[Bug target/112568] New: [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[Bug target/112568] New: [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

            Bug ID: 112568
           Summary: [14 Regression] Miscompilation of radeonsi (mesa) with
                    -march=raptorlake (-mavx) since
                    r14-5355-g3cd3a09b3f91a1
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kocelfc at tutanota dot com
  Target Milestone: ---
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu

Created attachment 56602
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56602&action=edit
objdump -d -Mintel of function at trunk

Since r14-5355-g3cd3a09b3f91a1, there is a bug causing gcc to miscompile
radeonsi when using -march=raptorlake, which leads to segfaults. I tried making
a testcase but I've got no idea where to even start looking.

The function appears in the preprocessed file as 
static uint16_t
tc_call_flush(struct pipe_context *pipe, void *call, uint64_t *last)
{
   struct tc_flush_call *p = ((struct tc_flush_call *)to_call_check((void
*)(call), ( ((sizeof(struct tc_flush_call)) + (8) - 1) / (8) )));
   struct pipe_screen *screen = pipe->screen;

   pipe->flush(pipe, p->fence ? &p->fence : 
# 3619 "../src/gallium/auxiliary/util/u_threaded_context.c" 3 4
                                           ((void *)0)
# 3619 "../src/gallium/auxiliary/util/u_threaded_context.c"
                                               , p->flags);
   screen->fence_reference(screen, &p->fence, 
# 3620 "../src/gallium/auxiliary/util/u_threaded_context.c" 3 4
                                             ((void *)0)
# 3620 "../src/gallium/auxiliary/util/u_threaded_context.c"
                                                 );

   tc_flush_queries(p->tc);

   return ( ((sizeof(struct tc_flush_call)) + (8) - 1) / (8) );
}

[Bug target/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #1 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56603
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56603&action=edit
objdump -d -Mintel of function with the commit reverted

[Bug target/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #2 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56604
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56604&action=edit
preprocessed files compressed with xz -9e

[Bug target/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Kostadin Shishmanov <kocelfc at tutanota dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kocelfc at tutanota dot com

--- Comment #3 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Using built-in specs.
COLLECT_GCC=gcc-14
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-pc-linux-gnu/14/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with:
/var/tmp/notmpfs/portage/sys-devel/gcc-14.0.0.9999/work/gcc-14.0.0.9999/configure
--host=x86_64-pc-linux-gnu --build=x86_64-pc-linux-gnu --prefix=/usr
--bindir=/usr/x86_64-pc-linux-gnu/gcc-bin/14
--includedir=/usr/lib/gcc/x86_64-pc-linux-gnu/14/include
--datadir=/usr/share/gcc-data/x86_64-pc-linux-gnu/14
--mandir=/usr/share/gcc-data/x86_64-pc-linux-gnu/14/man
--infodir=/usr/share/gcc-data/x86_64-pc-linux-gnu/14/info
--with-gxx-include-dir=/usr/lib/gcc/x86_64-pc-linux-gnu/14/include/g++-v14
--disable-silent-rules --disable-dependency-tracking
--with-python-dir=/share/gcc-data/x86_64-pc-linux-gnu/14/python
--enable-languages=c,c++ --enable-obsolete --enable-secureplt --disable-werror
--with-system-zlib --enable-nls --without-included-gettext
--disable-libunwind-exceptions --enable-checking=yes,extra,rtl
--with-bugurl=https://bugs.gentoo.org/ --with-pkgversion='Gentoo 14.0.0 p,
commit 35bb529f420ff5863fdae51049e6935ab927ff3d' --with-gcc-major-version-only
--enable-libstdcxx-time --enable-lto --disable-libstdcxx-pch --enable-shared
--enable-threads=posix --enable-__cxa_atexit --enable-clocale=gnu
--enable-multilib --with-multilib-list=m32,m64 --disable-fixed-point
--enable-targets=all --disable-libgomp --disable-libssp --disable-libada
--disable-cet --disable-systemtap --disable-valgrind-annotations
--disable-vtable-verify --disable-libvtv --with-zstd --without-isl
--enable-default-pie --enable-host-pie --enable-host-bind-now
--enable-default-ssp --with-build-config=bootstrap-O3
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.0 20231116 (experimental)
7a496b7ce105185be12e023e749c90d5f7561879 (Gentoo 14.0.0 p, commit
35bb529f420ff5863fdae51049e6935ab927ff3d)

