[PATCH] do not lower a/b to a*(1/b)

[Paolo Bonzini]

> 	If the patch that affects -ffast-math has not been tested to
> ensure that it does not introduce any new miscompares to SPEC, then it has
> not been sufficiently tested for any maintainer to approve the patch.

While I agree in principle, SPEC is non-free, the patch has testcases, 
and in the worst case there's always the 48-hours rule.

I'm asking on IRC whether somebody can test it for me.

Paolo