[Bug target/109140] New: ICE when building qemu (qemu-user, MIPS component) on sparc64-unknown-linux-gnu with -march=niagara4

[Bug target/109140] New: ICE when building qemu (qemu-user, MIPS component) on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

            Bug ID: 109140
           Summary: ICE when building qemu (qemu-user, MIPS component) on
                    sparc64-unknown-linux-gnu with -march=niagara4
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: sam at gentoo dot org
  Target Milestone: ---
              Host: sparc64-unknown-linux-gnu
            Target: sparc64-unknown-linux-gnu
             Build: sparc64-unknown-linux-gnu

Hit when building QEMU 7.2.0's qemu-user component for mips on sparc64. This
one is reproducible with cross.

It needs -O2 -mcpu=niagara4 at least and -O2 -march=ultrasparc is fine.

```
$ sparc64-unknown-linux-gnu-gcc target_mips_tcg_lmmi_helper.c.i -O2
-mcpu=niagara4 -c
../target/mips/tcg/lmmi_helper.c: In function ‘helper_pcmpgtb’:
../target/mips/tcg/lmmi_helper.c:554:1: error: unrecognizable insn:
(insn 9 8 10 2 (set (reg:DI 118)
        (unspec:DI [
                (ltu:V8QI (subreg:V8QI (reg/v:DI 115 [ ft ]) 0)
                    (subreg:V8QI (reg/v:DI 114 [ fs ]) 0))
            ] UNSPEC_FUCMP)) "../target/mips/tcg/lmmi_helper.c":551:18 -1
     (nil))
during RTL pass: vregs
../target/mips/tcg/lmmi_helper.c:554:1: internal compiler error: in
extract_insn, at recog.cc:2791
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
See <https://bugs.gentoo.org/> for instructions.

$ sparc64-unknown-linux-gnu-gcc --version
sparc64-unknown-linux-gnu-gcc (Gentoo Hardened 12.2.1_p20230304 p13) 12.2.1
20230304
Copyright (C) 2022 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE
```

Attached the preprocessed source, not yet reduced or bisected, but I'll be
doing both later.

[Bug target/109140] ICE when building qemu (qemu-user, MIPS component) on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #1 from Sam James <sam at gentoo dot org> ---
Created attachment 54669
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54669&action=edit
target_mips_tcg_lmmi_helper.c.i.xz

[Bug target/109140] ICE when building qemu (qemu-user, MIPS component) on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #2 from Sam James <sam at gentoo dot org> ---
Created attachment 54670
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54670&action=edit
reduced.i

Attached reduced version.

```
# sparc64-unknown-linux-gnu-gcc reduced.i -O2 -mcpu=niagara4 -c
reduced.i:3:1: warning: no semicolon at end of struct or union
    3 | } LMIValue;
      | ^
reduced.i:6:1: warning: return type defaults to ‘int’ [-Wimplicit-int]
    6 | helper_pcmpgtb() {
      | ^~~~~~~~~~~~~~
reduced.i: In function ‘helper_pcmpgtb’:
reduced.i:10:1: error: unrecognizable insn:
   10 | }
      | ^
(insn 24 23 25 2 (set (reg:DI 130)
        (unspec:DI [
                (gtu:V8QI (reg:V8QI 125)
                    (reg:V8QI 119))
            ] UNSPEC_FUCMP)) "reduced.i":9:55 -1
     (nil))
during RTL pass: vregs
reduced.i:10:1: internal compiler error: in extract_insn, at recog.cc:2791
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
See <https://bugs.gentoo.org/> for instructions.
```

[Bug target/109140] ICE when building qemu (qemu-user, MIPS component) on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #3 from Sam James <sam at gentoo dot org> ---
I'll bisect later but someone is free to beat me if they want. I think 11 was
fine, but need to check.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

Sam James <sam at gentoo dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |11.3.1
      Known to fail|                            |12.2.1

--- Comment #4 from Sam James <sam at gentoo dot org> ---
so far:
11.3.1 20230303 is OK.
12.2.1 20230304 isn't.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

Mikael Pettersson <mikpelinux at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikpelinux at gmail dot com

--- Comment #5 from Mikael Pettersson <mikpelinux at gmail dot com> ---
I can reproduce with a cross to sparc64-linux-gnu hosted on x86_64-linux-gnu.

git bisect between 11.3.0 (good) and 12.2.0 (bad) identified:

2b8453c401b699ed93c085d0413ab4b5030bcdb8 is the first bad commit
commit 2b8453c401b699ed93c085d0413ab4b5030bcdb8
Author: liuhongt <hongtao.liu@intel.com>
Date:   Mon Sep 6 13:48:49 2021 +0800

    Enable auto-vectorization at O2 with very-cheap cost model.

