[Bug middle-end/46916] New: gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[Bug middle-end/46916] New: gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

           Summary: gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs
                    compiler from profiledbootstrap
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: howarth@nitro.med.uc.edu


When gcc trunk is built with profiledbootstrap, the resulting compiler exhibits
ICEs in...

FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-1.c  -O1  compilation failed
to produce executable
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (internal compiler
error)
FAIL: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  (test for excess
errors)
WARNING: gcc.dg/torture/stackalign/non-local-goto-2.c  -O1  compilation failed
to produce executable

These are of the form...

/Users/howarth/darwin_objdir/gcc/xgcc -B/Users/howarth/darwin_objdir/gcc/
/Users/howarth/gcc/gcc/testsuite/gcc.dg/torture/stackalign/non-local-goto-1.c  
-O1  -mstackrealign -mpreferred-stack-boundary=5 -mno-mmx  -lm   -m32 -o
./non-local-goto-1.exe
xgcc: internal compiler error: Segmentation fault (program cc1)

and backtrace as...

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00007fff5bc00fe0
0x0000000100829fa8 in htab_find_slot_with_hash (htab=0x143200af0,
element=<value temporarily unavailable, due to optimizations>, hash=<value
temporarily unavailable, due to optimizations>, insert=INSERT) at
../../gcc/libiberty/hashtab.c:668
668      else if ((*htab->eq_f) (entry, element))
(gdb) bt
#0  0x0000000100829fa8 in htab_find_slot_with_hash (htab=0x143200af0,
element=<value temporarily unavailable, due to optimizations>, hash=<value
temporarily unavailable, due to optimizations>, insert=INSERT) at
../../gcc/libiberty/hashtab.c:668
#1  0x00000001007bc701 in cgraph_node (decl=0x143304e00) at
../../gcc/gcc/cgraph.c:502
(gdb)

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Jack Howarth <howarth at nitro dot med.uc.edu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-apple-darwin10
               Host|                            |x86_64-apple-darwin10
              Build|                            |x86_64-apple-darwin10

--- Comment #1 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-12 23:13:28 UTC ---
Using built-in specs.
COLLECT_GCC=/Users/howarth/dist/bin/gcc
COLLECT_LTO_WRAPPER=/Users/howarth/dist/libexec/gcc/x86_64-apple-darwin10.5.0/4.6.0/lto-wrapper
Target: x86_64-apple-darwin10.5.0
Configured with: ../gcc/configure --prefix=/Users/howarth/dist --with-gmp=/sw
--with-ppl=/sw --with-cloog=/sw --with-mpc=/sw --with-libiconv-prefix=/sw
--with-system-zlib --x-includes=/usr/X11R6/include --x-libraries=/usr/X11R6/lib
--enable-languages=c
Thread model: posix
gcc version 4.6.0 20101212 (experimental) (GCC)

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jh at suse dot cz

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-13 07:21:42 UTC ---
> When gcc trunk is built with profiledbootstrap, ...

I don't think I am using profiledbootstrap (since I don't know what it is, I
can do it without knowing), but I also see these failures that appeared between
revisions 167717 and 167731 (likely r167727). The backtrace is

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00007fff5bc00ff8
0x0000000100947f7f in htab_find_slot_with_hash (htab=0x141e00af0,
element=0x7fff5bc01040, hash=1619, insert=INSERT) at
../../work/libiberty/hashtab.c:644
644    {
(gdb) bt
#0  0x0000000100947f7f in htab_find_slot_with_hash (htab=0x141e00af0,
element=0x7fff5bc01040, hash=1619, insert=INSERT) at
../../work/libiberty/hashtab.c:644
#1  0x00000001008c2c28 in cgraph_node (decl=0x141f29500) at
../../work/gcc/cgraph.c:502

Configuration

Using built-in specs.
COLLECT_GCC=gcc46
COLLECT_LTO_WRAPPER=/opt/gcc/gcc4.6w/libexec/gcc/x86_64-apple-darwin10.5.0/4.6.0/lto-wrapper
Target: x86_64-apple-darwin10.5.0
Configured with: ../work/configure --prefix=/opt/gcc/gcc4.6w
--enable-languages=c,c++,fortran,objc,obj-c++,java,lto --with-gmp=/opt/sw64
--with-libiconv-prefix=/opt/sw64 --with-system-zlib --with-cloog=/opt/sw64
Thread model: posix
gcc version 4.6.0 20101212 (experimental) [trunk revision 167731p4] (GCC) 

On powerpc-apple-darwin9 I get

[karma] f90/bug% /opt/gcc/gcc4.6w/bin/gcc -O1
/opt/gcc/_gcc_clean/gcc/testsuite/gcc.dg/torture/stackalign/non-local-goto-1.c
Out of stack space.
Try running 'limit stacksize unlimited' in the shell to raise its limit.

with the maximum available stack size: (kbytes, -s) 65532.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-13 08:00:52 UTC ---
The failures are gone if I revert revision 167727.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #4 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-13 14:30:04 UTC ---
I don't see r167727 in gcc-patches for an approval from a darwin maintainer.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler from profiledbootstrap

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #5 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-13 15:32:34 UTC ---
I can confirm that reverting r167727 also eliminates the failures in
gcc.dg/torture/stackalign/non-local-goto-[1,2].c with profiledbootstrap as
well.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #6 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-13 15:34:01 UTC ---
The offending change is...

Author: hubicka
Date: Sun Dec 12 19:12:08 2010
New Revision: 167727

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=167727
Log:


    * varasm.c (default_function_section): Check flag_reorder_functions
    and targetm.have_named_sections.
    * config/darwin.c (darwin_function_section): Check
    flag_reorder_functions.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/darwin.c
    trunk/gcc/varasm.c

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.12.13 18:22:21
                 CC|                            |iains at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #7 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-13 18:22:21 UTC ---
the problem is infinite recursion -- the bailing of darwin_function_section
when reodering is off causes a fallback to darwin_text_section ()  -- which
calls unlikely_text_section () ... and we get a stack meltdown...

here is fix ... but I dunno if it's the "Right" fix --- so I'll let Mike &|
Honza comment... 


Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c    (revision 167755)
+++ gcc/config/darwin.c    (working copy)
@@ -1150,7 +1150,7 @@ darwin_text_section (int reloc, int weak)
   if (reloc)
     return (weak
         ? darwin_sections[text_unlikely_coal_section]
-        : unlikely_text_section ());
+        : darwin_sections[text_unlikely_section]);
   else
     return (weak
         ? darwin_sections[text_coal_section]
Index: gcc/config/darwin-sections.def
===================================================================
--- gcc/config/darwin-sections.def    (revision 167755)
+++ gcc/config/darwin-sections.def    (working copy)
@@ -30,6 +30,8 @@ along with GCC; see the file COPYING3.  If not see
 /* .text handled in varasm.c  */
 DEF_SECTION (text_coal_section, SECTION_CODE|SECTION_NO_ANCHOR,
          ".section __TEXT,__textcoal_nt,coalesced,pure_instructions", 0)
+DEF_SECTION (text_unlikely_section, SECTION_CODE|SECTION_NO_ANCHOR,
+         ".section __TEXT,__unlikely,regular,pure_instructions", 0)
 DEF_SECTION (text_unlikely_coal_section, SECTION_CODE|SECTION_NO_ANCHOR,
          ".section __TEXT,__text_unlikely_coal,"
          "coalesced,pure_instructions", 0)

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[mrs at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

mrs@gcc.gnu.org <mrs at gcc dot gnu.org> changed:

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

--- Comment #8 from mrs at gcc dot gnu.org <mrs at gcc dot gnu.org> 2010-12-13 18:32:56 UTC ---
Ok.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #9 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-13 19:10:55 UTC ---
> here is fix ... but I dunno if it's the "Right" fix --- so I'll let Mike &|
> Honza comment... 

The patch fixed the failures. I am regstrapping. Thanks!

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #10 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-13 20:16:23 UTC ---

perhaps it would be safer to name the fallback section:

>>	    : darwin_sections[text_unlikely_fallback_section]);


>> DEF_SECTION (text_unlikely_fallback_section, SECTION_CODE|SECTION_NO_ANCHOR,
         ".section __TEXT,__unlikely_flbk,regular,pure_instructions", 0)

(limit of 16chars for the section name == "unlikely fallback" )

then there's no danger of a section name clash when someone tries to build 
named_section ( "__TEXT,__unlikely,regular,pure_instruction", .... )

(If the regtest works w/out this, then perhaps not necessary ... )

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[mikestump at comcast dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #11 from Mike Stump <mikestump at comcast dot net> 2010-12-13 22:55:12 UTC ---
I don't think this should be necessary.  One section should be enough.  If you
have specific concerns, let me know, I could just be missing something.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #12 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 08:54:08 UTC ---
I have regstrapped the patch in comment #7 on top of revision 167770. The
failures corresponding to this PR are gone. However I see the following for
both -m32 and -m64:

FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -g  -fprofile-use
FAIL: g++.dg/tree-prof/partition2.C compilation,  -O3 -g  -fprofile-use
UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -O3 -g  -fprofile-use

the corresponding typical log entry being

Executing on host: /opt/gcc/build_w/gcc/testsuite/g++/../../g++
-B/opt/gcc/build_w/gcc/testsuite/g++/../../
/opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/partition2.C  -nostdinc++
-I/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/include/x86_64-apple-darwin10.5.0
-I/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/include
-I/opt/gcc/work/libstdc++-v3/libsupc++
-I/opt/gcc/work/libstdc++-v3/include/backward
-I/opt/gcc/work/libstdc++-v3/testsuite/util -fmessage-length=0  -g 
-fnon-call-exceptions -freorder-blocks-and-partition -fprofile-use   
-L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs 
-B/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs 
-L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs
-L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libiberty  -multiply_defined
suppress -lm   -m32 -o /opt/gcc/build_w/gcc/testsuite/g++/partition2.x02   
(timeout = 300)
warning: no debug symbols in executable (-arch i386)^M
output is:
warning: no debug symbols in executable (-arch i386)^M

FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -g  -fprofile-use

The key option is -g (the other tests of g++.dg/tree-prof/partition2.C without
-g succeed). I don't know if these failures come from the patch or have been
introduced between revisions 167731 and 167770 by an other commit.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #13 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 09:09:07 UTC ---
(In reply to comment #12)
> I have regstrapped the patch in comment #7 on top of revision 167770. The
> failures corresponding to this PR are gone. However I see the following for
> both -m32 and -m64:
> 
> FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
> UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -g  -fprofile-use
> FAIL: g++.dg/tree-prof/partition2.C compilation,  -O3 -g  -fprofile-use
> UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -O3 -g  -fprofile-use
> 
> the corresponding typical log entry being
> 
> Executing on host: /opt/gcc/build_w/gcc/testsuite/g++/../../g++
> -B/opt/gcc/build_w/gcc/testsuite/g++/../../
> /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/partition2.C  -nostdinc++
> -I/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/include/x86_64-apple-darwin10.5.0
> -I/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/include
> -I/opt/gcc/work/libstdc++-v3/libsupc++
> -I/opt/gcc/work/libstdc++-v3/include/backward
> -I/opt/gcc/work/libstdc++-v3/testsuite/util -fmessage-length=0  -g 
> -fnon-call-exceptions -freorder-blocks-and-partition -fprofile-use   
> -L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs 
> -B/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs 
> -L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libstdc++-v3/src/.libs
> -L/opt/gcc/build_w/x86_64-apple-darwin10.5.0/i386/libiberty  -multiply_defined
> suppress -lm   -m32 -o /opt/gcc/build_w/gcc/testsuite/g++/partition2.x02   
> (timeout = 300)
> warning: no debug symbols in executable (-arch i386)^M
> output is:
> warning: no debug symbols in executable (-arch i386)^M
> 
> FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
> UNRESOLVED: g++.dg/tree-prof/partition2.C execution,    -g  -fprofile-use
> 
> The key option is -g (the other tests of g++.dg/tree-prof/partition2.C without
> -g succeed). I don't know if these failures come from the patch or have been
> introduced between revisions 167731 and 167770 by an other commit.

