[Bug bootstrap/54718] New: [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[Bug bootstrap/54718] New: [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

             Bug #: 54718
           Summary: [4.8 regression] ICE in remap_gimple_stmt, at
                    tree-inline.c:1468
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: dehao@gcc.gnu.org, ebotcazou@gcc.gnu.org
              Host: sparc-sun-solaris2.11
            Target: sparc-sun-solaris2.11
             Build: sparc-sun-solaris2.11


Since 20120921 (r191613), go fails to bootstrap on Solaris 11.1/SPARC:

/vol/gcc/src/hg/trunk/local/gcc/go/gofrontend/expressions.cc: In member
function 'virtual bool
Type_conversion_expression::_ZNK26Type_conversion_expression24do_string_constant_valueEPSs.part.172(std::string*)
const':
/vol/gcc/src/hg/trunk/local/gcc/go/gofrontend/expressions.cc:14226:1: internal
compiler error: in remap_gimple_stmt, at tree-inline.c:1468
 }
 ^

The failure can be reproduced with the attached expression.ii like this:

cc1plus -fpreprocessed expressions.ii -quiet -mcpu=v9 -g -gtoggle -O2
-fno-exceptions -fno-rtti -fno-common -o expressions.s


A reghunt revealed that this was caused by the following patch:
The first bad revision is:
changeset:   3438:f36a4197405d
user:        dehao@138bc75d-0d04-0410-961f-82ee72b054a4
date:        Wed Sep 19 19:56:42 2012 +0000
summary:     Integrate lexical block into source_location.

2012-09-19  Dehao Chen  <dehao@google.com>

        * toplev.c (general_init): Init block_locations.
[...]

The strange thing is that the failure only occurs on Solaris 11.1, not 11.0 or
10, but can be reliably reproduced on the former.  Maybe related to the VM
system rewrite in 11.1 in some way?

  Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> 2012-09-26 13:06:56 UTC ---
Created attachment 28281
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28281
preprocessed input file

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2012-09-26
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-26 13:54:35 UTC ---
This means we have a stale gimple_block that is no longer contained in the
BLOCK tree (DECL_INITIAL of the function inlined).  Probably caused by
remove_unused_block_scopes.

Does the preprocessed source trigger the bug on all Solaris cc1plus'en?

(it seems to work on x86_64).

What's the last bad revision you tested?

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #3 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-09-26 13:58:58 UTC ---
> --- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-26 13:54:35 UTC ---
> This means we have a stale gimple_block that is no longer contained in the
> BLOCK tree (DECL_INITIAL of the function inlined).  Probably caused by
> remove_unused_block_scopes.
>
> Does the preprocessed source trigger the bug on all Solaris cc1plus'en?

No, only Solaris/SPARC 11.1, which let me suspect it could be related to
the VM system rewrite, which can change the details of address space
layout.

> (it seems to work on x86_64).

Right.

> What's the last bad revision you tested?

I don't have the build handy since I did a bootstrap without Go
afterwards, but I'm pretty sure it still failed in r191695.

    Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |iant at google dot com
         Depends on|                            |54688

--- Comment #4 from Rainer Orth <ro at gcc dot gnu.org> 2012-09-28 08:37:57 UTC ---
The fix for PR bootstrap/54688 also fixed this one.  I can't say if there's
another
unrelated issue here, though.

There's one other problem (which is most likely different and has occured
before):

Since 20120919, go1 might run into an infinite loop compiling one or more
sparcv9
libgo modules:

pstack shows this stacktrace:

