[Bug c++/108422] New: [13 Regression] ICE: base pointer cycle detected

[Bug c++/108422] New: [13 Regression] ICE: base pointer cycle detected

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

            Bug ID: 108422
           Summary: [13 Regression] ICE: base pointer cycle detected
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gscfq@t-online.de
  Target Milestone: ---

Started between 20220911 and 20220918 :


$ cat z1.cc
void foo (int a)
{
  auto bar = [&]()
  {
    #pragma omp target parallel
    #pragma omp target
    ++a;
  };
  bar();
}


$ cat z2.cc
void foo (int a)
{
  auto bar = [&]()
  {
    #pragma omp target
    #pragma omp target
    ++a;
  };
  bar();
}


$ g++-13-20230115 -c z1.cc -fopenmp
when processing group:
map(to:*__closure)
z1.cc: In lambda function:
z1.cc:5:13: internal compiler error: base pointer cycle detected
    5 |     #pragma omp target parallel
      |             ^~~
0xc0a6f3 omp_tsort_mapping_groups_1
        ../../gcc/gimplify.cc:9603
0xc0a515 omp_tsort_mapping_groups_1
        ../../gcc/gimplify.cc:9646
0xc0a515 omp_tsort_mapping_groups_1
        ../../gcc/gimplify.cc:9646
0xc1ee00 omp_tsort_mapping_groups
        ../../gcc/gimplify.cc:9684
0xc1ee00 gimplify_scan_omp_clauses
        ../../gcc/gimplify.cc:10771
0xc24fce gimplify_omp_workshare
        ../../gcc/gimplify.cc:15390
0xc10cca gimplify_expr(tree_node**, gimple**, gimple**, bool (*)(tree_node*),
int)
        ../../gcc/gimplify.cc:16903
0xc14308 gimplify_stmt(tree_node**, gimple**)
        ../../gcc/gimplify.cc:7219
0xc14861 gimplify_bind_expr
        ../../gcc/gimplify.cc:1430
0xc10eea gimplify_expr(tree_node**, gimple**, gimple**, bool (*)(tree_node*),
int)
        ../../gcc/gimplify.cc:16562
0xc14308 gimplify_stmt(tree_node**, gimple**)
        ../../gcc/gimplify.cc:7219
0xc11ccb gimplify_statement_list
        ../../gcc/gimplify.cc:2019
0xc11ccb gimplify_expr(tree_node**, gimple**, gimple**, bool (*)(tree_node*),
int)
        ../../gcc/gimplify.cc:16806
0xc14308 gimplify_stmt(tree_node**, gimple**)
        ../../gcc/gimplify.cc:7219
0xc14861 gimplify_bind_expr
        ../../gcc/gimplify.cc:1430
0xc10eea gimplify_expr(tree_node**, gimple**, gimple**, bool (*)(tree_node*),
int)
        ../../gcc/gimplify.cc:16562
0xc14308 gimplify_stmt(tree_node**, gimple**)
        ../../gcc/gimplify.cc:7219
0xc153ab gimplify_body(tree_node*, bool)
        ../../gcc/gimplify.cc:17623
0xc157fd gimplify_function_tree(tree_node*)
        ../../gcc/gimplify.cc:17822
0xa988e7 cgraph_node::analyze()
        ../../gcc/cgraphunit.cc:676

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |13.0

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[13 Regression] ICE: base   |[13 Regression] ICE: base
                   |pointer cycle detected      |pointer cycle detected
                   |                            |since
                   |                            |r13-2661-gb57abd072dd319a7
   Last reconfirmed|                            |2023-01-17
     Ever confirmed|0                           |1
                 CC|                            |julian at codesourcery dot com,
                   |                            |marxin at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r13-2661-gb57abd072dd319a7.

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |burnus at gcc dot gnu.org
           Keywords|                            |ice-on-invalid-code

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
    #pragma omp target parallel
    #pragma omp target
    ++a;

This violates:

"Restrictions to the target construct are as follows:
• Device-affecting constructs, other than target constructs for which the
ancestor device-modifier is specified, must not be encountered during execution
of a target region."

And GCC should print (existing check):
   warning: ‘target’ construct inside of ‘target’ region


However, when making it valid by adding
  #pragma omp requires reverse_offload
+ adding 'device(ancestor:1)' to the second 'omp target',
it still fails with the ICE.

