Current design decisions for my game engine:

Rie

I guess there isn't actually a reason to give each window its own command buffer. And it's not like there's any multithreading involved or that windows would need to run at different framerates. So the window-specific data will be the surface, the swapchain, and the related release semaphore. Frames-in-flight will have one command pool/buffer each, and they'll record commands for all windows. Then each frame there's a single vkQueuePresent, handling all swapchains. Yeah.

Rie

Wait! But one of my monitors is 60Hz while the other is 144Hz! I _do_ want windows to run at different framerates or else windows on my main monitor will be bottlenecked by the other one! Ahhhhh

Rie

Ok after doing some research my preliminary conclusion is that the only way to reliably have multiple windows with independent refresh rates is to use multithreading, one render thread per window. Threads are a blind spot of mine that I've been meaning to overcome anyway, guess the time has come 🥴

Rie

Once I have my game engine where every panel in the editor is detachable I'll get a third monitor to make it worth the effort

Rie

Hmm I wonder if it's worth checking out Qt as well as an option for editor GUI... it seems like a heavy dependency but it might be more suited for what I want to do. ImGui is great for in-game debug UI but is not really meant to be used for a full application

Rie

Should... should I write my own UI library? I'm gonna need one for ingame UI anyway, could do it like Godot did and make it so it's usable for an editor as well.

But I was kind of hoping my ingame UI tooling wouldn't need to be constrained by having to support regular desktop widgets and interactions, so I really should just pick a third party solution for the editor... nyeh

Rie

Ok I think I'll stop shedding the bike for now regarding editor gui and multi-window support. Just thinking about it was valuable and clarified some future requirements, but they can be implemented later. Next steps:
- Hardcode a 3D pipeline (like a cube)
- Load a gltf and render it bindlessly
- Dehardcode the pipeline

Rie

Last time the internal mesh data format I used was huge - it was just the vertex and index buffers dumped into a file so they could be memcopied into the GPU quickly without processing. This time I'll also add a (de)compression step, that should still be pretty fast. And use meshoptimizer during the import step.

Rie

ok forget hardcoding a cube mesh in the shader, going straight for glTF model loading. I can copy paste parts of that from the previous iteration of the engine, while improving it at the same time.

Rie

glTF model loading and 3D rendering with depth work! There's some strange bug when I try to load in larger models, and there are a bunch of validation errors about synchronization issues, but those are for another day

Rie

Progress of the evening: fixed the problem that prevented some models from loading (it was just a wrong variable), loading all meshes / surfaces in a glTF now, loading normals. Textures will be next!

Rie

biblically accurate bindless textures

Rie

Engine rewrite progress update after an extended weekend:
- Bindless textures and materials
- Normal mapping
- Scene tree with transform hierarchy
- Camera and scene data in per-frame uniform buffer instead of push constants
- MSAA
- Lots of refactoring and new abstractions

Rie

Some stuff was mostly copy-pasted from my previous project. The really new thing is the bindless paradigm. This finally helped me understand descriptor sets - I was really confused about them last time when I followed vkguide.dev.

Not sure if I'm doing bindless "right" or if I'm bound for troubles down the road but so far it works pretty well! I only do a single vkCmdBindDescriptorSets per frame, binding global resources, and push material IDs per object in push constants.

Rie

I don't know yet how this will change once I add more shaders / material types though. I only have a single graphics pipeline right now, figuring out a system to expand on that will be the next big task.

Also Slang has some stuff dedicated to "bindless" but I didn't really understand it and also it seems to work without it? Might refactor the shader once I figure out the advantage of DescriptorHandle and how to use it

https://shader-slang.org/slang/user-guide/convenience-features.html#descriptorhandle-for-bindless-descriptor-access

Rie

Anyway, possible next steps:
- Mipmaps
- Basic lighting
- Finally start cleaning up resources correctly
- Maybe look into batching the uploads of assets the GPU

Rie

With the combined power of albedo, normal maps, and a simple dot product I present: basic lighting

Rie

Got a bunch of stuff in mind for where to go next (Qt based editor, copying over more systems from the old codebase, etc) but I have to try to slow down development for at least the next month to focus on other priorities

Rie

Despite trying to put my focus elsewhere I've done some experimentation and thinking lately regarding editor UI.

I want to use Qt since it's a mature cross-platform UI framework. But it actually consists of two frameworks: QtWidgets (traditional desktop UI, software-rendered, uses procedural C++) and QtQuick (GPU rendered, mobile support, uses declarative QML and optional JavaScript and/or C++).

QtQuick seemed more "modern" so that's what I started with.

Rie

First major task was getting a 3D viewport rendered by my engine into a Qt window. Qt is quite flexible and there were a lot of different avenues that seemed feasible for this. There were Most of them didn't work out in the end or were intended for different use cases.

What worked for me in the end was:

Piero Bosio Social Web Site Personale

Current design decisions for my game engine:

Feed RSS

Gli ultimi otto messaggi ricevuti dalla Federazione

Post suggeriti

Che poi, spacciano per Made in Italy ciò che non lo è, ma il vero Made in Italy costa.

Send your message now and tell #Democrats they were not elected to fold.

This post did not contain any content.

Jó éjszakát, Budapest