[Bug c/52883] New: ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[Bug c/52883] New: ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[regehr at cs dot utah.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

             Bug #: 52883
           Summary: ICE in simplify_const_unary_operation, at
                    simplify-rtx.c:1464
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: regehr@cs.utah.edu
                CC: chenyang@cs.utah.edu


[regehr@dyson r2]$ current-gcc -c -O small.c
small.c: In function 'fn2':
small.c:21:1: internal compiler error: in simplify_const_unary_operation, at
simplify-rtx.c:1464
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.


[regehr@dyson r2]$ cat small.c 
int a, b, d, e, f, i, j, k, l, m;
unsigned c;
int g[] = { }
          , h[0];
int
fn1 () {
  return 0;
}
void
fn2 () {
  c = 0;
  e = 0;
  for (;; e = 0)
    if (f > j) {
      k = fn1 ();
      l = (d || k) * b;
      m = l * a;
      h[0] = m <= i;
    } else
      i = g[c];
}


[regehr@dyson r2]$ current-gcc -v
Using built-in specs.
COLLECT_GCC=current-gcc
COLLECT_LTO_WRAPPER=/uusoc/exports/scratch/regehr/z/compiler-install/gcc-r186167-install/bin/../libexec/gcc/x86_64-unknown-linux-gnu/4.8.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../configure --with-libelf=/usr/local --enable-lto
--prefix=/home/regehr/z/compiler-install/gcc-r186167-install
--program-prefix=r186167- --enable-languages=c,c++
Thread model: posix
gcc version 4.8.0 20120405 (experimental) (GCC)

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2012-04-06
         AssignedTo|unassigned at gcc dot       |ubizjak at gmail dot com
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-06 09:13:23 UTC ---
Confirmed.

simplify_const_unary_operation tries to simplify:

(zero_extend:DI (const_int 0 [0]))

and trips on assert where:

      /* When zero-extending a CONST_INT, we need to know its
             original mode.  */

There is nothing wrong with the above RTX, so there is no justification to
trigger the ICE. The compiler should simply reject to simplify this RTX.

I am testing following patch:

--cut here--
Index: simplify-rtx.c
===================================================================
--- simplify-rtx.c      (revision 186182)
+++ simplify-rtx.c      (working copy)
@@ -1461,7 +1461,9 @@ simplify_const_unary_operation (enum rtx_code code
        case ZERO_EXTEND:
          /* When zero-extending a CONST_INT, we need to know its
              original mode.  */
-         gcc_assert (op_mode != VOIDmode);
+         if (op_mode == VOIDmode)
+           return 0;
+
          if (op_width == HOST_BITS_PER_WIDE_INT)
            {
              /* If we were really extending the mode,
@@ -1628,9 +1630,8 @@ simplify_const_unary_operation (enum rtx_code code
          break;

        case ZERO_EXTEND:
-         gcc_assert (op_mode != VOIDmode);
-
-         if (op_width > HOST_BITS_PER_WIDE_INT)
+         if (op_mode == VOIDmode
+             || op_width > HOST_BITS_PER_WIDE_INT)
            return 0;

          hv = 0;
--cut here--

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #2 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-06 09:26:33 UTC ---
Uncovered by 4.8 change to allow immediate operands in *zero_extendsidi2_rex64:

(define_insn "*zero_extendsidi2_rex64"
  [(set (match_operand:DI 0 "nonimmediate_operand"
            "=r ,o,?*Ym,?*y,?*Yi,!*x")
    (zero_extend:DI
     (match_operand:SI 1 "x86_64_zext_general_operand"
                "rmZ,0,r   ,m  ,r   ,m*x")))]
  "TARGET_64BIT"
  "@
   mov{l}\t{%1, %k0|%k0, %1}
   #
  ...)

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

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

--- Comment #3 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-04-06 10:55:59 UTC ---
> simplify_const_unary_operation tries to simplify:
> 
> (zero_extend:DI (const_int 0 [0]))
> 
> and trips on assert where:
> 
>       /* When zero-extending a CONST_INT, we need to know its
>              original mode.  */
> 
> There is nothing wrong with the above RTX, so there is no justification to
> trigger the ICE.

