Gabriel Paubert wrote: > Now in practice what would be the cost of checking that the divisor > is -1 and take an alternate path that computes the correct > results (in modulo arithmetic) for this case ? We actually had to do this on the x86 early on for GNAT, UGH!