nx_fp32_q14.nx

nx_fp32_q14.nx -- IEEE 754 binary32 bit pattern -> Q14 fixed-point i64, all-integer. Reads STL coords + any other binary format that stores 32-bit floats in millimetres-scale magnitudes. Why integer-only conversion: Per cardinal feedback-bits-up-exceed-never-match, slicing math should be deterministic. Float ops drift per-platform; integer shift+add ops give bit-exact results on every backend nxc2 targets (x86_64, riscv64, arm64, ...). We pay the float price exactly once at parse-time and then operate in Q14 fixed-point for the entire slicer pipeline. Why Q14: Q14 = 1 unit = 1/16384. Matches nx_mesh's existing coord convention (defined in nx_mesh.nx) so STL-loaded meshes compose directly with nx_voxel_mesh / nx_render_pass / our slicer. Range note: At Q14, max representable mm value before i64 overflow is ~5.6e14 mm = 5.6e11 m. Any STL coord with magnitude > 2^40 mm = ~1e9 mm is clamped to the saturation sentinel. Real-world 3D-print STL coords are < 1000mm typically; Christus is 305mm. No risk. IEEE 754 binary32 layout (little-endian byte order in STL files): bit 31 : sign bits 30..23 : exponent (8 bits, biased by 127) bits 22..0 : mantissa (23 bits, implicit leading 1) value = (-1)^sign * (1 + mantissa/2^23) * 2^(exp - 127) For Q14 scaling we want: q14_value = value * 2^14 = (-1)^sign * (2^23 + mantissa) * 2^(exp - 127 - 23 + 14) = (-1)^sign * (2^23 + mantissa) * 2^(exp - 136) If exp - 136 >= 0: shift full_mantissa left by (exp - 136) If exp - 136 < 0: shift full_mantissa right by (136 - exp)