[Bug middle-end/49210] New: verify_gimple fails building ada/einfo.o at -O3

[Bug middle-end/49210] New: verify_gimple fails building ada/einfo.o at -O3

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

           Summary: verify_gimple fails building ada/einfo.o at -O3
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: aoliva@gcc.gnu.org
                CC: rguenth@gcc.gnu.org


The patch for bug 48989 regressed BUILD_CONFIG=bootstrap-O3 ada builds on i686-
and x86_64-linux-gnu.  While building stage2 we hit this error:

../../gcc/ada/einfo.adb: In function ‘Einfo.Is_String_Type.Part’:
../../gcc/ada/einfo.adb:6377:4: error: invalid conversion in gimple call
boolean

einfo__b___XDLU_0__1

# .MEM_50 = VDEF <.MEM_49>
iftmp.45_26 = einfo.is_string_type.part (id_1(D));

+===========================GNAT BUG DETECTED==============================+
| 4.7.0 20110528 (experimental) [trunk revision
8219ad7:2a33c3d:4b51a1cd8cb5f4ff6a659aa55ce963497a624bd3] (x86_64-pc-linux-gnu)
GCC error:|
| verify_gimple failed                                                     |
| Error detected around ../../gcc/ada/einfo.adb:6377:4                     |

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.05.28 21:12:50
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
            Version|unknown                     |4.7.0
   Target Milestone|---                         |4.7.0
            Summary|verify_gimple fails         |[4.7 regression]
                   |building ada/einfo.o at -O3 |verify_gimple fails
                   |                            |building ada/einfo.o at -O3
     Ever Confirmed|0                           |1

--- Comment #1 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-05-28 21:12:50 UTC ---
Yep.  I'm going to reduce it into a testcase for gnat.dg.

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

--- Comment #2 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-05-29 08:04:42 UTC ---
Created attachment 24390
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24390
Reduced testcase for x86

Run gnatchop and compile the adb file at -O3 -gnata.  Suitable for gnat.dg.

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-05-30 10:47:43 UTC ---
We have calls that assign

  iftmp.2_37 = boolean_subtype2.is_string_type.part (id_3(D));

 <integer_type 0x7ffff5b5c0a8 boolean_subtype2__b___XDLU_0__1
    type <boolean_type 0x7ffff7ee3dc8 boolean sizes-gimplified public visited
unsigned QI
        size <integer_cst 0x7ffff7ed3f80 constant visited 8>
        unit size <integer_cst 0x7ffff7ed3fa0 constant 1>
        align 8 symtab 0 alias set 5 canonical type 0x7ffff7ee3dc8 precision 8
min <integer_cst 0x7ffff7ee85e0 0> max <integer_cst 0x7ffff7ee8600 255> RM size
<integer_cst 0x7ffff7ee8640 1> RM max <integer_cst 0x7ffff7ee8620 1>>
    sizes-gimplified public visited unsigned QI size <integer_cst
0x7ffff7ed3f80 8> unit size <integer_cst 0x7ffff7ed3fa0 1>
    align 8 symtab 0 alias set 5 canonical type 0x7ffff5b5c0a8 precision 8 min
<integer_cst 0x7ffff5b6c800 0> max <integer_cst 0x7ffff5b6c820 255> RM size
<integer_cst 0x7ffff7ee8640 1> RM min <integer_cst 0x7ffff7ee85e0 0> RM max
<integer_cst 0x7ffff7ee8620 1>
    chain <type_decl 0x7ffff7f128a0 boolean_subtype2__b___XDLU_0__1>>

(the function result type) to

 <boolean_type 0x7ffff7ee3dc8 boolean sizes-gimplified public visited unsigned
QI
    size <integer_cst 0x7ffff7ed3f80 type <integer_type 0x7ffff7ee30a8
bit_size_type> constant visited 8>
    unit size <integer_cst 0x7ffff7ed3fa0 type <integer_type 0x7ffff7ee3000
size_type> constant 1>
    align 8 symtab 0 alias set 5 canonical type 0x7ffff7ee3dc8 precision 8 min
<integer_cst 0x7ffff7ee85e0 0> max <integer_cst 0x7ffff7ee8600 255> RM size
<integer_cst 0x7ffff7ee8640 1> RM max <integer_cst 0x7ffff7ee8620 1>>

where only for the type of the LHS we know it's two-valued.

This is exposed during inlining triggered by function splitting.

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-05-30 11:13:41 UTC ---
I think a similar case could be constructed with restrict qualified pointers.
I have a patch.

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-05-30 13:12:27 UTC ---
Author: rguenth
Date: Mon May 30 13:12:23 2011
New Revision: 174435

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=174435
Log:
2011-05-30  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/49210
    * ipa-split.c (split_function): Care for the case where the
    call result is not trivially convertible to the result holding
    variable.

    * gnat.dg/boolean_subtype2.adb: New testcase.
    * gnat.dg/boolean_subtype2.ads: Likewise.
    * gnat.dg/boolean_subtype2_pkg.ads: Likewise.

Added:
    trunk/gcc/testsuite/gnat.dg/boolean_subtype2.adb
    trunk/gcc/testsuite/gnat.dg/boolean_subtype2.ads
    trunk/gcc/testsuite/gnat.dg/boolean_subtype2_pkg.ads
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-split.c
    trunk/gcc/testsuite/ChangeLog

[Bug middle-end/49210] [4.7 regression] verify_gimple fails building ada/einfo.o at -O3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49210

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

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

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-05-30 13:12:45 UTC ---
Fixed.