represented. If I wasn't merging faces and each surface was a square of equal size then this would have worked; As it stands combining multiple faces is far more advantageous than using identical materials. Faces with the same texture are still combined into one mesh of course, the split only occurs when and where a different images is used.

Sorry I couldn't provide an answer. But hopefully the things I posted here in my attempt will prove useful to someone that finds them. Oh, and the issue I mentioned above was caused by accidentally using "int" instead of "float" which was rounding the number and erasing its decimals.