nx_fpga_cmp.nx

nx_fpga_cmp.nx -- LIB: RUNG 8b -- the ALU COMPARE ops (SLT signed, SLTU unsigned) on the LUT4 fabric. New mechanic: derive a result from a subtractor's INTERNAL signals (carry-out + sign bit), and use CONSTANT LUTs (init 0xffff = always 1, the carry-in; init 0 = always 0, the fill). Computes a-b = a + ~b + 1, then: SLTU = NOT(carry-out) -- unsigned a<b iff the subtraction borrows. SLT = (a_sign XOR b_sign) ? a_sign : (a-b)_sign -- signed a<b, overflow-correct (differ-signs -> a is neg; same-signs -> the (non-overflowing) diff's sign). A control bit `unsigned` selects which (result in bit 0; bits 1..W-1 are 0). Verified vs nx_rv64im_alu_compute. FABRIC (npi = 2W+1: PI 0..W-1=a, W..2W-1=b, 2W=unsigned). ncells = 3W+6: cell 0 = const1 (init 0xffff) cell 1 = const0 (init 0) cells 2..W+1 = notb_i = NOT(b_i) cells W+2..3W+1 = ripple adder a_i + notb_i, carry_0 = const1 (sum cell W+2+2i, carry cell W+2+2i+1) cell 3W+2 = sltu = NOT(carry-out) 3W+3 = diffsign = XOR(a_sign,b_sign) cell 3W+4 = slt = MUX(diffsign, subsign, a_sign) 3W+5 = res = MUX(unsigned, slt, sltu) NEVER-BRICK (#26): pure integer, bounded, total, deterministic, zero hardware writes. license_tier: ORIGINAL