https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122660
Bug ID: 122660
Summary: uninit predicated analysis not strong enough (derived
from uninit-pred-7_a.c)
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Target Milestone: ---
Created attachment 62779
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62779&action=edit
gimple testcase
Take the attached gimple testcase which derived from uninit-pred-7_a.c with a
pass_mergephi added after pass_dominator (the one right after pass_sra):
```
diff --git a/gcc/passes.def b/gcc/passes.def
index fac04cd86c7..2eb5b948618 100644
--- a/gcc/passes.def
+++ b/gcc/passes.def
@@ -258,6 +258,7 @@ along with GCC; see the file COPYING3. If not see
/* Threading can leave many const/copy propagations in the IL.
Clean them up. Failure to do so well can lead to false
positives from warnings for erroneous code. */
+ NEXT_PASS (pass_merge_phi);
NEXT_PASS (pass_copy_prop);
/* Identify paths that should never be executed in a conforming
program and isolate those paths. */
```
We get a bogus uninitialized warning:
```
<source>: In function 'foo':
<source>:105:3: warning: 'v' may be used uninitialized [-Wmaybe-uninitialized]
105 | blah (v_28);
| ^~~~
<source>:9:7: note: 'v' was declared here
9 | int v;
| ^
```
I will be xfailing uninit-pred-7_a.c when I submit the patches for PR122493 but
I wanted to file this first as it can be reproduced using a gimple testcase.
The main difference the merge_phi here does here is allows tail_merge to find
the bb containing the call to `blah (v);`:
```
find_duplicates: <bb 13> duplicate of <bb 11>
...
Removing basic block 13
;; basic block 13, loop depth 0
;; pred:
blah (v_34);
;; succ: 14
```
