[Bug middle-end/97902] New: x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[Bug middle-end/97902] New: x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[jan.smets at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

            Bug ID: 97902
           Summary: x86 frame pointer missing with -fno-omit-frame-pointer
                    (-mno-omit-leaf-frame-pointer)
           Product: gcc
           Version: 10.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jan.smets at nokia dot com
  Target Milestone: ---

int strcmpTEST(
    const char * s1,
    const char * s2)
{
    while (*s1++ == *s2++)
        if (*(s1-1) == '\0')
            return (0);

    return (*(s1-1) - *(s2-1));
}

Target: x86_64-linux-gnu

GCC 6.5/7.5: OK 
GCC 8.1/8.4/9.3/10.2: Missing.

[Bug middle-end/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-11-19
             Status|UNCONFIRMED                 |WAITING
     Ever confirmed|0                           |1
                 CC|                            |marxin at gcc dot gnu.org

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
I can't reproduce that with GCC 10.2.1 with -fno-omit-frame-pointer:

strcmpTEST:
.LFB0:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
...
.L3:
        popq    %rbp
        .cfi_def_cfa 7, 8
        ret

What problem exactly do you see?

[Bug middle-end/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[jan.smets at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #2 from Jan Smets <jan.smets at nokia dot com> ---
Apologies, I omitted the -O1 / -O2 

$ docker run --privileged --rm -it -v /tmp:/tmp gcc:10.2 bash -c "gcc -c
/tmp/test4.c -S -o - -O2 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer 
| grep rbp"

$ docker run --privileged --rm -it -v /tmp:/tmp gcc:7.5 bash -c "gcc -c
/tmp/test4.c -S -o - -O2 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer 
| grep rbp"
        pushq   %rbp
        movq    %rsp, %rbp
        popq    %rbp
        popq    %rbp

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
      Known to work|                            |7.5.0
             Status|WAITING                     |NEW
   Last reconfirmed|2020-11-19 00:00:00         |
      Known to fail|                            |10.2.1
          Component|middle-end                  |target

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r8-2488-g8e941ae950ddce17

i386: Don't use frame pointer without stack access

When there is no stack access, there is no need to use frame pointer
even if -fno-omit-frame-pointer is used and caller's frame pointer is
unchanged.

which seems as a reasonable explanation to not emit stack pointer?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
The documentation mentions that as well:

Note that -fno-omit-frame-pointer doesn't force a new stack
frame for all functions if it isn't otherwise needed, and hence doesn't
guarantee a new frame pointer for all functions.

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[jan.smets at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

Jan Smets <jan.smets at nokia dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |VERIFIED

--- Comment #6 from Jan Smets <jan.smets at nokia dot com> ---
As discussed on IRC, Martin asked to reopen:  The frame pointer is required for
backtracing on some embedded platforms.

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|VERIFIED                    |NEW
         Resolution|INVALID                     |---

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
Can you please H.J. take a look?
Maybe we can add a param that will drive the beviour?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING

--- Comment #8 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Martin Liška from comment #7)
> Can you please H.J. take a look?
> Maybe we can add a param that will drive the beviour?

I consider this optimization similar inlining or tail call. Should
-fno-omit-frame-pointer disable them?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |marxin at gcc dot gnu.org
             Status|WAITING                     |ASSIGNED

--- Comment #9 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to H.J. Lu from comment #8)
> (In reply to Martin Liška from comment #7)
> > Can you please H.J. take a look?
> > Maybe we can add a param that will drive the beviour?
> 
> I consider this optimization similar inlining or tail call.

How is that related? I don't see the connection in behavior.

> Should
> -fno-omit-frame-pointer disable them?

I would prefer to revert the patch 8e941ae950ddce1745b4d6819a7131908dd7de24.
The reporter requested to preserve the frame pointer (with
-fno-omit-frame-pointer) for backtracing capability and we optimized it out.

What do you think H.J. ?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #9)
> (In reply to H.J. Lu from comment #8)
> > (In reply to Martin Liška from comment #7)
> > > Can you please H.J. take a look?
> > > Maybe we can add a param that will drive the beviour?
> > 
> > I consider this optimization similar inlining or tail call.
> 
> How is that related? I don't see the connection in behavior.

