[Bug c/50506] New: gcc fails at assembly with -O1 while inlining is forced

[Bug c/50506] New: gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

             Bug #: 50506
           Summary: gcc fails at assembly with -O1 while inlining is
                    forced
    Classification: Unclassified
           Product: gcc
           Version: 4.6.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: galtgendo@o2.pl


Created attachment 25354
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25354
preprocesed code

This bug simply separates the latest comments of bug 39333 for better
visibility.

Using built-in specs.
COLLECT_GCC=gcc-4.6.1
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/i686-pc-linux-gnu/4.6.1/lto-wrapper
Target: i686-pc-linux-gnu
Configured with:
/mnt/workbox/builds/portage/sys-devel/gcc-4.6.1-r1/work/gcc-4.6.1/configure
--prefix=/usr --bindir=/usr/i686-pc-linux-gnu/gcc-bin/4.6.1
--includedir=/usr/lib/gcc/i686-pc-linux-gnu/4.6.1/include
--datadir=/usr/share/gcc-data/i686-pc-linux-gnu/4.6.1
--mandir=/usr/share/gcc-data/i686-pc-linux-gnu/4.6.1/man
--infodir=/usr/share/gcc-data/i686-pc-linux-gnu/4.6.1/info
--with-gxx-include-dir=/usr/lib/gcc/i686-pc-linux-gnu/4.6.1/include/g++-v4
--host=i686-pc-linux-gnu --build=i686-pc-linux-gnu --disable-altivec
--disable-fixed-point --with-ppl --with-cloog --disable-ppl-version-check
--with-cloog-include=/usr/include/cloog-ppl --enable-lto --enable-nls
--without-included-gettext --with-system-zlib --disable-werror
--enable-secureplt --disable-multilib --enable-libmudflap --disable-libssp
--enable-libgomp
--with-python-dir=/share/gcc-data/i686-pc-linux-gnu/4.6.1/python
--enable-checking=release --disable-libgcj --with-arch=i686 --enable-objc-gc
--enable-languages=c,c++,objc,fortran --enable-shared --enable-threads=posix
--enable-__cxa_atexit --enable-clocale=gnu
--with-bugurl=http://bugs.gentoo.org/ --with-pkgversion='Gentoo 4.6.1-r1 p1.0,
pie-0.4.5' --disable-libstdcxx-pch
Thread model: posix
gcc version 4.6.1 (Gentoo 4.6.1-r1 p1.0, pie-0.4.5) 
COLLECT_GCC_OPTIONS='-v' '-Wall' '-O1' '-c' '-o' 'fprintf-bug-code-4.6.o'
'-save-temps' '-mtune=generic' '-march=i686'
 /usr/libexec/gcc/i686-pc-linux-gnu/4.6.1/cc1 -fpreprocessed
fprintf-bug-code-4.6.i
-quiet -dumpbase fprintf-bug-code-4.6.i -mtune=generic -march=i686
-auxbase-strip
fprintf-bug-code-4.6.o -O1 -Wall -version -o fprintf-bug-code-4.6.s
GNU C (Gentoo 4.6.1-r1 p1.0, pie-0.4.5) version 4.6.1 (i686-pc-linux-gnu)
    compiled by GNU C version 4.6.1, GMP version 5.0.2, MPFR version 3.0.1-p4,
MPC version 0.8.2
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
GNU C (Gentoo 4.6.1-r1 p1.0, pie-0.4.5) version 4.6.1 (i686-pc-linux-gnu)
    compiled by GNU C version 4.6.1, GMP version 5.0.2, MPFR version 3.0.1-p4,
MPC version 0.8.2
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: 5f6717c7397ac8a8888626933ce7eb52
fprintf-bug-code-4.6.i: In function ‘write_cfg’:
fprintf-bug-code-4.6.i:71:1: sorry, unimplemented: inlining failed in call to
‘fprintf’: indirect function call with a yet undetermined callee
fprintf-bug-code-4.6.i:78:7: sorry, unimplemented: called from here

The code builds fine is either -fipa-cp is added or '__attribute__
((__always_inline__))' is removed.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

Rafał Mużyło <galtgendo at o2 dot pl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #25354|0                           |1
        is obsolete|                            |