I think this is a different problem - those messages are characteristic of
dsymutil missing an input file.
This might be related to the changes made to call dsymutil within collect2.

Is it possible to isolate the command and run it with -v ? and/or -Wl,-debug ?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

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

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

--- Comment #14 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-14 10:08:25 UTC ---
Hi,
does the following patch help?
Index: opts.c
===================================================================
--- opts.c      (revision 167752)
+++ opts.c      (working copy)
@@ -738,6 +738,9 @@ finish_options (struct gcc_options *opts
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If the target requested unwind info, then turn off the partitioning
      optimization with a different message.  Likewise, if the target does not
      support named sections.  */

it seems that with -freorder-functions-and-partition we should always imply
-freorder-functions. Without this the partitioning code will now partition in
between the same function section that is no-op and probably leads to failure.

This however also imply a semi-latent bug - if 
-freorder-functions-and-partition breaks on cases where both hot and cold
sections are the same, it ought to disable itself, because for some functions
(like ones in named sections) we can not partition then.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #15 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-14 10:09:53 UTC ---
Also it is clear who starts the infinite recursion?  I find it odd, that
disabling the function partitioning lead to need to explicitely declare
unlikely section.  Is someone else trying to access it and fail for some
reason?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #16 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 10:39:10 UTC ---
(In reply to comment #15)
> Also it is clear who starts the infinite recursion?  I find it odd, that
> disabling the function partitioning lead to need to explicitely declare
> unlikely section.  Is someone else trying to access it and fail for some
> reason?

well.... the pre-existing implementation of :

static section *
darwin_text_section (int reloc, int weak)
{
  if (reloc)
    return (weak
        ? darwin_sections[text_unlikely_coal_section]
        : unlikely_text_section ());  <<<<<<<<+++++++++
  else
    return (weak
        ? darwin_sections[text_coal_section]
        : text_section);
}

...  is invoked when darwin_function_section () returns NULL.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #17 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-14 10:58:40 UTC ---
> well.... the pre-existing implementation of :
> 
> static section *
> darwin_text_section (int reloc, int weak)
> {
>   if (reloc)
>     return (weak
>         ? darwin_sections[text_unlikely_coal_section]
>         : unlikely_text_section ());  <<<<<<<<+++++++++
>   else
>     return (weak
>         ? darwin_sections[text_coal_section]
>         : text_section);
> }
> 
> ...  is invoked when darwin_function_section () returns NULL.
The code probably should also disable itself with -fno-reorder-functions...

Honza

Re: [Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[Jan Hubicka]

> well.... the pre-existing implementation of :
> 
> static section *
> darwin_text_section (int reloc, int weak)
> {
>   if (reloc)
>     return (weak
>         ? darwin_sections[text_unlikely_coal_section]
>         : unlikely_text_section ());  <<<<<<<<+++++++++
>   else
>     return (weak
>         ? darwin_sections[text_coal_section]
>         : text_section);
> }
> 
> ...  is invoked when darwin_function_section () returns NULL.
The code probably should also disable itself with -fno-reorder-functions...

Honza

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #18 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 12:17:00 UTC ---
(In reply to comment #17)
> > well.... the pre-existing implementation of :
> > 
> > static section *
> > darwin_text_section (int reloc, int weak)
> > {
> >   if (reloc)
> >     return (weak
> >         ? darwin_sections[text_unlikely_coal_section]
> >         : unlikely_text_section ());  <<<<<<<<+++++++++
> >   else
> >     return (weak
> >         ? darwin_sections[text_coal_section]
> >         : text_section);
> > }
> > 
> > ...  is invoked when darwin_function_section () returns NULL.
> The code probably should also disable itself with -fno-reorder-functions...

well, this also fixes the issue .. 

but, the original implementation of darwin_text_section () pre-dates my
involvement 
-- I am not sure why it partitions, and ...
--  if we should preserve some partitioning even when the option is off ... 

Mike?

Index: gcc/opts.c
===================================================================
--- gcc/opts.c    (revision 167793)
+++ gcc/opts.c    (working copy)
@@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If user requested unwind info, then turn off the partitioning
      optimization.  */

Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c    (revision 167793)
+++ gcc/config/darwin.c    (working copy)
@@ -1147,7 +1147,7 @@ darwin_mark_decl_preserved (const char *name)
 static section *
 darwin_text_section (int reloc, int weak)
 {
-  if (reloc)
+  if (flag_reorder_functions && reloc)
     return (weak
         ? darwin_sections[text_unlikely_coal_section]
         : unlikely_text_section ());

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #19 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-14 13:18:30 UTC ---
Hi,
does the following patch fix the problem?
darwin_text_section no longer needs to care about hot/cold code since this is
already done in darwin_function_section.

In fact you might additionally consider putting 
  return (DECL_WEAK (decl)
      ? darwin_sections[text_coal_section]
      : text_section);
into places returning NULL in darwin_function_section, removing case
SECCAT_TEXT
from machopic_select_section undefining USE_SELECT_SECTION_FOR_FUNCTIONS
as suggested by rth as cleanup.

Honza

Index: opts.c
===================================================================
--- opts.c    (revision 167794)
+++ opts.c    (working copy)
@@ -738,6 +738,9 @@ finish_options (struct gcc_options *opts
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If the target requested unwind info, then turn off the partitioning
      optimization with a different message.  Likewise, if the target does not
      support named sections.  */
Index: config/darwin.c
===================================================================
--- config/darwin.c    (revision 167794)
+++ config/darwin.c    (working copy)
@@ -1145,16 +1145,11 @@ darwin_mark_decl_preserved (const char *
 }

 static section *
-darwin_text_section (int reloc, int weak)
+darwin_text_section (int weak)
 {
-  if (reloc)
-    return (weak
-        ? darwin_sections[text_unlikely_coal_section]
-        : unlikely_text_section ());
-  else
-    return (weak
-        ? darwin_sections[text_coal_section]
-        : text_section);
+  return (weak
+      ? darwin_sections[text_coal_section]
+      : text_section);
 }

 static section *

Re: [Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[Jan Hubicka]

Hi,
does the following patch fix the problem?
darwin_text_section no longer needs to care about hot/cold code since this is
already done in darwin_function_section.

In fact you might additionally consider putting 
  return (DECL_WEAK (decl)
	  ? darwin_sections[text_coal_section]
	  : text_section);
into places returning NULL in darwin_function_section, removing case SECCAT_TEXT
from machopic_select_section undefining USE_SELECT_SECTION_FOR_FUNCTIONS
as suggested by rth as cleanup.

Honza

Index: opts.c
===================================================================
--- opts.c	(revision 167794)
+++ opts.c	(working copy)
@@ -738,6 +738,9 @@ finish_options (struct gcc_options *opts
       opts->x_flag_reorder_blocks = 1;
     }
 
+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If the target requested unwind info, then turn off the partitioning
      optimization with a different message.  Likewise, if the target does not
      support named sections.  */
Index: config/darwin.c
===================================================================
--- config/darwin.c	(revision 167794)
+++ config/darwin.c	(working copy)
@@ -1145,16 +1145,11 @@ darwin_mark_decl_preserved (const char *
 }
 
 static section *
-darwin_text_section (int reloc, int weak)
+darwin_text_section (int weak)
 {
-  if (reloc)
-    return (weak
-	    ? darwin_sections[text_unlikely_coal_section]
-	    : unlikely_text_section ());
-  else
-    return (weak
-	    ? darwin_sections[text_coal_section]
-	    : text_section);
+  return (weak
+	  ? darwin_sections[text_coal_section]
+	  : text_section);
 }
 
 static section *

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #20 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 14:02:47 UTC ---
(In reply to comment #13)
> I think this is a different problem - those messages are characteristic of
> dsymutil missing an input file.  This might be related to the changes 
> made to call dsymutil within collect2.

This is also seen in
http://gcc.gnu.org/ml/gcc-testresults/2010-12/msg01197.html

> Is it possible to isolate the command and run it with -v ? and/or -Wl,-debug ?

Well, it seems to be another "Heisenfailure"!-(private joke with Iain): if I
run the test alone, it succeeds but fails when tested as part of the full g++
suite.


[macbook] build_w/gcc% make -k check-g++ RUNTESTFLAGS="tree-prof.exp
--target_board=unix'{-m32,-m64}'"
... Skip common boring details ...
    `if [ -f ${srcdir}/../dejagnu/runtest ] ; then echo
${srcdir}/../dejagnu/runtest ; else echo runtest; fi` --tool g++ tree-prof.exp
--target_board=unix'{-m32,-m64}' $runtestflags)
WARNING: Couldn't find the global config file.
Test Run By dominiq on Tue Dec 14 14:03:25 2010
Native configuration is x86_64-apple-darwin10.5.0

        === g++ tests ===

Schedule of variations:
    unix/-m32
    unix/-m64

Running target unix/-m32
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...

        === g++ Summary for unix/-m32 ===

# of expected passes        24
Running target unix/-m64
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...

        === g++ Summary for unix/-m64 ===

# of expected passes        24

        === g++ Summary ===

# of expected passes        48
/opt/gcc/build_w/gcc/testsuite/g++/../../g++  version 4.6.0 20101213
(experimental) [trunk revision 167770p4] (GCC) 

[macbook] build_w/gcc% make -k check-g++
RUNTESTFLAGS="--target_board=unix'{-m32,-m64}'"
... Skip common boring details ...
    `if [ -f ${srcdir}/../dejagnu/runtest ] ; then echo
${srcdir}/../dejagnu/runtest ; else echo runtest; fi` --tool g++
--target_board=unix'{-m32,-m64}' $runtestflags)
WARNING: Couldn't find the global config file.
Test Run By dominiq on Tue Dec 14 14:03:59 2010
Native configuration is x86_64-apple-darwin10.5.0

        === g++ tests ===

Schedule of variations:
    unix/-m32
    unix/-m64

Running target unix/-m32
... Skip boring details ...
Running /opt/gcc/work/gcc/testsuite/g++.dg/dg.exp ...
WARNING: g++.dg/ext/label13.C compilation failed to produce executable
... Skip boring details ...
Running /opt/gcc/work/gcc/testsuite/g++.dg/torture/stackalign/stackalign.exp
...
FAIL: g++.dg/torture/stackalign/eh-alloca-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-2.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-alloca-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-1.C  -O3 -g  execution test
FAIL: g++.dg/torture/stackalign/eh-vararg-2.C  -O3 -g  execution test
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
FAIL: g++.dg/tree-prof/partition2.C compilation,  -O3 -g  -fprofile-use
Running /opt/gcc/work/gcc/testsuite/g++.dg/vect/vect.exp ...
Running /opt/gcc/work/gcc/testsuite/g++.old-deja/old-deja.exp ...

