[Bug middle-end/55235] New: [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[Bug middle-end/55235] New: [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

             Bug #: 55235
           Summary: [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ubizjak@gmail.com
                CC: ebotcazou@gcc.gnu.org
            Target: i686


Revision 192641 [1] introduced following testsuite failure on i686:

FAIL: gcc.target/i386/pr44948-2a.c (internal compiler error)
FAIL: gcc.target/i386/pr44948-2a.c (test for excess errors)

[1] http://gcc.gnu.org/ml/gcc-cvs/2012-10/msg00764.html

 ~/gcc-build-fast/gcc/cc1 -O -m32 -mno-sse -quiet pr44948-2a.c 
pr44948-2a.c: In function ‘main’:
pr44948-2a.c:18:12: internal compiler error: in emit_block_move_hints, at
expr.c:1153
   struct A a = { { 0, 1, 2, 3 } };
            ^
0x6b9f68 emit_block_move_hints(rtx_def*, rtx_def*, rtx_def*, block_op_methods,
unsigned int, long)
        ../../gcc-svn/trunk/gcc/expr.c:1153
0x6ba261 emit_block_move(rtx_def*, rtx_def*, rtx_def*, block_op_methods)
        ../../gcc-svn/trunk/gcc/expr.c:1205
0x6c5e57 store_expr(tree_node*, rtx_def*, int, bool)
        ../../gcc-svn/trunk/gcc/expr.c:5261
0x6ca835 store_field
        ../../gcc-svn/trunk/gcc/expr.c:6531
0x6c3f57 expand_assignment(tree_node*, tree_node*, bool)
        ../../gcc-svn/trunk/gcc/expr.c:4838
0x5fc845 expand_gimple_stmt_1
        ../../gcc-svn/trunk/gcc/cfgexpand.c:2108
0x5fc845 expand_gimple_stmt
        ../../gcc-svn/trunk/gcc/cfgexpand.c:2204
0x5fe137 expand_gimple_basic_block
        ../../gcc-svn/trunk/gcc/cfgexpand.c:3979
0x60034b gimple_expand_cfg
        ../../gcc-svn/trunk/gcc/cfgexpand.c:4498
Please submit a full bug report,

(gdb) f 2
#2  0x00000000006b9f69 in emit_block_move_hints (x=0x7ffff1aca150,
y=0x7ffff1ab55b0, size=0x7ffff19a0570, method=BLOCK_OP_NORMAL,
expected_align=0, expected_size=-1)
    at ../../gcc-svn/trunk/gcc/expr.c:1153
1153      gcc_assert (MEM_P (x) && MEM_P (y));
(gdb) lis
1148
1149        default:
1150          gcc_unreachable ();
1151        }
1152
1153      gcc_assert (MEM_P (x) && MEM_P (y));
1154      align = MIN (MEM_ALIGN (x), MEM_ALIGN (y));
1155      gcc_assert (align >= BITS_PER_UNIT);
1156
1157      /* Make sure we've got BLKmode addresses; store_one_arg can decide
that

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-07 17:39:52 UTC ---
(gdb) p debug_rtx (x)
(mem/j/c:BLK (plus:SI (reg/f:SI 54 virtual-stack-vars)
        (const_int -16 [0xfffffffffffffff0])) [0 a.V4SF+0 S16 A128])
$1 = void
(gdb) p debug_rtx (y)
(const_vector:V4SF [
        (const_double:SF 0.0 [0x0.0p+0])
        (const_double:SF 1.0e+0 [0x0.8p+1])
        (const_double:SF 2.0e+0 [0x0.8p+2])
        (const_double:SF 3.0e+0 [0x0.cp+2])
    ])
$2 = void

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-11-07
     Ever Confirmed|0                           |1

--- Comment #2 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-11-07 18:57:10 UTC ---
The test looks nonsensical: it mixes vectors defined with -mno-sse and later
used with -msse2.  Moving struct A to within 'main' eliminates the ICE,
presumably the structure somehow sees the -msse2.

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

--- Comment #3 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-07 19:05:33 UTC ---
It looks to me that this is the problematic change:

--- trunk/gcc/expr.c    2012/10/20 20:16:27    192640
+++ trunk/gcc/expr.c    2012/10/20 21:00:23    192641

(...)

@@ -5225,21 +5244,26 @@
      supposed to be bit-copied or bit-initialized.  */
       && expr_size (exp) != const0_rtx)
     {
-      if (GET_MODE (temp) != GET_MODE (target)
-      && GET_MODE (temp) != VOIDmode)
+      if (GET_MODE (temp) != GET_MODE (target) && GET_MODE (temp) != VOIDmode)
     {
-      int unsignedp = TYPE_UNSIGNED (TREE_TYPE (exp));
-      if (GET_MODE (target) == BLKmode
-          && GET_MODE (temp) == BLKmode)
-        emit_block_move (target, temp, expr_size (exp),
-                 (call_param_p
-                  ? BLOCK_OP_CALL_PARM
-                  : BLOCK_OP_NORMAL));
-      else if (GET_MODE (target) == BLKmode)
-        store_bit_field (target, INTVAL (expr_size (exp)) * BITS_PER_UNIT,
-                 0, 0, 0, GET_MODE (temp), temp);
+      if (GET_MODE (target) == BLKmode)
+        {
+          if (REG_P (temp))
+            {
+          if (TREE_CODE (exp) == CALL_EXPR)
+            copy_blkmode_from_reg (target, temp, TREE_TYPE (exp));
+          else
+            store_bit_field (target,
+                     INTVAL (expr_size (exp)) * BITS_PER_UNIT,
+                     0, 0, 0, GET_MODE (temp), temp);
+        }
+          else
+        emit_block_move (target, temp, expr_size (exp),
+                 (call_param_p
+                  ? BLOCK_OP_CALL_PARM : BLOCK_OP_NORMAL));
+        }

emit_block_move was guarded with BLKmode guard on both operands, all other went
through store_bit_field.

