[Bug ipa/65502] New: pure-const should play well with clobbers.

[Bug ipa/65502] New: pure-const should play well with clobbers.

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65502

            Bug ID: 65502
           Summary: pure-const should play well with clobbers.
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hubicka at gcc dot gnu.org

I just noticed that C++ destructors are not conisdered const/pure:

 local analysis of Smarts::IterateScheduler<Smarts::Stub>::~IterateScheduler()
   scanning: MEM[(struct  &)this_2(D)] ={v} {CLOBBER};
    Indirect ref write is not const/pure
  scanning: return;
Smarts::IterateScheduler<Smarts::Stub>::~IterateScheduler() (struct
IterateScheduler * const this)
{
  <bb 2>:
  MEM[(struct  &)this_2(D)] ={v} {CLOBBER};
  return;

}

This is of course wrong, empty function is const.  One issue however is that we
want to preserve CLOBBER when inlining, so during early passes we may want to
keep clobbers nonpure.

[Bug ipa/65502] pure-const should play well with clobbers.

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65502

--- Comment #1 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
The following patch makes ipa-pure-const to detect these functions as
pure/const.
There are two issues
 1) I think we should preserve clobber semantic when removing a call to
pure/const
    destructor.  I wonder if we should output CLOBBER after every destructor
call
    from the gimplifier (that may be bit costy) or if we want tree-ssa-dce
    to insert the clobber when removing the statement that calls function with
    DECL_CXX_DESTRUCTOR (that is cheaper but drags in more of C++ semantics)

    Any solution to this would permit me to remove the conditional preventing
    clobbers from being ignored at compile time.
 2) The patch makes us to ICE at chromium build. The reason is that ipa-comdats
    sees symbol that should have been removed by remove_unreacable_functions
    and is not.  I suppose this is an semi-latent bug where 
    remove_unreachable_functions may need iteration in some side corners.

Fun...

Index: ipa-pure-const.c
===================================================================
--- ipa-pure-const.c    (revision 221541)
+++ ipa-pure-const.c    (working copy)
@@ -710,6 +710,16 @@ check_stmt (gimple_stmt_iterator *gsip,
   if (is_gimple_debug (stmt))
     return;

+  /* Do consider clobber as side effects before IPA, so we rather inline
+     C++ destructors and keep clobber semantics than eliminate them.
+
+     TODO: We may get smarter during early optimizations on these and let
+     functions containing only clobbers to be optimized more.  This is a
common
+     case of C++ destructors.  */
+
+  if ((ipa || cfun->after_inlining) && gimple_clobber_p (stmt))
+    return;
+
   if (dump_file)
     {
       fprintf (dump_file, "  scanning: ");

[Bug ipa/65502] pure-const should play well with clobbers.

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65502

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2015-03-22
           Assignee|unassigned at gcc dot gnu.org      |hubicka at gcc dot gnu.org
   Target Milestone|---                         |6.0
     Ever confirmed|0                           |1
           Severity|normal                      |enhancement

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
The ICE is now solved. Changing to enhancement for ipa-pure-const/Clobber
generation.

[Bug ipa/65502] pure-const should play well with clobbers.

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
   Target Milestone|6.0                         |---

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think we can safely ignore clobbers when scanning functions for
pure/constness.

[Bug ipa/65502] pure-const should play well with clobbers.

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65502

--- Comment #5 from Jan Hubicka <hubicka at ucw dot cz> ---
> I think we can safely ignore clobbers when scanning functions for
> pure/constness.

Yes (it is what the patch does), but doing so may cause worse code in the
function calling these destructors.  DCE will remove the destructor call and we
will lose information about memory location being dead.
I think we may want to output the clobber after destructors calls in this case.