[Bug target/49714] New: [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[Bug target/49714] New: [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

           Summary: [4.7 Regression] Revision 176128  introduced many ICEs
                    in insn_default_length, at config/i386/i386.md:600
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dominiq@lps.ens.fr
                CC: rth@gcc.gnu.org, iains@gcc.gnu.org
              Host: x86_64-apple-darwin10
            Target: x86_64-apple-darwin10
             Build: x86_64-apple-darwin10


On x86_64-apple-darwin10, revision 176128 introduced a lot of failures in the
test suite of the kind:

[macbook] f90/bug% /opt/gcc/gcc4.7p/bin/g++
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C: In member
function 'void E::_ZThn8_N1ED1Ev()':
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C:123:1: error:
unrecognizable insn:
(jump_insn 3 2 4 (set (pc)
        (mem:QI (const:DI (unspec:DI [
                        (symbol_ref/i:DI ("_ZN1ED1Ev") [flags 0x1]
<function_decl 0x1428b7400 __comp_dtor >)
                    ] UNSPEC_GOTPCREL)) [0 S1 A8])) -1
     (nil))
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C:123:1: internal
compiler error: in insn_default_length, at config/i386/i386.md:600

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.07.12 15:40:24
     Ever Confirmed|0                           |1

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2011-07-12 15:40:24 UTC ---
Can you please regtest this patch:

--cut here--
Index: i386.c
===================================================================
--- i386.c    (revision 176207)
+++ i386.c    (working copy)
@@ -29399,7 +29399,7 @@ x86_output_mi_thunk (FILE *file,
     {
       tmp = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, fnaddr), UNSPEC_GOTPCREL);
       tmp = gen_rtx_CONST (Pmode, tmp);
-      fnaddr = gen_rtx_MEM (QImode, tmp);
+      fnaddr = gen_rtx_MEM (Pmode, tmp);
     }
     }
   else
--cut here--

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

--- Comment #2 from Richard Henderson <rth at gcc dot gnu.org> 2011-07-12 15:46:13 UTC ---
Author: rth
Date: Tue Jul 12 15:46:02 2011
New Revision: 176208

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=176208
Log:
PR target/49714
        * config/i386/i386.c (x86_output_mi_thunk): Fix mode for
        destination address in memory on some paths.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

Richard Henderson <rth at gcc dot gnu.org> changed:

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

--- Comment #3 from Richard Henderson <rth at gcc dot gnu.org> 2011-07-12 15:47:53 UTC ---
Fixed.  Both the instance you noticed, Uros, but also another
along the i386 mach-o path.

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

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

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

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-07-12 16:28:28 UTC ---
Still fails with -m32:

[macbook] f90/bug% g++47 -m32
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C: In member
function 'void E::_ZThn4_N1ED1Ev()':
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C:123:1: error:
unrecognizable insn:
(jump_insn 3 2 4 (set (pc)
        (mem:SI (symbol_ref/i:SI ("_ZN1ED1Ev") [flags 0x1] <function_decl
0x142cb3600 __comp_dtor >) [0 S4 A8])) -1
     (nil))
/opt/gcc/work/gcc/testsuite/g++.old-deja/g++.other/dyncast1.C:123:1: internal
compiler error: in insn_default_length, at config/i386/i386.md:600

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |WAITING
         AssignedTo|unassigned at gcc dot       |rth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #5 from Richard Henderson <rth at gcc dot gnu.org> 2011-07-12 17:18:56 UTC ---
Created attachment 24748
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24748
i686-darwin cleanup

This appears as if it might work.

I can't seem to trigger the branch island path here.  It looks like
only darwin8 needs this, and even configuring for i686-apple-darwin8
doesn't trigger.

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

--- Comment #6 from Iain Sandoe <iains at gcc dot gnu.org> 2011-07-12 17:28:35 UTC ---
(In reply to comment #5)
> Created attachment 24748 [details]
> i686-darwin cleanup
> 
> This appears as if it might work.
> 
> I can't seem to trigger the branch island path here.  It looks like
> only darwin8 needs this, and even configuring for i686-apple-darwin8
> doesn't trigger.

hm. doing a cross to darwin-8 should produce the requirement for stubs and
branch islands (the latter, I think are ppc only).

----

You should also be able to trigger the branch island path by using
-mmacosx-version-min=10.4 on Darwin 9 or 10 (when building for PPC).  This
should also also trigger generation of stubs for function calls on m32 x86 and
ppc.

(will try and look later - but tied up with other hassles right now)

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

--- Comment #7 from Richard Henderson <rth at gcc dot gnu.org> 2011-07-12 18:20:33 UTC ---
Author: rth
Date: Tue Jul 12 18:20:29 2011
New Revision: 176211

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=176211
Log:
PR target/49714
        * config/i386/i386.c (x86_output_mi_thunk): Use
        machopic_indirect_call_target instead of machopic_indirection_name
        directly.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c

[Bug target/49714] [4.7 Regression] Revision 176128 introduced many ICEs in insn_default_length, at config/i386/i386.md:600

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49714

Richard Henderson <rth at gcc dot gnu.org> changed:

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

--- Comment #8 from Richard Henderson <rth at gcc dot gnu.org> 2011-07-12 18:37:29 UTC ---
Fixed.  Again.