Documentation
¶
Overview ¶
Command threshold reads BenchmarkGetUint32_Approaches output files (one per SIMD level) and suggests deltaFullUnpackThreshold values.
It compares single-lane walk cost at each position range against full-unpack cost (pos_00_15 with scratch is the cheapest; full unpack for the same block config is the reference). The crossover point where single-lane walk exceeds full unpack suggests the optimal threshold.
Usage:
go run ./internal/threshold benchmarks/get-scalar.txt benchmarks/get-sse2.txt \
benchmarks/get-avx2.txt benchmarks/get-avx512.txt
Click to show internal directories.
Click to hide internal directories.