which tells me the issue is pre-existing.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #6 from Mikael Pettersson <mikpelinux at gmail dot com> ---
With -O2 -ftree-vectorize -mcpu=niagara4 the ICE reproduces with gcc-10.4.0 but
not with gcc-9.5.0.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #7 from Mikael Pettersson <mikpelinux at gmail dot com> ---
With -O2 -ftree-vectorize -mcpu=niagara4 a bisect between 9.5.0 (good) and
10.4.0 (bad) found

6271dd984d7f920d4fb17ad37af6a1f8e6b796dc is the first bad commit
commit 6271dd984d7f920d4fb17ad37af6a1f8e6b796dc
Author: Wilco Dijkstra <wdijkstr@arm.com>
Date:   Wed Nov 20 16:29:23 2019 +0000

    PR85678: Change default to -fno-common

And sure enough, with -O2 -ftree-vectorize -fno-common -mcpu=niagara4 gcc-9.5.0
also ICEs.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Mikael Pettersson from comment #7)
> With -O2 -ftree-vectorize -mcpu=niagara4 a bisect between 9.5.0 (good) and
> 10.4.0 (bad) found
> 
> 6271dd984d7f920d4fb17ad37af6a1f8e6b796dc is the first bad commit
> commit 6271dd984d7f920d4fb17ad37af6a1f8e6b796dc
> Author: Wilco Dijkstra <wdijkstr@arm.com>
> Date:   Wed Nov 20 16:29:23 2019 +0000
> 
>     PR85678: Change default to -fno-common
> 
> And sure enough, with -O2 -ftree-vectorize -fno-common -mcpu=niagara4
> gcc-9.5.0 also ICEs.

Right still latent from before that commit ...

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #9 from Mikael Pettersson <mikpelinux at gmail dot com> ---
I can reproduce all the way down to gcc-4.7.4, gcc-4.6.4 doesn't support
-mcpu=niagara4.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #10 from Mikael Pettersson <mikpelinux at gmail dot com> ---
A bisect between 4.6.4 (good) and 4.7.4 (bad) found:

1f9ed162eb30f1b40b65d164b3a40ac78e1f006e is the first bad commit
commit 1f9ed162eb30f1b40b65d164b3a40ac78e1f006e
Author: David S. Miller <davem@davemloft.net>
Date:   Tue Nov 1 08:42:57 2011 +0000

    Add vcond/vcondu patterns to sparc backend.

If I disable the "vconduv8qiv8qi" expander this test case doesn't cause an ICE.

[Bug target/109140] ICE (during RTL pass: internal compiler error: in extract_insn, at recog.cc:2791) when building qemu on sparc64-unknown-linux-gnu with -march=niagara4

[sam at gentoo dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

Sam James <sam at gentoo dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |davem at davemloft dot net

--- Comment #11 from Sam James <sam at gentoo dot org> ---
Thank you for the extremely thorough bisect!

[Bug target/109140] internal error on array element comparison in loop with -O2 -mvis3

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Target Milestone|---                         |13.0
                 CC|                            |ebotcazou at gcc dot gnu.org
   Last reconfirmed|                            |2023-03-27
     Ever confirmed|0                           |1
            Summary|ICE (during RTL pass:       |internal error on array
                   |internal compiler error: in |element comparison in loop
                   |extract_insn, at            |with -O2 -mvis3
                   |recog.cc:2791) when         |
                   |building qemu on            |
                   |sparc64-unknown-linux-gnu   |
                   |with -march=niagara4 since  |
                   |r0-112855-g1f9ed162eb30f1   |

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
I can reproduce.

[Bug target/109140] internal error on array element comparison in loop with -O2 -mvis3

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

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

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

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Investigating.

[Bug target/109140] [12/13 regression] internal error on array element comparison in loop with -O2 -mvis3

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.0                        |12.3
            Summary|internal error on array     |[12/13 regression] internal
                   |element comparison in loop  |error on array element
                   |with -O2 -mvis3             |comparison in loop with -O2
                   |                            |-mvis3

[Bug target/109140] [12/13 regression] internal error on array element comparison in loop with -O2 -mvis3

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #14 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Eric Botcazou <ebotcazou@gcc.gnu.org>:

https://gcc.gnu.org/g:a21bd7faba67997a6da457dbda16f15bca1a9156

commit r13-6893-ga21bd7faba67997a6da457dbda16f15bca1a9156
Author: Eric Botcazou <ebotcazou@adacore.com>
Date:   Tue Mar 28 10:13:24 2023 +0200

    Fix PR target/109140

    This is a regression present on the mainline and 12 branch at -O2, but the
    issue is related to vectorization so was present at -O3 in earlier
versions.

    The vcondu expander that was added for VIS 3 more than a decade ago does
not
    fully work, because it does not filter out the unsigned condition codes
(the
    instruction is an UNSPEC that accepts only signed condition codes).