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

--- Comment #4 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-07 20:17:45 UTC ---
Please note that before your patch, emit_block_move was never called.

Following patch fixes testcase for me:

--cut here--
Index: expr.c
===================================================================
--- expr.c      (revision 193296)
+++ expr.c      (working copy)
@@ -5246,19 +5246,12 @@ store_expr (tree exp, rtx target, int call_param_p
        {
          if (GET_MODE (target) == BLKmode)
            {
-             if (REG_P (temp))
-               {
-                 if (TREE_CODE (exp) == CALL_EXPR)
-                   copy_blkmode_from_reg (target, temp, TREE_TYPE (exp));
-                 else
-                   store_bit_field (target,
-                                    INTVAL (expr_size (exp)) * BITS_PER_UNIT,
-                                    0, 0, 0, GET_MODE (temp), temp);
-               }
+             if (REG_P (temp) && TREE_CODE (exp) == CALL_EXPR)
+               copy_blkmode_from_reg (target, temp, TREE_TYPE (exp));
              else
-               emit_block_move (target, temp, expr_size (exp),
-                                (call_param_p
-                                 ? BLOCK_OP_CALL_PARM : BLOCK_OP_NORMAL));
+               store_bit_field (target,
+                                INTVAL (expr_size (exp)) * BITS_PER_UNIT,
+                                0, 0, 0, GET_MODE (temp), temp);
            }
          else
            convert_move (target, temp, TYPE_UNSIGNED (TREE_TYPE (exp)));
--cut here--

I am running bootstrap and regression test.

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-11-07 22:07:00 UTC ---
> Please note that before your patch, emit_block_move was never called.

Indeed, I missed that.

> Following patch fixes testcase for me:
> 
> --cut here--
> Index: expr.c
> ===================================================================
> --- expr.c      (revision 193296)
> +++ expr.c      (working copy)
> @@ -5246,19 +5246,12 @@ store_expr (tree exp, rtx target, int call_param_p
>         {
>           if (GET_MODE (target) == BLKmode)
>             {
> -             if (REG_P (temp))
> -               {
> -                 if (TREE_CODE (exp) == CALL_EXPR)
> -                   copy_blkmode_from_reg (target, temp, TREE_TYPE (exp));
> -                 else
> -                   store_bit_field (target,
> -                                    INTVAL (expr_size (exp)) * BITS_PER_UNIT,
> -                                    0, 0, 0, GET_MODE (temp), temp);
> -               }
> +             if (REG_P (temp) && TREE_CODE (exp) == CALL_EXPR)
> +               copy_blkmode_from_reg (target, temp, TREE_TYPE (exp));
>               else
> -               emit_block_move (target, temp, expr_size (exp),
> -                                (call_param_p
> -                                 ? BLOCK_OP_CALL_PARM : BLOCK_OP_NORMAL));
> +               store_bit_field (target,
> +                                INTVAL (expr_size (exp)) * BITS_PER_UNIT,
> +                                0, 0, 0, GET_MODE (temp), temp);
>             }
>           else
>             convert_move (target, temp, TYPE_UNSIGNED (TREE_TYPE (exp)));

This looks fine, although this is papering over the real problem:

(gdb) p debug_tree(field)
 <field_decl 0x7ffff6dfe390 V4SF
    type <vector_type 0x7ffff6df75e8
        type <real_type 0x7ffff6d1de70 float SF
            size <integer_cst 0x7ffff6d09d40 constant 32>
            unit size <integer_cst 0x7ffff6d09d60 constant 4>
            align 32 symtab 0 alias set -1 canonical type 0x7ffff6d1de70
precision 32
            pointer_to_this <pointer_type 0x7ffff6d250a8>>
        sizes-gimplified V4SF
        size <integer_cst 0x7ffff6d212a0 constant 128>
        unit size <integer_cst 0x7ffff6d212c0 constant 16>
        align 128 symtab 0 alias set -1 canonical type 0x7ffff6df75e8 nunits 4>
    BLK file pr44948-2a.c line 12 col 9 size <integer_cst 0x7ffff6d212a0 128>
unit size <integer_cst 0x7ffff6d212c0 16>
    align 128 offset_align 128
    offset <integer_cst 0x7ffff6d09d80 type <integer_type 0x7ffff6d1d000
sizetype> constant 0>
    bit offset <integer_cst 0x7ffff6d09e00 type <integer_type 0x7ffff6d1d0a8
bitsizetype> constant 0> context <record_type 0x7ffff6e342a0 A>>

(gdb) p field->decl_common.mode
$20 = BLKmode

(gdb) p field->typed.type->type_common.mode
$21 = V4SFmode

The field has BLKmode as DECL_MODE but V4SFmode as TYPE_MODE (TREE_TYPE) so a
VIEW_CONVERT_EXPR is missing somewhere if we need to support this nonsense...

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[uros at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

--- Comment #6 from uros at gcc dot gnu.org 2012-11-07 22:34:53 UTC ---
Author: uros
Date: Wed Nov  7 22:34:48 2012
New Revision: 193314

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193314
Log:
    PR middle-end/55235
    * expr.c (store_expr): Do not call emit_block_move for
    non-BLKmode values.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expr.c

[Bug middle-end/55235] [4.8 Regression] FAIL: gcc.target/i386/pr44948-2a.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55235

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-08 13:12:56 UTC ---
Fixed.