        === g++ Summary for unix/-m32 ===

# of expected passes        26164
# of unexpected failures    8
# of expected failures        157
# of unresolved testcases    2
# of unsupported tests        176
Running target unix/-m64
... Skip boring details ...
Running /opt/gcc/work/gcc/testsuite/g++.dg/dg.exp ...
WARNING: g++.dg/ext/label13.C compilation failed to produce executable
... Skip boring details ...
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
FAIL: g++.dg/tree-prof/partition2.C compilation,  -g  -fprofile-use
FAIL: g++.dg/tree-prof/partition2.C compilation,  -O3 -g  -fprofile-use
Running /opt/gcc/work/gcc/testsuite/g++.dg/vect/vect.exp ...
Running /opt/gcc/work/gcc/testsuite/g++.old-deja/old-deja.exp ...

        === g++ Summary for unix/-m64 ===

# of expected passes        26410
# of unexpected failures    2
# of expected failures        157
# of unresolved testcases    2
# of unsupported tests        370

        === g++ Summary ===

# of expected passes        52574
# of unexpected failures    10
# of expected failures        314
# of unresolved testcases    4
# of unsupported tests        546
/opt/gcc/build_w/gcc/testsuite/g++/../../g++  version 4.6.0 20101213
(experimental) [trunk revision 167770p4] (GCC) 

make[1]: [check-parallel-g++] Error 1 (ignored)

I'll test the patch in comment #19 (I love the - in the diff files!-).

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #21 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 14:04:58 UTC ---
(In reply to comment #19)
> Hi,
> does the following patch fix the problem?
> darwin_text_section no longer needs to care about hot/cold code since this is
> already done in darwin_function_section.
> 
> In fact you might additionally consider putting 
>   return (DECL_WEAK (decl)
>       ? darwin_sections[text_coal_section]
>       : text_section);
> into places returning NULL in darwin_function_section, removing case
> SECCAT_TEXT
> from machopic_select_section undefining USE_SELECT_SECTION_FOR_FUNCTIONS
> as suggested by rth as cleanup.

like this?
 (fixes the specific problem - but not reg-tested):


Index: gcc/opts.c
===================================================================
--- gcc/opts.c    (revision 167793)
+++ gcc/opts.c    (working copy)
@@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If user requested unwind info, then turn off the partitioning
      optimization.  */

Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c    (revision 167793)
+++ gcc/config/darwin.c    (working copy)
@@ -1145,19 +1145,6 @@ darwin_mark_decl_preserved (const char *name)
 }

 static section *
-darwin_text_section (int reloc, int weak)
-{
-  if (reloc)
-    return (weak
-        ? darwin_sections[text_unlikely_coal_section]
-        : unlikely_text_section ());
-  else
-    return (weak
-        ? darwin_sections[text_coal_section]
-        : text_section);
-}
-
-static section *
 darwin_rodata_section (int weak, bool zsize)
 {
   return (weak
@@ -1267,17 +1254,7 @@ machopic_select_section (tree decl,
   switch (categorize_decl_for_section (decl, reloc))
     {
     case SECCAT_TEXT:
-      {
-    struct cgraph_node *node;
-    if (decl && TREE_CODE (decl) == FUNCTION_DECL
-        && (node = cgraph_get_node (decl)) != NULL)
-      base_section = darwin_function_section (decl,
-                          node->frequency,
-                          node->only_called_at_startup,
-                          node->only_called_at_exit);
-    if (!base_section)
-          base_section = darwin_text_section (reloc, weak);
-      }
+      gcc_unreachable ();
       break;

     case SECCAT_RODATA:
@@ -2969,7 +2946,10 @@ darwin_function_section (tree decl, enum node_freq
               bool startup, bool exit)
 {
   if (!flag_reorder_functions)
-    return NULL;
+    return (DECL_WEAK (decl)
+        ? darwin_sections[text_coal_section]
+        : text_section);
+
   /* Startup code should go to startup subsection unless it is
      unlikely executed (this happens especially with function splitting
      where we can split away unnecesary parts of static constructors.  */
@@ -2994,7 +2974,9 @@ darwin_function_section (tree decl, enum node_freq
     return get_named_text_section
          (decl, "__TEXT,__hot,regular,pure_instructions", "_hot");
       default:
-    return NULL;
+    return (DECL_WEAK (decl)
+        ? darwin_sections[text_coal_section]
+        : text_section);
     }
 }

Index: gcc/config/darwin.h
===================================================================
--- gcc/config/darwin.h    (revision 167793)
+++ gcc/config/darwin.h    (working copy)
@@ -664,7 +664,7 @@ extern GTY(()) section * darwin_sections[NUM_DARWI

 #undef    TARGET_ASM_SELECT_SECTION
 #define TARGET_ASM_SELECT_SECTION machopic_select_section
-#define USE_SELECT_SECTION_FOR_FUNCTIONS
+
 #undef    TARGET_ASM_FUNCTION_SECTION
 #define TARGET_ASM_FUNCTION_SECTION darwin_function_section

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #22 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 14:17:53 UTC ---
I have found the reason of the unexpected pass when testing tree-prof.exp
alone: -g is not passed to the tests. If I force it I get

        === g++ tests ===

Schedule of variations:
    unix/-m32/-g
    unix/-m64/-g

Running target unix/-m32/-g
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
FAIL: g++.dg/tree-prof/partition1.C compilation,  -fprofile-use
FAIL: g++.dg/tree-prof/partition2.C compilation,  -fprofile-use

        === g++ Summary for unix/-m32/-g ===

# of expected passes        20
# of unexpected failures    2
# of unresolved testcases    2
Running target unix/-m64/-g
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
FAIL: g++.dg/tree-prof/partition1.C compilation,  -fprofile-use
FAIL: g++.dg/tree-prof/partition2.C compilation,  -fprofile-use

        === g++ Summary for unix/-m64/-g ===

# of expected passes        20
# of unexpected failures    2
# of unresolved testcases    2

        === g++ Summary ===

# of expected passes        40
# of unexpected failures    4
# of unresolved testcases    4
/opt/gcc/build_w/gcc/testsuite/g++/../../g++  version 4.6.0 20101213
(experimental) [trunk revision 167770p4] (GCC) 

make[1]: [check-parallel-g++] Error 1 (ignored)

Is this expected or should I fill another PR?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #23 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 14:23:42 UTC ---
(In reply to comment #22)
> I have found the reason of the unexpected pass when testing tree-prof.exp
> alone: -g is not passed to the tests. If I force it I get


> Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
> Using /opt/gcc/work/gcc/testsuite/config/default.exp as
> tool-and-target-specific interface file.
> Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
> FAIL: g++.dg/tree-prof/partition1.C compilation,  -fprofile-use
> FAIL: g++.dg/tree-prof/partition2.C compilation,  -fprofile-use


> Is this expected or should I fill another PR?

I suspect it is a different issue...  can we look at the output with "-v" and
"-Wl,-debug" ?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #24 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-14 14:30:55 UTC ---
(In reply to comment #23)
> (In reply to comment #22)
> > I have found the reason of the unexpected pass when testing tree-prof.exp
> > alone: -g is not passed to the tests. If I force it I get
> 
> 
> > Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
> > Using /opt/gcc/work/gcc/testsuite/config/default.exp as
> > tool-and-target-specific interface file.
> > Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
> > FAIL: g++.dg/tree-prof/partition1.C compilation,  -fprofile-use
> > FAIL: g++.dg/tree-prof/partition2.C compilation,  -fprofile-use
> 
> 
> > Is this expected or should I fill another PR?
> 
> I suspect it is a different issue...  can we look at the output with "-v" and
> "-Wl,-debug" ?

Also look at PR45646 which may be related and could have gone latent rather
than have been fixed.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #25 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-14 14:35:46 UTC ---
(In reply to comment #21)

> like this?
>  (fixes the specific problem - but not reg-tested):
> 
> 

This is in addition to the patch from Comment 7, right?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #26 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-14 14:37:52 UTC ---
(In reply to comment #25)
> (In reply to comment #21)
> 
> > like this?
> >  (fixes the specific problem - but not reg-tested):
> > 
> > 
> 
> This is in addition to the patch from Comment 7, right?

Nevermind. I see it replaces it.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #27 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 16:39:29 UTC ---
(In reply to comment #23)
> (In reply to comment #22)

> > Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...
> > FAIL: g++.dg/tree-prof/partition1.C compilation,  -fprofile-use
> > FAIL: g++.dg/tree-prof/partition2.C compilation,  -fprofile-use

hm.  AFAIU, these should not work because exceptions code is incompatible with
reordering - or has this changed?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #28 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 16:41:02 UTC ---
more of a problem is:

FAIL: g++.dg/pch/system-1.C  -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 assembly comparison
FAIL: g++.dg/pch/system-2.C  -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 assembly comparison

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #29 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 16:44:09 UTC ---
Partial tests (x86_64-apple-darwin10.5.0) with the patch in comment #21 (with
the one in comment #7 reverted) show that the pr is fixed, but it introduces at
least one regression for both -m32/64:

FAIL: g++.dg/other/pr22003.C (test for excess errors)

[macbook] f90/bug% g++46 -O2 -fno-exceptions -freorder-blocks-and-partition
/opt/gcc/work/gcc/testsuite/g++.dg/other/pr22003.C
/opt/gcc/work/gcc/testsuite/g++.dg/other/pr22003.C: In destructor 'virtual
c3::~c3()':
/opt/gcc/work/gcc/testsuite/g++.dg/other/pr22003.C:17:8: error: virtual
c3::~c3() causes a section type conflict
/opt/gcc/work/gcc/testsuite/g++.dg/other/pr22003.C:17:8: error: virtual
c3::~c3() causes a section type conflict
/opt/gcc/work/gcc/testsuite/g++.dg/other/pr22003.C:17:8: error: virtual
c3::~c3() causes a section type conflict

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #30 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 16:46:19 UTC ---
PASS: g++.dg/pch/system-1.C  -O2 -g -I. (test for excess errors)
line #35
<
__GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_0966702C:
> __GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_1FC588D5:
line #72
<       .long  
__GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_0966702C
>       .long   __GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_1FC588D5
line #251
<
__GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_0966702C.eh:
> __GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_1FC588D5.eh:
line #1373
<       .ascii
"_GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_0966702C\0"
>       .ascii "_GLOBAL__sub_I__GCC_gcc_live_trunk_gcc_testsuite_g__.dg_pch_system_1.C_867072EB_1FC588D5\0"
FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #31 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-14 17:09:47 UTC ---
With the patch from comment 21 applied on x86_64-apple-darwin10, I am seeing...

FAIL: gcc.dg/darwin-weakimport-3.c scan-assembler-not coalesced
FAIL: gcc.dg/pr25376.c scan-assembler my_named_section

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #32 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 17:16:39 UTC ---
minor update (this removes a test for reorder & partition + exceptions that is
carried out too early)

Tests for reorder + exceptions and reorder + unwind are carried out in opts.c
now.



