Will it be performance enhancing to seperate calculation of SH and cov3d from vertex shader(namely projection.comp and inverse_index.comp)? I mean it requires additional buffer memory and you can directly use gl_InstanceId to gain gaussian id rather than calculating inverse index. And recently I implemented a webgpu based viewer using two methods: one seperating calculation of SH and cov3d from vertex shader and the other not. Then I found that the FPS dropped by about 16 using the former one. (Or maybe there are bugs in my implementation)