[Bug target/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Sam James <sjames at gcc dot gnu.org> changed:

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

--- Comment #4 from Sam James <sjames at gcc dot gnu.org> ---
I have an LLVM miscompilation x86-64 which seems to bisect to the same. Not
looked into it more yet.

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |blocker
   Target Milestone|---                         |14.0
           Keywords|                            |wrong-code
          Component|target                      |rtl-optimization
           Priority|P3                          |P1

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The first thing to start with is figure out which file is being miscompiled.
Once we have that it might be easier to track down what is going wrong (we can
even look at the RTL dumps before and after to understand the issue).

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the thing to look for the placement of vzeroupper. That is what mode
switching is used for inside the x86 back-end.

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #7 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56612
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56612&action=edit
bad object file

I've attached the bad object file (the one from the compilation with the "bad"
commit). Going to also upload a good one in a bit. What flags do you need me to
add for the rtl dumps?

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Kostadin Shishmanov from comment #7)
> Created attachment 56612 [details]
> bad object file
> 
> I've attached the bad object file (the one from the compilation with the
> "bad" commit). Going to also upload a good one in a bit. What flags do you
> need me to add for the rtl dumps?

-da should be enough. I would only attach the first one that is different. Also
attach the preprocessed source if you have not yet (-save-temps and either a .i
or .ii file compressed if needed).

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #9 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56613
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56613&action=edit
dump of the first bad pass with the problematic commit

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #10 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56614
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56614&action=edit
dump of the first bad pass with the commit reverted

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #11 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
I think the gcc builds might be a bit more different than just the commit
reverted, so I am going to redo everything and upload the attachments again
with just that difference.

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Kostadin Shishmanov <kocelfc at tutanota dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56613|0                           |1
        is obsolete|                            |

--- Comment #12 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56616
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56616&action=edit
dump of the first different pass with the problematic commit

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Kostadin Shishmanov <kocelfc at tutanota dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56614|0                           |1
        is obsolete|                            |

--- Comment #13 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56617
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56617&action=edit
dump of the first different pass with the commit reverted

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Kostadin Shishmanov <kocelfc at tutanota dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56603|0                           |1
        is obsolete|                            |

--- Comment #14 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56618
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56618&action=edit
good object file

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Kostadin Shishmanov <kocelfc at tutanota dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56602|0                           |1
        is obsolete|                            |
  Attachment #56612|0                           |1
        is obsolete|                            |

--- Comment #15 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
Created attachment 56619
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56619&action=edit
bad object file

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #16 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Kostadin Shishmanov from comment #13)
> Created attachment 56617 [details]
> dump of the first different pass with the commit reverted

Note using -fdump-unnumbered might help to avoid dumping memory addresses that
were from memory addresses inside gcc itself. I forgot about that when I
mentioned the -da option.

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #17 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Can you try the patch in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112572#c18 ?

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[kocelfc at tutanota dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

--- Comment #18 from Kostadin Shishmanov <kocelfc at tutanota dot com> ---
(In reply to Andrew Pinski from comment #17)
> Can you try the patch in
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112572#c18 ?

The patch does fix it.

[Bug rtl-optimization/112568] [14 Regression] Miscompilation of radeonsi (mesa) with -march=raptorlake (-mavx) since r14-5355-g3cd3a09b3f91a1

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112568

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

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

--- Comment #19 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Even though this is an older bug, pr 112572 has more analysis of what is going
wrong so marking as a dup.

*** This bug has been marked as a duplicate of bug 112572 ***