Index: gcc/opts.c
===================================================================
--- gcc/opts.c    (revision 167793)
+++ gcc/opts.c    (working copy)
@@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If user requested unwind info, then turn off the partitioning
      optimization.  */

Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c    (revision 167793)
+++ gcc/config/darwin.c    (working copy)
@@ -1145,19 +1145,6 @@ darwin_mark_decl_preserved (const char *name)
 }

 static section *
-darwin_text_section (int reloc, int weak)
-{
-  if (reloc)
-    return (weak
-        ? darwin_sections[text_unlikely_coal_section]
-        : unlikely_text_section ());
-  else
-    return (weak
-        ? darwin_sections[text_coal_section]
-        : text_section);
-}
-
-static section *
 darwin_rodata_section (int weak, bool zsize)
 {
   return (weak
@@ -1267,17 +1254,7 @@ machopic_select_section (tree decl,
   switch (categorize_decl_for_section (decl, reloc))
     {
     case SECCAT_TEXT:
-      {
-    struct cgraph_node *node;
-    if (decl && TREE_CODE (decl) == FUNCTION_DECL
-        && (node = cgraph_get_node (decl)) != NULL)
-      base_section = darwin_function_section (decl,
-                          node->frequency,
-                          node->only_called_at_startup,
-                          node->only_called_at_exit);
-    if (!base_section)
-          base_section = darwin_text_section (reloc, weak);
-      }
+      gcc_unreachable ();
       break;

     case SECCAT_RODATA:
@@ -2559,17 +2536,6 @@ darwin_override_options (void)
   if (!global_options_set.x_dwarf_strict) 
     dwarf_strict = 1;

-  /* Disable -freorder-blocks-and-partition for darwin_emit_unwind_label.  */
-  if (flag_reorder_blocks_and_partition 
-      && (targetm.asm_out.emit_unwind_label == darwin_emit_unwind_label))
-    {
-      inform (input_location,
-              "-freorder-blocks-and-partition does not work with exceptions "
-              "on this architecture");
-      flag_reorder_blocks_and_partition = 0;
-      flag_reorder_blocks = 1;
-    }
-
   if (flag_mkernel || flag_apple_kext)
     {
       /* -mkernel implies -fapple-kext for C++ */
@@ -2969,7 +2935,10 @@ darwin_function_section (tree decl, enum node_freq
               bool startup, bool exit)
 {
   if (!flag_reorder_functions)
-    return NULL;
+    return (DECL_WEAK (decl)
+        ? darwin_sections[text_coal_section]
+        : text_section);
+
   /* Startup code should go to startup subsection unless it is
      unlikely executed (this happens especially with function splitting
      where we can split away unnecesary parts of static constructors.  */
@@ -2994,7 +2963,9 @@ darwin_function_section (tree decl, enum node_freq
     return get_named_text_section
          (decl, "__TEXT,__hot,regular,pure_instructions", "_hot");
       default:
-    return NULL;
+    return (DECL_WEAK (decl)
+        ? darwin_sections[text_coal_section]
+        : text_section);
     }
 }

Index: gcc/config/darwin.h
===================================================================
--- gcc/config/darwin.h    (revision 167793)
+++ gcc/config/darwin.h    (working copy)
@@ -664,7 +664,7 @@ extern GTY(()) section * darwin_sections[NUM_DARWI

 #undef    TARGET_ASM_SELECT_SECTION
 #define TARGET_ASM_SELECT_SECTION machopic_select_section
-#define USE_SELECT_SECTION_FOR_FUNCTIONS
+
 #undef    TARGET_ASM_FUNCTION_SECTION
 #define TARGET_ASM_FUNCTION_SECTION darwin_function_section

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #33 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-14 17:32:16 UTC ---
> minor update (this removes a test for reorder & partition + exceptions that is
> carried out too early)
This seem sane to me.  I've also posted more fixed to function partitioning
that
should handle named functions and other side cases.

Thanks for looking into this,
Honza

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #34 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 19:00:43 UTC ---
If I did not make any mistake, the updated patch in comment #32 does not fix
the failures in comments #29 and #31.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #35 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 19:53:34 UTC ---
(In reply to comment #34)
> If I did not make any mistake, the updated patch in comment #32 does not fix
> the failures in comments #29 and #31.

no, you are correct - I think we now have to wait for Honza's "additional
fixes" as per comment #33

fails in 28, 29 and 31 are still present.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #36 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 20:05:43 UTC ---
> fails in 28, 29 and 31 are still present.

On which platform do you see the failures in 28? I don't see them on
x86_64-apple-darwin10.5.0 when doing a full test of g++.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|x86_64-apple-darwin10       |*-apple-darwin{9,10}
               Host|x86_64-apple-darwin10       |*-apple-darwin{9,10}
              Build|x86_64-apple-darwin10       |*-apple-darwin{9,10}

--- Comment #37 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 20:17:41 UTC ---
(In reply to comment #36)
> > fails in 28, 29 and 31 are still present.
> 
> On which platform do you see the failures in 28? I don't see them on
> x86_64-apple-darwin10.5.0 when doing a full test of g++.

on i686-darwin9 @m32 and m64.

(adjusting the target/host/build triplets - the overall bug also affects
powerpc)

FAIL: g++.dg/other/pr22003.C (test for excess errors)
FAIL: g++.dg/pch/system-1.C  -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 assembly comparison
FAIL: g++.dg/pch/system-2.C  -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 assembly comparison

+ partition{1,2}.C  *  -fprofile-use

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #38 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-14 21:28:12 UTC ---
I posted the fixes, but it should not affect -fno-reorder-blocks-and-partition.
Any idea what is going wrong here?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #39 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-14 21:32:15 UTC ---
The patch I had in mind
http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01129.html

The hunk
Index: gcc/opts.c
===================================================================
--- gcc/opts.c    (revision 167793)
+++ gcc/opts.c    (working copy)
@@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_blocks = 1;
+
   /* If user requested unwind info, then turn off the partitioning
      optimization.  */

should be
Index: gcc/opts.c
===================================================================
--- gcc/opts.c    (revision 167793)
+++ gcc/opts.c    (working copy)
@@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
       opts->x_flag_reorder_blocks = 1;
     }

+  if (opts->x_flag_reorder_blocks_and_partition)
+    opts->x_flag_reorder_functions = 1;
+
   /* If user requested unwind info, then turn off the partitioning
      optimization.  */

sorry for typo.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #40 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 21:37:53 UTC ---
(In reply to comment #38)
> I posted the fixes, but it should not affect -fno-reorder-blocks-and-partition.
> Any idea what is going wrong here?

well, unless some interaction from the typo.. (possible) .. AFAICT most of the
fallout is multiple definitions of .eh symbols... 

I was just looking in the debugger .. 
...  wondering if I can intercept the cold ones and give them different names. 
;-)

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #41 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-14 22:46:09 UTC ---
> well, unless some interaction from the typo.. (possible) .. AFAICT most of the
> fallout is multiple definitions of .eh symbols... 
Actually forcing hot&cold sections to coincide might result in multiple
definition of eh symbol.
The typo should fix it for -freorder-blocks-and-partition compilations and the
patch I posed to ML
might solve other cases by disabling the partitioning logic when we don't
really partition the function
(that was point of the patch)

Honza

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #42 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 22:55:06 UTC ---
I have reverted any of the above patches and updated to revision 167812. First,
I now see the failures in comment #28 (they are probably a different PR).
Second, any of the patches caused the failures in comment #12 by enabling the
tests to be run. Without patch, I get 

[macbook] build_w/gcc% make -k check-g++
RUNTESTFLAGS="tree-prof.exp=partition2.C
--target_board=unix'{-m32/-g,-m64/-g}'"
...

        === g++ tests ===

Schedule of variations:
    unix/-m32/-g
    unix/-m64/-g

Running target unix/-m32/-g
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...

        === g++ Summary for unix/-m32/-g ===

# of unsupported tests        1
Running target unix/-m64/-g
Using /sw64/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw64/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/work/gcc/testsuite/g++.dg/tree-prof/tree-prof.exp ...

        === g++ Summary for unix/-m64/-g ===

# of unsupported tests        1

        === g++ Summary ===

# of unsupported tests        2
/opt/gcc/build_w/gcc/testsuite/g++/../../g++  version 4.6.0 20101214
(experimental) [trunk revision 167812p3] (GCC) 

I'll apply the Honza's patch and regtest during the night.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #43 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-14 23:21:35 UTC ---
(In reply to comment #39)
> The patch I had in mind
> http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01129.html
> 
> The hunk
> Index: gcc/opts.c
> ===================================================================
> --- gcc/opts.c    (revision 167793)
> +++ gcc/opts.c    (working copy)
> @@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
>        opts->x_flag_reorder_blocks = 1;
>      }
> 
> +  if (opts->x_flag_reorder_blocks_and_partition)
> +    opts->x_flag_reorder_blocks = 1;
> +
>    /* If user requested unwind info, then turn off the partitioning
>       optimization.  */
> 
> should be
> Index: gcc/opts.c
> ===================================================================
> --- gcc/opts.c    (revision 167793)
> +++ gcc/opts.c    (working copy)
> @@ -723,6 +723,9 @@ finish_options (struct gcc_options *opts, struct g
>        opts->x_flag_reorder_blocks = 1;
>      }
> 
> +  if (opts->x_flag_reorder_blocks_and_partition)
> +    opts->x_flag_reorder_functions = 1;
> +
>    /* If user requested unwind info, then turn off the partitioning
>       optimization.  */
> 
> sorry for typo.

The cited patch doesn't have any such change in that particular context. Was
this entire hunk missing from the posted patch?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #44 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 23:39:34 UTC ---
Created attachment 22758
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22758
merge between Honza's and Iain's patches

combined patch.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #45 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 23:43:04 UTC ---
the majority of fails are now:

warning: no debug symbols in executable (-arch xxxxx)

which is probably not good news .. 
(but possibly a dsymutil issue)...

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #46 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-14 23:44:20 UTC ---
a better prune....

Index: gcc/testsuite/lib/prune.exp
===================================================================
--- gcc/testsuite/lib/prune.exp    (revision 167813)
+++ gcc/testsuite/lib/prune.exp    (working copy)
@@ -56,6 +56,7 @@ proc prune_gcc_output { text } {
     regsub -all "(^|\n)\[^\n\]*ld: warning: can't add line info to anonymous
symbol\[^\n\]*" $text "" text
     regsub -all "(^|\n)\[^\n\]*warning:
DWARFDebugInfoEntry::AppendDependants\[^\n\]*AT_\[^\n\]*_bound\[^\n\]*FORM_ref4\[^\n\]*"
$text "" text
     regsub -all "(^|\n)\[^\n\]*warning:\[^\n\]*TAG_variable: 
AT_location\[^\n\]*didn't have valid function low pc\[^\n\]*" $text "" text
+    regsub -all "(^|\n)\[^\n\]*warning:\[^\n\]*no debug symbols in executable
.-arch .\[^\n8\]*86.\[^\n\]*" $text "" text

     #send_user "After:$text\n"

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #47 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-14 23:51:20 UTC ---
(In reply to comment #39)
> The patch I had in mind
> http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01129.html

With this patch applied on top of revision 167812 without any of the above
patches I have a lot of failures of the kind:

FAIL: gcc.dg/torture/stackalign/alloca-1.c  -O0  execution test

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #48 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-15 00:06:12 UTC ---
Uh, this is getting confusing.  We are really tracking 3 problems iteracting
with each other (and prevosly hidding each other)

1) darwin_text_section returns unlikely_text section that leads to infinite
recursion after the patch preventing darwin_function_section to partition
functions with -fno-partition-functions

This is preferably fixed by making darwin_function_section to return only the
default section, but we do have alternative fix adding unlikely function as
predefined section.

We also have alternative patch here that scratch the
USE_SELECT_SECTION_FOR_FUNCTIONS hack. That one should test independently with
only problem demonstrated by 2).

