[Bug rtl-optimization/110867] New: ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[Bug rtl-optimization/110867] New: ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[prathamesh3492 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

            Bug ID: 110867
           Summary: ICE in combine after
                    7cdd0860949c6c3232e6cff1d7ca37bb5234074c
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: prathamesh3492 at gcc dot gnu.org
  Target Milestone: ---

For the following test-case adapted from libgcc/fixed-bit.c:

typedef int DItype __attribute__ ((mode (DI)));

void
__gnu_saturate1sq (DItype *a)
{
  DItype max, min;
  max = (DItype)1 << (31 + 0);
  max = max - 1;

  min = (DItype)1 << (2 * (4 * 8) - 1);
  min = min >> (2 * (4 * 8) - 1 - (31 + 0));



  if (*a > max)
    *a = max;
  else if (*a < min)
    *a = min;
}

Compiling with -O2 on armv8l-unknown-linux-gnueabihf results in following ICE:
typedef int DItype __attribute__ ((mode (DI)));

during RTL pass: combine
foo.c: In function '__gnu_saturate1sq':
foo.c:19:1: internal compiler error: in decompose, at rtl.h:2297
   19 | }
      | ^
0xaa23e3 wi::int_traits<std::pair<rtx_def*, machine_mode> >::decompose(long
long*, unsigned int, std::pair<rtx_def*, machine_mode> const&)
        ../../gcc/gcc/rtl.h:2297
0xaf5ab3 wide_int_ref_storage<false,
true>::wide_int_ref_storage<std::pair<rtx_def*, machine_mode>
>(std::pair<rtx_def*, machine_mode> const&)
        ../../gcc/gcc/wide-int.h:1030
0xaf5023 generic_wide_int<wide_int_ref_storage<false, true>
>::generic_wide_int<std::pair<rtx_def*, machine_mode> >(std::pair<rtx_def*,
machine_mode> const&)
        ../../gcc/gcc/wide-int.h:788
0xf916f9 simplify_const_unary_operation(rtx_code, machine_mode, rtx_def*,
machine_mode)
        ../../gcc/gcc/simplify-rtx.cc:2131
0xf8bad5 simplify_context::simplify_unary_operation(rtx_code, machine_mode,
rtx_def*, machine_mode)
        ../../gcc/gcc/simplify-rtx.cc:889
0xf8a591 simplify_context::simplify_gen_unary(rtx_code, machine_mode, rtx_def*,
machine_mode)
        ../../gcc/gcc/simplify-rtx.cc:360
0x9bd1b7 simplify_gen_unary(rtx_code, machine_mode, rtx_def*, machine_mode)
        ../../gcc/gcc/rtl.h:3520
0x1bd5677 simplify_comparison
        ../../gcc/gcc/combine.cc:13125
0x1bc2b2b simplify_set
        ../../gcc/gcc/combine.cc:6848
0x1bc1647 combine_simplify_rtx
        ../../gcc/gcc/combine.cc:6353
0x1bbf97f subst
        ../../gcc/gcc/combine.cc:5609
0x1bb864b try_combine
        ../../gcc/gcc/combine.cc:3302
0x1bb30fb combine_instructions
        ../../gcc/gcc/combine.cc:1264
0x1bd8d25 rest_of_handle_combine
        ../../gcc/gcc/combine.cc:15059
0x1bd8dd5 execute
        ../../gcc/gcc/combine.cc:15103
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.


