[PATCH] Transform C - zext(bool) -> bool ? C - 1 : C

[Stephen Lin]

> Basically, the question is whether or not the code generators prefer
> having selects of binops or binops of selects, when the two are
> equivalent. Most binops are canonicalized to the latter already, so I
> would like to fill in some gaps that I noticed. This would yield code
> of the same basic form as your proposed canonicalization (a select
> with a binop as an operand) and I would like to know if there are any
> backends that generate suboptimal code from this form, in general.

Sorry, I meant canonicalized to the *former* already.