[Bug tree-optimization/90307] -Wuninitialized only at -O1, not at -O2

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90307

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2019-05-02
          Component|c++                         |tree-optimization
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think SRA split this according to external_type layout even though accesses
are using internal_type layout.  For the full remat of the aggregate,
that is, the scalarization itself is triggered by the filed accesses later.
Doesn't look very clever overall.

@@ -24,7 +38,13 @@
   s.u.internal.size = 5;
   MEM[(struct  &)&x] ={v} {CLOBBER};
   x.u.internal = MEM[(const struct sstring &)&s].u.internal;
+  x$u$external$str_24 = MEM[(union contents *)&s];
+  x$u$external$size_64 = MEM[(union contents *)&s + 8B];
+  x$u$internal$size_65 = MEM[(union contents *)&s + 15B];
   MEM[(struct  &)&D.2505] ={v} {CLOBBER};
+  MEM[(union contents *)&x] = x$u$external$str_24;
+  MEM[(union contents *)&x + 8B] = x$u$external$size_64;
+  MEM[(union contents *)&x + 15B] = x$u$internal$size_65;
   D.2505.u.internal = MEM[(const struct sstring &)&x].u.internal;
   use (&D.2505);