This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Improve VRP range intersection
- From: Richard Biener <rguenther at suse dot de>
- To: gcc-patches at gcc dot gnu dot org
- Date: Fri, 30 Sep 2016 09:08:59 +0200 (CEST)
- Subject: [PATCH] Improve VRP range intersection
- Authentication-results: sourceware.org; auth=none
I noticed we intersect ~[a_1, a_1] and [2, 2] to ~[a_1, a_1]. While
we don't generally want to choose an integral range a singleton integral
range is always preferable.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2016-09-30 Richard Biener <rguenther@suse.de>
* tree-vrp.c (intersect_ranges): If we failed to handle
the intersection choose a constant singleton range if available.
Index: gcc/tree-vrp.c
===================================================================
--- gcc/tree-vrp.c (revision 240645)
+++ gcc/tree-vrp.c (working copy)
@@ -8555,7 +8555,16 @@ intersect_ranges (enum value_range_type
/* As a fallback simply use { *VRTYPE, *VR0MIN, *VR0MAX } as
result for the intersection. That's always a conservative
- correct estimate. */
+ correct estimate unless VR1 is a constant singleton range
+ in which case we choose that. */
+ if (vr1type == VR_RANGE
+ && is_gimple_min_invariant (vr1min)
+ && vrp_operand_equal_p (vr1min, vr1max))
+ {
+ *vr0type = vr1type;
+ *vr0min = vr1min;
+ *vr0max = vr1max;
+ }
return;
}