Adding two numbers in one register

two_8.8_one_reg.txt

Square by Pulse does this trick of adding two 8.8 fixed numbers in
"one 40-bit register" in the inner loop of it's pin-mapping routine. I
wonder if this is a known bit-twiddling trick documented some where?
(It feels like Lamport's 1976 paper where u is a 23 bit number)

Say we have u and v in 8.8 fixed point format and we want to increment
those by some signed deltas du and dv also in 8.8 fixed point.

I'll only consider a negative du because that seems to be an
interesting case in the code.  Eg. let u = v = 64 = 64.00 = 0x4000 and
du -2 = -2.0 = 0xFE00, dv = 2 = 2.0 = 0x0200

Square sets up the 40-bit delta increment in CL:EBP registers as:
  0xFFFFFFFE00 - 40-bit signed-extend du
+ 0x0200000000 - 40-bit dv << 24
--------------
= 0x01FFFFFE00

and the 40-bit value in BL:EDX registers as:
  0x0000004000 - 40-bit u & 0xFFFF
+ 0x4000000000 - 40-bit v << 24
+ 0x0000800000
--------------
= 0x4000804000

Addition proceeds as:
  0x4000804000   64.00 64.00
+ 0x01FFFFFE00
--------------
  0x4200803E00   66.00 62.00
+ 0x01FFFFFE00
--------------
  0x4400803C00   68.00 60.00
+ 0x01FFFFFE00
--------------
  0x4600803A00   70.00 58.00