[Bug tree-optimization/103127] New: ICE in fold_convert_loc, at fold-const.c:2573

[Bug tree-optimization/103127] New: ICE in fold_convert_loc, at fold-const.c:2573

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

            Bug ID: 103127
           Summary: ICE in fold_convert_loc, at fold-const.c:2573
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
                CC: qing.zhao at oracle dot com
  Target Milestone: ---
              Host: x86_64-linux-gnu
            Target: ppc64le-linux-gnu

The following fails:

$ ppc64le-linux-gnu-gcc
/home/marxin/Programming/gcc/gcc/testsuite/gcc.target/powerpc/pr98872.c
-ftrivial-auto-var-init=pattern -c
during RTL pass: expand
/home/marxin/Programming/gcc/gcc/testsuite/gcc.target/powerpc/pr98872.c: In
function ‘foo’:
/home/marxin/Programming/gcc/gcc/testsuite/gcc.target/powerpc/pr98872.c:10:17:
internal compiler error: in fold_convert_loc, at fold-const.c:2573
   10 |   __vector_quad acc;
      |                 ^~~
0x5de137 fold_convert_loc(unsigned int, tree_node*, tree_node*)
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/fold-const.c:2573
0x90c44f expand_DEFERRED_INIT
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/internal-fn.c:3100
0x738fb7 expand_call_stmt
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/cfgexpand.c:2749
0x738fb7 expand_gimple_stmt_1
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/cfgexpand.c:3876
0x738fb7 expand_gimple_stmt
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/cfgexpand.c:4040
0x73d437 expand_gimple_basic_block
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/cfgexpand.c:6085
0x73f466 execute
       
/home/marxin/BIG/buildbot/buildworker/marxinbox-gcc-trunk-ppc64le/build/gcc/cfgexpand.c:6811
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

[Bug middle-end/103127] ICE in fold_convert_loc, at fold-const.c:2573

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |middle-end
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=98872

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The types are OPAQUE_TYPE.

/* This is for types that will use MODE_OPAQUE in the back end.  They are meant
   to be able to go in a register of some sort but are explicitly not to be
   converted or operated on like INTEGER_TYPE.  They will have size and
   alignment information only.  */
DEFTREECODE (OPAQUE_TYPE, "opaque_type", tcc_type, 0)

So if I understand this correctly and PR 98872 correctly. We should not expand
a DEFERRED_INIT for this type.

[Bug middle-end/103127] ICE in fold_convert_loc, at fold-const.c:2573

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-11-08

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #3 from Peter Bergner <bergner at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #1)
> The types are OPAQUE_TYPE.
[snip]
> So if I understand this correctly and PR 98872 correctly. We should not
> expand a DEFERRED_INIT for this type.

So something like this?  Ie, it's ok to just skip this altogether?  It does fix
this particular ICE.


diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
index 9e10da0ad5c..200d14337d7 100644
--- a/gcc/internal-fn.c
+++ b/gcc/internal-fn.c
@@ -3041,6 +3041,9 @@ expand_DEFERRED_INIT (internal_fn, gcall *stmt)
   tree var_type = TREE_TYPE (lhs);
   gcc_assert (init_type > AUTO_INIT_UNINITIALIZED);

+  if (OPAQUE_TYPE_P (var_type))
+    return;
+
   if (TREE_CODE (lhs) == SSA_NAME)
     reg_lhs = true;
   else

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[qinzhao at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

qinzhao at gcc dot gnu.org changed:

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

--- Comment #4 from qinzhao at gcc dot gnu.org ---
with -ftrivial-auto-var-init=zero, failed at the same place.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #5 from Peter Bergner <bergner at gcc dot gnu.org> ---
(In reply to qinzhao from comment #4)
> with -ftrivial-auto-var-init=zero, failed at the same place.

You mean without the patch from Comment #3?  With the patch, I don't see an ICE
using -ftrivial-auto-var-init=zero either.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[qinzhao at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #6 from qinzhao at gcc dot gnu.org ---
(In reply to Peter Bergner from comment #5)
> (In reply to qinzhao from comment #4)
> > with -ftrivial-auto-var-init=zero, failed at the same place.
> 
> You mean without the patch from Comment #3? 

Yes.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[qinzhao at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #7 from qinzhao at gcc dot gnu.org ---
(In reply to Peter Bergner from comment #3)
> (In reply to Andrew Pinski from comment #1)
> > The types are OPAQUE_TYPE.
> [snip]
> > So if I understand this correctly and PR 98872 correctly. We should not
> > expand a DEFERRED_INIT for this type.
> 
> So something like this?  Ie, it's ok to just skip this altogether?  It does
> fix this particular ICE.

