[Bug tree-optimization/65797] New: [5.0 regression] IPA ICF causes function to be emitted with no debug line info

[Bug tree-optimization/65797] New: [5.0 regression] IPA ICF causes function to be emitted with no debug line info

[ian at airs dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

            Bug ID: 65797
           Summary: [5.0 regression] IPA ICF causes function to be emitted
                    with no debug line info
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: hubicka at ucw dot cz
          Reporter: ian at airs dot com

This stand alone test case is boiled down from libgo/runtime/go-make-slice.c.


typedef unsigned int uintptr_t __attribute__ ((mode (pointer)));
typedef unsigned int uint64_t __attribute__ ((mode (DI)));

extern int f1 (char *, uintptr_t, uintptr_t);

int
f2 (char *p, uintptr_t s)
{
  return f1 (p, s, s);
}

int
f4 (char *p, uint64_t s1, uint64_t s2)
{
  uintptr_t t1, t2;

  t1 = (uintptr_t) s1;
  if ((uint64_t) t1 != s1)
    return 0;

  t2 = (uintptr_t) s2;
  if ((uint64_t) t2 != s2)
    return 0;

  return f1 (p, t1, t2);
}

int
f3 (char *p, uint64_t s)
{
  return f4 (p, s, s);
}


When I compile this -g -O2 -fno-optimize-sibling-calls on x86_64 f2 is
converted into a function that does nothing but call f3.  From the standpoint
of execution that is fine.  What is not fine is that f2 is emitted with no
debug line information.  When compiling for x86_64, the .s output is:

        .p2align 4,,15
        .globl  f2
        .type   f2, @function
f2:
.LFB4:
        .cfi_startproc
        subq    $8, %rsp
        .cfi_def_cfa_offset 16
        call    f3
        addq    $8, %rsp
        .cfi_def_cfa_offset 8
        ret
        .cfi_endproc
.LFE4:
        .size   f2, .-f2

Note the absence of any .loc directives.

This is causing a failure in the libgo testsuite (runtime/pprof on x86_64). 
The failure is, essentially, a failure of the stack backtrace to determine the
name of the function f2.  In the absence of line information the library is
unable to know the name of the function.

[Bug tree-optimization/65797] [5.0 regression] IPA ICF causes function to be emitted with no debug line info

[ian at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #2 from ian at gcc dot gnu.org <ian at gcc dot gnu.org> ---
Author: ian
Date: Fri Apr 17 21:29:20 2015
New Revision: 222200

URL: https://gcc.gnu.org/viewcvs?rev=222200&root=gcc&view=rev
Log:
    PR go/64683
runtime/pprof: Assume function with no name is in runtime.

GCC PR 65797 causes some of the runtime functions to be
compiled with no name in the debug info.  This in turn causes
the runtime/pprof test to fail as reported in GCC PR 64683.

There are no good choices when a function has no name in the
debug info, but here we assume that if we see such a function
while reading the runtime functions, we assume that it is also
a runtime function.

Modified:
    trunk/libgo/go/runtime/pprof/pprof.go

[Bug tree-optimization/65797] [5.0 regression] IPA ICF causes function to be emitted with no debug line info

[ian at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #1 from ian at gcc dot gnu.org <ian at gcc dot gnu.org> ---
Author: ian
Date: Fri Apr 17 21:29:08 2015
New Revision: 222199

URL: https://gcc.gnu.org/viewcvs?rev=222199&root=gcc&view=rev
Log:
    PR go/64683
runtime/pprof: Assume function with no name is in runtime.

GCC PR 65797 causes some of the runtime functions to be
compiled with no name in the debug info.  This in turn causes
the runtime/pprof test to fail as reported in GCC PR 64683.

There are no good choices when a function has no name in the
debug info, but here we assume that if we see such a function
while reading the runtime functions, we assume that it is also
a runtime function.

Modified:
    branches/gcc-5-branch/libgo/go/runtime/pprof/pprof.go

[Bug tree-optimization/65797] [5.0 regression] IPA ICF causes function to be emitted with no debug line info

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #3 from Jan Hubicka <hubicka at ucw dot cz> ---
Hi,
the ICF wrapper are created same way as thunks (by expand_thunk) which probably
suppress debug info because we do not want to see it for thunks. I suppose it
is:
DECL_IGNORED_P (thunk_fndecl) = 1
I suppose we can add a parameter that will enable debug info production and
pass it
as true via the create_wrapper path (can't hack it right now as I am on a trip
in China)
I am adding Martin to CC.

In many cases we however create a wrapper where keeping function is backtrace
is even harder. If go needs functions in backtrace even in this case (when they
are not exported nor have address taken) perhaps we have to disable ICF?
We may add a way to disable ICF per function basis.

