[Bug gcov/profile/24069] New: ice during feedback stage of profiledbootstrap on powerpc

[Bug gcov/profile/24069] New: ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1031 bytes --]

Ran into an ice today while attempting a profiled bootstrap of mainline:

/> gcc -c -O2 -fprofile-use -freorder-blocks-and-partition dwarf2out.i
dwarf2out.c: In function âoutput_call_frame_infoâ:
dwarf2out.c:2410: internal compiler error: in make_edges, at cfgbuild.c:350
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

I'll attach .i, .gcda, and .gcno files

-- 
           Summary: ice during feedback stage of profiledbootstrap on
                    powerpc
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gcov/profile
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: steinmtz at us dot ibm dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc64-linux
  GCC host triplet: powerpc64-linux
GCC target triplet: powerpc64-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|gcov/profile                |rtl-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-26 19:57 -------
Created an attachment (id=9809)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9809&action=view)
dwarf2out.gcda

Attached dwarf2out.gcda

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-26 19:58 -------
Created an attachment (id=9810)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9810&action=view)
dwarf2out.gcno

Attached dwarf2out.gcno

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-26 20:01 -------
Is gcc really being used or is "stage1/xgcc -B stage1" being used?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build, ice-on-valid-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-26 20:14 -------
When I first saw it, it was with "stage1/xgcc -B stage1", however I've since 
done a non-profiled bootstrap of mainline, and the resulting gcc exhibits the 
same problem.

Working on getting the .i attached.  Unfortunately the file is a bit too large 
so I'm working to reduce it.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1059 bytes --]


------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-26 20:28 -------
Until I attach a .i file, this could be reproduced by anyone with a copy of 
the mainline source and the gcda/gcno attachments I've already provided.  
You'd need to to something like this:

.../build-gcc-mline/gcc> set MYSRC="/home/steinmtz/src/mainline/gcc"
.../build-gcc-mline/gcc> cd ~/build-gcc-mline/gcc
.../build-gcc-mline/gcc> /home/steinmtz/install/mainline/bin/gcc -c -O2 -
DIN_GCC -DHAVE_CONFIG_H -I. -I$MYSRC/gcc -I$MYSRC/gcc/../include -
I$MYSRC/gcc/../libcpp/include -fprofile-use -freorder-blocks-and-partition 
$MYSRC/gcc/dwarf2out.c -o dwarf2out.o
/home/steinmtz/work/src/mainline/gcc/gcc/dwarf2out.c: In function 
âoutput_call_frame_infoâ:
/home/steinmtz/work/src/mainline/gcc/gcc/dwarf2out.c:2410: internal compiler 
error: in make_edges, at cfgbuild.c:350
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-26 21:49 -------
Created an attachment (id=9811)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9811&action=view)
dwarf2out.i

Here's the .i.	It results in a little "noise" in the form of some warnings
during the compilation, but the resulting failure is the same.	I'll update it
if I get a chance to clean it up more.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-26 22:48 -------
Just a note:
I was able to profiledbootstrap on x86_64-pc-linux-gnu (but that was over a week ago).
On powerpc-darwin, it failed, and I got PR 23396 out of it, from what I heard from someone else using 
a newer glibc on i686-pc-linux-gnu and it will fail the same way.
I will try x86_64-pc-linux-gnu and see if I get something like this.

Also:
> When I first saw it, it was with "stage1/xgcc -B stage1", however I've since 
> done a non-profiled bootstrap of mainline, and the resulting gcc exhibits the 
> same problem.
I just wantted to make sure that the build mechanism was not messed up, because that would actually 
be easier to fix :).

