[Bug middle-end/66984] New: ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[Bug middle-end/66984] New: ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gary at intrepid dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

            Bug ID: 66984
           Summary: ICE: fold_binary changes type of operand, causing
                    failure in verify_gimple_assign_binary
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gary at intrepid dot com
  Target Milestone: ---

This problem was previously reported in bug #46679 and bug #56363.  I'm logging
it as new issue because it occurred as a recent regression in 6.0, though the
changes made to 6.0 contributed only in that improved tree folding exposed this
problem in our testing of UPC on the GUPC branch.  As in the two bug reports
mentioned above, we can't easily replicate the issue, because the UPC code
generator uses FLOOR_DIV_EXPR, but "C" generally does not.

The tree verify failure occurs using -m32 (x86), because on that target the
size of 'int' equals the size of 'size_t' and they have the same mode.

The test case fails as follows:

 timer_reduce.upc: In function ‘timers_reduce’:
 timer_reduce.upc:3:6: error: type mismatch in binary expression
  void timers_reduce(double *max_t, int j){
       ^
 int
 sizetype
 int
 D.1986 = D.1984 /[ex] THREADS.4;

 timer_reduce.upc:3:6: internal compiler error: verify_gimple failed
 0xeb38bb verify_gimple_in_seq(gimple_statement_base*)
        /eng/upc/dev/gary/gupc-merge/src/gupc/gcc/tree-cfg.c:4805
 0xbc91b6 gimplify_body(tree_node*, bool)
        /eng/upc/dev/gary/gupc-merge/src/gupc/gcc/gimplify.c:9260
 0xbc960d gimplify_function_tree(tree_node*)
        /eng/upc/dev/gary/gupc-merge/src/gupc/gcc/gimplify.c:9349
 [...]

The operand type is changed here in fold-const.c:

10820         /* If arg0 is a multiple of arg1, then rewrite to the fastest div
10821            operation, EXACT_DIV_EXPR.
10822
10823            Note that only CEIL_DIV_EXPR and FLOOR_DIV_EXPR are rewritten
now.
10824            At one time others generated faster code, it's not clear if
they do
10825            after the last round to changes to the DIV code in expmed.c. 
*/
10826         if ((code == CEIL_DIV_EXPR || code == FLOOR_DIV_EXPR)
10827             && multiple_of_p (type, arg0, arg1))
10828           return fold_build2_loc (loc, EXACT_DIV_EXPR, type, arg0, arg1);

(gdb) p code
$24 = FLOOR_DIV_EXPR
(gdb) p type
$25 = (tree) 0x7ffff188c7e0
(gdb) pt
warning: Expression is not an assignment (and might have no effect)
 <integer_type 0x7ffff188c7e0 int public SI
    size <integer_cst 0x7ffff1889ca8 type <integer_type 0x7ffff188c2a0
bitsizetype> constant 32>
    unit size <integer_cst 0x7ffff1889cc0 type <integer_type 0x7ffff188c1f8
sizetype> constant 4>
    align 32 symtab 0 alias set -1 canonical type 0x7ffff188c7e0 precision 32
min <integer_cst 0x7ffff1889ee8 -2147483648> max <integer_cst 0x7ffff1889f00
2147483647>
    pointer_to_this <pointer_type 0x7ffff18a6a80>>

'op0' is the first operand,  'arg0' is 'op0' after applying STRIP_NOPS().

(gdb) p op0
$26 = (tree) 0x7ffff19e15c0
(gdb) pt
warning: Expression is not an assignment (and might have no effect)
 <nop_expr 0x7ffff19e15c0
    type <integer_type 0x7ffff188c7e0 int public SI
        size <integer_cst 0x7ffff1889ca8 constant 32>
        unit size <integer_cst 0x7ffff1889cc0 constant 4>
        align 32 symtab 0 alias set -1 canonical type 0x7ffff188c7e0 precision
32 min <integer_cst 0x7ffff1889ee8 -2147483648> max <integer_cst 0x7ffff1889f00
2147483647>
        pointer_to_this <pointer_type 0x7ffff18a6a80>>
    side-effects
    arg 0 <save_expr 0x7ffff19abee0
        type <integer_type 0x7ffff188c1f8 sizetype public unsigned SI size
<integer_cst 0x7ffff1889ca8 32> unit size <integer_cst 0x7ffff1889cc0 4>
            align 32 symtab 0 alias set -1 canonical type 0x7ffff188c1f8
precision 32 min <integer_cst 0x7ffff1889cd8 0> max <integer_cst 0x7ffff1889000
4294967295>>
        side-effects
        arg 0 <mult_expr 0x7ffff19de668 type <integer_type 0x7ffff188c1f8
sizetype>

