Today Torque 3D released version 3.9 and in a somewhat odd release blog post they spoke mostly about the plans for the next release. Torque is an MIT licensed open sourced 3D game engine written in C++. This release constitutes 515 commits with the major releases being proper roll over to deferred renderer, an initial implementation of entity/component and DirectX 11 integration.
As mentioned earlier, the majority of the release blog post covers features intended for the 4.0 release:
So, here's a main features list that we're looking at bringing into Torque for 4.0:
- Physically-Based Rendering. This is, unsurprisingly a pretty big one. Luckily, it's also most of the way done! The only main piece that needs finishing is reflection probes, and the core features are in. From there it would just be a matter of dialing in the supporting art pipeline and tweaking the math to make it look as good as possible.
- MacOS support. This was originally slated for 3.9, if you look at the 3.8 release blog, but a lack of machines to test on, and the age-addled old platform code made it impractical to really tackle initially. The good news is, near the end of 3.9's development, with SDL and epoxy to take a LOT of the workload off, huge strides were made and it's actually almost completed. There's still some issues with rendering on some machines and catching errors and warnings that crop up, but it's looking very practical to have MacOS as an official platform for 4.0! You can see Timmy's screenshot in the 3.9 RC thread showing off T3D running on MacOS with his and JeffH's work now.
- SDL as the main platform layer. Torque3D's supported SDL for quite a while now, and it's naturally the main platform layer for Linux(and MacOS). However, it's been a little bit of a red-headed stepchild to the Win32 platform code. For 4.0, that's changing and SDL will be adopted as the core platform layer with whatever glue needed sitting atop it. This will drastically simplify the platform code and make it easier to maintain. It also offers a good bit of future proofing should any other platforms come into the equation down the line. SDL's stuff is very nearly at parity with the Windows side, it's pretty much just a matter of jumping over the last bits and doing cleanup/bugfixes to make it rock-solid.
- New Project Manager The old project manager was useful, but was hacked together with QT and php, so maintaining it was a little sketchy and fell by the wayside for the more standard CMake gui. However, there's lots of management things CMake can't do, which is why the Project Manager will be making a comeback, it'll just be a graphical frontend sitting upon the more standard cmake project generation. This new PM will let you manage your updates, and hopefully in the future hook into online repositories for easier grabbing and integration of updates as well as fetching content packages without hassle.
- New Shadergen As great a job as the current workhorse does of generating shaders for all the many materials you need to make a game, updating and expanding it has proven to be a pretty annoying thing to deal with due to how it's structured. Excellent peice of tech, but a chore to maintain, just like ye olde German tanks in WW2. As such, we'll be looking into restructuring shadergen to not only make it easier to update and expand upon in the backend-sense, but also easier to build materials/shaders for the end user as well, up to, and probably including a node-based approach for engineering your fancy shaders for your fancy materials.
- Graphics API refinements DirectX 9 has proven to be a monstrous workhorse of the graphical APIs, reliably serving well beyond what anyone though it would. However, while it has proven to be a rock-solid API, the time has come to retire the poor girl and send her off to the glue facto-eeeer, the farm. Yes. As such, DirectX 11 will be taking over as the main Dx rendering API, and will continue to see refinements to bring it up to speed now that Dx9 isn't hampering it or OpenGL. Timmy's been poking at this and has noted improved performance in Dx11 by quite a lot, and some gains in OpenGL as well.
- Threads. Threads for days One major blocker to rendering performance that even dropping Dx9 won't help is that we curretly don't thread or do much to optimize out the render calls themselves. That'll change in 4.0. We've been brainstorming the optimal approach for a few months now and have a pretty solid plan of attack that will see the render calls threaded out, and also proper batching of rendered geometry where appropriate. So fewer, faster drawcalls for the same workload. We'll also be looking at threads for handling resource loading and spawning of objects to cut out all those hitches when the map starts or stuff is created.
- Hardware Skinning Another thing that's been pretty much complete outside a few oddball behaviors we'll be getting in there is Hardware Skinning. When it's behaving, the current implementation already notes a huge improvement in performance for animated meshes, so getting it polished up and in should be a huge, immediate performance boon.
- Physics API T3D's had a physics abstraction layer for a long time now, and it's been pretty useful in letting the end user decide if they wanted PhysX or Bullet. However, Torque's stock physics has still served a useful niche in being network-reliable, or lightweight for basic physics mechanics, and some people find it to be a pretty workable thing. As such, the Torque physics will be converted over into a Physics plugin, so all physics and collisions will go through the Physics API and standardize all the behavior for that across the engine, cleaning up quite a lot and making it easier to maintain.
- Entities, Components, and Assets This is the big one from non-rendering side. 3.9 already has the initial implementation of the entity/component stuff, as well as the Assets/Modules systems, but they're not yet fully utilized. That'll change in 4.0. If you've been checking out my work blog with the recent updates, I've been covering work I've been doing on the improved asset pipeline, as well as continued work with the Entity/Component stuff. For 4.0, this will become the standard, and all existing gameplay classes(Vehicle, Player, Item, Shapebase) will be replaced with GameObjects built of entities and components to do the same work. The idea is to give users a similar base to what is in Torque now in terms of starting objects, but remove all the bulky, hardcoded functionality. It should be a standard point, not an anchor, after all. Tying to that, we also will have:
- Assimp support This was actually mostly done back in the day, just some issues with animation stuffs and a few other minor problems. I'd poked at this with some R&D time a while back, so it shouldn't take much tweaking to get it polished up and hooked into the assets system, which will allow quite a few new 3d model formats, including FBX.
- A new base template This is also basically complete. This will be a much more streamlined starting template intended to easily drop in modules, assets and the like into to build up your game project, as opposed to having to spend time stripping out the stuff you don't need. Numerous ancillary improvements also help load times, easier to read and comprehend code and the like. This will replace the empty and full templates, cutting down on the effort needed to maintain(no need to duplicate changes).
- Vive support This comes from our own Mango's efforts. It's pretty much done, but didn't quite squeak in for 3.9. It will be going into 4.0 for sure.