2) -freorder-blocks-and-partition does not imply -freorder-functions. This
means that -O0 we do not actual partition, since hot and cold section is forced
by be same by my change to function_section that triggered the PR

This demonstrate as the partition failure and should be solved by the hunk
enabling reorder functions when -freorder-blocks-and-partition is given.

I had typo in the original change I pasted here that it enabled
-freorder-blocks instead.  Just scratch that vairant.

3) the fact that we produce wrong assembly with 2 instead of missing
optimization actually uncovers bigger problem that we seem to screw up when
function partition is enabled but we do not really partition the given
function. In this case we do most of the code pesimization and worse yet, we
tend to mess up EH.

This should be fixed by the patch I sent to PR but for some reason it breaks
stackalign.  Do we have any idea why?

I guess we ought to be able to solve 1) and 2) in parallel and 3)
incrementally.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #49 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 00:28:04 UTC ---
(In reply to comment #48)
> Uh, this is getting confusing.  We are really tracking 3 problems iteracting
> with each other (and prevosly hidding each other)

indeed.

> 1) darwin_text_section returns unlikely_text section that leads to infinite
> recursion after the patch preventing darwin_function_section to partition
> functions with -fno-partition-functions
> 
> This is preferably fixed by making darwin_function_section to return only the
> default section, but we do have alternative fix adding unlikely function as
> predefined section.

well, I think we can consider that variant withdrawn - replaced by...

> We also have alternative patch here that scratch the
> USE_SELECT_SECTION_FOR_FUNCTIONS hack. That one should test independently with
> only problem demonstrated by 2).

... OK, at the moment I have it combined (I thought that was the suggestion @
comment 19 ;-))

> 2) -freorder-blocks-and-partition does not imply -freorder-functions. This
> means that -O0 we do not actual partition, since hot and cold section is forced
> by be same by my change to function_section that triggered the PR
> 
> This demonstrate as the partition failure and should be solved by the hunk
> enabling reorder functions when -freorder-blocks-and-partition is given.
> 
> I had typo in the original change I pasted here that it enabled
> -freorder-blocks instead.  Just scratch that vairant.

The patch @44 has the correct variant... 

> 3) the fact that we produce wrong assembly with 2 instead of missing
> optimization actually uncovers bigger problem that we seem to screw up when
> function partition is enabled but we do not really partition the given
> function. In this case we do most of the code pesimization and worse yet, we
> tend to mess up EH.

we get a lot of "warning: no debug symbols in executable" messages from
dsymutil -- that might indicate bad debug info -- or be a latent bug in
dsymutil (not clear at the moment).


> This should be fixed by the patch I sent to PR but for some reason it breaks
> stackalign.  Do we have any idea why?

I don't think it breaks stackalign on i686-darwin9.

> I guess we ought to be able to solve 1) and 2) in parallel and 3)
> incrementally.

I have no more time today --  tomorrow I'll remove the
USE_SELECT_SECTION_FOR_FUNCTIONS change.

===== result from c44  (modified as per 46) i686-darwin9

Running target unix/-m32
[for some reason the prune fails here... ]
FAIL: g++.dg/bprob/g++-bprob-2.C compilation,  -g  -fprofile-use
UNRESOLVED: g++.dg/bprob/g++-bprob-2.C execution,    -g  -fprofile-use
FAIL: g++.dg/bprob/g++-bprob-2.C compilation,  -O3 -g  -fprofile-use
UNRESOLVED: g++.dg/bprob/g++-bprob-2.C execution,    -O3 -g  -fprofile-use
 ++     warning: no debug symbols in executable (-arch i386) 

FAIL: g++.dg/debug/dwarf2/pubnames-1.C scan-assembler-times
"main.0"[^\\n]*external name 1

FAIL: g++.dg/other/pr22003.C (test for excess errors)

as per Comment #28/30:
FAIL: g++.dg/pch/system-1.C  -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 assembly comparison
FAIL: g++.dg/pch/system-2.C  -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 assembly comparison

+ PR46904

Running target unix/-m64
FAIL: g++.dg/debug/dwarf2/pubnames-1.C scan-assembler-times
"main.0"[^\\n]*external name 1

FAIL: g++.dg/other/pr22003.C (test for excess errors)

as per Comment #28/30
FAIL: g++.dg/pch/system-1.C  -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-1.C  -O2 assembly comparison
FAIL: g++.dg/pch/system-2.C  -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 -g assembly comparison
FAIL: g++.dg/pch/system-2.C  -O2 assembly comparison


+ PR46904

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #50 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 01:18:18 UTC ---
I noticed that the proposed patch at
http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01129.html also seems to have an
error here...

*************** machopic_select_section (tree decl,
*** 1276,1282 ****
                            node->only_called_at_startup,
                            node->only_called_at_exit);
      if (!base_section)
!           base_section = darwin_text_section (reloc, weak);
        }
        break;

--- 1271,1277 ----
                            node->only_called_at_startup,
                            node->only_called_at_exit);
      if (!base_section)
!           base_section = darwin_text_section (reloc);
        }
        break;


which I assume should have be a change to...

!           base_section = darwin_text_section (weak);

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #51 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-15 01:52:59 UTC ---
> which I assume should have be a change to...
> 
> !           base_section = darwin_text_section (weak);

Ah, yes. Sorry for the typo.  Does it make any difference?

Honza

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #52 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 03:14:56 UTC ---
Created attachment 22760
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22760
reduced patch

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #53 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 03:19:51 UTC ---
(In reply to comment #51)
> > which I assume should have be a change to...
> > 
> > !           base_section = darwin_text_section (weak);
> 
> Ah, yes. Sorry for the typo.  Does it make any difference?
> 
> Honza

Honza,
    This correction added to the attached reduced patch passes both...

make -k check RUNTESTFLAGS="tree-prof.exp --target_board=unix'{-m32,-m64}'"

and

make -k check RUNTESTFLAGS="stackalign.exp --target_board=unix'{-m32,-m64}'"

Running a complete regtest now but this patch still generates code with -g that
confuses dsymutil.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #54 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 10:22:27 UTC ---
(In reply to comment #49)

 As Dominique surmised, this is a different problem:

> as per Comment #28/30:
> FAIL: g++.dg/pch/system-1.C  -g assembly comparison
> FAIL: g++.dg/pch/system-1.C  -O2 -g assembly comparison
> FAIL: g++.dg/pch/system-1.C  -O2 assembly comparison
> FAIL: g++.dg/pch/system-2.C  -g assembly comparison
> FAIL: g++.dg/pch/system-2.C  -O2 -g assembly comparison
> FAIL: g++.dg/pch/system-2.C  -O2 assembly comparison

now PR46956

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #55 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 12:36:00 UTC ---
this solves 

FAIL: g++.dg/other/pr22003.C (test for excess errors)

but we have total failure with "-O3 -g" for gcc.
[all fails are warning: no debug symbols in executable (-arch xxxx)]
(looking into that next)

----- your patch + replacing the Darwin bits with this.

Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c (revision 167849)
+++ gcc/config/darwin.c (working copy)
@@ -1145,19 +1145,6 @@ darwin_mark_decl_preserved (const char *name)
 }

 static section *
-darwin_text_section (int reloc, int weak)
-{
-  if (reloc)
-    return (weak
-           ? darwin_sections[text_unlikely_coal_section]
-           : unlikely_text_section ());
-  else
-    return (weak
-           ? darwin_sections[text_coal_section]
-           : text_section);
-}
-
-static section *
 darwin_rodata_section (int weak, bool zsize)
 {
   return (weak
@@ -1267,17 +1254,7 @@ machopic_select_section (tree decl,
   switch (categorize_decl_for_section (decl, reloc))
     {
     case SECCAT_TEXT:
-      {
-       struct cgraph_node *node;
-       if (decl && TREE_CODE (decl) == FUNCTION_DECL
-           && (node = cgraph_get_node (decl)) != NULL)
-         base_section = darwin_function_section (decl,
-                                                 node->frequency,
-                                                 node->only_called_at_startup,
-                                                 node->only_called_at_exit);
-       if (!base_section)
-          base_section = darwin_text_section (reloc, weak);
-      }
+      gcc_unreachable ();
       break;

     case SECCAT_RODATA:
@@ -2559,17 +2536,6 @@ darwin_override_options (void)
   if (!global_options_set.x_dwarf_strict) 
     dwarf_strict = 1;

-  /* Disable -freorder-blocks-and-partition for darwin_emit_unwind_label.  */
-  if (flag_reorder_blocks_and_partition 
-      && (targetm.asm_out.emit_unwind_label == darwin_emit_unwind_label))
-    {
-      inform (input_location,
-              "-freorder-blocks-and-partition does not work with exceptions "
-              "on this architecture");
-      flag_reorder_blocks_and_partition = 0;
-      flag_reorder_blocks = 1;
-    }
-
   if (flag_mkernel || flag_apple_kext)
     {
       /* -mkernel implies -fapple-kext for C++ */
@@ -2969,32 +2935,42 @@ darwin_function_section (tree decl, enum node_freq
                          bool startup, bool exit)
 {
   if (!flag_reorder_functions)
-    return NULL;
+    return (DECL_WEAK (decl))
+           ? darwin_sections[text_coal_section]
+           : text_section;
+
   /* Startup code should go to startup subsection unless it is
      unlikely executed (this happens especially with function splitting
      where we can split away unnecesary parts of static constructors.  */
   if (startup && freq != NODE_FREQUENCY_UNLIKELY_EXECUTED)
-    return get_named_text_section
-            (decl, "__TEXT,__startup,regular,pure_instructions", "_startup");
+    return (DECL_WEAK (decl))
+           ? darwin_sections[text_startup_coal_section]
+           : darwin_sections[text_startup_section];

   /* Similarly for exit.  */
   if (exit && freq != NODE_FREQUENCY_UNLIKELY_EXECUTED)
-    return get_named_text_section (decl,
-                                  "__TEXT,__exit,regular,pure_instructions",
-                                  "_exit");
+    return (DECL_WEAK (decl))
+           ? darwin_sections[text_exit_coal_section]
+           : darwin_sections[text_exit_section];

   /* Group cold functions together, similarly for hot code.  */
   switch (freq)
     {
       case NODE_FREQUENCY_UNLIKELY_EXECUTED:
-       return get_named_text_section
-                (decl,
-                 "__TEXT,__unlikely,regular,pure_instructions", "_unlikely");
+       return (DECL_WEAK (decl))
+               ? darwin_sections[text_cold_coal_section]
+               : darwin_sections[text_cold_section];
+       break;
       case NODE_FREQUENCY_HOT:
-       return get_named_text_section
-                (decl, "__TEXT,__hot,regular,pure_instructions", "_hot");
+       return (DECL_WEAK (decl))
+               ? darwin_sections[text_hot_coal_section]
+               : darwin_sections[text_hot_section];
+       break;
       default:
-       return NULL;
+       return (DECL_WEAK (decl))
+               ? darwin_sections[text_coal_section]
+               : text_section;
+       break;
     }
 }

Index: gcc/config/darwin-sections.def
===================================================================
--- gcc/config/darwin-sections.def      (revision 167849)
+++ gcc/config/darwin-sections.def      (working copy)
@@ -34,6 +34,24 @@ DEF_SECTION (text_unlikely_coal_section, SECTION_C
             ".section __TEXT,__text_unlikely_coal,"
             "coalesced,pure_instructions", 0)

