Rendering strategy for 2d

Started by
1 comment, last by Fagertveit 7 years, 12 months ago

Hi folks!

Been struggling with finding good information on how to set up my rendering strategy for a home rolled html5 webgl game engine that I'm working on.

I'm doing 2d mostly, but might mix some 3d into it at a later stage. I wonder if there's some good practice I should keep to when it comes to render calls, shader programs and textures.

Right now I'm trying to cram as much into one rendercall and use a large texture map for most of my textures, this works pretty well and I'm getting a great performance, so it makes me wonder how far I can push the GPU and perhaps do some crazy/fun things when it comes to shaders in future development. I'm also thinking that I should perhaps see the renderer through a more shader centric viewpoint, which means that I have some fun rewriting ahead of me.

Right now I have a pretty general purpose 2d shader that handles most of my draw calls, it can blend colors or just render primitives with colors, or it can render textures. Maybe I can do more specific shaders for specific tasks and add some more shader programs when rendering a scene, what amount of shaders is common for a single scene? can I use 1-5 or maybe 10-20 without worrying?

Same goes with textures, now I'm pretty hardcore, everything I render comes from like 1-3 textures these days, in one game that I'm working on the texture map is pretty large as well (4096^2) Is it good to have large texture maps like that, or should I keep them around 512/1024^2'ish instead?

What are common when it comes to texture binding within a specific scene?

I started OpenGL when it was fixed pipeline, and I've mostly learned through basic OpenGL literature, it seems hard to get good info about a real implementation as most is basic stuff on that level. I really want to get to the next stage when it comes to graphical programming, and those questions have bothered me to no end! Hope to get some good insight from you guys!

Cheerz!

Advertisement

Hi Fagertveit,

Your questions are a little tricky to answer, because they're fairly general. I don't have an experience with webgl, but I'll try to give you what advice I can.

It sounds like you're concerned about state thrashing. Some of the approaches you mention were used in DX9 generation engines for reducing state changes -- back in those days, state changes were such a huge concern that they would cast shadows on other forms of optimisation. That was caused by CPU overhead within the DX9 drivers, and also by pipeline flushes in the GPU hardware.

I'm not sure about webgl, but most of the time it's less of a concern nowadays. If you have a very big scene, with thousands of objects and lot of detail, then maybe it would be an issue... But most of the time, you should be fine even with hundreds of shaders and many textures. Modern hardware and drivers are pretty forgiving.

My feeling is that you're best bet is to make these decisions based on what is easiest for creating content -- not performance. That is, if using many smaller textures is easier to create models, then just do it. As long as you have mipmaps, small or large textures will rasterize at similar speeds.

You can use just a single shader, if you want, just with a bunch of "[branch] if(...)" statements. This is called an ubershader, and it's a viable approach. But you could also just have many small shaders.

A great way to get more information is to use RenderDoc or GPA to profile another application. These can be used to profile and debug your own work; but they can also be used to see what other games do. If you can find a project that is similar to what you want to do, then just open it up in a debugger and see how they approached these questions.

You also (sometimes) open up commercial games like that... And you'll probably see that they are using thousands of resources, with all different formats and dimensions. That can be a great way to learn practical approaches to things. And I think you'll find that sometimes even very popular games often use very simple methods!

@dpj: Thanks for the answer, it's pretty much what I wanted to know as well, I think I've been a bit to restrictive when it comes to state changes, and need to lighten up some, I think my greatest bottleneck has been to prepare the data for the shaders really, and I will redesign this somewhat.

I've done one ubershader for all my 2d rendering, but I think I can start dividing these to more specialized ones to get the most oumph out of the system, I don't think I'll reach more than 20 programs at tops. And 2d games works well with texture atlases, so textures might be around that count as well, but as I've done in the past I've pretty much tried to keep textures in the count of 2-3 hehe, with some being really big!

Good advice to check out other applications and games, I think Firefox has great debugtools for WebGL where you can se which shaders are in use and how they impact performance, so I'll try to find some games that work in similar way like mine and do some digging.

The boring things with most webgl games is that they are made with some of the big frameworks that have somewhat singular rendering pipelines, but the is bound to be some golden nuggets out there somewhere! :)

Many thanks for the reply!

This topic is closed to new replies.

Advertisement