Honza

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[ian at airs dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #4 from Ian Lance Taylor <ian at airs dot com> ---
I think we should have a goal of making backtraces always work.  I don't know
why we would ever want backtraces to fail.  Every function should have a name
and a file name.  I can accept that in some cases there may be no useful line
information, but it seems clear to me that a function should always have a
name.

I would guess that thunks do not appear in backtraces because they end with a
jump instruction rather than a call.  Note that my test case used
-fno-optimize-sibling-calls.

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #5 from Jan Hubicka <hubicka at ucw dot cz> ---
Well, if you turn one function to alias of another, there is no way to preserve
it (like Gold's ICF doesn't). With dwarf extensions we can restore some of the
info based on context where the function is called, but not all of it.  In this
case I guess it would be an improvement to have wrapper function in the
backtrace followed by the unexepected function we redirected to.

Honza
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797
> 
> --- Comment #4 from Ian Lance Taylor <ian at airs dot com> ---
> I think we should have a goal of making backtraces always work.  I don't know
> why we would ever want backtraces to fail.  Every function should have a name
> and a file name.  I can accept that in some cases there may be no useful line
> information, but it seems clear to me that a function should always have a
> name.
> 
> I would guess that thunks do not appear in backtraces because they end with a
> jump instruction rather than a call.  Note that my test case used
> -fno-optimize-sibling-calls.
> 
> -- 
> You are receiving this mail because:
> You are the assignee for the bug.

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #6 from Jan Hubicka <hubicka at ucw dot cz> ---
The following untested patch could help.  We may need to set location of the
debug statement
etc.  I probably won't be able to do much more on this till Monday evening

Honza

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[ian at airs dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #7 from Ian Lance Taylor <ian at airs dot com> ---
With gold's ICF, as I understand it, there is a function name and file/line
information for every function in the backtrace.  It may not be the name or the
file/line info you expect, but it's there.

What broke the Go testcase is that there was no information at all.

It's not urgent for Go, by the way, as I've committed a workaround.  But I
think it would be nice to fix for GCC 5 if possible.

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

--- Comment #8 from Jan Hubicka <hubicka at ucw dot cz> ---
> With gold's ICF, as I understand it, there is a function name and file/line
> information for every function in the backtrace.  It may not be the name or the

ICF does not do the wrappers as far as I know. It either redirect to alias or
not
at all.  If debugger behaves in a way ignoring functions with DECL_IGNORED=1,
then
the backtraces are the same with alias reidrection and wrapper.
In both cases original function disappears and is replaced by merged-to
function.

With this patch we preserve original function but insert the merged-to
function.
Both is wrong, but I guess having both functions where possible is better htan
completely dropping the original.

> file/line info you expect, but it's there.
> 
> What broke the Go testcase is that there was no information at all.
> 
> It's not urgent for Go, by the way, as I've committed a workaround.  But I
> think it would be nice to fix for GCC 5 if possible.

It is Jakub's call. I think the patch can go in if someone gets around testing
it.
What I am concerned about is that expand_thunk may not completely set up the
datastructures
for debug info as this path was never excercised. But htat should be easy to
test.

Honza
> 
> -- 
> You are receiving this mail because:
> You are the assignee for the bug.

[Bug tree-optimization/65797] [5 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |5.2

[Bug ipa/65797] [5/6 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
          Component|tree-optimization           |ipa
            Summary|[5 regression] IPA ICF      |[5/6 regression] IPA ICF
                   |causes function to be       |causes function to be
                   |emitted with no debug line  |emitted with no debug line
                   |info                        |info

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
The underlying issue wasn't fixed on trunk just the symptom in libgo.

[Bug ipa/65797] [5/6 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|5.2                         |5.3

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 5.2 is being released, adjusting target milestone to 5.3.

[Bug ipa/65797] [9/10/11/12 regression] IPA ICF causes function to be emitted with no debug line info

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|8.5                         |9.4

--- Comment #20 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 8 branch is being closed.

[Bug ipa/65797] [9/10/11/12 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.4                         |9.5

--- Comment #21 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9.4 is being released, retargeting bugs to GCC 9.5.

[Bug ipa/65797] [10/11/12/13 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.5                         |10.4

--- Comment #22 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9 branch is being closed

[Bug ipa/65797] [10/11/12/13 regression] IPA ICF causes function to be emitted with no debug line info

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.4                        |10.5

--- Comment #23 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 10.4 is being released, retargeting bugs to GCC 10.5.

[Bug ipa/65797] [11/12/13/14 regression] IPA ICF causes function to be emitted with no debug line info

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65797

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #24 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.