Gabriel Paubert wrote:
Now in practice what would be the cost of checking that the divisoris -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!