Re: int vs. float divide optimization on AXP EV45

[Torbjorn Granlund <tege at matematik dot su dot se>]

  GCC has code to implement integer division by performing multiplications
  in the FPU.  It's code Torbjorn contributed to GCC several years ago.  I
  believe you can find details about it in the 1992 or 1993 PLDI conference
  proceedings.

Not quite.

GCC has code for doing integer diviision with compile-time constant divisors
by using run-time integer multiply.  No floating-point operations are
involved.

An article was published in the 1994 PLDI proceedings ("Division by
Invariant Integers using Multiplication").  Not all of what is described in
that article is implemented in GCC.  In particular, invariant divisors that
are not compile-time constants are not handled.  Also, we never use
floating-point operations.

Torbjörn