_i16dot_minrepro.nx

_i16dot_minrepro.nx -- the feature program for __i16_dot (R0r-b, 2026-09-17): the compiler's whole-chunk integer dot, a:*i16[n] . b:*i16[n] -> i64 with the int32x8 accumulator register-resident and one horizontal sum per call. nx_cc_equiv_gate builds this with the CHALLENGER only (a baseline that lacks the builtin cannot), runs it, and requires exit 0. Every arm is checked against an exact scalar reference read back from the same bytes: exit 1 = the 1024-lane signed ramp disagrees with the scalar sum exit 2 = the smallest legal call (one 16-lane trip) disagrees exit 3 = a planted +1 in one lane of a does not move the dot by exactly that lane of b (a lane was skipped) exit 4 = the reference is zero (a vacuous fixture; the ramp is built so it cannot be) exit 5 = the sum of eight int32 lanes exceeds int32 and the builtin's int64 horizontal sum lost it The contract the builtin holds the caller to: n is a POSITIVE multiple of 16, and n is bounded so no int32 lane overflows (nx_nofloat_llm keeps n at NF_CHUNK_K = 1024 lanes of 12-bit values). expect_exit: 0 license_tier: ORIGINAL