Older games used precompiled shaders. These are inaccessible to the game devs and usually handled by the hardware makers, so the platform OEM for consoles and the video card OEM on PCs. Game devs have begged for the ability to write their own shaders for years and finally got it with DX11 and Vulkan. And that's when things went to hell. Instead of the shaders being written and compiled for the specific hardware, they now have to be compiled for your GPU at run time. It's a messy and imperfect process. EA, Ubisoft or anyone else is never going to have the same level of understanding of a GPU that Nvidia or AMD will have. Often the stuttering is due to the shaders having to be recompiled in game, something that never happened before.
Compiling shaders directly from a high level representation to the GPU ISA only really happens on consoles.
In DirectX on PC, shaders have been compiled into an intermediate form going back to Direct3D 8. All of these intermediate forms are lowered into an ISA-specific instruction set by the drivers.
This final compilation step is triggered lazily when a draw happens, so if you are working on a "modern" engine that uses thousands of different material types your choices to handle this are to a) endure a hiccup as these shaders are compiled the first time they are used, b) force compilation at a load stage (usually by doing like a 1x1 pixel draw), or c) restructure the shader infrastructure by going to a megashader or similar.
> Compiling shaders directly from a high level representation to the GPU ISA only really happens on consoles.
When targeting Apple platforms, you can use the metal-tt tool to precompile your shaders to ISA. You give it a list of target triples and a JSON file that describes your PSO. metal-tt comes with Xcode, and is also available for Windows as part of the Game Porting Toolkit.
Unfortunately, most people don’t do that. They’re spoiled by the Steam monoculture, in which Steam harvests the compiled ISA from gamers’ machines and makes it available on Valve’s CDN.
metal-tt/metal-nt require you to specify the exact architecture, and that's not forward-compatible unless you update your application for every new device release. Even minor SKU revisions like applegpu_g13p/applegpu_g13g/applegpu_g13s/applegpu_g13c are different compilation targets, and that doesn't help you when Apple releases applegpu_g14.
Yep. That’s why the artifact includes the IR, which can be compiled on first use on new architectures.
> Compiling shaders directly from a high level representation to the GPU ISA only really happens on consoles.
No, that's not correct. In fact, it's mostly the other way around. Consoles have known hardware and thus games can ship with precompiled shaders. I know this has been done since at least PS2 era since I enjoy taking apart game assets.
While on PC, you can't know what GPU is in the consumer device.
For example, Steam has this whole concept of precompiled shader downloads in order to mitigate the effect for the end user.
> Consoles have known hardware and thus games can ship with precompiled shaders. I know this has been done since at least PS2 era since I enjoy taking apart game assets.
That's what I said. Consoles ship GPU machine code, PCs ship textual shaders (in the case of OpenGL) or some intermediate representation (DXIL, DXBC, SPIRV, ...)