Optimizations when drawing large number of shapes?

[PetrGlad]

I have implemented an app that should be able to draw a large amount of small simple objects (say, up to 40k of rectangles). The performance turned out to be surprisingly good even without much of optimizations already. So it is usable as-is. But since all rendering loop runs on single core, for the largest datasets CPU load sometimes already hits 100%. Hence I am pondering whether it make sense to implement some optimizations but wonder if those will be worth it.

I have already cached the colors calculation since the profiler had shown it takes substantial time (that made it ~10% faster).

So here are the questions:

• Can I draw the heavy part of the scene to an image (or texture) and then re-use it (painting a clipped version for scrolls) until zoom/resolution/window borders/data model change? Or maybe there is a way to have scalable representation (like SVG) that I could keep even for different zoom factors? For example if tesselation is expensive, it looks like Context::tesselate may help. But since most of the load is in scrolling I would like to clip to visible rectangle afterwards, just before painting. Can I use a layer or an Area for that?
• Does egui do tesselation and clipping for all objects at once, or does it paint-as-you-go?
• What happens if an object is too small to display? E.g. what happens to a rectangle whose dimensions are way smaller than 1 pixel? In other words, if an object is too small to display, does it make sense to skip it explicitly, or it will not help much?
• When window is minimized or otherwise not visible, I may prefer to pause painting (e.g. to save battery). Should I handle it or egui has some helpers for this? Also, is egui essentially single-threaded or some work can be offloaded to other threads/cores?

[PetrGlad]

In my case I probably could even use a Mesh directly. It seems that I can use Painter::add(&self, shape: impl Into<Shape>) where Shape parameter is a Shape::Shape(Mesh). I see the Mesh can be either directly constructed or made with Context::tesselate.

But the other questions still hold, in particular, does it make sense to do build a mesh and re-render it or maybe render the mesh into a image (how) and re-use it?