如何将纹理应用于 OpenGL 中的子网格？

Question

I have one single mesh, which I've divided into submeshes and draw individually by using the indices of the vertices. However, I was wondering what the most efficient way of applying a different image to a different to each submesh. These submeshes are being drawn seperately. For example, this is one submesh.

mat4 view = identity_mat4();
mat4 persp_proj = perspective(100.0f, (float)width / (float)height, 0.1f, 100.0f);
mat4 model = identity_mat4();
model = rotate_y_deg(model, turn);
model = translate(model, vec3(ahead_x, 0.0f, ahead_z));
view = translate(view, vec3(0.0, 0.0, -5.0f));

// update uniforms & draw
glUniformMatrix4fv(proj_mat_location, 1, GL_FALSE, persp_proj.m);
glUniformMatrix4fv(view_mat_location, 1, GL_FALSE, view.m);
glUniformMatrix4fv(matrix_location, 1, GL_FALSE, model.m);
//glDrawElements(GL_TRIANGLES, mesh_data.mPointCount, GL_UNSIGNED_INT, 0);
glDrawArrays(GL_TRIANGLES, wheels[0], wheels[1] - wheels[0]);

Is there a simple way of achieving this? I was considering binding the appropriate image for every mesh that is being drawn, but seeing as all my VBO managing is done in a different function, I thought it might be overly complicated. Thanks.

Answer 1

In general, all sub-/meshes will have different materials and image maps. Often they will require also different shaders, because the surface needs different BRDF (bidirectional reflectance distribution function) to convey the object's appearance.

So you shouldn't think that having separate textures is something inherently bad. If there's not much objects or submeshes, you might set texture with every draw, and leave it like that.

As your draw count will approach some hundreds calls per frame, you might see that cpu load getting higher, and fps dropping. This is when you need to rethink the approach, and there's no silver bullet. The "most efficient" is every time will be something different, depending on the things you need to draw.

For simplest optimization, you'll need to reorder objects drawing, so the objects with the same shader/texture will be drawn first, and you don't have to change textures between drawing multiple submeshes.

More complex optimizations will require you to merge multiple submeshes with same material into one bigger mesh, so it would be draw in a single draw call. And at this point you might even create bigger texture on the fly (it's called "texture atlas"), combine multiple images into one texture, and reposition submeshes' texture coordinates to point in the correct portion of this new bigger image. This works especially good when there's many objects with small textures, like sprites.

In short, start doing complex design when you know you'll have to draw hundreds of objects every frame. With simpler scene it's ok to bind submeshes' textures with each draw call.