--- Comment #1 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-24 13:05:00 UTC ---
Created attachment 25355
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25355
reduced testcase

Sorry, that's full preprocessed code - attaching reduced version.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

Rafał Mużyło <galtgendo at o2 dot pl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #25355|0                           |1
        is obsolete|                            |

--- Comment #2 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-24 13:07:45 UTC ---
Created attachment 25356
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25356
the really reduced testcase

I should really check what I'm attaching.
This is the result of final reduction.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #3 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-24 13:17:59 UTC ---
OK, that's a bit surprising, -fipa-cp does help for attachment from comment 1,
but not for my final reduction - removing the attribute still works there,
though.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #4 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-24 13:38:47 UTC ---
Well,...
The actual result was quite a bit more interesting, cause:
Using built-in specs.
COLLECT_GCC=/usr/i686-pc-linux-gnu/gcc-bin/4.5.3/gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/i686-pc-linux-gnu/4.5.3/lto-wrapper
Target: i686-pc-linux-gnu
Configured with:
/mnt/workbox/builds/portage/sys-devel/gcc-4.5.3-r1/work/gcc-4.5.3/configure
--prefix=/usr --bindir=/usr/i686-pc-linux-gnu/gcc-bin/4.5.3
--includedir=/usr/lib/gcc/i686-pc-linux-gnu/4.5.3/include
--datadir=/usr/share/gcc-data/i686-pc-linux-gnu/4.5.3
--mandir=/usr/share/gcc-data/i686-pc-linux-gnu/4.5.3/man
--infodir=/usr/share/gcc-data/i686-pc-linux-gnu/4.5.3/info
--with-gxx-include-dir=/usr/lib/gcc/i686-pc-linux-gnu/4.5.3/include/g++-v4
--host=i686-pc-linux-gnu --build=i686-pc-linux-gnu --disable-altivec
--disable-fixed-point --with-ppl --with-cloog --disable-ppl-version-check
--with-cloog-include=/usr/include/cloog-ppl --enable-lto --enable-nls
--without-included-gettext --with-system-zlib --disable-werror
--enable-secureplt --disable-multilib --enable-libmudflap --disable-libssp
--enable-libgomp
--with-python-dir=/share/gcc-data/i686-pc-linux-gnu/4.5.3/python
--enable-checking=release --disable-libgcj --with-arch=i686 --enable-objc-gc
--enable-languages=c,c++,objc,fortran --enable-shared --enable-threads=posix
--enable-__cxa_atexit --enable-clocale=gnu
--with-bugurl=http://bugs.gentoo.org/ --with-pkgversion='Gentoo 4.5.3-r1 p1.0,
pie-0.4.5' --disable-libstdcxx-pch
Thread model: posix
gcc version 4.5.3 (Gentoo 4.5.3-r1 p1.0, pie-0.4.5) 
COLLECT_GCC_OPTIONS='-v' '-O1' '-fipa-cp' '-Wall' '-c' '-o'
'fprintf-mini-bug-4.6.o' '-mtune=generic' '-march=i686'
 /usr/libexec/gcc/i686-pc-linux-gnu/4.5.3/cc1 -fpreprocessed
fprintf-mini-bug-4.6.i -quiet -dumpbase fprintf-mini-bug-4.6.i -mtune=generic
-march=i686 -auxbase-strip fprintf-mini-bug-4.6.o -O1 -Wall -version -fipa-cp
-o /tmp/ccAnJvmq.s
GNU C (Gentoo 4.5.3-r1 p1.0, pie-0.4.5) version 4.5.3 (i686-pc-linux-gnu)
    compiled by GNU C version 4.5.3, GMP version 5.0.2, MPFR version 3.0.1-p4,
MPC version 0.8.2
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
GNU C (Gentoo 4.5.3-r1 p1.0, pie-0.4.5) version 4.5.3 (i686-pc-linux-gnu)
    compiled by GNU C version 4.5.3, GMP version 5.0.2, MPFR version 3.0.1-p4,
MPC version 0.8.2
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: 8436f5776e6fdb86b8b69daea94dff9c
fprintf-mini-bug-4.6.i: In function ‘write_cfg’:
fprintf-mini-bug-4.6.i:4:1: sorry, unimplemented: inlining failed in call to
‘foo’: originally indirect function call not considered for inlining
fprintf-mini-bug-4.6.i:11:7: sorry, unimplemented: called from here

This is with the code from comment 2 and '-fipa-cp'
 doesn't help in that case.

