How to draw lines in unity efficiently?

As we all know, we can set a lot of triangles no more than 65536 faces in one mesh. Now, I want to draw a lot of lines(more than 10000). When I call SetIndices(int[] indices, MeshTopology.LineStrip, int submesh), I only can set one line in one mesh. If I need to draw 10000 lines, do I need to use 10000 meshes? It's crazy!
It's really appreciate if anyone could give me some advice!

Thanks,
Dobbie

 

If I use MeshTopology.Lines, there will be a lot of duplicate vertices. Because I need to draw curves actually.
For example, If I have 10 points in each curve, I will add perhaps 2 times vertices than using Line_Strip. Does it influence the performance? I think it's still not efficient.

 

Thanks! I will have a test about it.
But I can still set no more than 65536 vertices or indices in one mesh？

 

Unity has 32 bit indices from 2017.3

 

Does it also support 32 bit vertices? I only see a indexformat in Mesh.

 

The indices are for the vertices.
https://docs.unity3d.com/ScriptReference/Mesh-indexFormat.html

 

ok. It means vertices are also 32 bit if indexfromat is 32 bit.
Thanks!

 

Thanks! I will take a look. I'm not familiar about LineRenderer.
But I think it just draw the linestrip. If so, each line needs a LineRenderer?

 

OK. You mean I need to preprocess the vertices data before rendering.
1. But for public static void Simplify(List<Vector3> points, float tolerance, List<int> pointsToKeep), the input points must be a list of line_strip points? If I want to simplify a lot of lines, could I finish it in one simplify command?
2. Do you know if there are some tools to simplify the lines before put it into Unity3d?

Thanks!

 

But the simplify function will think the input data is a continuous line. E.g a line from a to b to c would be 3 points. it's the same as line strip.
So each simplify function can only deal with 1 continuous line. I need to call a lot of times simplify functions to preprocess my lines. Right?

 

Every lines have perhaps 32 points and there are maybe 10000 to 50000 lines. So I need to call thousands of simplify functions. Is it too much?

 

Sorry for necroposting.
I've just tried to use degenerate vertices in a cube and doesn't seem to work with LINE_SPRIP. I do it like below:

Code (CSharp):

1.        
2.         /*
3.          *  4 . . . 7
4.          *  . \     . \
5.          *  .   5 . . . 6
6.          *  .   .   .   .
7.          *  0 . . . 3   .
8.          *    \ .     \ .
9.          *      1 . . . 2
10.          */
11.         public static readonly int[] CubeIndicesLineStrip =
12.         {
13.             0, 1, 2, 3, 0, 0,  // Draw bottom
14.             4, 4, 5, 6, 7, 4,  // Draw top
15.         };
16.  

The result:

 

For Line Strip, how does it work exactly? You can terminate a line by putting for instance line 1 = 0,1,2,3,4,5. so 0,1,2,3,4,5,5 and then respectively 6,6,7,8,9,10? Or would you double the vertices and not the buffer?

edited again to add: Does MeshTopology.StripLines look for repeating indices? Every time it finds a new index (does not equal the previous one) it continues the line, and if it finds the same index it ends it there and if there are more it makes new lines?

Seems like from a memory management perspective there doesnt need be an index buffer, I would just mark 0,5 and then 6,10. Unity could divide the buffer size by 2 and get the number of lines.

Also do I have to specify the color for each vertex using MeshTopology.Line and StripLines?

Edit To Add:

I would like to consider using multiple lines of differnt colors but it would seem easier if you had an array of lines and line[1].color = color.something;

So Line1: 0,1,2,3,4,5,6,7,8,9
Line2: 10,11,12,13,14,15,16

LineColors[0] = color.red;
LineColors[1] = color.green;

etc etc where each element in the array represents all of the colors in the line array.

Line1 could be an int vector 2 as well so that...

Line1.x = 0, y = 9;

Better yet:

LineArray[0].color = red;
LineArray[0].vertices = Vector3[];

Mesh myMesh = new Mesh(bool isLineArray, vector3[] vertices, vector2[] buffers, color[] colorsForEachLine)


Could be made so that if it recieves a color instead of a color array, all lines are of this color, or if it receives a color array that does not match the indices, it knows that the colors are for ranges of lines. IF the GPU requires indices, Im sure it would be simple enough to write a shader that generates that on load? Or if it requires a full array of colors that match each vertex, it could just copy the value for each one. It would in theory save on data transfer, also if clever, I am sure it would save GPU processing time too.

*** Last edit
Also is there any under the hood stuff going on so that we can create our own lines? is it like the GL.Draw type of stuff? Or what would be the absolutely most efficient way to render lines?

 

They don't want the vertical line connecting the top and bottom halves of the cube. Specifically I believe they're asking about your old comment here:

To me this sounds like a driver implementation quirk / bug as it's not part of any spec for LINE_STRIP I know of for either OpenGL or Direct3D. So it's possible on some hardware degenerate line segments would create a break a line strip, but it certainly doesn't appear to on my RTX 2080 or the GPU @xotonic is using.

Generally speaking there's not a lot of reason to use a line strip, explicitly because to make a break in a line strip is supposed to require creating a new sub mesh which is going to be bad for performance if you need a lot of separate lines. On very old GPUs (90's) there used to be a noticeable performance improvement when using Triangles Strip vs Triangle based indices, but on modern GPUs there's no benefit, and possibly even a performance hit from using triangle strips so basically all engines expunged support for them. Including Unity.

LineStrip

remains, but I suspect on modern GPUs there's a similar lack of performance improvement from using them over

Lines

. I just setup a test case with a 65535 vertex spiral sphere that I could switch between

Lines

and

LineStrip

topology & matching indices and I could not see a performance difference even though the

Line

has twice as many indices. That's because the "work" the GPU has to do is per vertex and per pixel, not per index or index pair. And the difference between 0.125MB of 16 bit

LineStrip

index data and 0.25MB of

Line

index data is inconsequential compared to the constant 0.75MB of vertex positional data.

TLDR: Stop using

MeshTopology.LineStrip

and use

MeshTopology.Line

if you want multiple unconnected lines in a single line mesh.

That depends on the material, or more specifically the shader you're using on the mesh. A line mesh is still "just" a mesh. Basically any shader you can use on a triangle mesh you can use on a line mesh. As long as you provide the vertex data that shader needs, it'll render however you want. Just as 1 pixel thin lines instead of as triangles. If you want to assign the color per vertex, then assign the vertex colors. If you need to have one line that's a solid color and another line that's a different solid color but which shares a vertex position, then you need a duplicate vertex at that position with two colors. Or you need to draw two completely separate line (sub)meshes with different materials that have a different constant color assigned. If you want lines that accept the vertex color, you probably want to use a particle shader.