v4.2 Threat Inputs

Non-functional changes

• dev version numbers now contain commit sha
• added GitHub Actions to automatically publish pre-releases
• added go perft [depth] command
• added arch autodetection in Makefile
• removed Skylake target
• added option to compile HCE and singlelayer version

v4.2

• implemented threat inputs for NNUE
• implemented threat input weights prefetching
• SPSA tuned at ltc
• fixed various incorrect draw evaluation for winning positions
• added upcoming cycle detection
• use tt score in probcut
• mix corrplexity in ldse margin
• mix history into lmp margin
• sf style thread voting
• shared corrhists

Performance

• +57 STC against v4.1 (3740 estimated CCRL blitz)
• +75 LTC against v4.1 (3600 estimated CCRL 40/15)

Bench: 10998114

Binary Guide

If you are unsure which binary to use, start from the top and go down until you find a binary that doesn't crash.
Note that all binaries are for x86 systems.

• avx512: Intel Icelake / AMD Zen 4 or later
• avx2-bmi2: Intel Haswell / AMD Zen 3 or later
• avx2: Intel Haswell / AMD Zen 1 or later

Generally avoid avx2-bmi2 on Zen 1/2 AMD CPUs as they can be slower than the avx2 binary.