Thanks,
Prathamesh

[Bug rtl-optimization/110867] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[stefansf at linux dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |stefansf at linux dot ibm.com

--- Comment #1 from Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> ---
The optimization introduced by r14-2879-g7cdd0860949c6c hits during combination
of insn

(insn 31 3 32 2 (set (reg:SI 118 [ _1 ])
        (mem:SI (reg/v/f:SI 115 [ a ]) [1 *a_4(D)+0 S4 A64])) "t.c":15:7 758
{*arm_movsi_vfp}
     (nil))

and

(insn 9 32 10 2 (set (reg:CC 100 cc)
        (compare:CC (reg:SI 118 [ _1 ])
            (const_int -2147483648 [0xffffffff80000000]))) "t.c":15:6 272
{*arm_cmpsi_insn}
     (nil))

The idea of r14-2879-g7cdd0860949c6c is to get rid of large constants while
performing an unsigned comparison.  In this case it looks like a 32-bit
constant is sign-extended into a 64-bit constant and then a 32-bit comparison
is done.  While writing the optimization I always assumed that the constant
does fit into int_mode which is apparently not the case here.  Thus one
possible solution would be to simply bail out in those cases:

diff --git a/gcc/combine.cc b/gcc/combine.cc
index 0d99fa541c5..e46d202d0a7 100644
--- a/gcc/combine.cc
+++ b/gcc/combine.cc
@@ -11998,11 +11998,15 @@ simplify_compare_const (enum rtx_code code,
machine_mode mode,
      x0 >= 0x40.  */
   if ((code == LEU || code == LTU || code == GEU || code == GTU)
       && is_a <scalar_int_mode> (GET_MODE (op0), &int_mode)
+      && HWI_COMPUTABLE_MODE_P (int_mode)
       && MEM_P (op0)
       && !MEM_VOLATILE_P (op0)
       /* The optimization makes only sense for constants which are big enough
         so that we have a chance to chop off something at all.  */
       && (unsigned HOST_WIDE_INT) const_op > 0xff
+      /* Bail out, if the constant does not fit into INT_MODE.  */
+      && (unsigned HOST_WIDE_INT) const_op
+        < ((HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1) << 1) - 1)
       /* Ensure that we do not overflow during normalization.  */
       && (code != GTU || (unsigned HOST_WIDE_INT) const_op <
HOST_WIDE_INT_M1U))
     {

Does this resolve the problem for you?

[Bug rtl-optimization/110867] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[prathamesh3492 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #2 from prathamesh3492 at gcc dot gnu.org ---
(In reply to Stefan Schulze Frielinghaus from comment #1)
> The optimization introduced by r14-2879-g7cdd0860949c6c hits during
> combination of insn
> 
> (insn 31 3 32 2 (set (reg:SI 118 [ _1 ])
>         (mem:SI (reg/v/f:SI 115 [ a ]) [1 *a_4(D)+0 S4 A64])) "t.c":15:7 758
> {*arm_movsi_vfp}
>      (nil))
> 
> and
> 
> (insn 9 32 10 2 (set (reg:CC 100 cc)
>         (compare:CC (reg:SI 118 [ _1 ])
>             (const_int -2147483648 [0xffffffff80000000]))) "t.c":15:6 272
> {*arm_cmpsi_insn}
>      (nil))
> 
> The idea of r14-2879-g7cdd0860949c6c is to get rid of large constants while
> performing an unsigned comparison.  In this case it looks like a 32-bit
> constant is sign-extended into a 64-bit constant and then a 32-bit
> comparison is done.  While writing the optimization I always assumed that
> the constant does fit into int_mode which is apparently not the case here. 
> Thus one possible solution would be to simply bail out in those cases:
> 
> diff --git a/gcc/combine.cc b/gcc/combine.cc
> index 0d99fa541c5..e46d202d0a7 100644
> --- a/gcc/combine.cc
> +++ b/gcc/combine.cc
> @@ -11998,11 +11998,15 @@ simplify_compare_const (enum rtx_code code,
> machine_mode mode,
>       x0 >= 0x40.  */
>    if ((code == LEU || code == LTU || code == GEU || code == GTU)
>        && is_a <scalar_int_mode> (GET_MODE (op0), &int_mode)
> +      && HWI_COMPUTABLE_MODE_P (int_mode)
>        && MEM_P (op0)
>        && !MEM_VOLATILE_P (op0)
>        /* The optimization makes only sense for constants which are big
> enough
>          so that we have a chance to chop off something at all.  */
>        && (unsigned HOST_WIDE_INT) const_op > 0xff
> +      /* Bail out, if the constant does not fit into INT_MODE.  */
> +      && (unsigned HOST_WIDE_INT) const_op
> +        < ((HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1) << 1) -
> 1)
>        /* Ensure that we do not overflow during normalization.  */
>        && (code != GTU || (unsigned HOST_WIDE_INT) const_op <
> HOST_WIDE_INT_M1U))
>      {
> 
> Does this resolve the problem for you?

