Faster than std:sort and pdqsort
Beats std::sort by 30% on M1 using sorting networks to eliminate conditional branches.

Beats std::sort and pdqsort by replacing branches with sorting networks.
Systems programmers, performance engineers
pdqsort · ips4o
Beats std::sort by 30% on M1 using sorting networks to eliminate conditional branches.
Beats std::sort by 24% on M1 using sorting networks for small subsets and branchless partitioning.
Removing branches from Quicksort cuts sort time in half on Apple Silicon.
Branch-avoidant stores beat std::sort on M1, but it's a micro-optimization of a solved algorithm.
Branchless partitioning with 512-element stack buffer beats pdqsort on M1 and Ryzen.
Zone-mapped sorted heap with 30x smaller IVF index than HNSW, no pgvector limits.