Slick Forums

I've been toying around a bit with the TWL demos, and I noticed that they use a lot of CPU on my little subnotebook. Particularly the TextAreaDemo uses around 40% of my 1.3 ghz core 2 duo, and that for just one smallish text display area.

I did a bit of profiling with the (probably not so accurate) jvisualvm, and the results were that a lot of that time is spent in the actual openGL calls to render the text. Here's part of the snapshot from the sampler.



So I'm not entirely sure, but I believe there is room for optimization there. I had the idea of something like the option of caching text as textures or images, which seem to render significantly faster. But I haven't really looked into the code much yet, so I can't say.

In either case, it'd be really nice if drawing large amounts of text could be sped up to the point where an extensive GUI can actually run on a low-end PC, and with some pointers I'm fine making the necessary modifications myself, if viable at all.

You can write another Renderer implementation which uses different approaches / GL calls to render text.

I don't think that caching whole text as texture is a good idea - as you this might creates lots of textures.

In your case it looks more likely that using display lists is causing the issue for you - you can test that by disabling the current caching in de.matthiasmann.twl.renderer.lwjg.LWJGLFont.cacheText() by always returning null.

_________________
TWL - The Themable Widget Library

Thanks for the quick response! ^^

I tried your suggestion, but it only made rendering slightly slower. (from 40% / 25% on my two cores to 50% / 40%). Something I don't quite understand is that the calls to glClear() also seem to get more expensive, the more expensive the other gl calls get(glClear always takes around 50% in the profiler).

This sounds like your GPU runs into a fill rate limit - as nearly everything in TWL (using the provided default renderer) is alpha blended.
So it's unlikely that you can make it much faster. You could try to disable blending (use alpha testing) for rendering text and use a font without AA.

_________________
TWL - The Themable Widget Library

Ditching immediate mode might also help reduce some CPU load (on my machine, fill-rate bottlenecks lead to high self-time in nSwapBuffers, not glVertex and glTexCoord).