Most efficient way to use cosmic-text with GPU rendering?

[BillyDM]

I'm creating a high-performance low-level UI library https://github.com/MeadowlarkDAW/firewheel. It uses femtovg for rendering 2D shapes on the GPU using OpenGL, however the built-in text rendering is poor quality so I want to replace it with cosmic-text.

I understand that cosmic-text renders on the CPU, and that's fine. My question is what is the recommended and/or most efficient way to send textures to the GPU?

In your examples, it appears you render the whole line of text into a single texture on the CPU and then send that texture to the GPU https://github.com/pop-os/cosmic-text/blob/main/examples/editor-libcosmic/src/text.rs.

The Vizia UI library takes a different approach by caching each glyph into a texture atlas and then sending commands to the GPU to sample each glyph one at a time from that atlas https://github.com/vizia/vizia/blob/main/crates/vizia_core/src/text/cosmic.rs.