[Bug rtl-optimization/60141] ICE in i386.c distance_non_agu_define_in_bb

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60141

--- Comment #1 from Teresa Johnson <tejohnson at google dot com> ---
This appears to have been exposed, not caused, by the change to the
pass ordering. It is in fact a duplicate of PR middle-end/43631, which
was fixed on trunk by r197994 on 4/15/13. I will backport the fix to
google-4_8.

Teresa

On Mon, Feb 10, 2014 at 4:49 PM, shenhan at google dot com
<gcc-bugzilla@gcc.gnu.org> wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60141
>
>             Bug ID: 60141
>            Summary: ICE in i386.c distance_non_agu_define_in_bb
>            Product: gcc
>            Version: 4.8.3
>             Status: UNCONFIRMED
>           Severity: normal
>           Priority: P3
>          Component: rtl-optimization
>           Assignee: unassigned at gcc dot gnu.org
>           Reporter: shenhan at google dot com
>                 CC: llozano at google dot com, tejohnson at google dot com
>
> This only fails for google/gcc-4_8
>
> To reproduce -
>   configure:
>      ~/gcc.git/configure --disable-bootstrap --enable-languages='c,c++'
> --disable-multilib --disable-libatomic     --disable-libmudflap
> --disable-libssp --enable-libgomp --enable-__cxa_atexit
> --enable-checking=release     --disable-libquadmath --disable-libitm
> --enable-esp --with-arch=atom --with-tune=atom --enable-frame-pointer
>   Build -
>      make -j40
>   Compile a testcase -
>    ./gcc/xgcc -B./gcc -g -O2 -fpic -c -m32 ~/SROA.pp.cpp
>   Output -
>    SROA.cpp: In member function ‘void
> {anonymous}::AllocaPartitioning::UseBuilder::operator()()’:
> SROA.cpp:873:3: internal compiler error: Segmentation fault
> 0x8d0baf crash_signal
>         ../../../gcc.git/gcc/toplev.c:346
> 0xa798c8 distance_non_agu_define_in_bb
>         ../../../gcc.git/gcc/config/i386/i386.c:17506
> 0xa79cad distance_non_agu_define
>         ../../../gcc.git/gcc/config/i386/i386.c:17564
> 0xa79cad ix86_lea_outperforms
>         ../../../gcc.git/gcc/config/i386/i386.c:17730
> 0xb8e6ca output_89
>         ../../../gcc.git/gcc/config/i386/i386.md:2152
> 0x72a5c3 final_scan_insn(rtx_def*, _IO_FILE*, int, int, int*)
>         ../../../gcc.git/gcc/final.c:2895
> 0x72ae84 final(rtx_def*, _IO_FILE*, int)
>         ../../../gcc.git/gcc/final.c:1986
> 0x72b46d rest_of_handle_final
>         ../../../gcc.git/gcc/final.c:4432
> 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.
>
> Debugging shows that in distance_non_agu_define_in_bb (i386.c),
> BLOCK_FOR_INSN(start) returns zero, which caused this ICE.
>
> A preliminary bisecting in google/gcc-4_8 shows that the follow CL introduces
> this ICE. (So I cc'ed Teresa Johnson)
>
>
> commit 907fca49eae2da51a328eb192c691cc24f69e6af
> Author: tejohnson <tejohnson@138bc75d-0d04-0410-961f-82ee72b054a4>
> Date:   Mon Dec 2 18:28:19 2013 +0000
>
>     Backport the last few fixes for -freorder-blocks-and-partition from
>     trunk to google/4_8: r204985, r205057 and r205243
>
>     ------------------------------------------------------------------------
>         r204985 | tejohnson | 2013-11-18 14:38:51 -0800 (Mon, 18 Nov 2013) | 15
> lines
>
>         This patch fixes an lto profiledbootstrap failure with
>         -freorder-blocks-and-partition enabled. Currently compgotos
>         is the only pass that goes into cfglayout mode after bb reordering,
>         which is undesireable (and in the case of
> -freorder-blocks-and-partition
>         can cause illegal partitioning) because of the optimizations performed
>         on the cfg when going into cfglayout mode. Moved compgoto before
>         bb reordering to avoid these problems.
>
>         2013-11-18  Teresa Johnson  <tejohnson@google.com>
>
>         * gcc/cfgrtl.c (cfg_layout_initialize): Assert if we
>         try to go into cfglayout after bb reordering.
>         * gcc/passes.def: Move compgotos before bb reordering
>         since it goes into cfglayout.
>
>     ------------------------------------------------------------------------
>     ------------------------------------------------------------------------
>         r205057 | tmsriram | 2013-11-19 14:12:21 -0800 (Tue, 19 Nov 2013) | 7
> lines
>
>         Emit a label for the split cold function part.  Label name is formed by
>         suffixing the original function name with "cold".
>
>         Patch tested for bootstrap on all default languages on x86_64 and
>         regression testsuite checked for parity with RUNTESTFLAGS -m32 and m64.
>
>         2013-11-19  Sriraman Tallam  <tmsriram@google.com>
>
>         * final.c (final_scan_insn): Emit a label for the split
>         cold function part.  Label name is formed by suffixing
>         the original function name with "cold".
>
>         2013-11-19  Sriraman Tallam  <tmsriram@google.com>
>
>         * gcc.dg/tree-prof/cold_partition_label.c: New testcase.
>
>     ------------------------------------------------------------------------
>     ------------------------------------------------------------------------
>         r205243 | tejohnson | 2013-11-21 20:16:47 -0800 (Thu, 21 Nov 2013) | 6
> lines
>
>         2013-11-21  Teresa Johnson  <tejohnson@google.com>
>
>         * cfgcleanup.c (outgoing_edges_match): Walk up past note instructions
>         not understood by old_insns_match_p.
>
>     ------------------------------------------------------------------------
>
>
>     git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/google/gcc-4_8@205594
> 138bc75d-0d04-0410-961f-82ee72b054a4
>
> --
> You are receiving this mail because:
> You are on the CC list for the bug.