not sure whether this is too conservative or not.
if a OPAQUE_TYPE variable is in memory, we might be able to initialize it with
memset. 
however, when it is in register, we might need to skip the initialization. 
> 
> 
> diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
> index 9e10da0ad5c..200d14337d7 100644
> --- a/gcc/internal-fn.c
> +++ b/gcc/internal-fn.c
> @@ -3041,6 +3041,9 @@ expand_DEFERRED_INIT (internal_fn, gcall *stmt)
>    tree var_type = TREE_TYPE (lhs);
>    gcc_assert (init_type > AUTO_INIT_UNINITIALIZED);
>  
> +  if (OPAQUE_TYPE_P (var_type))
> +    return;
> +
>    if (TREE_CODE (lhs) == SSA_NAME)
>      reg_lhs = true;
>    else

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #8 from Peter Bergner <bergner at gcc dot gnu.org> ---
(In reply to qinzhao from comment #7)
> (In reply to Peter Bergner from comment #3)
> > (In reply to Andrew Pinski from comment #1)
> > > The types are OPAQUE_TYPE.
> > [snip]
> > > So if I understand this correctly and PR 98872 correctly. We should not
> > > expand a DEFERRED_INIT for this type.
> > 
> > So something like this?  Ie, it's ok to just skip this altogether?  It does
> > fix this particular ICE.
> 
> not sure whether this is too conservative or not.
> if a OPAQUE_TYPE variable is in memory, we might be able to initialize it
> with memset. 
> however, when it is in register, we might need to skip the initialization. 
> > 
> > 
> > diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
> > index 9e10da0ad5c..200d14337d7 100644
> > --- a/gcc/internal-fn.c
> > +++ b/gcc/internal-fn.c
> > @@ -3041,6 +3041,9 @@ expand_DEFERRED_INIT (internal_fn, gcall *stmt)
> >    tree var_type = TREE_TYPE (lhs);
> >    gcc_assert (init_type > AUTO_INIT_UNINITIALIZED);
> >  
> > +  if (OPAQUE_TYPE_P (var_type))
> > +    return;
> > +
> >    if (TREE_CODE (lhs) == SSA_NAME)
> >      reg_lhs = true;
> >    else

So this then?

diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
index 9e10da0ad5c..783781aed6a 100644
--- a/gcc/internal-fn.c
+++ b/gcc/internal-fn.c
@@ -3070,6 +3070,9 @@ expand_DEFERRED_INIT (internal_fn, gcall *stmt)
     }
   else
     {
+      if (OPAQUE_TYPE_P (var_type))
+       return;
+
       /* If this variable is in a register use expand_assignment.  */
       tree init;
       if (tree_fits_uhwi_p (var_size)

This also fixes the ICE and generates a memset when the test case looks like:

void
foo (__vector_quad *dst)
{
  __vector_quad acc[16];
  *dst = acc[0];
}

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[qing.zhao at oracle dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #9 from Qing Zhao <qing.zhao at oracle dot com> ---
> On Nov 8, 2021, at 1:56 PM, bergner at gcc dot gnu.org <gcc-bugzilla@gcc.gnu.org> wrote:
> 
> 
> So this then?

This is better, I think.
You can send a patch review request to gcc-patch alias for more comments or
approval.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #10 from Peter Bergner <bergner at gcc dot gnu.org> ---
(In reply to Qing Zhao from comment #9)
> This is better, I think.
> You can send a patch review request to gcc-patch alias for more comments or
> approval.

Ok, if people are fine with this, I'll kick off a bootstrap and regtest of this
patch and submit if clean.  Thanks.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[qinzhao at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #11 from qinzhao at gcc dot gnu.org ---
Please see the details of the discussion of the patch for this bug:

https://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg275509.html

as a summary, we decide to exclude OPAQUE_TYPE from auto-init at this time.
so the fix will be in the routine “is_var_need_auto_init” in gimplify.c to
exclude OPAQUE_TYPE from auto-init.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

--- Comment #12 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Peter Bergner <bergner@gcc.gnu.org>:

https://gcc.gnu.org/g:5b1ef8b9db964ec2375df29a73d2b1651afe7ea9

commit r12-5698-g5b1ef8b9db964ec2375df29a73d2b1651afe7ea9
Author: Peter Bergner <bergner@linux.ibm.com>
Date:   Wed Dec 1 13:17:52 2021 -0600

    middle-end: Skip initialization of opaque type variables [PR103127]

    For -ftrivial-auto-var-init=*, skip initializing the variable if it is an
    opaque type, because CONST0_RTX(mode) is not defined for opaque modes.

    2021-12-01  Peter Bergner  <bergner@linux.ibm.com>

    gcc/
            PR middle-end/103127
            * gimplify.c (is_var_need_auto_init): Handle opaque types.

    gcc/testsuite/
            PR middle-end/103127
            * gcc.target/powerpc/pr103127.c: New test.

[Bug middle-end/103127] ICE in fold_convert_loc with __vector_quad and -ftrivial-auto-var-init=pattern on powerpc64*

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103127

Peter Bergner <bergner at gcc dot gnu.org> changed:

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

--- Comment #13 from Peter Bergner <bergner at gcc dot gnu.org> ---
Fixed on trunk.