Notes:
The answer is at the place where the sign bit is set, at the upper arrow. We see there, that the actual address of the constant 10 for positive exponents or 0.1 for negative exponents is jammed into the address field of the multiply instruction. Sure it is non-reentrant self-modifying code that could not work in the modern architectures, but it is effective.
Today we would call this an extreme form of hoisting the "if" expression from an important loop all the way up to the two places where the Boolean was set. We may wonder if this was a good idea or not. It was, in the context of the early sixties.