+DEF_SECTION (text_hot_section, SECTION_CODE,
+            ".section __TEXT,__text_hot,regular,pure_instructions", 0)
+DEF_SECTION (text_cold_section, SECTION_CODE,
+            ".section __TEXT,__text_cold,regular,pure_instructions", 0)
+DEF_SECTION (text_startup_section, SECTION_CODE,
+            ".section __TEXT,__text_startup,regular,pure_instructions", 0)
+DEF_SECTION (text_exit_section, SECTION_CODE,
+            ".section __TEXT,__text_exit,regular,pure_instructions", 0)
+
+DEF_SECTION (text_hot_coal_section, SECTION_CODE,
+            ".section __TEXT,__text_hot_coal,coalesced,pure_instructions", 0)
+DEF_SECTION (text_cold_coal_section, SECTION_CODE,
+            ".section __TEXT,__text_cold_coal,coalesced,pure_instructions", 0)
+DEF_SECTION (text_startup_coal_section, SECTION_CODE,
+            ".section __TEXT,__text_stt_coal,coalesced,pure_instructions", 0)
+DEF_SECTION (text_exit_coal_section, SECTION_CODE,
+            ".section __TEXT,__text_exit_coal,coalesced,pure_instructions", 0)
+
 /* const */
 DEF_SECTION (const_section, 0, ".const", 0)
 DEF_SECTION (const_coal_section, SECTION_NO_ANCHOR,
Index: gcc/config/darwin.h
===================================================================
--- gcc/config/darwin.h (revision 167849)
+++ gcc/config/darwin.h (working copy)
@@ -664,7 +664,7 @@ extern GTY(()) section * darwin_sections[NUM_DARWI

 #undef TARGET_ASM_SELECT_SECTION
 #define TARGET_ASM_SELECT_SECTION machopic_select_section
-#define USE_SELECT_SECTION_FOR_FUNCTIONS
+
 #undef TARGET_ASM_FUNCTION_SECTION
 #define TARGET_ASM_FUNCTION_SECTION darwin_function_section

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #56 from Jan Hubicka <hubicka at gcc dot gnu.org> 2010-12-15 13:10:15 UTC ---
Hi,
thanks for testing.  I wonder why we need to add those +DEF_SECTION fields?
I think the infinite recursion problem should be fixed by dropping the whole
machinery on selecting sections.

With the darwin bits alone, do we get clear run except for the two partition
tests?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #57 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 13:25:07 UTC ---
(In reply to comment #56)
>   I wonder why we need to add those +DEF_SECTION fields?
> I think the infinite recursion problem should be fixed by dropping the whole
> machinery on selecting sections.

nothing to do with the recursion - that's solved early in this thread --- 

-- we need a coalesced section to match each non-coalesced one.
since "darwin_function_section()" is called quite frequently, it seemed more
efficient to pre-declare the sections rather than keep finding them from
matching against a named section.

[ the key change is that there need to be coalesced sections for the DECL_WEAK
() -- I am not strongly saying the sections need to be pre-declared - if you
think that named sections would be better]

we would need :

   if (exit && freq != NODE_FREQUENCY_UNLIKELY_EXECUTED)
    return (DECL_WEAK (decl))
                  ? get_named_text_section (decl,
                                                                
"__TEXT,__exit_coal,coalesced,pure_instructions", "_exit_coal");
                  : get_named_text_section (decl,
                                                                
"__TEXT,__exit,regular,pure_instructions", "_exit");

.. or some more efficient equivalent.

> With the darwin bits alone, do we get clear run except for the two partition
> tests?

tests are running ... looking OK so far,
... but we have a serious problem with dsymutil && || debug for "-O3 -g".

Still needs examining as to why...

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #58 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-15 13:33:15 UTC ---
> [ the key change is that there need to be coalesced sections for the DECL_WEAK
> () -- I am not strongly saying the sections need to be pre-declared - if you
> think that named sections would be better]

No, this seems fine.  I was just curious.
> 
> tests are running ... looking OK so far,
> ... but we have a serious problem with dsymutil && || debug for "-O3 -g".

Hmm, so it is independent of my bb-reorder patch...

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22758|0                           |1
        is obsolete|                            |

--- Comment #59 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 13:41:40 UTC ---
Created attachment 22764
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22764
Honza+Darwin mods

Honza, 
sorry, I think I misled you ... 

... I have your patch applied + the changes to Darwin I posted.

If you think I should test with just the darwin bits, OK I can do that.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #60 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-15 13:49:51 UTC ---
Hi,
yes please try with the darwin bits alone and the hunk in opts.c enabling
function reordering when partitioning is on.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #61 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 13:56:07 UTC ---
(In reply to comment #59)
> Created attachment 22764 [details]
> Honza+Darwin mods

> If you think I should test with just the darwin bits, OK I can do that.

with *only* the darwin patch (comment #55) -- the debug problem is _not_
present.

----

With the total patch  [bb + darwin, comment #59]

+ every "-O3 -g" case fails 

+ gcc:

FAIL: gcc.dg/darwin-weakimport-3.c scan-assembler-not coalesced
FAIL: gcc.dg/pr25376.c scan-assembler my_named_section

g++:
+ FAIL: g++.dg/debug/dwarf2/pubnames-1.C scan-assembler-times
"main.0"[^\\n]*external name 1

[not tested yet - fortran/ObjC*/Java]

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #62 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 14:00:40 UTC ---
(In reply to comment #61)
> (In reply to comment #59)
> > Created attachment 22764 [details] [details]
> > Honza+Darwin mods
> 
> > If you think I should test with just the darwin bits, OK I can do that.
> 
> with *only* the darwin patch (comment #55) -- the debug problem is _not_
> present.

adding the hunk for opts.c does _not_ appear to cause the failure either [I'll
rebuild & confirm].

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #63 from Jan Hubicka <hubicka at ucw dot cz> 2010-12-15 14:07:16 UTC ---
OK,
if the darwin changes + opts.c change solves the problem, lets submit the patch
and I will try to re-review the bb-reorder patch and we can deal with the
problem
incrementally. Thanks!

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #64 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 14:34:03 UTC ---
Iain,
    My old radar bug report numbered 7289379, "linker warnings when no unwind
labels are emitted while targeting 10.6", my be have some useful information.
The report was...

08-Oct-2009 11:17 PM Jack Howarth:
When the following patch is used in FSF gcc trunk to leverage the fact that
unwind labels are no longer required when targeting 10.6...


2009-10-08  Jack Howarth  <howarth@bromo.med.uc.edu>

        PR c++/41313
        * gcc/config/darwin.c: Suppress unwind labels when targeting 10.6 or
later.
        Disable -freorder-blocks-and-partition for exception handling and when
        the target requested unwind info. 


Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c (revision 152583)
+++ gcc/config/darwin.c (working copy)
@@ -1454,7 +1454,8 @@
 {
   char *lab;

-  if (! for_eh)
+  /* Suppress unwind labels when targeting 10.6 or later.  */ 
+  if ((! for_eh) || (darwin_macosx_version_min &&
strverscmp(darwin_macosx_version_min, "10.6") >= 0))
     return;

   lab = concat (IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (decl)), ".eh", NULL);
@@ -1697,6 +1698,16 @@
   if (dwarf_strict < 0) 
     dwarf_strict = 1;

+  /* Disable -freorder-blocks-and-partition for exception handling or when
+     the target requested unwind info.  */
+  if (flag_reorder_blocks_and_partition && (flag_exceptions ||
flag_unwind_tables))
+    {
+      inform (input_location,
+              "-freorder-blocks-and-partition does not work with exceptions on
this architecture");
+      flag_reorder_blocks_and_partition = 0;
+      flag_reorder_blocks = 1;
+    }
+
   if (flag_mkernel || flag_apple_kext)
     {
       /* -mkernel implies -fapple-kext for C++ */

Two FSF gcc fail their compilation excessive errors test due to warnings of the
form...

ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
output is:
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o
ld: warning: can't add line info to anonymous symbol anon-func-0xF40 from
/var/tmp//ccrw73YL.o

The resulting binaries run fine however which suggest these warnings are
non-fatal. They should be suppressed if the warnings are incorrectly being
emitted by the compiler. I have attached an archive,
no_unwind_label_link_bug.tar.bz2, which contains both failing test cases...

FAIL: g++.dg/tree-prof/partition1.C compilation,  -g  -fprofile-use
UNRESOLVED: g++.dg/tree-prof/partition1.C execution,    -g  -fprofile-use

FAIL: g++.dg/tree-prof/partition1.C compilation,  -O3 -g  -fprofile-use
UNRESOLVED: g++.dg/tree-prof/partition1.C execution,    -O3 -g  -fprofile-use

as case1 and case2 subdirectories respectively. A link_it shell script is
provided in each to allow the linker warning to be reproduced. The common
feature in the failing test cases appear to be the addition of the -g
flag. Other similar partition1.C testcase runs which don't use -g do not
produce these unexpected linker warnings.


'no_unwind_label_link_bug.tar.bz2' was successfully uploaded

08-Oct-2009 11:43 PM Jack Howarth:
Minor correction, the actual patch applied to FSF gcc trunk was...

Index: gcc/config/darwin.c
===================================================================
--- gcc/config/darwin.c    (revision 152583)
+++ gcc/config/darwin.c    (working copy)
@@ -1454,7 +1454,8 @@
 {
   char *lab;

-  if (! for_eh)
+  /* Suppress unwind labels when targeting 10.6 or later.  */
+  if ((! for_eh) || (darwin_macosx_version_min &&
strverscmp(darwin_macosx_version_min, "10.6") >= 0))
     return;

   lab = concat (IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (decl)), ".eh", NULL);
@@ -1697,6 +1698,18 @@
   if (dwarf_strict < 0) 
     dwarf_strict = 1;

+  /* Disable -freorder-blocks-and-partition for exception handling or when
+     the target requested unwind info when targeting earlier than 10.6.  */
+  if (flag_reorder_blocks_and_partition 
+      && (darwin_macosx_version_min && strverscmp(darwin_macosx_version_min,
"10.6") < 0)
+          && (flag_exceptions || flag_unwind_tables))
+    {
+      inform (input_location,
+              "-freorder-blocks-and-partition does not work with exceptions on
this architecture");
+      flag_reorder_blocks_and_partition = 0;
+      flag_reorder_blocks = 1;
+    }
+
   if (flag_mkernel || flag_apple_kext)
     {
       /* -mkernel implies -fapple-kext for C++ */


16-Nov-2009 03:17 PM Jack Howarth:
See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42070#c1.

and Apple's response to the uploaded test case was...

11-Nov-2009 06:25 PM YVONNE  VILLA  :
Engineering has provided the following information:

In the sample you supplied, the warning is because there is code in the
__TEXT/__unlikely section and there is dwarf debug information that says it
part of a function.  The linker sanity checks how it broke up the .o file into
"atoms"  by checking it against the dwarf debug info.  

I think the __unlikely section should be avoided on darwin until it is shown to
work with the linker.  At a minimum each chunk in the __unlikely section should
have a label on it with a name based on the function it came from.  Those
labels would also help debugging.