Yes, it worked thanks! I will do a full bootstrap+test with your fix and let
you know the results.

Thanks,
Prathamesh

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build, ice-on-valid-code
            Summary|ICE in combine after        |[14 Regression] ICE in
                   |7cdd0860949c6c3232e6cff1d7c |combine after
                   |a37bb5234074c               |7cdd0860949c6c3232e6cff1d7c
                   |                            |a37bb5234074c
   Target Milestone|---                         |14.0

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[prathamesh3492 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #3 from prathamesh3492 at gcc dot gnu.org ---
(In reply to prathamesh3492 from comment #2)
> (In reply to Stefan Schulze Frielinghaus from comment #1)
> > The optimization introduced by r14-2879-g7cdd0860949c6c hits during
> > combination of insn
> > 
> > (insn 31 3 32 2 (set (reg:SI 118 [ _1 ])
> >         (mem:SI (reg/v/f:SI 115 [ a ]) [1 *a_4(D)+0 S4 A64])) "t.c":15:7 758
> > {*arm_movsi_vfp}
> >      (nil))
> > 
> > and
> > 
> > (insn 9 32 10 2 (set (reg:CC 100 cc)
> >         (compare:CC (reg:SI 118 [ _1 ])
> >             (const_int -2147483648 [0xffffffff80000000]))) "t.c":15:6 272
> > {*arm_cmpsi_insn}
> >      (nil))
> > 
> > The idea of r14-2879-g7cdd0860949c6c is to get rid of large constants while
> > performing an unsigned comparison.  In this case it looks like a 32-bit
> > constant is sign-extended into a 64-bit constant and then a 32-bit
> > comparison is done.  While writing the optimization I always assumed that
> > the constant does fit into int_mode which is apparently not the case here. 
> > Thus one possible solution would be to simply bail out in those cases:
> > 
> > diff --git a/gcc/combine.cc b/gcc/combine.cc
> > index 0d99fa541c5..e46d202d0a7 100644
> > --- a/gcc/combine.cc
> > +++ b/gcc/combine.cc
> > @@ -11998,11 +11998,15 @@ simplify_compare_const (enum rtx_code code,
> > machine_mode mode,
> >       x0 >= 0x40.  */
> >    if ((code == LEU || code == LTU || code == GEU || code == GTU)
> >        && is_a <scalar_int_mode> (GET_MODE (op0), &int_mode)
> > +      && HWI_COMPUTABLE_MODE_P (int_mode)
> >        && MEM_P (op0)
> >        && !MEM_VOLATILE_P (op0)
> >        /* The optimization makes only sense for constants which are big
> > enough
> >          so that we have a chance to chop off something at all.  */
> >        && (unsigned HOST_WIDE_INT) const_op > 0xff
> > +      /* Bail out, if the constant does not fit into INT_MODE.  */
> > +      && (unsigned HOST_WIDE_INT) const_op
> > +        < ((HOST_WIDE_INT_1U << (GET_MODE_PRECISION (int_mode) - 1) << 1) -
> > 1)
> >        /* Ensure that we do not overflow during normalization.  */
> >        && (code != GTU || (unsigned HOST_WIDE_INT) const_op <
> > HOST_WIDE_INT_M1U))
> >      {
> > 
> > Does this resolve the problem for you?
> 
> Yes, it worked thanks! I will do a full bootstrap+test with your fix and let
> you know the results.
Bootstrap+testing works fine with your fix. Thanks!
> 
> Thanks,
> Prathamesh