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-27 15:06 -------
Problem goes away if the -freorder-blocks-and-partition option is removed.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-27 20:05 -------
Created an attachment (id=9822)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9822&action=view)
Smaller dwarf2out.i

Attached a smaller .i file.  Any further reduction causes profile data
mismatches.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
Attachment #9811 is|0                           |1
           obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-27 20:52 -------
(In reply to comment #7)
> Just a note:
> I was able to profiledbootstrap on x86_64-pc-linux-gnu (but that was over a week ago).
> On powerpc-darwin, it failed, and I got PR 23396 out of it, from what I heard from someone
> else using 
> a newer glibc on i686-pc-linux-gnu and it will fail the same way.
> I will try x86_64-pc-linux-gnu and see if I get something like this.
I tried x86-64-pc-linux-gnu and I did not see it and I don't have access to powerpc64-linux but if you 
could give the backtrace, that would be nice.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ctice at apple dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-29 00:13 -------
On powerpc-darwin, I was able to profiledbootstrap so this most likely 64bit related.  By the way how 
did you configure gcc?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|ice during feedback stage of|[4.1 Regression] ice during
                   |profiledbootstrap on powerpc|feedback stage of
                   |                            |profiledbootstrap on powerpc
   Target Milestone|---                         |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-29 04:36 -------
./> gcc -v
Using built-in specs.
Target: powerpc64-linux
Configured with: /home/steinmtz/work/src/mainline/gcc/configure --
prefix=/home/steinmtz/work/install/mainline --build=powerpc64-linux --
host=powerpc64-linux --target=powerpc64-linux --with-cpu=default32 --with-
as=/home/steinmtz/tools/binutils-20041116/bin/as --with-
ld=/home/steinmtz/tools/binutils-20041116/bin/ld --enable-threads=posix --
enable-shared --enable-__cxa_atexit --enable-languages=c,c++,f95 --with-
gmp=/home/pthaugen/work/install/gmp
Thread model: posix
gcc version 4.1.0 20050926 (experimental)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-29 19:26 -------
Here's the backtrace:

#0  make_edges (min=0x402ebe70, max=0x4070edd0, update_p=1) 
at /home/steinmtz/work/src/mainline/gcc/gcc/cfgbuild.c:350
#1  0x1048d144 in find_many_sub_basic_blocks (blocks=0x0) 
at /home/steinmtz/work/src/mainline/gcc/gcc/cfgbuild.c:763
#2  0x103d0d00 in tree_expand_cfg () 
at /home/steinmtz/work/src/mainline/gcc/gcc/cfgexpand.c:1537
#3  0x103cc1a4 in execute_one_pass (pass=0x106b3514) 
at /home/steinmtz/work/src/mainline/gcc/gcc/passes.c:809
#4  0x103cc2d0 in execute_pass_list (pass=0x106b3514) 
at /home/steinmtz/work/src/mainline/gcc/gcc/passes.c:841
#5  0x1009844c in tree_rest_of_compilation (fndecl=0x4081ab00)
    at /home/steinmtz/work/src/mainline/gcc/gcc/tree-optimize.c:419
#6  0x10017b10 in c_expand_body (fndecl=0x4081ab00) 
at /home/steinmtz/work/src/mainline/gcc/gcc/c-decl.c:6636
#7  0x104280a0 in cgraph_expand_function (node=0x40826e70) 
at /home/steinmtz/work/src/mainline/gcc/gcc/cgraphunit.c:1053
#8  0x104282b8 in cgraph_expand_all_functions () 
at /home/steinmtz/work/src/mainline/gcc/gcc/cgraphunit.c:1119
#9  0x10428968 in cgraph_optimize () 
at /home/steinmtz/work/src/mainline/gcc/gcc/cgraphunit.c:1276
#10 0x10019a58 in c_write_global_declarations () 
at /home/steinmtz/work/src/mainline/gcc/gcc/c-decl.c:7634
#11 0x10390370 in compile_file () 
at /home/steinmtz/work/src/mainline/gcc/gcc/toplev.c:1003
#12 0x10392094 in do_compile () 
at /home/steinmtz/work/src/mainline/gcc/gcc/toplev.c:1933
#13 0x10392130 in toplev_main (argc=14, argv=0xffffdd34) 
at /home/steinmtz/work/src/mainline/gcc/gcc/toplev.c:1965
#14 0x10074644 in main (argc=0, argv=0x0) 
at /home/steinmtz/work/src/mainline/gcc/gcc/main.c:35


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug rtl-optimization/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-30 17:47 -------
Here's the offending jump:

