[Bug c/97493] New: generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[Bug c/97493] New: generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[suochenyao at 163 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

            Bug ID: 97493
           Summary: generate wrong code with "-Os -fno-toplevel-reorder
                    -frename-registers"
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: suochenyao at 163 dot com
  Target Milestone: ---

Description:
I think gcc generates wrong code with -Os -fno-toplevel-reorder
-frename-registers.
But no difference is found with optimization levels.
No message and difference occurs with "gcc -Wall -Wextra" and
"-fno-strict-aliasing -fwrapv"
*******************************************************************************
OS and Platform:
CentOS Linux release 7.8.2003 (Core), x86_64 GNU/Linux
*******************************************************************************
Program:
int printf(const char *, ...);
union uuu{
  short a;
  long b;
} c={0};
long *d = &c.b;
short *e = &c.a;
short f=0;
int main() {
  *d = 3;
  f = ++*e;
  printf("%d\n", c.a);
}
*******************************************************************************
Gcc Version:
$ gcc -v
Using built-in specs.
COLLECT_GCC=/home/suocy/bin/gcc-dev/bin/gcc
COLLECT_LTO_WRAPPER=/home/suocy/bin/gcc-dev/libexec/gcc/x86_64-pc-linux-gnu/11.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../configure --prefix=/home/suocy/bin/gcc-dev
--disable-multilib --enable-languages=c,c++
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 11.0.0 20201018 (experimental) (GCC)
*******************************************************************************
Command Line:
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -O0 a.c -o a.out0
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -O1 a.c -o a.out1
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -O2 a.c -o a.out2
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -O3 a.c -o a.out3
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -Os a.c -o a.outs
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -Os -fno-toplevel-reorder
-frename-registers a.c -o a.outf
$ ./a.out0 > out0
$ ./a.out1 > out1
$ ./a.out2 > out2
$ ./a.out3 > out3
$ ./a.outs > outs
$ ./a.outf > outf
$ diff out0 out1
$ diff out0 out2
$ diff out0 out3
$ diff out0 outs
$ diff out0 outf
1c1
< 4
---
> 1

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|---                         |INVALID

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The testcase is invalid.  In standard C only one union member may be active at
a time, and while GCC as extension allows type puning through unions, the union
must be visible in the expressions accessing it, so taking addresses of the
union elts and dereferencing them through those pointers where the union is not
visible is UB.

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |WAITING
   Last reconfirmed|                            |2020-10-19
     Ever confirmed|0                           |1
         Resolution|INVALID                     |---

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
But he's using -fno-strict-aliasing so at least that part isn't relevant?
That said, the C rules still make it an INVALID testcase but I think
GCC-wise it could be well-defined, so understanding what goes wrong would be
nice.

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I can't reproduce it with -fno-strict-aliasing, only when that option is
missing.

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[suochenyao at 163 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

--- Comment #4 from suochenyao at 163 dot com <suochenyao at 163 dot com> ---
I made mistake, Jelinek is right.
This testcase is invalid and I can not reproduce it as well.
Sorry for wasting your time, my fault.
I apologize for wasting you time.

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[suochenyao at 163 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

suochenyao at 163 dot com <suochenyao at 163 dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|---                         |INVALID

[Bug c/97493] generate wrong code with "-Os -fno-toplevel-reorder -frename-registers"

[suochenyao at 163 dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97493

--- Comment #5 from suochenyao at 163 dot com <suochenyao at 163 dot com> ---
*******************************************************************************
I think this can be reproduced with "-fno-strict-aliasing"...
I am not sure whether it can be helpful...
*******************************************************************************
OS and Platform:
CentOS Linux release 7.8.2003 (Core), x86_64 GNU/Linux
*******************************************************************************
Program:
int printf(const char *, ...);
union {
  long a;
  int b;
} c;
int d=0;
long e=0;
long *f = &c.a;
short g=0;
int *h = &c.b;
short i(int j) {
  g = *h;
  *f = 3;
  for (;; e--) {
    if (*h)
      return d;
    *h = j;
  }
  return 0;
}
int main() {
  i(1);
  printf("%d\n", (int)e);
  return 0;
}
*******************************************************************************
gcc version:
$ gcc -v
Using built-in specs.
COLLECT_GCC=/home/suocy/bin/gcc-dev/bin/gcc
COLLECT_LTO_WRAPPER=/home/suocy/bin/gcc-dev/libexec/gcc/x86_64-pc-linux-gnu/11.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../configure --prefix=/home/suocy/bin/gcc-dev/
--disable-multilib --enable-languages=c,c++
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 11.0.0 20201029 (experimental) (GCC)
*******************************************************************************
Command Lines:
$ gcc a.c -o a.o1
$ gcc -Wall -Wextra -fno-strict-aliasing -fwrapv -Og -fgcse -fstrict-aliasing
a.c -o a.o2
$ gcc -Og -fgcse -fstrict-aliasing a.c -o a.o3
$ ./a.o1
0
$ ./a.o2
-1
$ ./a.o3
-1