This version seems to work but the allocation of segments has low
utilization. Probably best to allocate in chunks rather than try to
make them contiguous.
Handling f16 requires special work, compared to other scalars, as the minimum conversion operation for u32->f16 in GLSL (unpackHalf2x16) loads two f16s from one u32. This means that in order to minimize unnecessary calls to unpackHalf2x16, we should look-ahead to see if the current f16 has already been extracted in the process of dealing with the last f16. Similar considerations exist for write operations, where we want to pack, when possible, two f16s in one go (using packHalf2x16).
This patch adds a module that contains both scene and ptcl types (very
lightly adapted from piet-metal), as well as infrastructure for encoding
Rust-side.
WIP, it's not wired up in either the shader or on the Rust side.
