[Bug middle-end/63974] New: [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[Bug middle-end/63974] New: [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

            Bug ID: 63974
           Summary: [5 Regression] gcc.c-torture/compile/991213-3.c ICEs
                    with -mabi=ilp32
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: critical
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
            Target: aarch64

/home/apinski/src/gcc-cavium/toolchain-thunder/src/gcc/testsuite/gcc.c-torture/compile/991213-3.c:
In function 'jump':
/home/apinski/src/gcc-cavium/toolchain-thunder/src/gcc/testsuite/gcc.c-torture/compile/991213-3.c:1:15:
internal compiler error: in decompose, at rtl.h:1461
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://www.cavium.com/support.html> for instructions.
Preprocessed source stored into /tmp/ccgU2CUE.out file, please attach this to
your bugreport.
[reply] [-] Comment 1

--- CUT ---
This was introduced by the wide_int merge.

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |5.0

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

ktkachov at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-12-10
                 CC|                            |ktkachov at gcc dot gnu.org
     Ever confirmed|0                           |1
      Known to fail|                            |5.0

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Confirmed with -mabi=ilp32 (passes without it).
Also a longer stack trace:
$TOP/gcc/testsuite/gcc.c-torture/compile/991213-3.c:3:15: internal compiler
error: in decompose, at rtl.h:2007
0xa342d2 wi::int_traits<std::pair<rtx_def*, machine_mode> >::decompose(long*,
unsigned int, std::pair<rtx_def*, machine_mode> const&)
        $TOP/gcc/rtl.h:2005
0xa342d2 wide_int_ref_storage<std::pair<rtx_def*, machine_mode> >
        $TOP/gcc/wide-int.h:948
0xa342d2 generic_wide_int<std::pair<rtx_def*, machine_mode> >
        $TOP/gcc/wide-int.h:725
0xa342d2 simplify_const_unary_operation(rtx_code, machine_mode, rtx_def*,
machine_mode)
        $TOP/gcc/simplify-rtx.c:1731
0xa313a6 simplify_unary_operation(rtx_code, machine_mode, rtx_def*,
machine_mode)
        $TOP/gcc/simplify-rtx.c:822
0x73f4a2 convert_memory_address_addr_space_1
        $TOP/gcc/explow.c:357
0x74007e convert_memory_address_addr_space(machine_mode, rtx_def*, unsigned
char)
        $TOP/gcc/explow.c:426
0x980410 maybe_legitimize_operand
        $TOP/gcc/optabs.c:8309
0x980410 maybe_legitimize_operands(insn_code, unsigned int, unsigned int,
expand_operand*)
        $TOP/gcc/optabs.c:8346
0x9804fa maybe_gen_insn(insn_code, unsigned int, expand_operand*)
        $TOP/gcc/optabs.c:8364
0x981512 maybe_expand_jump_insn(insn_code, unsigned int, expand_operand*)
        $TOP/gcc/optabs.c:8422
0x981fb6 expand_jump_insn(insn_code, unsigned int, expand_operand*)
        $TOP/gcc/optabs.c:8448
0x981ffe emit_indirect_jump(rtx_def*)
        $TOP/gcc/optabs.c:4516
0x676436 expand_computed_goto
        $TOP/gcc/cfgexpand.c:3056
0x676436 expand_gimple_stmt_1
        $TOP/gcc/cfgexpand.c:3278
0x676436 expand_gimple_stmt
        $TOP/gcc/cfgexpand.c:3447
0x6792d7 expand_gimple_basic_block
        $TOP/gcc/cfgexpand.c:5280
0x679fb7 execute
        $TOP/gcc/cfgexpand.c:5889
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|                            |jakub at gcc dot gnu.org
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 34423
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34423&action=edit
gcc5-pr63974.patch

Untested fix.

The bug is that convert_memory_address (Pmode, ...) is called twice, which is
not possible for VOIDmode constants.
As optabs.c (indirectly from emit_indirect_jump) already calls it, because the
argument is built with create_address_operand and the EXPAND_ADDRESS case
in maybe_legitimize_operand calls convert_memory_address (Pmode, ...), the
other
convert_memory_address call is only harmful.

I don't have easy access to any POINTERS_EXTEND_UNSIGNED target though, so
can't test it easily.

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Tue Jan 13 13:44:06 2015
New Revision: 219529

URL: https://gcc.gnu.org/viewcvs?rev=219529&root=gcc&view=rev
Log:
    PR middle-end/63974
    * cfgexpand.c (expand_computed_goto): Don't call
    convert_memory_address here.

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

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
As it didn't regress on x86_64 x32, I've applied it.  If any problems show up
on aarch64, please reopen.

[Bug middle-end/63974] [5 Regression] gcc.c-torture/compile/991213-3.c ICEs with -mabi=ilp32

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63974

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #4)
> As it didn't regress on x86_64 x32, I've applied it.  If any problems show
> up on aarch64, please reopen.

I have not seen any yet.  I did a build for aarch64-elf with no regressions
(including this testcase now passing for -mabi=ilp32).