It does, however, for gcc 4.6.1.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-09-24 15:54:22 UTC ---
And this is not a bug in GCC really as GCC is correct it cannot inline
always_inline functions for indirect function calls.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #6 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-24 16:05:20 UTC ---
So, whose bug is it then ? glibc ?
cause before reduction, the code was
(http://bugs.gentoo.org/show_bug.cgi?id=365015):

#include <stdio.h>
static void w_i(void *fp, int (*outf)(void *, const char *, ...))
{
  outf(fp, ";%s\n", " - EOF -");
}

void write_cfg()
{
  FILE *fp = 0;
  w_i(fp, (int (*)(void *, const char *, ...))fprintf);
}

In gcc 4.5.3 it works likely due to internal fprint definition.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

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

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

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-09-25 12:25:17 UTC ---
And we can't inline varargs functions either.  Don't use the always_inline
attribute.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

Rafał Mużyło <galtgendo at o2 dot pl> changed:

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

--- Comment #8 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-25 12:52:53 UTC ---
(In reply to comment #7)
> And we can't inline varargs functions either.  Don't use the always_inline
> attribute.

One moment, please...
Last time I've checked, I was *not* one of glibc devs.
...No, I'm still not one.

There are already plenty unpleasant things Linus said about gcc optimizations,
you want to add glibc users too ?

Just check the previous two attachments, how the final testcase was produced.

Also, if the code is incorrect, why does adding '-fipa-cp' make it suddenly
correct ?

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #9 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-25 13:33:42 UTC ---
What I'm trying to say is that gcc should either:
- accept the code even with -fno-ipa-cp
- reject the code even with -fipa-cp
- print better diagnostics, if -fipa-cp should be the magic switch to make the
code work

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #10 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-25 13:51:14 UTC ---
So, I ran one more test:
gcc-4.6.1 -O2 -Wall -c -o fprintf-mini-bug-4.6.o fprintf-mini-bug-4.6.i
-fno-align-functions -fno-align-jumps -fno-align-labels -fno-align-loops
-fno-caller-saves -fno-crossjumping -fno-cse-follow-jumps -fno-devirtualize
-fno-expensive-optimizations -fno-gcse -fno-inline-small-functions -fno-ipa-cp
-fno-ipa-sra -fno-optimize-register-move -fno-optimize-sibling-calls
-fno-peephole2 -fno-regmove -fno-reorder-blocks -fno-reorder-functions
-fno-rerun-cse-after-loop -fno-schedule-insns2 -fno-strict-aliasing
-fno-thread-jumps -fno-tree-builtin-call-dce -fno-tree-pre
-fno-tree-switch-conversion -fno-tree-vrp

This makes '-Q --help=optimizers' output equal to '-O1' case, but
the testcase from comment 2 succeeds with this command.

So either '-Q --help=optimizers' output is incomplete or it's actually not any
specific option that triggers the change...
Well, unless there's a bug in the command line parser or a problem with the
specs.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

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

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

--- Comment #11 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-09-25 17:16:16 UTC ---
(In reply to comment #9)
> What I'm trying to say is that gcc should either:
> - accept the code even with -fno-ipa-cp
> - reject the code even with -fipa-cp
> - print better diagnostics, if -fipa-cp should be the magic switch to make the
> code work

only -fipa-cp makes us see that the indirect call resolves to an always-inline
function.  We could issue an error whenever the out-of-line body of an
always-inline function cannot be removed after optimization, but that would
upset even more people.

[Bug c/50506] gcc fails at assembly with -O1 while inlining is forced

[galtgendo at o2 dot pl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50506

--- Comment #12 from Rafał Mużyło <galtgendo at o2 dot pl> 2011-09-25 18:02:58 UTC ---
(In reply to comment #11)

Does the WONTFIX resolution here mean that glibc will need a fix then ?