[Bug rtl-optimization/94368] New: [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu

[Bug rtl-optimization/94368] New: [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

            Bug ID: 94368
           Summary: [9/10 Regression] ICE in final_scan_insn_1, at
                    final.c:3074(error: could not split insn) on
                    aarch64-linux-gnu
           Product: gcc
           Version: 9.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at debian dot org
  Target Milestone: ---

seen with gcc-9 branch 20200325, with -O1 on aarch64-linux-gnu.

$ cat partr.i
# 5 "" 3
b, c, d, e, f;
short g;
a() __attribute__((__const__));
h() {
i:
  for (;;) {
    __atomic_load_n(&e, d);
    long j = a(2) && __sync_val_compare_and_swap(&c, 0, j);
    b = 1;
    if (h == e)
      break;
  }
  __sync_val_compare_and_swap(&g, -1, f);
  goto i;
}

$ gcc-9 -c -O1 -Wall partr.i 
: In function ‘h’:
:19:1: error: could not split insn
(insn 51 49 110 (parallel [
            (set (reg:CC 66 cc)
                (unspec_volatile:CC [
                        (const_int 0 [0])
                    ] UNSPECV_ATOMIC_CMPSW))
            (set (reg:SI 1 x1 [125])
                (zero_extend:SI (mem/v:HI (reg/f:DI 27 x27 [132]) [-1  S2
A16])))
            (set (mem/v:HI (reg/f:DI 27 x27 [132]) [-1  S2 A16])
                (unspec_volatile:HI [
                        (const_int -1 [0xffffffffffffffff])
                        (reg:HI 0 x0 [orig:122 f ] [122])
                        (const_int 0 [0])
                        (const_int 32773 [0x8005]) repeated x2
                    ] UNSPECV_ATOMIC_CMPSW))
            (clobber (reg:SI 2 x2 [138]))
        ]) "":17:3 3636 {aarch64_compare_and_swaphi}
     (expr_list:REG_DEAD (reg:HI 0 x0 [orig:122 f ] [122])
        (expr_list:REG_UNUSED (reg:CC 66 cc)
            (expr_list:REG_UNUSED (reg:SI 2 x2 [138])
                (expr_list:REG_UNUSED (reg:SI 1 x1 [125])
                    (nil))))))
during RTL pass: final
:19:1: internal compiler error: in final_scan_insn_1, at final.c:3074
Please submit a full bug report,
with preprocessed source if appropriate.


$ gcc-10 -c -O1 -Wall partr.i 
: In function ‘h’:
:19:1: error: could not split insn
(insn 49 47 5 (parallel [
            (set (reg:CC 66 cc)
                (unspec_volatile:CC [
                        (const_int 0 [0])
                    ] UNSPECV_ATOMIC_CMPSW))
            (set (reg:SI 1 x1 [127])
                (zero_extend:SI (mem/v:HI (reg/f:DI 27 x27 [120]) [-1  S2
A16])))
            (set (mem/v:HI (reg/f:DI 27 x27 [120]) [-1  S2 A16])
                (unspec_volatile:HI [
                        (const_int -1 [0xffffffffffffffff])
                        (reg:HI 0 x0 [orig:124 f ] [124])
                        (const_int 0 [0])
                        (const_int 32773 [0x8005]) repeated x2
                    ] UNSPECV_ATOMIC_CMPSW))
            (clobber (reg:SI 2 x2 [137]))
        ]) "":17:3 3908 {aarch64_compare_and_swaphi}
     (expr_list:REG_DEAD (reg:HI 0 x0 [orig:124 f ] [124])
        (expr_list:REG_UNUSED (reg:CC 66 cc)
            (expr_list:REG_UNUSED (reg:SI 2 x2 [137])
                (expr_list:REG_UNUSED (reg:SI 1 x1 [127])
                    (nil))))))
during RTL pass: final
:19:1: internal compiler error: in final_scan_insn_1, at final.c:3073
0x60be4f _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
        ../../src/gcc/rtl-error.c:108
0x8380bb final_scan_insn_1
        ../../src/gcc/final.c:3073
0x8382d3 final_scan_insn(rtx_insn*, _IO_FILE*, int, int, int*)
        ../../src/gcc/final.c:3152
0x8385a7 final_1
        ../../src/gcc/final.c:2020
0x838d23 rest_of_handle_final
        ../../src/gcc/final.c:4658
0x838d23 execute
        ../../src/gcc/final.c:4736
Please submit a full bug report,
with preprocessed source if appropriate.


this is a compiler with hardening flags enabled by default.

[Bug rtl-optimization/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-03-29
             Status|UNCONFIRMED                 |WAITING
                 CC|                            |marxin at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
I can't reproduce that with x86-64 cross compiler.

[Bug rtl-optimization/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Can't reproduce it either.  Any specific -march= etc. options?

