good in general, but bad nested loops

[Henry Cejtin]

Sorry, I had assumed that just deleting the `jo' instructions would give me
the no-overflow code (not when I tested it for speed, just when I was looking
at it).  I wonder what the OCaml people do about register allocation on the
Intel side.
The scary thing is that all the examples using nested loops (nested loops
itself, matrix multiply and severl others) we are slow on.  I don't quite
know why, or if it is a single problem.
Still, the timings are certainly not bad.  Your backend is clearly doing a
good job, especially considering how young it is.