Just that it makes no sense.  Please fix whoever generated it instead.

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #4 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-06 11:18:35 UTC ---
(In reply to comment #3)
> > simplify_const_unary_operation tries to simplify:
> > 
> > (zero_extend:DI (const_int 0 [0]))
> > 
> > and trips on assert where:
> > 
> >       /* When zero-extending a CONST_INT, we need to know its
> >              original mode.  */
> > 
> > There is nothing wrong with the above RTX, so there is no justification to
> > trigger the ICE.
> 
> Just that it makes no sense.  Please fix whoever generated it instead.

It is allowed in *zero_extendsidi2_rex64, and "0" propagates into the pattern
for some corner case. For the affected part, 4.7 generates:

.L3:
        movl    $0, %eax        # 42    *zero_extendsidi2_rex64/1
        movl    g(,%rax,4), %ebx        # 44    *movsi_internal/1
        jmp     .L8     # 60    jump

while 4.8 (with the above fix) generates:

.L3:
        movl    $0, %eax        # 42    *zero_extendsidi2_rex64/1
        movl    g(,%rax,4), %ebx        # 44    *movsi_internal/1
        jmp     .L8     # 60    jump

I fail to see why zero_extending a constant should trigger an ICE, while at the
same time sign_extending a constant is allowed (please see following lines in
both cases).

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-06 12:00:56 UTC ---
(In reply to comment #4)

> It is allowed in *zero_extendsidi2_rex64, and "0" propagates into the pattern
> for some corner case. For the affected part, 4.7 generates:

Oops, this is the correct sequence:

.L3:
        movl    $0, %edx        # 66    *movsi_internal/1
        movl    %edx, %eax      # 42    *zero_extendsidi2_rex64/1
        movl    g(,%rax,4), %ebx        # 44    *movsi_internal/1
        jmp     .L8     # 62    jump    [length = 2]

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |NEW
         AssignedTo|ubizjak at gmail dot com    |unassigned at gcc dot
                   |                            |gnu.org

--- Comment #6 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-06 12:36:06 UTC ---
(In reply to comment #3)

> Just that it makes no sense.  Please fix whoever generated it instead.

OK, tracing this a bit deeper:

#0  internal_error (gmsgid=0x111f5e0 "in %s, at %s:%d") at
../../gcc-svn/trunk/gcc/diagnostic.c:839
#1  0x0000000000e1cda4 in fancy_abort (file=<optimized out>, line=1464,
function=0xef4380 "simplify_const_unary_operation")
    at ../../gcc-svn/trunk/gcc/diagnostic.c:899
#2  0x0000000000881783 in simplify_const_unary_operation (code=<optimized out>,
mode=DImode, op=0x7ffff19ad470, op_mode=<optimized out>)
    at ../../gcc-svn/trunk/gcc/simplify-rtx.c:1464
#3  0x000000000087f4d0 in simplify_unary_operation (code=ZERO_EXTEND,
mode=DImode, op=0x7ffff19ad470, op_mode=VOIDmode)
    at ../../gcc-svn/trunk/gcc/simplify-rtx.c:579
#4  0x00000000005ec860 in cselib_expand_value_rtx_1 (orig=<optimized out>,
evd=0x7fffffffd950, max_depth=<optimized out>)
    at ../../gcc-svn/trunk/gcc/cselib.c:1805

The problem starts at cselib.c, where:

  /* If an operand has been simplified into CONST_INT, which doesn't
     have a mode and the mode isn't derivable from whole rtx's mode,
     try simplify_*_operation first with mode from original's operand
     and as a fallback wrap CONST_INT into gen_rtx_CONST.  */

We do have specializations for this case:

  switch (GET_RTX_CLASS (code))
    {
    case RTX_UNARY:
      if (CONST_INT_P (XEXP (copy, 0))
      && GET_MODE (XEXP (orig, 0)) != VOIDmode)
    {
      scopy = simplify_unary_operation (code, mode, XEXP (copy, 0),
                        GET_MODE (XEXP (orig, 0)));
      if (scopy)
        return scopy;
    }
      break;
  ...

