Most time gui spends in femtovg::path::cache::PathCache::tesselate_bezier, is this optimal?

[qarmin]

When testing my GTK app I found that a lot of time app spend in different libraries like librsvg, pango etc.

I wanted to check how much time spend app with slint frontend and I found that most of ther time, slint spends in femtovg tesselate_bezier function.

App is compiled with release profile

Is this really needed to call it so much time? - it is executed few times per second

This is not really problem for me, since app works quite smooth on my PC(but I use old i7 processor), so maybe performance problems may be visible on slower computers

simplescreenrecorder-2023-12-08_20.12.37.mp4

With software renderer I see that CPU usage is distributed in several functions and I don't see anything similar to femotvg function

simplescreenrecorder-2023-12-08_20.24.13.mp4

With skia also I don't see similar thing

simplescreenrecorder-2023-12-08_20.35.13.mp4

When counting only app code, this function takes ~50% of time

Also reported to femtovg/femtovg#186

[tronical]

Thanks again for your detailed report. You're poking at the right issue :). I believe the other renderers (Qt, Skia) have caching going on to avoid re-doing the tessellation. Plus I guess the elements rendered here, likely border rectangles, are drawn using a dedicated code path perhaps, while with FemtoVG it's the general path renderer. There's a path cache in FemtoVG as well, but it's dependent on the global transformation. So once you start moving a Rectangle it starts falling apart (due to resize or scrolling).

As you point out in femtovg/femtovg#186, perhaps there's a way to also make the function itself faster in FemtoVG - but I don't know that code path well enough to make that call :(