Wire NNUE eval into search, build the training pipeline, and add a fast

move/eval-agreement benchmark against Stockfish (tests/lop)

Implements the architecture decisions in src/docs/NNUE.md: HalfKAv2_hm
feature set, int16/int8 quantization, per-ply accumulator backup for
king moves, and a lazy-eval-ready integration into search.c/eval.c.
Adds the full annotate_pgns/ training pipeline (extraction, K-fit/
labeling, quiet filtering, tensor building, training, export) plus an
end-to-end orchestrator (run_nnue_pipeline.py) and validation tooling
(sanity_check_net.py, compare_vs_stockfish_net.py). Adds tests/lop:
a 10k-position list-of-positions benchmark comparing engine eval rank
and move choice against pre-computed Stockfish/lc0 references, much
cheaper than a full self-play match for day-to-day progress checks.

First sd10-matched self-play result against classic HCE: LLR +2.28
(H1=+2.94), -72=119+169 over 360 games, confirming the net is
"smarter per node" at matched depth. A real-time (--st 4) match is
currently weaker (candidate loses ~3 plies of depth to a ~3.3-3.7x
nps gap against an unrecalibrated search), consistent with
NNUE.md's adoption-criterion section -- pruning/reduction constants
still need retuning for the new eval's score scale before a fair
real-time comparison is possible.

Co-Authored-By: Claude Sonnet 5 <[email protected]>

1 files changed, 14 insertions, 0 deletions

diff --git a/src/GNUmakefile b/src/GNUmakefile
index bbe79d1..7dbb143 100644
--- a/src/GNUmakefile
+++ b/src/GNUmakefile
@@ -98,6 +98,20 @@ ifdef GPROF
  PROFILE	+=	-pg
 endif
 
+# Scoped ISA opt-in for NNUE SIMD work (int8/int16 dot products, and
+# the FC head's existing float32 loops which clang already
+# autovectorizes to SSE only without this). Default ON since 2026-10-16
+# -- confirmed a real, measured win for both classic HCE (~3%) and
+# (once real NNUE weights are loaded) the NNUE FC head (~7%, see
+# docs/NNUE.md), and the build box (Zen 2) has full-width 256-bit
+# AVX2 + FMA3. Pass AVX2=0 explicitly to disable (e.g. building on a
+# pre-Haswell/pre-Zen box, 2013 and earlier) -- unlike -march=native,
+# this doesn't assume the BUILD box matches the RUN box, but it does
+# assume AVX2 availability by default now rather than opt-in.
+ifneq ($(AVX2),0)
+ PROFILE	+=	-mavx2 -mfma
+endif
+
 ifdef DEBUG
  PROFILE 	+=	-g -DDEBUG
  BINARY 	=	_typhoon