[Bug middle-end/94465] New: [10 Regression] fnsplit dump file ICEs

[Bug middle-end/94465] New: [10 Regression] fnsplit dump file ICEs

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94465

            Bug ID: 94465
           Summary: [10 Regression] fnsplit dump file ICEs
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---

Created attachment 48179
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48179&action=edit
c-decl.i

The attached source ICEs when using `-c -o foo.o -Ofast -fdump-tree-fnsplit-all
-w`.

with

c-decl.c:9540: internal compiler error: tree check: expected class 'type', have
'exceptional' (error_mark) in build_int_cst, at tree.c:1366
0x6cb76b tree_class_check_failed(tree_node const*, tree_code_class, char
const*, int, char const*)
        ../../gcc-fsf/gcc/tree.c:9790
0x1119357 tree_class_check(tree_node*, tree_code_class, char const*, int, char
const*)
        ../../gcc-fsf/gcc/tree.h:3401
0x1119357 build_int_cst(tree_node*, poly_int<2u, long>)
        ../../gcc-fsf/gcc/tree.c:1366
0x111a64b array_ref_low_bound(tree_node*)
        ../../gcc-fsf/gcc/tree.c:13429
0xed1d87 dump_generic_node(pretty_printer*, tree_node*, int, dump_flag, bool)
        ../../gcc-fsf/gcc/tree-pretty-print.c:2301
0xaa98e7 dump_unary_rhs
        ../../gcc-fsf/gcc/gimple-pretty-print.c:416
0xaa98e7 dump_gimple_assign
        ../../gcc-fsf/gcc/gimple-pretty-print.c:673
0xaab8d7 gimple_dump_bb_buff
        ../../gcc-fsf/gcc/gimple-pretty-print.c:2964
0xaab8d7 gimple_dump_bb(_IO_FILE*, basic_block_def*, int, dump_flag)
        ../../gcc-fsf/gcc/gimple-pretty-print.c:2988
0x8bedff dump_bb(_IO_FILE*, basic_block_def*, int, dump_flag)
        ../../gcc-fsf/gcc/cfghooks.c:286
0xe24ca3 dump_function_to_file(tree_node*, _IO_FILE*, dump_flag)
        ../../gcc-fsf/gcc/tree-cfg.c:8098
0xce70f7 execute_function_dump
        ../../gcc-fsf/gcc/passes.c:1750
0x1933813 diagnostic_report_diagnostic(diagnostic_context*, diagnostic_info*)
        ../../gcc-fsf/gcc/diagnostic.c:1142
0x1933be7 diagnostic_impl
        ../../gcc-fsf/gcc/diagnostic.c:1309
0x1934943 internal_error(char const*, ...)
        ../../gcc-fsf/gcc/diagnostic.c:1711
0x6cb76b tree_class_check_failed(tree_node const*, tree_code_class, char
const*, int, char const*)
        ../../gcc-fsf/gcc/tree.c:9790
0x1119357 tree_class_check(tree_node*, tree_code_class, char const*, int, char
const*)
        ../../gcc-fsf/gcc/tree.h:3401
0x1119357 build_int_cst(tree_node*, poly_int<2u, long>)
        ../../gcc-fsf/gcc/tree.c:1366
0x111a64b array_ref_low_bound(tree_node*)
        ../../gcc-fsf/gcc/tree.c:13429
0xed1d87 dump_generic_node(pretty_printer*, tree_node*, int, dump_flag, bool)
        ../../gcc-fsf/gcc/tree-pretty-print.c:2301
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

It looks like it's assuming the array offset is an integer constant but in this
case it's an ssa_name:

 <array_ref 0xffffb7bace00
    type <integer_type 0xffffb7baf5e8 int public SI
        size <integer_cst 0xffffb7b9efd8 constant 32>
        unit-size <integer_cst 0xffffb7bb3000 constant 4>
        align:32 warn_if_not_align:0 symtab:0 alias-set 3 canonical-type
0xffffb7baf5e8 precision:32 min <integer_cst 0xffffb7b9ef90 -2147483648> max
<integer_cst 0xffffb7b9efa8 2147483647>
        pointer_to_this <pointer_type 0xffffb7bb7a80>>

    arg:0 <var_decl 0xffffb7ff52d0 x0
        type <array_type 0xffffb79aac78 type <integer_type 0xffffb7baf5e8 int>
            BLK
            align:32 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0xffffb79aac78
            pointer_to_this <pointer_type 0xffffb79dbc78>>
        addressable used public external read BLK c-decl.i:1:8
        align:32 warn_if_not_align:0 context <translation_unit_decl
0xffffb79de0f0 c-decl.i>
        chain <var_decl 0xffffb7ff5360 x6 type <integer_type 0xffffb7baf5e8
int>
            used public static read SI c-decl.i:9:1 size <integer_cst
0xffffb7b9efd8 32> unit-size <integer_cst 0xffffb7bb3000 4>
            align:32 warn_if_not_align:0 context <translation_unit_decl
0xffffb79de0f0 c-decl.i> chain <var_decl 0xffffb7ff53f0 x7>>>
    arg:1 <ssa_name 0xffffb7ba1480 type <error_mark 0xffffb7b9ed80>
        nothrow
        def_stmt
        version:3 in-free-list>

when it's doing 

#3  0x000000000111a64c in array_ref_low_bound (exp=exp@entry=0xffffb7bace00) at
../../gcc-fsf/gcc/tree.c:13429
13429     return build_int_cst (TREE_TYPE (TREE_OPERAND (exp, 1)), 0);


But I don't know what the correct fix here is.

[Bug middle-end/94465] [10 Regression] fnsplit dump file ICEs

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94465

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
   Target Milestone|---                         |10.0
   Last reconfirmed|                            |2020-04-03
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that the SSA name is released:

    arg:1 <ssa_name 0xffffb7ba1480 type <error_mark 0xffffb7b9ed80>
        nothrow
        def_stmt
        version:3 in-free-list>

so we're dumping something that was freed.  Indeed we're dumping
from delete_basic_block and there from cleanup_control_flow_pre
which fails to delete unreachable basic-blocks in postorder
which likely only matters for dumping.

[Bug middle-end/94465] [10 Regression] fnsplit dump file ICEs

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94465

--- Comment #2 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:ef6631051d4bb9519f3f0dd5f3b3eadbd67bc10a

commit r10-7536-gef6631051d4bb9519f3f0dd5f3b3eadbd67bc10a
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Apr 3 13:46:49 2020 +0200

    middle-end/94465 - handle released SSA names in array_ref_low_bound

    array_ref_low_bound is used in dumping ARRAY_REFs which in turn
    is called when basic blocks are deleted.  cleanup_control_flow_pre
    consciously decides to remove unreachable basic-blocks in arbitrary
    order so the following makes array_ref_low_bound forgiving in the
    case the SSA name with the index definition has been released
    already.

    2020-04-03  Richard Biener  <rguenther@suse.de>

            PR middle-end/94465
            * tree.c (array_ref_low_bound): Deal with released SSA names
            in index position.

[Bug middle-end/94465] [10 Regression] fnsplit dump file ICEs

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94465

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

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

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.

[Bug middle-end/94465] [10 Regression] fnsplit dump file ICEs

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94465

--- Comment #4 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
Thanks Richard!