    While I was at it, I also added the missing vcond and vcondu expanders for
    the new comparison instructions that were added in VIS 4.

    gcc/
            PR target/109140
            * config/sparc/sparc.cc (sparc_expand_vcond): Call signed_condition
            on operand #3 to get the final condition code.  Use std::swap.
            * config/sparc/sparc.md (vcondv8qiv8qi): New VIS 4 expander.
            (fucmp<gcond:code>8<P:mode>_vis): Move around.
            (fpcmpu<gcond:code><GCM:gcm_name><P:mode>_vis): Likewise.
            (vcondu<GCM:mode><GCM:mode>): New VIS 4 expander.

    gcc/testsuite/
            * gcc.target/sparc/20230328-1.c: New test.
            * gcc.target/sparc/20230328-2.c: Likewise.
            * gcc.target/sparc/20230328-3.c: Likewise.
            * gcc.target/sparc/20230328-4.c: Likewise.

[Bug target/109140] [12/13 regression] internal error on array element comparison in loop with -O2 -mvis3

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #15 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Eric Botcazou
<ebotcazou@gcc.gnu.org>:

https://gcc.gnu.org/g:6f2bb57f909014618aaf770bfaaf49e28e410311

commit r12-9324-g6f2bb57f909014618aaf770bfaaf49e28e410311
Author: Eric Botcazou <ebotcazou@adacore.com>
Date:   Tue Mar 28 10:13:24 2023 +0200

    Fix PR target/109140

    This is a regression present on the mainline and 12 branch at -O2, but the
    issue is related to vectorization so was present at -O3 in earlier
versions.

    The vcondu expander that was added for VIS 3 more than a decade ago does
not
    fully work, because it does not filter out the unsigned condition codes
(the
    instruction is an UNSPEC that accepts only signed condition codes).

    While I was at it, I also added the missing vcond and vcondu expanders for
    the new comparison instructions that were added in VIS 4.

    gcc/
            PR target/109140
            * config/sparc/sparc.cc (sparc_expand_vcond): Call signed_condition
            on operand #3 to get the final condition code.  Use std::swap.
            * config/sparc/sparc.md (vcondv8qiv8qi): New VIS 4 expander.
            (fucmp<gcond:code>8<P:mode>_vis): Move around.
            (fpcmpu<gcond:code><GCM:gcm_name><P:mode>_vis): Likewise.
            (vcondu<GCM:mode><GCM:mode>): New VIS 4 expander.

    gcc/testsuite/
            * gcc.target/sparc/20230328-1.c: New test.
            * gcc.target/sparc/20230328-2.c: Likewise.
            * gcc.target/sparc/20230328-3.c: Likewise.
            * gcc.target/sparc/20230328-4.c: Likewise.

[Bug target/109140] [12/13 regression] internal error on array element comparison in loop with -O2 -mvis3

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

--- Comment #16 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Eric Botcazou
<ebotcazou@gcc.gnu.org>:

https://gcc.gnu.org/g:002470a073ca0742820991ee221986a722848ff2

commit r11-10601-g002470a073ca0742820991ee221986a722848ff2
Author: Eric Botcazou <ebotcazou@adacore.com>
Date:   Tue Mar 28 10:13:24 2023 +0200

    Fix PR target/109140

    This is a regression present on the mainline and 12 branch at -O2, but the
    issue is related to vectorization so was present at -O3 in earlier
versions.

    The vcondu expander that was added for VIS 3 more than a decade ago does
not
    fully work, because it does not filter out the unsigned condition codes
(the
    instruction is an UNSPEC that accepts only signed condition codes).

    While I was at it, I also added the missing vcond and vcondu expanders for
    the new comparison instructions that were added in VIS 4.

    gcc/
            PR target/109140
            * config/sparc/sparc.c (sparc_expand_vcond): Call signed_condition
            on operand #3 to get the final condition code.  Use std::swap.
            * config/sparc/sparc.md (vcondv8qiv8qi): New VIS 4 expander.
            (fucmp<gcond:code>8<P:mode>_vis): Move around.
            (fpcmpu<gcond:code><GCM:gcm_name><P:mode>_vis): Likewise.
            (vcondu<GCM:mode><GCM:mode>): New VIS 4 expander.

    gcc/testsuite/
            * gcc.target/sparc/20230328-1.c: New test.
            * gcc.target/sparc/20230328-2.c: Likewise.
            * gcc.target/sparc/20230328-3.c: Likewise.
            * gcc.target/sparc/20230328-4.c: Likewise.

[Bug target/109140] [12/13 regression] internal error on array element comparison in loop with -O2 -mvis3

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109140

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

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

--- Comment #17 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Thanks for reporting the problem.