I used this in a C++ big integer library I wrote in the '90s. One issue I came across is how to best handle it when the operands are different sizes. I asked on sci.math, but never found an answer. 10 years later, a similar question arose on gmp-discuss [1], also sans answer. The second message in that thread quotes my old sci.math post [2], which includes some experimental results that suggest that there are patterns here, but the patterns escaped me.
[1] https://gmplib.org/list-archives/gmp-discuss/2003-January/00...
[2] https://gmplib.org/list-archives/gmp-discuss/2003-January/00...