(jump_insn:HI 2539 2830 2305 120 (set (pc)
        (reg:SI 66 ctr [ targetm.asm_out.unwind_label ])) 504 
{*indirect_jumpsi} (insn_list:REG_DEP_TRUE 2537 (insn_list:REG_DEP_TRUE 2538 
(insn_list:REG_DEP_TRUE 1498 (nil))))
    (insn_list:REG_LABEL 1504 (expr_list:REG_CROSSING_JUMP (nil)
            (nil))))


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 17:57 -------
(In reply to comment #14)
> Here's the offending jump:
> 
> (jump_insn:HI 2539 2830 2305 120 (set (pc)
>         (reg:SI 66 ctr [ targetm.asm_out.unwind_label ])) 504 
> {*indirect_jumpsi} (insn_list:REG_DEP_TRUE 2537 (insn_list:REG_DEP_TRUE 2538 
> (insn_list:REG_DEP_TRUE 1498 (nil))))
>     (insn_list:REG_LABEL 1504 (expr_list:REG_CROSSING_JUMP (nil)
>             (nil))))

Huh, this should be an indirect call and not a jump.
Can you attach the dumps from -fdump-rtl-expand-all?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|rtl-optimization            |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[steinmtz at us dot ibm dot com]

------- Additional Comments From steinmtz at us dot ibm dot com  2005-09-30 20:16 -------
Created an attachment (id=9849)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9849&action=view)
Dump from -fdump-rtl-expand-all

Output from -fdump-rtl-expand-all attached as requested.

Note that the jump instruction I posted was obtained by placing the following
call directly before the failing gcc_assert.

      print_rtl_single (dump_file, insn);
      gcc_assert (JUMP_LABEL (insn)); /* failing assert at cfgbuild.c line 350
*/

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 20:34 -------
Actually the jump_insn and the backtrace does not make sense because it should have triggered the 
computed_jump_p part in make_edge.
Can you add a fflush(dump_file); right after the print_rtl_single.  Also can you attach all of the dumps 
for -fdump-tree-all-all because the backtrace does not make sense between that the jump_insn having 
registers allocated already.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 22:21 -------
The jump:
(jump_insn:HI 2539 2830 2305 120 (set (pc)
        (reg:SI 66 ctr)) 482 {*indirect_jumpsi} (insn_list:REG_DEP_TRUE 2537 (insn_list:REG_DEP_TRUE 
2538 (insn_list:REG_DEP_TRUE 1498 (nil))))
    (insn_list:REG_LABEL 1504 (expr_list:REG_CROSSING_JUMP (nil)
            (nil))))

And the backtrace:
#0  fancy_abort (file=0x92cb54 "../../gcc/cfgbuild.c", line=350, function=0x8e5374 "make_edges") at 
../../gcc/diagnostic.c:590
#1  0x007db014 in make_edges (min=0x2dd87e8, max=0x2df70b0, update_p=1) at ../../gcc/
cfgbuild.c:350
#2  0x007dc378 in find_many_sub_basic_blocks (blocks=0x1b6d240) at ../../gcc/cfgbuild.c:763
#3  0x0059a7f8 in split_all_insns (upd_life=0) at ../../gcc/recog.c:2795
#4  0x003898ec in rest_of_handle_flow2 () at ../../gcc/flow.c:4624
#5  0x0069f964 in execute_one_pass (pass=0x9fdef4) at ../../gcc/passes.c:809


This makes me feel better.  Plus this is a bug really related to/exposed by -freorder-blocks-and-
partition, we are changing an direct into an indirect one but make_edges is not thinking it should get 
an indirect to a known label here or at least as far as I can tell.

Looking more into it.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 22:22 -------
Note I could reproduce this with a cross crompiler from powerpc-darwin7.9 to powerpc64-linux with 
the following configure options:
Configured with: ../configure --target=powerpc64-linux-gnu --with-cpu=default32


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 22:28 -------
This is the same problem as listed in PR 20606 which is not exposed by -freorder-blocks-and-partition 
at all.  I am looking into this further.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |20606
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069

[Bug middle-end/24069] [4.1 Regression] ice during feedback stage of profiledbootstrap on powerpc

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-30 22:41 -------
Created an attachment (id=9851)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9851&action=view)
Patch which needs testing

Could do you do a profiledbootstrap with this patch?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pinskia at gcc dot gnu dot
                   |dot org                     |org
             Status|UNCONFIRMED                 |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24069