But, since we have VOIDmode operand, we should do as the comment says and wrap
CONST_INT with CONST. However, we directly hit:

  scopy = simplify_rtx (copy);
  if (scopy)
    return scopy;
  return copy;

where simplify_rtx just passes problematic RTX to simplify_unary_operation.

Leaving this to RTX experts, since the code does not match the comment.

[Bug rtl-optimization/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |ebotcazou at gcc dot
                   |gnu.org                     |gnu.org

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |NEW
          Component|rtl-optimization            |target
         AssignedTo|ebotcazou at gcc dot        |unassigned at gcc dot
                   |gnu.org                     |gnu.org

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-04-08 13:36:16 UTC ---
> Leaving this to RTX experts, since the code does not match the comment.

The problem is elsewhere though, namely in the zero_extendsidi2_rex64 pattern:

(define_insn "*zero_extendsidi2_rex64"
  [(set (match_operand:DI 0 "nonimmediate_operand"
            "=r ,o,?*Ym,?*y,?*Yi,!*x")
    (zero_extend:DI
     (match_operand:SI 1 "x86_64_zext_general_operand"
                "rmZ,0,r   ,m  ,r   ,m*x")))]
  "TARGET_64BIT"

The problematic RTL is generated by reload because the Z constraint is invalid
in the above pattern.  Excerpt from the manual:

"For all conversion operations, X must not be `VOIDmode' because the
mode in which to do the conversion would not be known.  The conversion
must either be done at compile-time or X must be placed into a register.

`(sign_extend:M X)'
     Represents the result of sign-extending the value X to machine
     mode M.  M must be a fixed-point mode and X a fixed-point value of
     a mode narrower than M.

`(zero_extend:M X)'
     Represents the result of zero-extending the value X to machine
     mode M.  M must be a fixed-point mode and X a fixed-point value of
     a mode narrower than M."

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |ubizjak at gmail dot com
                   |gnu.org                     |

--- Comment #8 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-09 07:29:10 UTC ---
(In reply to comment #7)

> "For all conversion operations, X must not be `VOIDmode' because the
> mode in which to do the conversion would not be known.  The conversion
> must either be done at compile-time or X must be placed into a register.

I was not aware of this little detail...

Taking the bug.

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #9 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-09 08:07:47 UTC ---
Created attachment 27116
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27116
Untested patch

Patch in testing.

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[uros at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #10 from uros at gcc dot gnu.org 2012-04-09 09:57:18 UTC ---
Author: uros
Date: Mon Apr  9 09:57:13 2012
New Revision: 186243

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=186243
Log:
    PR target/52883
    * config/i386/predicates.md (x86_64_zext_general_operand): Prevent
    VOIDmode immediate operands.
    * config/i386/constraints.md (Wz): New constraint.
    * config/i386/i386.md (*zero_extendsidi2_rex64): Use Wz instead of Z.

testsuite/ChangeLog:

    PR target/52883
    * gcc.target/i386/pr52883.c: New testcase.


Added:
    trunk/gcc/testsuite/gcc.target/i386/pr52883.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/constraints.md
    trunk/gcc/config/i386/i386.md
    trunk/gcc/config/i386/predicates.md
    trunk/gcc/testsuite/ChangeLog

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86
             Status|ASSIGNED                    |RESOLVED
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2012-04/msg00424.htm
                   |                            |l
         Resolution|                            |FIXED

--- Comment #11 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-09 10:02:40 UTC ---
Fixed.

[Bug target/52883] ICE in simplify_const_unary_operation, at simplify-rtx.c:1464

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52883

--- Comment #12 from Uros Bizjak <ubizjak at gmail dot com> 2012-04-09 18:59:50 UTC ---
*** Bug 52914 has been marked as a duplicate of this bug. ***