Finally figured out a nice way to fix BC1 indices.

https://github.com/wolfpld/tracy/commit/b6eb08befef6ddf72b578111b2248749a3e8f366

Four table lookups per block are no longer required. Yay!

Just some bit twiddling involved. Yay!

The compiler can vectorize this. Yay!

It emits scatters. Abort! Abort!

https://gcc.godbolt.org/z/1bn4ezafb