https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98169
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I guess the question is if we should consider == or ord as canonical form for
the self-comparisons in GIMPLE, or shouldn't canonicalize at all.
Ditto for != and unord.
And depending whatever we choose make sure the targets emit optimal
instructions for that.
In match.pd we have:
/* Simplify comparison of something with itself. For IEEE
floating-point, we can only do some of these simplifications. */
(for cmp (eq ge le)
(simplify
(cmp @0 @0)
(if (! FLOAT_TYPE_P (TREE_TYPE (@0))
|| ! HONOR_NANS (@0))
{ constant_boolean_node (true, type); }
(if (cmp != EQ_EXPR)
(eq @0 @0)))))
(for cmp (ne gt lt)
(simplify
(cmp @0 @0)
(if (cmp != NE_EXPR
|| ! FLOAT_TYPE_P (TREE_TYPE (@0))
|| ! HONOR_NANS (@0))
{ constant_boolean_node (false, type); })))
etc. rules, so if we e.g. decide that a ord a is canonical form of a == a,
we'd need to add a (simplify (eq @0 @0) (if (FLOAT_TYPE_P (TREE_TYPE (@0)) &&
HONOR_NANS (@0)) (ordered @0 @0))).
Note that we already canonicalize the ge and le to eq, which I'm not sure is
correct, because a <= a should signal on qNaN but a == a will not.
