The approach here can be used to process strings in chunks without too much added complexity, and it has low setup/teardown costs. The lookup tables fit in thre

The approach here can be used to process strings in chunks without too much added complexity, and it has low setup/teardown costs. The lookup tables fit in three 128-bit SIMD registers so there's minimal cache pollution, which is my usual worry with clever table-driven schemes for accelerating things.