 * * *

For the failing case,

Breakpoint 4, gimplify_scan_omp_clauses (list_p=0x7ffff6fc5570,
pre_p=0x7fffffffd7a0, region_type=ORT_TARGET, code=OMP_TARGET) at
../../repos/gcc/gcc/gimplify.cc:10771

10771             outlist = omp_tsort_mapping_groups (groups, grpmap);
(gdb) p debug(groups->m_vec->m_vecdata.grp_start[0])

map(to:*(const struct ._anon_0 *) __closure)
map(firstprivate:__closure [pointer assign, bias: 0])
map(tofrom:*__closure->__a [len: 4])
map(always_pointer:__closure->__a [pointer assign, bias: 0])
map(to:*__closure)
map(to:*(const struct ._anon_0 *) __closure)

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> ---
  FOR_EACH_VEC_ELT (*groups, i, grp)
    {
      if (grp->mark != PERMANENT)
        if (!omp_tsort_mapping_groups_1 (&cursor, groups, grpmap, grp))
          return NULL;
    }

first processes:

map(to:*(const struct ._anon_0 *) __closure) map(firstprivate:__closure
[pointer assign, bias: 0]) map(tofrom:*__closure->__a [len: 4])
map(always_pointer:__closure->__a [pointer assign, bias: 0]) map(to:*__closure)
map(to:*(const struct ._anon_0 *) __closure)

which then has:
  *innerp == grp
in omp_tsort_mapping_groups_1.

The next VEC has:

(gdb) p debug(*grp.grp_start)
map(tofrom:*__closure->__a [len: 4]) map(always_pointer:__closure->__a [pointer
assign, bias: 0]) map(to:*__closure) map(to:*(const struct ._anon_0 *)
__closure)

But the inner one is again the previous one:

(gdb) p debug(**innerp.grp_start)
map(to:*(const struct ._anon_0 *) __closure) map(firstprivate:__closure
[pointer assign, bias: 0]) map(tofrom:*__closure->__a [len: 4])
map(always_pointer:__closure->__a [pointer assign, bias: 0]) map(to:*__closure)
map(to:*(const struct ._anon_0 *) __closure)

where:

(gdb) p (*innerp)->mark
$78 = PERMANENT


Untested patch:

--- a/gcc/gimplify.cc
+++ b/gcc/gimplify.cc
@@ -9643 +9643 @@ omp_tsort_mapping_groups_1 (omp_mapping_group ***outlist,
-      if (innerp && *innerp != grp)
+      if (innerp && *innerp != grp && (*innerp)->mark != PERMANENT)

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Created attachment 54491
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54491&action=edit
Testcase - compile with 'g++ -fopenmp' + run

(In reply to Tobias Burnus from comment #3)
> --- a/gcc/gimplify.cc
> +++ b/gcc/gimplify.cc
> @@ -9643 +9643 @@ omp_tsort_mapping_groups_1 (omp_mapping_group ***outlist,
> -      if (innerp && *innerp != grp)
> +      if (innerp && *innerp != grp && (*innerp)->mark != PERMANENT)

While COMPILING WORKS and the expected nested-target warning is printed,
it FAILS at RUNTIME (on a system not configured for offloading) with a
segfault.

That's related to the warning:
  test.cc:20:13: warning: '__closure' is used uninitialized [-Wuninitialized]
which appears at the location of the outer 'omp target (parallel)'.


Note: without inner target construct (reverse offload), the program runs fine.
(Independent of the one-line patch.)

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Priority|P3                          |P1

[Bug c++/108422] [13 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

Tobias Burnus <burnus at gcc dot gnu.org> changed:

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

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> ---
As discussed with richi on #gcc IRC: it shouldn't be P1 - and I could/should
mark it as P3, which I now did.

Reasoning:
- With GCC 12, it compiles without an ICE but as with mainline with the patch
in
  comment 4 applied, it will fail (segfault) at runtime
  (→ testcase in attachment 54491)
- If being strict, nested target is not permitted - except with
  'reverse offload', which fails in GCC 12 (sorry) as not supported.

(Still, we should try to fix the ICE - but best instead of creating invalid
code,
either a sorry or working code would be good.)

[Bug c++/108422] [13/14 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.0                        |13.2

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 13.1 is being released, retargeting bugs to GCC 13.2.

[Bug c++/108422] [13/14 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.2                        |13.3

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 13.2 is being released, retargeting bugs to GCC 13.3.

[Bug c++/108422] [13/14/15 Regression] ICE: base pointer cycle detected since r13-2661-gb57abd072dd319a7

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108422

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.3                        |13.4

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 13.3 is being released, retargeting bugs to GCC 13.4.