https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126704
Bug ID: 126704
Summary: Missing or not complete global range after VRP2 (with
and without dom)
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Blocks: 85316, 126010
Target Milestone: ---
Take:
```
int
Foo (int *p, int t)
{
int i = *p;
if (! (i <= 99999))
__builtin_unreachable ();
if (! (0 <= i && i <= 99999))
__builtin_unreachable ();
return i/10;
}
```
While working on changing tree_single_nonnegative_p over to use the ranger I
noticed this. This comes from `gcc.dg/pr80776-1.c` but changed to force the use
of the range after VRP2.
at -O2 at VRP2, the global range for the load will be `[-INF, 99999]` which is
what dom had set and is not updated afterwards.
With `-O2 -fno-tree-dominator-opts`, the global range will be gone.
This causes us not to notice that i is non-negative while expanding.
So the issue here is 2 fold, if we remove dom, we will remove sometimes partial
ranges but even those are partial ranges which is not updated by VRP2.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85316
[Bug 85316] [meta-bug] VRP range propagation missed cases
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126010
[Bug 126010] [meta-bug] Remove DOM
