The merging logic is highly cpu-intensive, thus causing unnecessary slowdown of the client thread that could be spent rendering more chunks instead
The CPU overhead of managing parts of the data for the megaquads was still present even if it was disabled (min/max XYZ, isRectangle)
The GPU-side performance and VRAM gains were negligible
Additionally, the smaller codebase around the vertex data shuffling will later allow for easier compatibility tweaks with mods that change the vertex data layout.
Additional changes in the commit not related to this feature:
Arrays no longer get zeroed first inside setState, they will all get overwritten immediately after anyways
Swapped the position of the brightness and color array fields to better align with the Tessellator's ordering (didn't change the vertex layout indices, that's planned for a separate PR in the future)
The reasons behind this are the following:
Additionally, the smaller codebase around the vertex data shuffling will later allow for easier compatibility tweaks with mods that change the vertex data layout.
Additional changes in the commit not related to this feature: