[Bug c/99431] New: Return value of __builtin_memset() incorrect if partially optimized away

[Bug c/99431] New: Return value of __builtin_memset() incorrect if partially optimized away

[jonathon.reinhart at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99431

            Bug ID: 99431
           Summary: Return value of __builtin_memset() incorrect if
                    partially optimized away
           Product: gcc
           Version: 9.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jonathon.reinhart at gmail dot com
  Target Milestone: ---

Created attachment 50319
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50319&action=edit
Minimal example

Please see attached code or https://godbolt.org/z/oTMhsd.

If the address of a struct object is passed to __builtin_memset(), and some
leading fields are assigned elsewhere in the function, GCC may effectively
adjust the portion of the object written to by the memset(), to eliminate
redundant writes. However, the return value of memset(), which should always
match the first argument, is also (incorrectly) adjusted, leading to an invalid
result.

Specifically, this happens with -Os. I attempted to narrow this down to a
specific set of optimizations by testing everything listed in the output of
this command, but that was insufficient to reproduce the problem:

  gcc -Q -Os --help=optimizers

I found it interesting, although not surprising, that adding this between the
__builtin_memset() call and the "other" assignments caused the problem to go
away: 

    __asm__("" :::"memory");

GCC versions tested:
  10.2 OK
  10.1 OK
  9.3  Broken
  8.3  Broken
  7.3  Broken
  7.2  Broke
  7.1  Broken
  6.3  OK

[Bug c/99431] Return value of __builtin_memset() incorrect if partially optimized away

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99431

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
On the trunk started with r7-5965-g7c9560a578a06125cb30458a26605f91feb29b0d
and got fixed with r10-7140-g349ab34dc64a10fe0b1eda66d13b62862878b73e

[Bug c/99431] Return value of __builtin_memset() incorrect if partially optimized away

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99431

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So dup.

*** This bug has been marked as a duplicate of bug 94130 ***

[Bug c/99431] Return value of __builtin_memset() incorrect if partially optimized away

[jonathon.reinhart at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99431

--- Comment #3 from Jonathon Reinhart <jonathon.reinhart at gmail dot com> ---
Yep, that's it. I figured there had to be a dup, but couldn't find it. Thanks
Jakub.