[Bug rtl-optimization/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
            Summary|[9/10 Regression] ICE in    |[9/10 Regression] ICE in
                   |final_scan_insn_1, at       |final_scan_insn_1, at
                   |final.c:3074(error: could   |final.c:3074(error: could
                   |not split insn) on          |not split insn) on
                   |aarch64-linux-gnu           |aarch64-linux-gnu since
                   |                            |r9-3744
   Target Milestone|---                         |9.4

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Actually, I can, -O2 -fpie is needed.
Started with r9-3744-gd400fda3a8c3330f77eb9d51874f5482d3819a9f

[Bug rtl-optimization/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug rtl-optimization/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Testcase that still FAILs on the trunk:

/* PR target/94368 */
/* { dg-do compile { target fpic } } */
/* { dg-options "-fpic -O1 -fcommon" } */

int b, c, d, e, f, h;
short g;
int foo (int) __attribute__ ((__const__));

void
bar (void)
{
  while (1)
    {
      while (1)
        {
          __atomic_load_n (&e, 0);
          if (foo (2))
            __sync_val_compare_and_swap (&c, 0, f);
          b = 1;
          if (h == e)
            break;
        }
      __sync_val_compare_and_swap (&g, -1, f);
    }
}

[Bug target/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 48143
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48143&action=edit
gcc10-pr94368.patch

Untested fix.

[Bug target/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The problem is that RA doesn't care about predicates, only constraints, and so
the constraints need to match the predicates, or it needs to be ensured in insn
conditions instead.

[Bug target/94368] [9/10 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:a27c534794dbe3530acae3427d2c58f937f1b050

commit r10-7472-ga27c534794dbe3530acae3427d2c58f937f1b050
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Tue Mar 31 11:08:22 2020 +0200

    aarch64: Fix up aarch64_compare_and_swaphi pattern [PR94368]

    The following testcase ICEs in final_scan_insn_1.  The problem is in the
    @aarch64_compare_and_swaphi define_insn_and_split, since 9 it uses
    aarch64_plushi_operand predicate for the "expected value" operand, which
    allows either 0..0xfff constants or 0x1000..0xf000 constants (i.e. HImode
    values which when zero extended are either 0..0xfff or (0..0xfff) << 12).
    The problem is that RA doesn't care about predicates, it honors just
    constraints and the used constraint on the operand is n, which means any
    HImode CONST_SCALAR_INT.  In the testcase LRA thus propagates the -1
    value into the insn.
    This is a define_insn_and_split which requires mandatory split.
    But during split2 pass, we check the predicate (and don't check
    constraints), which fails and thus we don't split it and during final ICE
    because the mandatory splitting didn't happen.

    The following patch fixes it by adding a matching constraint to the
    predicate and using it.

    2020-03-31  Jakub Jelinek  <jakub@redhat.com>

            PR target/94368
            * config/aarch64/constraints.md (Uph): New constraint.
            * config/aarch64/atomics.md (cas_short_expected_imm): New mode
attr.
            (@aarch64_compare_and_swap<mode>): Use it instead of n in operand
2's
            constraint.

            * gcc.dg/pr94368.c: New test.

[Bug target/94368] [9 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[9/10 Regression] ICE in    |[9 Regression] ICE in
                   |final_scan_insn_1, at       |final_scan_insn_1, at
                   |final.c:3074(error: could   |final.c:3074(error: could
                   |not split insn) on          |not split insn) on
                   |aarch64-linux-gnu since     |aarch64-linux-gnu since
                   |r9-3744                     |r9-3744

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed on the trunk so far.

[Bug target/94368] [9 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-9 branch has been updated by Kyrylo Tkachov
<ktkachov@gcc.gnu.org>:

https://gcc.gnu.org/g:13f6d5ac48a7d55b41927849aeebc5832f8c63f0

commit r9-8437-g13f6d5ac48a7d55b41927849aeebc5832f8c63f0
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date:   Wed Apr 1 13:53:05 2020 +0100

    aarch64: Fix up aarch64_compare_and_swaphi pattern [PR94368]

    2020-04-01  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

            Backport from mainline
            2020-03-31  Jakub Jelinek  <jakub@redhat.com>

            PR target/94368
            * config/aarch64/constraints.md (Uph): New constraint.
            * config/aarch64/atomics.md (cas_short_expected_imm): New mode
attr.
            (@aarch64_compare_and_swap<mode>): Use it instead of n in operand
2's
            constraint.

            * gcc.dg/pr94368.c: New test.

[Bug target/94368] [9 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-8 branch has been updated by Andre Simoes Dias Vieira
<avieira@gcc.gnu.org>:

https://gcc.gnu.org/g:163e0d03f2cdc80de4968c4be4694cdaeaf47fcd

commit r8-10212-g163e0d03f2cdc80de4968c4be4694cdaeaf47fcd
Author: Andre Vieira <andre.simoesdiasvieira@arm.com>
Date:   Thu Apr 16 10:16:17 2020 +0100

    aarch64: Fix up aarch64_compare_and_swaphi pattern [PR94368]

    2020-04-16  Andre Vieira  <andre.simoesdiasvieira@arm.com>

            Backport from mainline
            2020-03-31  Jakub Jelinek  <jakub@redhat.com>

            PR target/94368
            * config/aarch64/constraints.md (Uph): New constraint.
            * config/aarch64/atomics.md (cas_short_expected_imm): New mode
attr.
            (aarch64_compare_and_swap<mode>): Use it instead of n in operand
2's
            constraint.

            * gcc.dg/pr94368.c: New test.

[Bug target/94368] [9 Regression] ICE in final_scan_insn_1, at final.c:3074(error: could not split insn) on aarch64-linux-gnu since r9-3744

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94368

Wilco <wilco at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wilco at gcc dot gnu.org
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #11 from Wilco <wilco at gcc dot gnu.org> ---
Closing since the fix has been backported to all active branches.