[Bug c++/98932] New: Wrong output with -O3 on aarch64

[Bug c++/98932] New: Wrong output with -O3 on aarch64

[kristian.klausen at scoutdi dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

            Bug ID: 98932
           Summary: Wrong output with -O3 on aarch64
           Product: gcc
           Version: 8.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kristian.klausen at scoutdi dot com
  Target Milestone: ---

Hi, 

I have been struggling with a vectorization issue, and have managed to
re-create to a minimal example as attached. 

When compiled with `-O2` the correct output is produced, however when compiled
with `-O3` the output is bogus when compiled with gcc-8 and gcc-9. Gcc-7 gives
the correct output. 

Tested with the following versions under qemu for aarch64: 

```
g++-7 --version
g++-7 (Ubuntu/Linaro 7.5.0-6ubuntu2) 7.5.0
g++-8 --version
g++-8 (Ubuntu/Linaro 8.4.0-3ubuntu2) 8.4.0
g++-9 --version
g++-9 (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
```

Output: 
```
:~# g++-9 -O3 a.cpp && ./a.out
Length is: 26348240
:~# g++-8 -O3 a.cpp && ./a.out
Length is: 26348240
:~# g++-7 -O3 a.cpp && ./a.out
Length is: 294
```

Also gave it a go with the suggested flags before filing the report: 

```
g++-8 -O3  -Wall -Wextra -fno-strict-aliasing -fwrapv a.cpp && ./a.out
Length is: 26348240
```

The code gives the correct output with gcc-10: 

```
g++-10 -O3 a.cpp && ./a.out
Length is: 294
g++-10 --version
g++-10 (Ubuntu 10.2.0-5ubuntu1~20.04) 10.2.0
```

Let me know if I can provide more information. 

Best,
Kristian

[Bug c++/98932] Wrong output with -O3 on aarch64

[kristian.klausen at scoutdi dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

--- Comment #1 from Kristian <kristian.klausen at scoutdi dot com> ---
Created attachment 50115
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50115&action=edit
Test-case

[Bug c++/98932] Wrong output with -O3 on aarch64

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-02-02
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
                 CC|                            |marxin at gcc dot gnu.org

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
I'm going to bisect that..

[Bug tree-optimization/98932] Wrong output with -O3 on aarch64

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |11.0
                 CC|                            |rguenth at gcc dot gnu.org,
                   |                            |rsandifo at gcc dot gnu.org
             Status|ASSIGNED                    |NEW
           Assignee|marxin at gcc dot gnu.org          |unassigned at gcc dot gnu.org

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
Ok, so the issue was fixed in g:74166aabeb7f22990476b1169bba031b8323ee92, which
is about a preferred vector types.

So with the following patch:

diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index 9e88438b3c3..1e5a05a00b0 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -17351,14 +17351,14 @@ aarch64_autovectorize_vector_modes (vector_modes
*modes, bool)

        TODO: We could support a limited form of V4QImode too, so that
        we use 32-bit vectors for 8-bit elements.  */
-    V4HImode,
+//    V4HImode,

     /* Try using 64-bit vectors for 32-bit elements and 128-bit vectors
        for 64-bit elements.

        TODO: We could similarly support limited forms of V2QImode and V2HImode
        for this case.  */
-    V2SImode
+//    V2SImode
   };

   /* Try using N-byte SVE modes only after trying N-byte Advanced SIMD mode.
@@ -23901,8 +23901,8 @@ aarch64_libgcc_floating_mode_supported_p
 #define TARGET_VECTORIZE_VEC_PERM_CONST \
   aarch64_vectorize_vec_perm_const

-#undef TARGET_VECTORIZE_RELATED_MODE
-#define TARGET_VECTORIZE_RELATED_MODE aarch64_vectorize_related_mode
+//#undef TARGET_VECTORIZE_RELATED_MODE
+//#define TARGET_VECTORIZE_RELATED_MODE aarch64_vectorize_related_mode
 #undef TARGET_VECTORIZE_GET_MASK_MODE
 #define TARGET_VECTORIZE_GET_MASK_MODE aarch64_get_mask_mode
 #undef TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE

I tracked that it was fixed with g:04b99da898a9817e72fedb4063589648b7961ac5,
which is a vectorizer related revision.

Thus it's likely dup of PR97236.

Can we backport the fixed into 8 and 9 branch?

[Bug tree-optimization/98932] Wrong output with -O3 on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yes.

*** This bug has been marked as a duplicate of bug 97236 ***

[Bug tree-optimization/98932] Wrong output with -O3 on aarch64

[kristian.klausen at scoutdi dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

--- Comment #5 from Kristian <kristian.klausen at scoutdi dot com> ---
Thanks for such a swift reponse! Looking forward to testing the patches for
8.x.

[Bug tree-optimization/98932] Wrong output with -O3 on aarch64

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

--- Comment #6 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Kristian from comment #5)
> Thanks for such a swift reponse! Looking forward to testing the patches for
> 8.x.

You're welcome. Just a note, please try to use the latest GCC release (version
10). GCC 8 will goes out of support quite soon and you will more likely receive
backports for serious bugs.

[Bug tree-optimization/98932] Wrong output with -O3 on aarch64

[kristian.klausen at scoutdi dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98932

--- Comment #7 from Kristian <kristian.klausen at scoutdi dot com> ---
Thanks!

Yes, I agree. We are however bound to earlier versions due to CUDA-dependency
on the NVIDIA Jetson-platforms:
https://github.com/OE4T/meta-tegra/wiki/Compatibility-notes

Hopefully NVIDIA will update their CUDA-libraries in due time. 

Best,
Kristian