You can't do backtracing with inlining (well, you'll lose those frames).

Here one of the callers will have a correct frame and backtracing still
works, no?

> > Should
> > -fno-omit-frame-pointer disable them?
> 
> I would prefer to revert the patch 8e941ae950ddce1745b4d6819a7131908dd7de24.
> The reporter requested to preserve the frame pointer (with
> -fno-omit-frame-pointer) for backtracing capability and we optimized it out.
> 
> What do you think H.J. ?

I don't think that's good.  It's perfectly valid with optimization.

One could spec a -fforce-frame-pointer but then one needs to define what
its guarantee is.

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Richard Biener from comment #10)
> (In reply to Martin Liška from comment #9)
> > (In reply to H.J. Lu from comment #8)
> > > (In reply to Martin Liška from comment #7)
> > > > Can you please H.J. take a look?
> > > > Maybe we can add a param that will drive the beviour?
> > > 
> > > I consider this optimization similar inlining or tail call.
> > 
> > How is that related? I don't see the connection in behavior.
> 
> You can't do backtracing with inlining (well, you'll lose those frames).
> 
> Here one of the callers will have a correct frame and backtracing still
> works, no?

Backtrace works with inlining, tail call and this optimization.

> > > Should
> > > -fno-omit-frame-pointer disable them?
> > 
> > I would prefer to revert the patch 8e941ae950ddce1745b4d6819a7131908dd7de24.
> > The reporter requested to preserve the frame pointer (with
> > -fno-omit-frame-pointer) for backtracing capability and we optimized it out.
> > 
> > What do you think H.J. ?
> 
> I don't think that's good.  It's perfectly valid with optimization.

Agree.

> One could spec a -fforce-frame-pointer but then one needs to define what
> its guarantee is.


Should it disable tail call?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #12 from Martin Liška <marxin at gcc dot gnu.org> ---
Thanks for the feedback. So do you tend to close it again as invalid?

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[jan.smets at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #13 from Jan Smets <jan.smets at nokia dot com> ---
H.J, There are still some very basic backtrace implementations that rely on
frame pointers. (No DWARF based things or any other forms of 'assistance'). A
missing stack frame means the "previous" function is not visible in the trace.
That makes it fairly useless.

We do explicitly disable a (partial)inlining, sibling calls, use
-fno-omit-frame-pointer and -mno-omit-leaf-frame-pointer. The latter two
options makes someone (perhaps incorrectly) assume that frame pointers are not
being omitted.

I understand, technically, they're not being omitted because there is no stack
usage to begin with... If a new option -fforce-frame-pointer is what is
required, then so be it, but I personally think it just adds more confusion on
what (no-)omit-frame-(leaf-)pointer does.  All I want are stack frames :-)

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

H.J. Lu <hjl.tools at gmail dot com> changed:

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

--- Comment #14 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Jan Smets from comment #13)
> H.J, There are still some very basic backtrace implementations that rely on
> frame pointers. (No DWARF based things or any other forms of 'assistance').
> A missing stack frame means the "previous" function is not visible in the
> trace. That makes it fairly useless.
> 
> We do explicitly disable a (partial)inlining, sibling calls, use
> -fno-omit-frame-pointer and -mno-omit-leaf-frame-pointer. The latter two
> options makes someone (perhaps incorrectly) assume that frame pointers are
> not being omitted.
> 
> I understand, technically, they're not being omitted because there is no
> stack usage to begin with... If a new option -fforce-frame-pointer is what
> is required, then so be it, but I personally think it just adds more
> confusion on what (no-)omit-frame-(leaf-)pointer does.  All I want are stack
> frames :-)

Please open a new bug for a new option to disable all optimizations which
may skip frame pointer.

[Bug target/97902] x86 frame pointer missing with -fno-omit-frame-pointer (-mno-omit-leaf-frame-pointer)

[jan.smets at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97902

--- Comment #15 from Jan Smets <jan.smets at nokia dot com> ---
Thanks. See 98018.