On 3/28/23 10:50, Aldy Hernandez wrote:


On 3/28/23 09:54, Jakub Jelinek wrote:
Hi!

On Wed, Mar 22, 2023 at 07:32:44AM +0100, Aldy Hernandez wrote:
    * value-range.cc (frange::set): Add nan_state argument.
    * value-range.h (class nan_state): New.
    (frange::get_nan_state): New.

The following patch makes use of those changes in float_widen_lhs_range.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

LGTM.
Aldy

2023-03-28  Jakub Jelinek  <ja...@redhat.com>

    * range-op-float.cc (float_widen_lhs_range): Use pass get_nan_state
    as 4th argument to set to avoid clear_nan and union_ calls.

--- gcc/range-op-float.cc.jj    2023-03-23 15:25:47.119740274 +0100
+++ gcc/range-op-float.cc    2023-03-27 13:28:18.847264635 +0200
@@ -2262,12 +2262,7 @@ float_widen_lhs_range (tree type, const
       or real_max_representable (type) as upper bound.  */
    bool save_flag_finite_math_only = flag_finite_math_only;
    flag_finite_math_only = false;
-  ret.set (type, lb, ub);
-  if (lhs.kind () != VR_VARYING)
-    {
-      ret.clear_nan ();
-      ret.union_ (lhs);
-    }
+  ret.set (type, lb, ub, lhs.get_nan_state ());
    flag_finite_math_only = save_flag_finite_math_only;
    return ret;

I just noticed we're still doing the flag_finite_math_only hack. We should add some sort of flag to the setter to avoid reducing the range. I suppose this can wait to the next release.

Aldy

Reply via email to