1653:    /var/gcc/regression/trunk/11-gcc/build/./gcc/go1 fmt_test.go scan_test
 00229ec4 find_base_term(rtx_def*) (ba7c00, c, 27fb548, 33587b8, 94aef80,
e10dbf0) + 2a0
 00229d48 find_base_term(rtx_def*) (ba7c00, c, a6fc00, b43c00, 22a000, 201bbe0)
+ 124
 0022ac7c base_alias_check(rtx_def*, rtx_def*, machine_mode, machine_mode)
(eca9c840, ecac4080, c, 9, 22a000, 0) + 10
 0022c684 true_dependence_1(rtx_def const*, machine_mode, rtx_def*, rtx_def
const*, rtx_def*, bool) (1, 9, ecac4080, eca9c850, eca9c840, 1) + 150
 006a6500 drop_overlapping_mem_locs(void**, void*) (ba4400, ecac4080, 8604618,
8604608, eca9c850, 84878b0) + 1ac
 00887c24 htab_traverse_noresize (ff1180, 6a6354, ffbfc1bc, 0, 1bb1631c,
1bb04964) + 38
 0069ec30 clobber_overlapping_mems(dataflow_set_def*, rtx_def*) (e5c258,
ec1f6980, ffbfc224, 0, 1bb1631c, 1bb1631c) + 5c
 006a4870 val_store(dataflow_set_def*, rtx_def*, rtx_def*, rtx_def*, bool)
(e5c258, 107c898, ec1f6980, f6d97f68, 1, 107d7b8) + 240
 006a549c _ZL19compute_bb_dataflowP15basic_block_def.isra.55 (f7c66e48, e5c258,
107c898, ec1f69a0, f6d993d0, f6d97f68) + 6f8
 006a7a34 vt_find_locations() (cec5920, b52800, 10c58a8, f7c66e40, e5c0a8,
e5c0a0) + 3f0
 006b0858 variable_tracking_main() (6b07ec, b2f500, 1, 1, ba7e90, b52800) + 6c
 00443a54 execute_one_pass(opt_pass*) (1, b2f500, b52800, b72800, b9e800, 0) +
134
 00443f18 execute_pass_list(opt_pass*) (b2f500, b2d75c, b52800, b72800, b9e800,
0) + 14
 00443f3c execute_pass_list(opt_pass*) (b2d75c, b2d790, b52800, b72800, b9e800,
0) + 38
 00443f3c execute_pass_list(opt_pass*) (b2d790, b52800, b7292c, fb7ce8a0, 6,
b30334) + 38
 0028578c expand_function(cgraph_node*) (fb7ce8a0, b9e800, ba8800, fb48bd80,
b72800, b3d000) + bc
 00287ae0 compile() (ec0788, 78d, b3cc00, b3cc00, b3cc00, ec2f18) + c90
 00287eb0 finalize_compilation_unit() (1, 108, 1, ba8800, ba7e90, b9e800) + 80
 001ba404 Gogo::write_globals() (bcfaa8, ffbfc984, 0, d367c8, 0, ffbfc984) +
ac4
 004f385c compile_file() (2, 0, 7, b2b4c4, b9e800, ba7e90) + 88
 004f5368 toplev_main(int, char**) (ba8800, b9e800, b9e800, b2b4c4, b30400,
ba7e90) + aa8
 001823ec _start   (0, 0, 0, 0, 0, 0) + 5c

I'll hae to investigate more closely.

  Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2012-09-28 08:54:51 UTC ---
(In reply to comment #4)
> The fix for PR bootstrap/54688 also fixed this one.  I can't say if there's
> another unrelated issue here, though.
> 
> There's one other problem (which is most likely different and has occured
> before):
> 
> Since 20120919, go1 might run into an infinite loop compiling one or more
> sparcv9 libgo modules:

Related to PR54507. You can try to compile with -fno-var-tracking-assignments.

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #6 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-10-04 12:41:42 UTC ---
> --- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2012-09-28
> 08:54:51 UTC ---
> (In reply to comment #4)
>> The fix for PR bootstrap/54688 also fixed this one.  I can't say if there's
>> another unrelated issue here, though.
>> 
>> There's one other problem (which is most likely different and has occured
>> before):
>> 
>> Since 20120919, go1 might run into an infinite loop compiling one or more
>> sparcv9 libgo modules:
>
> Related to PR54507. You can try to compile with -fno-var-tracking-assignments.

That helps indeed: on a 1.2 GHz UltraSPARC-T2, the sparcv9 fmt test
takes

real    20:14:33.53
user    20:14:09.94
sys           12.42

without -fno-var-tracking-assignments, but

real       16:41.37
user       16:11.86
sys            4.08

with it.  This is still too slow for the default testsuite timeout of 5
min, but at least a much more reasonable range.

    Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-07 13:31:26 UTC ---
So, is this now fixed?

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #8 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-11-07 13:46:39 UTC ---
> --- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-07 13:31:26 UTC ---
> So, is this now fixed?

If this is really just the same issue as PR bootstrap/54688 and nothing
else, it is.

    Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-07 14:05:15 UTC ---
Did you mean some other PR?

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-11 08:57:24 UTC ---
Ping.

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

--- Comment #11 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-12-12 15:53:49 UTC ---
The extreme sparcv9 libgo compile times without
-fno-var-tracking-assignments are handled in PR debug/54402, it seems.

I haven't seen the ICE covered in this PR in quite some time.

    Rainer

[Bug bootstrap/54718] [4.8 regression] ICE in remap_gimple_stmt, at tree-inline.c:1468

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54718

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

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

--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-13 10:43:54 UTC ---
I believe the ICE has been fixed.