            arg 0 <nop_expr 0x7ffff19abe00 type <integer_type 0x7ffff188c1f8
sizetype>
                arg 0 <parm_decl 0x7ffff19e0280 j>>
            arg 1 <nop_expr 0x7ffff19abde0 type <integer_type 0x7ffff188c1f8
sizetype>
                readonly arg 0 <var_decl 0x7ffff198d7e0 THREADS>>>
        timer_reduce.upc:4:19>>
(gdb) p arg0
$27 = (tree) 0x7ffff19abee0
(gdb) pt
warning: Expression is not an assignment (and might have no effect)
 <save_expr 0x7ffff19abee0
    type <integer_type 0x7ffff188c1f8 sizetype public unsigned SI
        size <integer_cst 0x7ffff1889ca8 constant 32>
        unit size <integer_cst 0x7ffff1889cc0 constant 4>
        align 32 symtab 0 alias set -1 canonical type 0x7ffff188c1f8 precision
32 min <integer_cst 0x7ffff1889cd8 0> max <integer_cst 0x7ffff1889000
4294967295>>
    side-effects
    arg 0 <mult_expr 0x7ffff19de668 type <integer_type 0x7ffff188c1f8 sizetype>

        arg 0 <nop_expr 0x7ffff19abe00 type <integer_type 0x7ffff188c1f8
sizetype>
            arg 0 <parm_decl 0x7ffff19e0280 j>>
        arg 1 <nop_expr 0x7ffff19abde0 type <integer_type 0x7ffff188c1f8
sizetype>
            readonly arg 0 <var_decl 0x7ffff198d7e0 THREADS>>>
    timer_reduce.upc:4:19>

As shown, the type of arg0 is 'sizetype' and op0 is 'int'.

After this rewrite is applied: fold_build2_loc (loc, EXACT_DIV_EXPR, type,
arg0, arg1),
the 'rhs1' type will be 'sizetype', but 'lhstype' and 'rhs2_type' will be
'int'.


3930      if (!useless_type_conversion_p (lhs_type, rhs1_type)
3931          || !useless_type_conversion_p (lhs_type, rhs2_type))
3932        {
3933          error ("type mismatch in binary expression");
3934          debug_generic_stmt (lhs_type);
3935          debug_generic_stmt (rhs1_type);
3936          debug_generic_stmt (rhs2_type);
3937          return true;
3938        }

We need the type conversion encoded in op0:

(gdb) p useless_type_conversion_p(type, TREE_TYPE(op0))
$28 = true
(gdb) p useless_type_conversion_p(type, TREE_TYPE(arg0))
$29 = false

useless_type_conversion_p() requires that the types agree in signedness and
precesion.

Perhaps this issue can be fixed by adding CEIL_DIV_EXPR and FLOOR_DIV_EXPR to
the if statement at the beginning of fold_binary_loc().

Index: fold-const.c
===================================================================
--- fold-const.c        (revision 226135)
+++ fold-const.c        (working copy)
@@ -9151,7 +9151,8 @@ fold_binary_loc (location_t loc,
      cases, the appropriate type conversions should be put back in
      the tree that will get out of the constant folder.  */

-  if (kind == tcc_comparison || code == MIN_EXPR || code == MAX_EXPR)
+  if (kind == tcc_comparison || code == MIN_EXPR || code == MAX_EXPR
+      code == CEIL_DIV_EXPR || code == FLOOR_DIV_EXPR)
     {
       STRIP_SIGN_NOPS (arg0);
       STRIP_SIGN_NOPS (arg1);
>From gcc-bugs-return-493163-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Fri Jul 24 07:17:48 2015
Return-Path: <gcc-bugs-return-493163-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 39708 invoked by alias); 24 Jul 2015 07:17:48 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 39686 invoked by uid 48); 24 Jul 2015 07:17:44 -0000
From: "eric.niebler at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug c++/66985] New: [concept] template introduction with template template parameter and template parameter pack causes ICE
Date: Fri, 24 Jul 2015 07:17:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: new
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: c++
X-Bugzilla-Version: c++-concepts
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: eric.niebler at gmail dot com
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Resolution:
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: bug_id short_desc product version bug_status bug_severity priority component assigned_to reporter target_milestone
Message-ID: <bug-66985-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2015-07/txt/msg02053.txt.bz2
Content-length: 1196

https://gcc.gnu.org/bugzilla/show_bug.cgi?idf985

            Bug ID: 66985
           Summary: [concept] template introduction with template template
                    parameter and template parameter pack causes ICE
           Product: gcc
           Version: c++-concepts
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: eric.niebler at gmail dot com
  Target Milestone: ---

Code:

template <template <class...> class T, class... U>
concept bool _Valid = requires { typename T<U...>; };

template <template <class...> class T, class... U>
struct __defer { };

_Valid{T, ...U}
struct __defer<T, U...> {
  using type = T<U...>;
};


Result:

$ /usr/local/gcc-concepts/bin/g++.exe -std=gnu++1z -c bug.cpp
bug.cpp:8:15: internal compiler error: in cp_parser_single_declaration, at
cp/parser.c:24808
 _Valid{T, ...U}
               ^

bug.cpp:8:15: internal compiler error: Aborted
g++: internal compiler error: Aborted (program cc1plus)
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-07-24
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The usual fix in fold-const.c is to make sure to convert operands to the
required type when building the final expression.  Thus instead of

10828           return fold_build2_loc (loc, EXACT_DIV_EXPR, type, arg0, arg1);

do

   return fold_build2_loc (loc, EXACT_DIV_EXPR, type,
                           fold_convert (type, arg0), fold_convert (type,
arg1));

you can see this pattern in many places.

