>>2884989Cube accessory and stress test shader
https://litter.catbox.moe/v6fnuq.7zCube should be 96 faces, each with its own material. With 12 cubes, I get 30 fps default shader, 30 fps with 8 ops, 19 fps with 800 ops. 8 ops is actually slower than default shader: if I set my CPU to powersave at 1/3 frequency, default shader is 15 fps, 8 ops is is 11 fps, and 800 ops is 6 fps. These values don't change based on whether the cubes are visible in the camera.
So there are various limits. CPU can be a limiter, or there can be limits in the GPU. My understanding is that texture reads are often optimized to happen at the same time as shader math, which can mean that more texture reads or more shader operations can be 'free' until the other operation stops being the bottleneck. Probably something similar with cached textures and like L1, L2 cache at some point.
So a lot of the time, the CPU won't be the bottleneck. Especially with MMM if skinning is done in the vertex shader, meaning done by the GPU, instead of by the CPU as with MMD.
>https://files.catbox.moe/529sgz.rar (+カメラtest)I loaded up the same test effects as before, that apply the matrices from +Camera and +CameraTest to objects, and when rotating Camera bone, the output rotations are different. So that works, even in MMD. I guess the question becomes, can it be chained properly? So the parents, which might just be the All Parents and Viewpoint bones, also rotate the output in the inverse direction?
Note that I'm not bothering to look at the numbers, in part because due to a Wine bug (not sure if fixed with Wine 9.0, or if it's a system-specific bug) fills the window decoration with screen buffer garbage most of the time, hiding the numbers.
It's definitely not like old people who just ignore all the malware that has infected their browser and computer.