[Bug middle-end/96564] [11 Regression] New maybe use of uninitialized variable warning since r11-959

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96564

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=87313
           Keywords|                            |alias, diagnostic,
                   |                            |missed-optimization

--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> ---
r11-959 enabled -Wuninitialized for allocated objects (alloca/VLA and malloc). 
The underlying problem is the failure to take advantage of the non-aliasing
guarantee of the return value of these functions.  One report that describes
this limitation is pr87313.  A slightly simpler but even more obvious test case
is:

$ cat pr96564.c && gcc -O2 -S -Wall -Wextra -fdump-tree-optimized=/dev/stdout
pr96564.c
void f (int *x)
{
  int a[*x];

  if (a != x)
    *a = 0;

  *x = *a;
}

pr96564.c: In function ‘f’:
pr96564.c:8:6: warning: ‘*(<unknown>)[0]’ may be used uninitialized
[-Wmaybe-uninitialized]
    8 |   *x = *a;
      |   ~~~^~~~

;; Function f (f, funcdef_no=0, decl_uid=1931, cgraph_uid=1, symbol_order=0)

Removing basic block 5
f (int * x)
{
  int[0:D.1936] * a.0;
  sizetype _2;
  int _8;
  sizetype _9;
  int prephitmp_18;
  int pretmp_20;

  <bb 2> [local count: 1073741824]:
  _8 = *x_1(D);
  _2 = (sizetype) _8;
  _9 = _2 * 4;
  a.0_11 = __builtin_alloca_with_align (_9, 32);
  if (x_1(D) != a.0_11)
    goto <bb 4>; [70.00%]
  else
    goto <bb 3>; [30.00%]

  <bb 3> [local count: 322122544]:
  pretmp_20 = (*a.0_11)[0];

  <bb 4> [local count: 1073741824]:
  # prephitmp_18 = PHI <pretmp_20(3), 0(2)>
  *x_1(D) = prephitmp_18;
  return;

}