Displaying huge letters without memory overflow

[akmere]

Hello guys,
is there any straigtforward way to display huge letters without overflowing memory? Even at the cost of quality. At some level of magnification/scaling my program is getting super laggy.
. As of now if I want to make a text bigger, I'm just putting the bigger scale to TextArea.

let text_area = TextArea {
            buffer: self.buffer.as_ref().unwrap(),
            left: lefto + x_translation,
            top: topo + (configuration.height as f32 - y_translation),
            scale: **scaling**,
            (...)

I can put some fake lower resolution in text_rendere.prepare(), letters will be appropriately bigger, but it surely isn't pragmatic.

I'm using the current version of from github. Thanks for all help!

[grovesNL]

Hi! 👋 Are you able to reproduce this on the hello-world example with a large font size? #74 might help slightly with startup time

[Patiga]

Hey, so far I really enjoy using this library, however I'm stumped right now with text scaling, in the same boat as @akmere .

What I want to do is render text at different sizes/zoom levels, without populating the atlas with differently sized glyphs. Best case for me would be that I can simply specify at which resolution level the glyphs go into the atlas.
I'm okay with the text looking pixelated when zooming in and getting aliasing effects when zoomed out.
Like akmere, I thought that the scale field of TextArea would do exactly that, but that fills up my atlas very quickly.

Thanks for helping and maintaining the lib! :)

[BillyDM]

For rendering large text it's probably better to just use the swash crate to render large text directly into a single bitmap, and then draw that bitmap on the GPU as a regular image. It would be nice if there was a convenient way to do this.

[BillyDM]

More specifically, you only need to perform a render pass for text that has changed. Once the text has been rendered to a texture once, you can reuse that texture for future frames.

[BillyDM]

If you want to get fancy, you could even pack the rendered textures into a single texture to reduce texture switching overhead.

[BillyDM]

Oh actually, looking it up there is another solution, which is to tessellate the text into triangle meshes, and then just render the meshes on the GPU like normal. The epaint library from egui can do this: https://docs.rs/epaint/latest/epaint/tessellator/struct.Tessellator.html#method.tessellate_text

Though you might need to do MSAA to antialias the text.

[Patiga]

After reading up a bit on sdf font rendering it appears that their performance should actually be comparable.
Their drawback appears to be that it doesn't render the fonts perfectly due to rounded corners, which is a bit more complicated to fix.
Now I'm kinda motivated to write my own sdf wgpu text renderer. Like in kaku, single-sized text rendering can be provided as an alternative, as it is not much different.

kaku doesn't appear to use a texture atlas and there is some other stuff I want to do differently like allocations and dependencies, so I'll try to roll my own for a bit :)

[BillyDM]

Just use a rounded font :)