[Bug tree-optimization/66868] New: [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[Bug tree-optimization/66868] New: [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[doko at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

            Bug ID: 66868
           Summary: [5/6 Regression] wrong code generated with -O3 (dead
                    code removal?)
           Product: gcc
           Version: 5.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at gcc dot gnu.org
  Target Milestone: ---

Created attachment 35975
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35975&action=edit
preprocessed source

[forwarded from https://bugs.launchpad.net/ubuntu/+source/gcc-5/+bug/1473674]

seen in the apt package manager, when built with -O3. Works with -O2 or -O3
-fno-inline-functions -finline-small-functions, on at least x86_64 and
powerpc64le. The test CDROMTest.FindPackages from the apt testsuite then fails.

built with g++ -c -g -O3 -fPIE -fstack-protector-strong -fPIC 

upstream writes:

There is a Cdrom wrapper:
"""
class Cdrom : public pkgCdrom {
   public:
      bool FindPackages(std::string const &CD,
     std::vector<std::string> &List,
     std::vector<std::string> &SList,
     std::vector<std::string> &SigList,
     std::vector<std::string> &TransList,
     std::string &InfoDir) {
  std::string const startdir = SafeGetCWD();
  EXPECT_FALSE(startdir.empty());
  bool const result = pkgCdrom::FindPackages(CD, List, SList, SigList,
TransList, InfoDir, NULL, 0);
...
}
"""

and a unittest that calls it:
"""
TEST(CDROMTest,FindPackages)
{
   Cdrom cd;
  std::string InfoDir;
  EXPECT_TRUE(cd.FindPackages(path, Packages, Sources, Signatur, Translation,
InfoDir));
...
  EXPECT_EQ(path + "/.disk/", InfoDir);
}
"""
The actual code for this is apt-pkg/cdrom.cc:
"""
bool pkgCdrom::FindPackages(string CD,
       vector<string> &List,
       vector<string> &SList,
       vector<string> &SigList,
       vector<string> &TransList,
       string &InfoDir, pkgCdromStatus *log,
       unsigned int Depth)
{
...
   if (DirectoryExists(".disk") == true)
   {
      if (InfoDir.empty() == true)
  InfoDir = CD + ".disk/";
   }
...
"""

So I suspect that the optimizer gets confused that InfoDir is a reference or it
gets confused because InfoDir is not used in FindPackages anymore and it
assumes its dead code.

I tried to create a simplified testcase but failed so far. Whats interessting
is that if I add a std::cerr << "debug" line into cdrom.cc lines (or even a "CD
= CD"):
"""
   if (DirectoryExists(".disk") == true)
   {
      if (InfoDir.empty() == true) {
std::cerr << "debug" << std::endl;
  InfoDir = CD + ".disk/";
}
   }
"""
it works (which indicates dead-code elimination to me).

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

Markus Trippelsdorf <trippels at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |trippels at gcc dot gnu.org

--- Comment #1 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
Have you tried to build with -fsanitize=undefined?

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

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

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

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[doko at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #2 from Matthias Klose <doko at gcc dot gnu.org> ---
the testsuite passes with -fsanitize=undefined (no additional output) and both
-fsanitize=undefined -fsanitize-undefined-trap-on-error doesn't abort either.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

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

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

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 5.2 is being released, adjusting target milestone to 5.3.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #4 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Hm.  I compiled it as stated and I see a bunch of code that appears to be
storing the ".disk/" string.  So it doesn't look like dead code elimination. 
Perhaps a branch is short circuiting this, or the address of where to store it
has been corrupted.

I don't have any way to confirm that I'm reproducing the problem when I
compile, from what's given here.  It might be helpful to provide the full test
case without ellipses, and the failure message given by the test.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #5 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Using current trunk, I compared the 193t.optimized dumps for the original
cdrom.ii attachment and for the same attachment modified to add the debug
output statement from the end of comment 2.  Other than the added statements
for the debug output, there don't appear to be any changes.  This bug is marked
as 5.1/6 regression -- are you certain that it regresses on trunk?  Will check
5.1 the same way.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #6 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Sorry, not from comment 2 on this bugzilla; comment 2 from the launchpad bug.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #7 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
I tried the same thing with a snapshot of the 5 branch I had lying around: 
r225783 from 2015-07-14.  I also don't see any differences in the output from
the middle end, as I would expect since this bug has shown up on multiple
targets.  So maybe this has been fixed recently?  Let me see if I can find an
older snapshot.

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

--- Comment #8 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Same experiment with r222218 from 2015-04-18.  Same results.

At this point I can't reproduce anything from the information given.  Do you
have any local modifications that could be causing this?

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[doko at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

Matthias Klose <doko at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |powerpc64le-linux-gnu,
                   |                            |x86_64-linux-gnu
      Known to work|                            |6.0
      Known to fail|5.1.1, 6.0                  |5.2.0

--- Comment #9 from Matthias Klose <doko at gcc dot gnu.org> ---
this works ok with trunk r226068, still fails on the branch.

gcc is configured with
--enable-languages=c,ada,c++,java,go,d,fortran,objc,obj-c++ --prefix=/usr
--program-suffix=-5 --enable-shared --enable-linker-build-id
--libexecdir=/usr/lib --without-included-gettext --enable-threads=posix
--libdir=/usr/lib --enable-nls --with-sysroot=/ --enable-clocale=gnu
--enable-libstdcxx-debug --enable-libstdcxx-time=yes
--with-default-libstdcxx-abi=new --enable-gnu-unique-object
--disable-libquadmath --enable-plugin --with-system-zlib
--disable-browser-plugin --enable-java-awt=gtk --enable-gtk-cairo
--with-java-home=/usr/lib/jvm/java-1.5.0-gcj-5-ppc64el/jre --enable-java-home
--with-jvm-root-dir=/usr/lib/jvm/java-1.5.0-gcj-5-ppc64el
--with-jvm-jar-dir=/usr/lib/jvm-exports/java-1.5.0-gcj-5-ppc64el
--with-arch-directory=ppc64 --with-ecj-jar=/usr/share/java/eclipse-ecj.jar
--enable-objc-gc --enable-secureplt --with-cpu=power7 --with-tune=power8
--disable-multilib --enable-multiarch --disable-werror --with-long-double-128
--enable-checking=release --build=powerpc64le-linux-gnu
--host=powerpc64le-linux-gnu --target=powerpc64le-linux-gnu

[Bug tree-optimization/66868] [5/6 Regression] wrong code generated with -O3 (dead code removal?)

[doko at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66868

Matthias Klose <doko at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |5.1.0

--- Comment #11 from Matthias Klose <doko at gcc dot gnu.org> ---
rechecked with a 5.1.0 release tarball, fails with -O3