[Bug c/104316] New: Missing optimization for uninitialized local struct variable

[Bug c/104316] New: Missing optimization for uninitialized local struct variable

[roland.illig at gmx dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104316

            Bug ID: 104316
           Summary: Missing optimization for uninitialized local struct
                    variable
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: roland.illig at gmx dot de
  Target Milestone: ---

~~~c
#include <stdlib.h>

typedef struct Buffer {
    char *buf;
    size_t len;
    size_t cap;
} Buffer;

Buffer Buffer_Init_Return(void);
void Buffer_Init_Ref(Buffer *buf);
void Buffer_Add(Buffer *buf, const char *);

void
Use_Buffers(void)
{
    Buffer buf1 = Buffer_Init_Return();

    Buffer buf2;
    buf2 = Buffer_Init_Return();

    Buffer_Add(&buf1, "1");
    Buffer_Add(&buf2, "2");
}
~~~

https://godbolt.org/z/qTz36qP9r

The code snippets for initializing buf1 and buf2 are almost the same. From an
as-if perspective, they are identical.

GCC 12 generates several copy instructions when initializing buf2, while Clang
and ICC generate the straight-forward code.

Would it be difficult to tell GCC to optimize the above initialization for
buf2, for the case that buf2 is not accessed between the declaration and the
assignment?

[Bug tree-optimization/104316] RSO is not used when doing assignment rather than initialization and address taken afterwards

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104316

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Missing optimization for    |RSO is not used when doing
                   |uninitialized local struct  |assignment rather than
                   |variable                    |initialization and address
                   |                            |taken afterwards
           Keywords|                            |missed-optimization
           Severity|normal                      |enhancement
          Component|c                           |tree-optimization

[Bug tree-optimization/104316] RSO is not used when doing assignment rather than initialization and address taken afterwards

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104316

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
return slot optimization

[Bug tree-optimization/104316] RSO is not used when doing assignment rather than initialization and address taken afterwards

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104316

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2022-01-31
     Ever confirmed|0                           |1

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
NRV has:
static bool
dest_safe_for_nrv_p (gcall *call)
{
  tree dest = gimple_call_lhs (call);

  dest = get_base_address (dest);
  if (! dest)
    return false;

  if (TREE_CODE (dest) == SSA_NAME)
    return true;

  if (call_may_clobber_ref_p (call, dest, false)
      || ref_maybe_used_by_stmt_p (call, dest, false))
    return false;

  return true;
}

Which is fine, the problem is the clobber list is not exactly flow senative, I
think there is another bug about that.

[Bug tree-optimization/104316] RSO is not used when doing assignment rather than initialization and address taken afterwards

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104316

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |23384

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Yes PR 23384.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=23384
[Bug 23384] escaped set should be flow sensitive