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[stefansf at linux dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #4 from Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> ---
Thanks for testing so quickly :)

I've send a patch for review:

https://gcc.gnu.org/pipermail/gcc-patches/2023-August/626075.html

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Stefan Schulze Frielinghaus
<stefansf@gcc.gnu.org>:

https://gcc.gnu.org/g:41ef5a34161356817807be3a2e51fbdbe575ae85

commit r14-2932-g41ef5a34161356817807be3a2e51fbdbe575ae85
Author: Stefan Schulze Frielinghaus <stefansf@linux.ibm.com>
Date:   Wed Aug 2 21:43:22 2023 +0200

    rtl-optimization/110867 Fix narrow comparison of memory and constant

    In certain cases a constant may not fit into the mode used to perform a
    comparison.  This may be the case for sign-extended constants which are
    used during an unsigned comparison as e.g. in

    (set (reg:CC 100 cc)
        (compare:CC (mem:SI (reg/v/f:SI 115 [ a ]) [1 *a_4(D)+0 S4 A64])
            (const_int -2147483648 [0xffffffff80000000])))

    Fixed by ensuring that the constant fits into comparison mode.

    Furthermore, on some targets as e.g. sparc the constant used in a
    comparison is chopped off before combine which leads to failing test
    cases (see PR 110869).  Fixed by not requiring that the source mode has
    to be DImode, and excluding sparc from the last two test cases entirely
    since there the constant cannot be further reduced.

    gcc/ChangeLog:

            PR rtl-optimization/110867
            * combine.cc (simplify_compare_const): Try the optimization only
            in case the constant fits into the comparison mode.

    gcc/testsuite/ChangeLog:

            PR rtl-optimization/110869
            * gcc.dg/cmp-mem-const-1.c: Relax mode for constant.
            * gcc.dg/cmp-mem-const-2.c: Relax mode for constant.
            * gcc.dg/cmp-mem-const-3.c: Relax mode for constant.
            * gcc.dg/cmp-mem-const-4.c: Relax mode for constant.
            * gcc.dg/cmp-mem-const-5.c: Exclude sparc since here the
            constant is already reduced.
            * gcc.dg/cmp-mem-const-6.c: Exclude sparc since here the
            constant is already reduced.

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |panchenghui at loongson dot cn

--- Comment #6 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
*** Bug 110939 has been marked as a duplicate of this bug. ***

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xry111 at gcc dot gnu.org

--- Comment #7 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
Can we close it as fixed now?

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[stefansf at linux dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #8 from Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> ---
Created attachment 55716
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55716&action=edit
Really fix narrow comparison

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[stefansf at linux dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #9 from Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> ---
It looks like as if the first fix didn't entirely solve the problem.  It turns
out that the normal form of const_int is not always met.  Before releasing a
new patch, could you test it first in order to make sure that I do not break
bootstrapping again.  I already gave it a try against the reproducer but would
like to make sure that the whole bootstrap is successful.

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[prathamesh3492 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

--- Comment #10 from prathamesh3492 at gcc dot gnu.org ---
(In reply to Stefan Schulze Frielinghaus from comment #9)
> It looks like as if the first fix didn't entirely solve the problem.  It
> turns out that the normal form of const_int is not always met.  Before
> releasing a new patch, could you test it first in order to make sure that I
> do not break bootstrapping again.  I already gave it a try against the
> reproducer but would like to make sure that the whole bootstrap is
> successful.

Hi Stefan,
I bootstrapped+tested your patch from Comment 8 on arm, and it seems OK.

Thanks,
Prathamesh

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

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

--- Comment #11 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
The patch is pushed at r14-4355.

*** This bug has been marked as a duplicate of bug 110939 ***

[Bug rtl-optimization/110867] [14 Regression] ICE in combine after 7cdd0860949c6c3232e6cff1d7ca37bb5234074c

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110867

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|DUPLICATE                   |FIXED

--- Comment #12 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
Mistakenly marked it as dup.