[Bug c/114437] New: Inline asm with "+m,r" operand ignores input value

[Bug c/114437] New: Inline asm with "+m,r" operand ignores input value

[xog4nar4@a-n.cc]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

            Bug ID: 114437
           Summary: Inline asm with "+m,r" operand ignores input value
           Product: gcc
           Version: 13.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: xog4nar4@a-n.cc
  Target Milestone: ---

In the following snippet, GCC seems to assume that `arg` is not read by the asm
statement and optimizes out previous stores when compiling on -O3:
```
extern int external_fn(int);
extern int arg;

int fn() {
    arg = 123;
    asm volatile("" : "+m,r"(arg) : : "memory");
    return external_fn(arg);
}
```

Resulting asm (123 is not stored, leaving `arg` uninitialized):
```
0000000000000000 <fn>:
   0:   f3 0f 1e fa             endbr64
   4:   8b 3d 00 00 00 00       mov    edi,DWORD PTR [rip+0x0]        # a
<fn+0xa>
   a:   e9 00 00 00 00          jmp    f <fn+0xf>
```

The asm snippet comes from the `DoNotOptimize` function in Google Benchmark,
where it is intended to prevent the compiler from optimizing away parts of a
micro-benchmark:
https://github.com/google/benchmark/blob/main/include/benchmark/benchmark.h#L518


Godbolt reproduction: https://godbolt.org/z/3W97dM9eW

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
"+m" does not do what you think it does ...

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

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

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

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
These constraints don't do what think it does.

You can see the behavior by doing:
    asm volatile("#%0 %1" : "+m,r"(arg) : : "memory");


You get:
        #arg(%rip) .LC0(%rip)


Which is correct as `1` is stored in `.LC0(%rip)`.

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[xog4nar4@a-n.cc]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

--- Comment #3 from xog4nar4@a-n.cc ---
Just to confirm, is this an incorrect constraint for the intended goal of
convincing the compiler that the value is used and modified, without actually
doing either and preserving the value, or am I just using it wrong in this
particular example? I'm somewhat surprised since this inline asm snippet is
used in multiple well-known micro-benchmarking libraries.


Is the following version correct?
```
asm volatile("" : "=m,r"(arg) : "0,0"(arg) : "memory");
```

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|INVALID                     |DUPLICATE

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to xog4nar4 from comment #3)
> Just to confirm, is this an incorrect constraint for the intended goal of
> convincing the compiler that the value is used and modified, without
> actually doing either and preserving the value, or am I just using it wrong
> in this particular example? I'm somewhat surprised since this inline asm
> snippet is used in multiple well-known micro-benchmarking libraries.
> 
> 
> Is the following version correct?
> ```
> asm volatile("" : "=m,r"(arg) : "0,0"(arg) : "memory");
> ```

The correct way of implementing this inline-asm is don't. Oh and the above does
not work either because a constraint of `"=m":"0"` is also not going to work.
Inline-asm is really NOT designed for this purpose at all and abusing it this
way is just going to run into these issues.

*** This bug has been marked as a duplicate of bug 92597 ***

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[xog4nar4@a-n.cc]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

xog4nar4@a-n.cc changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|DUPLICATE                   |INVALID

--- Comment #5 from xog4nar4@a-n.cc ---
I would prefer to use something else, but I'm not aware of any less hacky way
to ensure that the compiler does not interfere with a benchmark. If there's any
explicitly supported way to achieve that in GCC, I'm all ears, but I don't
assume so given that each and every C++ benchmarking library I know uses inline
asm.

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
You can juse use "+mr" without the comma. There is no way to use alterantives
in inline-asm.

[Bug middle-end/114437] Inline asm with "+m,r" operand ignores input value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114437

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|INVALID                     |DUPLICATE

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Still a dup.

*** This bug has been marked as a duplicate of bug 92597 ***