Re: [lno] [RFC] if-conversion and auto vectorizer

[Diego Novillo <dnovillo at redhat dot com>]

On Fri, 2004-03-05 at 14:27, Devang Patel wrote:

> My worry is it would it cause confusion to gimple verifiers?
> Would it be possible that during verification this is flagged
> as non gimple expression ? If not and CFG routines want alter
> control flow when they see it then its fine. I do not know
> if someone else will be tickled by COND_EXPR or not.
> 
We just have to add the right bits in the GIMPLE grammar.  We will need
to define exactly what kind of arguments we want to allow.  In
principle, I'd say that we would do something like

	cond-assign : MODIFY_EXPR
			op0 -> is_gimple_min_lval
			op1 -> cond-assign-rhs
	cond-assign-rhs : COND_EXPR
			op0 -> is_gimple_condexpr
			op1 -> is_gimple_val


Diego.