Re: match.pd tweaks for vectors and issues with HONOR_NANS

[Marc Glisse <marc dot glisse at inria dot fr>]

On Sun, 16 Nov 2014, Marc Glisse wrote:

> On Sun, 16 Nov 2014, Richard Biener wrote:
>
> > I think the element_mode is the way to go.
>
> The following passed bootstrap+testsuite.
>
> 2014-11-16  Marc Glisse  <marc.glisse@inria.fr>
>
> 	* tree.c (element_mode, integer_truep): New functions.
> 	* tree.h (element_mode, integer_truep): Declare them.
> 	* fold-const.c (negate_expr_p, fold_negate_expr, combine_comparisons,
> 	fold_cond_expr_with_comparison, fold_real_zero_addition_p,
> 	fold_comparison, fold_ternary_loc, tree_call_nonnegative_warnv_p,
> 	fold_strip_sign_ops): Use element_mode.
> 	(fold_binary_loc): Use element_mode and element_precision.
> 	* match.pd: Use integer_truep, element_mode, element_precision,
> 	VECTOR_TYPE_P and build_one_cst. Extend some transformations to
> 	vectors. Simplify A/-A.

Hmm, that left before I wrote any explanation... There are occurences of 
HONOR_XXX (TYPE_MODE (t)) in other files (reassoc for instance), but we 
have to start somewhere, so I only touched match.pd and fold-const.c. I 
did not change the cases where it was clear that the type had to be a 
scalar.

--
Marc Glisse