Block a user
Performance with USE_SIMD_FALLBACK=ON is surprisingly close to the SIMD performance. Investigate
Use binary search for Node16 if simd is not available
This was recommended in the ART paper, but linear search looks to be faster.
Use binary search for Node16 if simd is not available
andrew
deleted branch simd-node16-insert-point from weaselab/conflict-set
2024-03-28 18:35:06 +00:00
Use simd to find insertion position for Node16
Fuse fast path switch into regular switch to reduce branches
Tried it, but it didn't seem to improve instruction count according to cachegrind
Fuse fast path switch into regular switch to reduce branches