This warning does not have anything to do removing labels from __eh_frame
section.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #65 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 15:15:42 UTC ---
Iain, 
   Can you try labeling the unlikely sections with the function names as Apple
suggested and see if that helps with the problems with -g using honza patch?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #66 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 15:17:29 UTC ---
(In reply to comment #64)
Thanks Jack, I'll take a look at this info, if necessary. 

a couple of points;

this below has subsequently been replaced with an #undef of the target hook in
config/darwin10.h

> -  if (! for_eh)
> +  /* Suppress unwind labels when targeting 10.6 or later.  */ 
> +  if ((! for_eh) || (darwin_macosx_version_min &&
> strverscmp(darwin_macosx_version_min, "10.6") >= 0))
>      return;

this below is no longer properly functional with the new arrangement of options
processing.
(the test is now done in opts.c - if you leave this in you get a false warning
when -fno-exceptions is given on a g++ command line)

> +  /* Disable -freorder-blocks-and-partition for exception handling or when
> +     the target requested unwind info.  */
> +  if (flag_reorder_blocks_and_partition && (flag_exceptions ||
> flag_unwind_tables))
> +    {
> +      inform (input_location,
> +              "-freorder-blocks-and-partition does not work with exceptions on
> this architecture");
> +      flag_reorder_blocks_and_partition = 0;
> +      flag_reorder_blocks = 1;
> +    }

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #67 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 17:11:28 UTC ---
Created attachment 22768
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22768
darwin-function-section-patch

OK - sorted out a couple of minor glitches...
...I'm going to reg-test the attached more widely (powerpc + other languages).

this does not suffer from the debug issue... but...

Honza, two points:

1a / opts.c - this should probably be placed _after_ all the tests that might
set x_flag_reorder_blocks_and_partition to 0.

1b / it should be conditional on the User _not_ having set
x_flag_reorder_functions ;)

something like --- (in the attached patch too)
  if (opts->x_flag_reorder_blocks_and_partition
      && !opts_set->x_flag_reorder_functions)
    opts->x_flag_reorder_functions = 1;

===

2.  The patch attached actually clears 46904 as well, ---- BUT there are
several new fails

FAIL: partition{1,2}.C   * -fprofile-use
FAIL: gcc.dg/tree-prof/bb-reorg.c compilation,  -fprofile-use -D_PROFILE_USE
FAIL: gcc.dg/tree-prof/pr34999.c compilation,  -fprofile-use -D_PROFILE_USE

these are 
FATAL:Symbol _foo.eh already defined.

can you separate the portion of your patch that would resolve this - without
introducing the elements we think are responsible for the debug problem?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #68 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-15 21:29:07 UTC ---
> With the total patch  [bb + darwin, comment #59]
> ...
> [not tested yet - fortran/ObjC*/Java]

With this patch I see ~78 new failures with -m32 -O3 -g of the kind "warning:
no debug symbols in executable (-arch i386)". See also

http://gcc.gnu.org/ml/gcc-testresults/2010-12/msg01287.html

I also confirm the lot of failures for libstdc++ tests.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22768|0                           |1
        is obsolete|                            |

--- Comment #69 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 21:37:22 UTC ---
Created attachment 22777
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22777
updated darwin candidate function sect. patch

OK, I've added a hack that (hopefully) gets around the multiple .eh symbol
issues.

>From my testing so far, there do not seem to be any "no symbols ... " warnings
generated .. 

(there are some dsymutil warnings in the torture suite, because that does not
use the prune facility AFAICT)

This is a candidate for solving PR46916 + PR46904.

I will test on *-darwin9 .. 
... would be interested to see how it behaves on x86_64-d10.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #70 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 21:52:05 UTC ---
Has anyone tried running dwarfdump on the object files from one of these
failing -g testcases? If dwarfdump works, perhaps a comparison of the output
compare with the same testcase compiled with current trunk might help clarify
the issue. If we are sure that the dwarf looks valid, I can run one of these
offending binaries past the dsymutil maintainer.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #71 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 22:02:39 UTC ---
Actually I would be surprised if the new patches don't have problems with eh
under darwin9. See...

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41313#c12

As I recall, the eh labels in darwin9 and earlier will cause duplicate symbols
for hot/cold partitioning. We left those releases broken (with
-freorder-blocks-and-partition disabled when darwin_emit_unwind_label is in
use) because Mike thought Jakub's solution would be very difficult or
impossible to implement. By using the new linker support in darwin10 to omit
the eh labels, we were able to enable -freorder-blocks-and-partition with eh on
darwin10.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #72 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-15 23:35:34 UTC ---
(In reply to comment #71)
> Actually I would be surprised if the new patches don't have problems with eh
> under darwin9. See...

Yeah, the symbol error is gone.. but
...  the linker warning remains.
ld warning: can't add line info to anonymous symbol __text_cold@0 from
/var/folders/OW/OW-PGOtgHbKakssxFpJpkU++-0E/-Tmp-//cck7RZLJ.o

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41313#c12

OK, will check that out.

>  We left those releases broken (with
> -freorder-blocks-and-partition disabled when darwin_emit_unwind_label is in
> use) because Mike thought Jakub's solution would be very difficult or
> impossible to implement. By using the new linker support in darwin10 to omit
> the eh labels, we were able to enable -freorder-blocks-and-partition with eh on
> darwin10.

well, the current way of disabling needs updating for the new options scheme ; 
have to look at that tomorrow .. 

However, the patch should still be considered as a fix for PR46916/46904.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #73 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-15 23:44:30 UTC ---
   RE:

+  /* FIXME: We should not be trying to output the same label for different
+     partitions of a function.  */

Actually this may be implementing exactly what Jakub suggested in Comment 12 of
PR41313
and thus isn't a FIXME at all. Darwin is the only target that uses eh labels so
this issue of
duplicate symbols for the eh labels from hot and cold partitions never came up
outside of darwin.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #74 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 00:17:20 UTC ---
Iain,
     Why don't you see the errors...

../../gcc/gcc/config/darwin.c: In function 'darwin_emit_unwind_label':
../../gcc/gcc/config/darwin.c:1664:21: error: initialization discards 'const'
qualifier from pointer target type [-Werror]
../../gcc/gcc/config/darwin.c:1666:10: error: type defaults to 'int' in
declaration of 'invok_count' [-Werror=implicit-int]
../../gcc/gcc/config/darwin.c:1686:8: error: assignment discards 'const'
qualifier from pointer target type [-Werror]
../../gcc/gcc/config/darwin.c:1690:9: error: assignment discards 'const'
qualifier from pointer target type [-Werror]

with your proposed patch from Comment 67?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #75 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 04:58:53 UTC ---
Created attachment 22780
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22780
fixed version of updated darwin candidate function sect. patch

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #76 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 05:01:08 UTC ---
The fixed version of Iain's patch works around the errors in Comment 74 on
x86_64-apple-darwin10.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Jack Howarth <howarth at nitro dot med.uc.edu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22780|0                           |1
        is obsolete|                            |

--- Comment #77 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 05:32:41 UTC ---
Created attachment 22781
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22781
fixed version of updated darwin candidate function sect. patch

Properly place the initialization of ext.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Jack Howarth <howarth at nitro dot med.uc.edu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22781|0                           |1
        is obsolete|                            |

--- Comment #78 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 06:26:13 UTC ---
Created attachment 22782
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22782
fixed version of updated darwin candidate function sect. patch

Made ext static to avoid...

In file included from ../../gcc-4.6-20101215/gcc/config/darwin.c:3015:0:
../../gcc-4.6-20101215/gcc/config/darwin.c: In function 'void
darwin_emit_unwind_label(FILE*, tree, int, int)':
../../gcc-4.6-20101215/gcc/config/darwin.c:1669:20: error: 'ext' is used
uninitialized in this function [-Werror=uninitialized]
cc1plus: all warnings being treated as errors

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #79 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 16:12:23 UTC ---
With the patch from Comment 78, I don't see dsymutil failues (at least for a
simple c language bootstrap on x86_64-apple-darwin10). However at -m32,  I am
now seeing libgomp testsuite execution failures of the form...

libgomp: Thread creation failed: Resource temporarily unavailable
FAIL: libgomp.c/appendix-a/a.15.1.c execution test

Oddly these don't occur at -m64. Any idea what change you made to the current
version of the patch which suppressed the dsymutil failures?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #80 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-16 16:25:47 UTC ---
(In reply to comment #79)
> With the patch from Comment 78, I don't see dsymutil failues (at least for a

sorry about the typos ...
..  I have a new patch which deals with the disabling of partitioning for
d9/exceptions.
(later, I'm doing a bit more testing first).

> libgomp: Thread creation failed: Resource temporarily unavailable
> FAIL: libgomp.c/appendix-a/a.15.1.c execution test

characteristic of running out of threads .. 
... is it repeatable ?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #81 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 17:07:37 UTC ---
(In reply to comment #80)

> characteristic of running out of threads .. 
> ... is it repeatable ?

Yes and it happens for a lot of tests in libgomp at -m32...

FAIL: libgomp.c/appendix-a/a.15.1.c execution test
FAIL: libgomp.c/appendix-a/a.16.1.c execution test
FAIL: libgomp.c/appendix-a/a.21.1.c execution test
FAIL: libgomp.c/appendix-a/a.26.1.c execution test
FAIL: libgomp.c/appendix-a/a.29.1.c execution test
FAIL: libgomp.c/appendix-a/a.36.1.c execution test
FAIL: libgomp.c/appendix-a/a.39.1.c execution test
FAIL: libgomp.c/appendix-a/a.4.1.c execution test
etc

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #82 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-16 17:33:30 UTC ---
(In reply to comment #80)
> Yes and it happens for a lot of tests in libgomp at -m32...

Confirmed here!-(so far the best fix for this pr is to revert 167727) Is there
a way to disable the new 'if'

+  if (!flag_reorder_functions
+      || !targetm.have_named_sections)
+    return NULL;

for darwin only?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #83 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-16 17:40:35 UTC ---
(In reply to comment #82)
> (In reply to comment #80)
> > Yes and it happens for a lot of tests in libgomp at -m32...
> 
> Confirmed here!-(so far the best fix for this pr is to revert 167727) Is there
> a way to disable the new 'if'
> 
> +  if (!flag_reorder_functions
> +      || !targetm.have_named_sections)
> +    return NULL;
> 
> for darwin only?

hang on guys ;) ... 
.. the patch I have in progress  [which is different from the one of yesterday
evening]

(a) resolves this and PR46904 
(b) has no libgomp issues against trunk 167948 

(on *-d9) .

-- I will check it on D10 .. before re-posting.
.. if you want to try in advance I can copy it to you by email.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #84 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-16 18:04:30 UTC ---
> .. if you want to try in advance I can copy it to you by email.

yes, please!

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22760|0                           |1
        is obsolete|                            |
  Attachment #22764|0                           |1
        is obsolete|                            |
  Attachment #22777|0                           |1
        is obsolete|                            |
  Attachment #22782|0                           |1
        is obsolete|                            |

--- Comment #85 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-16 20:33:59 UTC ---
Created attachment 22787
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22787
updated #2 darwin candidate function sect. patch


I've bootstrapped this patch on *-darwin9 and x86_64-darwin10... 
... there are no libgomp issues.

remaining points:

   o as of XCode 3.2.4 the "warning: cannot add symbol ... " is NOT fixed.

   The solution here is to make sure that each new block has a non-local
starting label  - it does not need to be external -- but it must be
non-anonymous (i.e. NOT begin with L)  ..... [ Honza any ideas on how we might
achieve that ???]

  o  I think that the "no debug symbols" is down to missing pub sections....
investigating.

  o  Darwin 9's linker appears to cater for generation of the .eh .. 

  o .. so at the moment Darwin10 fails on partition2.C ... for the reasons
above.