Care to post a patch?  It's pre-approved.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[jay.krell at cornell dot edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #2 from Jay <jay.krell at cornell dot edu> ---
1 please be sure that dividing the most negative number by -1 "works". 
Perhaps just don't optimize anything with negstive numbers.


2 I suggest that gcc's C/C++ frontends expose these other forms of division,
for the sake of testability.


Thank you,
 - Jay

On Jul 24, 2015, at 2:09 AM, "rguenth at gcc dot gnu.org"
<gcc-bugzilla@gcc.gnu.org> wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984
> 
> Richard Biener <rguenth at gcc dot gnu.org> changed:
> 
>           What    |Removed                     |Added
> ----------------------------------------------------------------------------
>             Status|UNCONFIRMED                 |NEW
>   Last reconfirmed|                            |2015-07-24
>     Ever confirmed|0                           |1
> 
> --- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
> The usual fix in fold-const.c is to make sure to convert operands to the
> required type when building the final expression.  Thus instead of
> 
> 10828           return fold_build2_loc (loc, EXACT_DIV_EXPR, type, arg0, arg1);
> 
> do
> 
>   return fold_build2_loc (loc, EXACT_DIV_EXPR, type,
>                           fold_convert (type, arg0), fold_convert (type,
> arg1));
> 
> you can see this pattern in many places.
> 
> Care to post a patch?  It's pre-approved.
> 
> -- 
> You are receiving this mail because:
> You are on the CC list for the bug.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gary at intrepid dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #3 from Gary Funck <gary at intrepid dot com> ---
(In reply to Richard Biener from comment #1)
> The usual fix in fold-const.c is to make sure to convert operands to the
> required type when building the final expression.  Thus instead of
> 
> 10828           return fold_build2_loc (loc, EXACT_DIV_EXPR, type, arg0,
> arg1);
> 
> do
> 
>    return fold_build2_loc (loc, EXACT_DIV_EXPR, type,
>                            fold_convert (type, arg0), fold_convert (type,
> arg1));
> 
> you can see this pattern in many places.
> 
> Care to post a patch?  It's pre-approved.

OK, I'll post a patch.  Thanks.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gary at intrepid dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #4 from Gary Funck <gary at intrepid dot com> ---
(In reply to Jay from comment #2)
> 1 please be sure that dividing the most negative number by -1 "works". 
> Perhaps just don't optimize anything with negstive numbers.

- Checking for negative numbers at compile-time can only be done on integer
constant values.

- Whether dividing the max negative number by -1 just "works" depends upon the
definition and implementation of FLOOR_DIV_EXPR.

- For UPC's use of FLOOR_DIV_EXPR, it isn't possible to construct that
scenario.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gary at intrepid dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #5 from Gary Funck <gary at intrepid dot com> ---
(In reply to Jay from comment #2)
> 2 I suggest that gcc's C/C++ frontends expose these other forms of division,
> for the sake of testability.

Perhaps defining a builtin for CEIL_DIV_EXPR and FLOOR_DIV_EXPR might work.

For the builtins, you could file an RFE and post a suggested patch (with test
cases) for review.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gfunck at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #6 from gfunck at gcc dot gnu.org ---
Author: gfunck
Date: Fri Jul 24 16:10:39 2015
New Revision: 226168

URL: https://gcc.gnu.org/viewcvs?rev=226168&root=gcc&view=rev
Log:
2015-07-24  Gary Funck  <gary@intrepid.com>

        PR middle-end/66984
        * fold-const.c (fold_binary_loc): Call fold_convert on arguments to
        fold_build2 for CEIL_DIV_EXPR and FLOOR_DIV_EXPR optimization.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[gary at intrepid dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #7 from Gary Funck <gary at intrepid dot com> ---
Don't know what the criteria is for closing bugs, but as far as I'm concerned,
this bug can be marked resolved and the other two referenced PR's marked as
duplicates of this one.  (They're against older rev's, so not sure if they're
exactly duplicates, but the chances of the fix being back-ported is probably
low.)

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |mpolacek at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #8 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Assuming fixed then.

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |6.0

[Bug middle-end/66984] ICE: fold_binary changes type of operand, causing failure in verify_gimple_assign_binary

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66984

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 46679 has been marked as a duplicate of this bug. ***