simd: agnostic gemm_u8_i8 surface, integer-slice-op lift, per-CPU matrix, BF16 AMX wiring#182

Merged

AdaWorldAPI merged 12 commits into

claude/continue-ndarray-x0Oaw

May 21, 2026

Commits on May 20, 2026

feat(simd): agnostic gemm_u8_i8 surface — compile-time dispatch chain
claude
committed
feat(simd): AVX-VNNI ymm arm for gemm_u8_i8 + bump avx512 cfg to SPR
claude
committed
test(simd): ignored timing harness for gemm_u8_i8 vs scalar
claude
committed
feat(simd): restore array_windows + add slice-level add_mul_f32/f64
claude
committed
docs(simd): per-CPU agnostic-surface resolution matrix + integration plan
claude
committed
docs(simd): expand per-CPU matrix — full integer-lane storage + Phase 6
claude
committed
feat(simd_int_ops): MX-T1a — lift add_i8 / sub_i8 / add_i16 to polyfilled lanes
claude
committed
docs(simd): flip MX-T1a cells + lock in asm-byte rule for AMX/F16
claude
committed