I will finish reg-tests and then post to gcc-patches -- unless there are more
matters arising?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #86 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-16 20:41:20 UTC ---
What is the reasoning behind emitting eh labels again on darwin10? In theory,
this makes darwin 'less strange' if we can avoid emitting eh labels as no other
target does.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #87 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-16 20:51:45 UTC ---
(In reply to comment #86)
> What is the reasoning behind emitting eh labels again on darwin10? In theory,
> this makes darwin 'less strange' if we can avoid emitting eh labels as no other
> target does.

NOT emitting them --- if you check the patch:

  /* We use the linker to emit the .eh labels for Darwin 9 and above.  */
  if (! for_eh || generating_for_darwin_version >= 9)
    return;


----

the fix you had before was wrong in the sense that it would prevent you from
emitting correct asm with -mmacosx-version-min=10.x (x<6).

now it will emit the .eh when the target OS is <9.

Iain

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #88 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-17 13:50:40 UTC ---
(In reply to comment #85)
> Created attachment 22787 [details]
> updated #2 darwin candidate function sect. patch
> 
> 
>   o  I think that the "no debug symbols" is down to missing pub sections....
> investigating.
> 
...
> 
>   o .. so at the moment Darwin10 fails on partition2.C ... for the reasons
> above.
> 

Iain,
   Do you think the "no debug symbols" warnings in the partition2.C test case
on darwin10 are the caused by the same issue (lack of pub symbols) as those
additional instances caused by honza'a patch? Also, do we see any additional
failures beyond those warnings when honza'a patch is added onto this current
darwin patch?

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #89 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-17 14:08:12 UTC ---
(In reply to comment #88)
> Iain,
>    Do you think the "no debug symbols" warnings in the partition2.C test case
> on darwin10 are the caused by the same issue (lack of pub symbols) as those
> additional instances caused by honza'a patch? Also, do we see any additional
> failures beyond those warnings when honza'a patch is added onto this current
> darwin patch?

I think you should analyse comment #42. If I did not do any mistake, the
partition2.C test is unsupported with current trunk, but becomes supported with
any of the patches in this PR. If this is correct, it means that these patches
change something in the testsuite machinery (e.g., a failing test becomes
successful).
In this case one should find what is this change, then assess if the test has
to be supported or not. For the former case, the failure will have to be fixed,
but in the later case only some test(s) in some *.exp file will have to be
fixed.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #90 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-17 14:21:00 UTC ---
(In reply to comment #89)
> (In reply to comment #88)
> > Iain,
> >    Do you think the "no debug symbols" warnings in the partition2.C test case
> > on darwin10 are the caused by the same issue (lack of pub symbols) as those
> > additional instances caused by honza'a patch? Also, do we see any additional
> > failures beyond those warnings when honza'a patch is added onto this current
> > darwin patch?
> 
> I think you should analyse comment #42. If I did not do any mistake, the
> partition2.C test is unsupported with current trunk, but becomes supported with
> any of the patches in this PR. If this is correct, it means that these patches
> change something in the testsuite machinery (e.g., a failing test becomes
> successful).

well, this change is the likely difference:

+  if (opts->x_flag_reorder_blocks_and_partition
+      && !opts_set->x_flag_reorder_functions)
+    opts->x_flag_reorder_functions = 1;

as far as whether Darwin10 supports -freorder-blocks-and-partition
- there should be no difference before/after the patch (darwin < 10 should not
support at present).

It would be interesting to know why the test claims to be 'unsupported' for
current trunk (except that it would fail because of this bug).

// { dg-require-effective-target freorder }
// { dg-options "-fnon-call-exceptions -freorder-blocks-and-partition" }

you would have to back out of the patches and the change that causes the fault
.. if the test still claims to be unsupported - that implies something
different (again, not related to the current bug).

In any event the two remaining issues need resolution -- I suspect they are
causing fallout elsewhere.
... I just don't see the point in swelling this patch to solve those .. 

the intention of this patch is to get correct function section selection for
Darwin.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #91 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-17 14:31:18 UTC ---
(In reply to comment #90)
> ... In any event the two remaining issues need resolution -- I suspect they are
> causing fallout elsewhere.
> ... I just don't see the point in swelling this patch to solve those .. 

I agree that it is likely an other issue for which a new PR should be opened
and that this PR has now grown out of proportion ans should be closed ASAP (BTW
x86_64-darwin10, full test, and powerpc-apple-darwin9, c++ and libstdc++-v3,
regtested fine).

> the intention of this patch is to get correct function section selection for
> Darwin.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #92 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-17 14:47:01 UTC ---
  Don't forget about PR45646 where we used to fail
g++.dg/tree-prof/partition2.C with a linker error of...

ld: in /var/folders/1C/1CdoNxmNFHyOIjNBLNuJh++++TM/-Tmp-//ccVrX7dM.o, section
not found for address 0x696765625F617863

This disappeared between r166079 and r166156 for unknown reasons. The bug may
have went latent and could be related.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #93 from Jack Howarth <howarth at nitro dot med.uc.edu> 2010-12-17 14:59:30 UTC ---
(In reply to comment #90)

> as far as whether Darwin10 supports -freorder-blocks-and-partition
> - there should be no difference before/after the patch (darwin < 10 should not
> support at present).
> 

Are you sure that we still need to disable -freorder-blocks-and-partition on
darwin <10? I thought the only reason we did that was because of the duplicate
eh label symbols emitted from darwin_emit_unwind_label. Now that you have those
properly renamed for the hot and cold partitions, this issue should be solved.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #94 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-17 15:03:38 UTC ---
(In reply to comment #93)
> (In reply to comment #90)
> 
> > as far as whether Darwin10 supports -freorder-blocks-and-partition
> > - there should be no difference before/after the patch (darwin < 10 should not
> > support at present).
> > 
> 
> Are you sure that we still need to disable -freorder-blocks-and-partition on
> darwin <10? I thought the only reason we did that was because of the duplicate
> eh label symbols emitted from darwin_emit_unwind_label. Now that you have those
> properly renamed for the hot and cold partitions, this issue should be solved.

I decided to maintain the status quo (modulo external changes).  

-- Some partitioning will fail on D9 for the same reasons as D10
-- once we've sorted out the other two problems 
    (I agree with Dominique, they should be two additional separate PRs)
-- then it is a one-line patch to enable for D9.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #95 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-17 22:07:55 UTC ---
For the reasons given below, I have reached the conclusion that the failures
for g++.dg/tree-prof/partition2.C are not caused by the above patches, but
exposed by any patch fixing this pr.

Looking at gcc/testsuite/lib/target-supports.exp, I see

proc check_effective_target_freorder {} {
    return [check_no_compiler_messages freorder object {
        void foo (void) { }
    } "-freorder-blocks-and-partition"]
}

Then I did the following tests:

[macbook] f90/bug% cat > order.c
void foo (void) { }
[macbook] f90/bug% /opt/gcc/gcc4.6p/bin/gcc -c -freorder-blocks-and-partition
order.c
gcc: internal compiler error: Segmentation fault (program cc1)

The backtrace

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00007fff5bc00ffc
htab_find_slot_with_hash (htab=0x142420d90, element=0x7fff5bc01050, hash=2086,
insert=INSERT) at ../../p_work/libiberty/hashtab.c:644
644    {
(gdb) bt
#0  htab_find_slot_with_hash (htab=0x142420d90, element=0x7fff5bc01050,
hash=2086, insert=INSERT) at ../../p_work/libiberty/hashtab.c:644
#1  0x00000001007eb558 in cgraph_node (decl=0x142591200) at
../../p_work/gcc/cgraph.c:502

is exactly the one reported in comment #0. So if the segmentation fault is
fixed, it is also fixed for the freorder test and the tests that were
UNSUPPORTED are then executed.

So the partition2.C failures are due to another bug of the generation of
debugging information on darwin. If nobody is faster, I'll open a new pr for it
when this pr will have been fixed. For the record note that the failure for -O3
-g is slightly different from the one reported in comment #12:

ld: warning: can't add line info to anonymous symbol anon-func-0x0 from
/var/folders/LW/LW1oufkMGIqlLpjYn45fBU+++TI/-Tmp-//cc5PMZA6.o
ld: warning: can't add line info to anonymous symbol anon-func-0x0 from
/var/folders/LW/LW1oufkMGIqlLpjYn45fBU+++TI/-Tmp-//cc5PMZA6.o
warning: no debug symbols in executable (-arch i386)

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #96 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-17 23:56:16 UTC ---
(In reply to comment #95)
> For the reasons given below, I have reached the conclusion that the failures
> for g++.dg/tree-prof/partition2.C are not caused by the above patches, but
> exposed by any patch fixing this pr.

In which case, I propose that we close this PR with 

http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01359.html  (still needs review, I
think)

and deal with "anonymous code atoms" and "missing pubnames sections" as two
different PRs.

thanks, Iain

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #97 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2010-12-20 09:05:05 UTC ---
The patch in attachment 22787 fixes also a "Stage 3 ada bootstrap error on
i686-apple-darwin9", see comment #3 of pr46950.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #98 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-20 10:01:07 UTC ---
Author: iains
Date: Mon Dec 20 10:01:02 2010
New Revision: 168083

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=168083
Log:

    PR c++/46904
    PR middle-end/46916
    PR target/46950
    * config/darwin.c (generating_for_darwin_version): New var.
    (darwin_text_section): Remove.
    (darwin_rodata_section): Do not check function section here.
    (darwin_emit_unwind_label): Do not emit for Darwin >= 9.
    Emit distinct labels for successive calls for the same decl.
    (darwin_override_options): Set generating_for_darwin_version.
    Add FIXME and disable -freorder-blocks-and-partition.
    Suppress automatic asynchronous unwind tables for m32.
    Switch off -freorder-blocks-and-partition when unwind tables
    will be generated.  Update to use generating_for_darwin_version.
    (darwin_function_section): Check for cases that need to be placed
    in coalesced sections.
    * config/darwin-sections.def: Define hot, cold, startup and exit sections
    for both coalesced and regular code.
    * config/darwin.h (USE_SELECT_SECTION_FOR_FUNCTIONS): Delete.
    * config/darwin10.h (TARGET_ASM_EMIT_UNWIND_LABEL): Delete.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/darwin-sections.def
    trunk/gcc/config/darwin.c
    trunk/gcc/config/darwin.h
    trunk/gcc/config/darwin10.h

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #99 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-20 10:14:39 UTC ---
Author: iains
Date: Mon Dec 20 10:14:36 2010
New Revision: 168084

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=168084
Log:

    PR middle-end/46916
    * opts.c (finish_options): Enable -freorder-functions
    when -freorder-blocks-and-partition is active.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/opts.c

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

--- Comment #100 from Iain Sandoe <iains at gcc dot gnu.org> 2010-12-20 11:28:44 UTC ---
(In reply to comment #39)
> The patch I had in mind
> http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01129.html

BTW: When the BB reordering is applied
 (together with http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01553.html and
http://gcc.gnu.org/ml/gcc-patches/2010-12/msg01554.html)

I see the code trashing reduced -- but also that there are no partition
switches created for either partition{1,2}.C on i686-darwin9.

Perhaps that is intended -- but it's a 'heads up' that the test-cases might
need adjustment to force section switches, if that is the intention.

I think this PR should be fixed now.

[Bug middle-end/46916] gcc.dg/torture/stackalign/non-local-goto-[1,2].c ICEs compiler due to r167727

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46916

Jack Howarth <howarth at nitro dot med.uc.edu> changed:

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

--- Comment #101 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-01-09 16:06:52 UTC ---
Fixed in current gcc trunk.