Range extension: bytes → words without overflow

200 + 100 in a byte already overflows. The fix: before doing arithmetic, "unpack" the bytes into 16-bit words. Interleave with zeros — zero extension; interleave with a sign mask — sign extension.

m0 — original 16 bytes
m1 — copy
m2 — second interleave operand

    

    

      
      
      
      
    

    

      low half → punpcklbw
      high half → punpckhbw
      negative byte / 0xFF mask