[Bug tree-optimization/61375] New: ICE in int_cst_value at -O3 in tree-ssa pass when compiling a reference to an __int128 value

["gary at intrepid dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61375

            Bug ID: 61375
           Summary: ICE in int_cst_value at -O3 in tree-ssa pass when
                    compiling a reference to an __int128 value
           Product: gcc
           Version: 4.10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gary at intrepid dot com
                CC: nenad at intrepid dot com
            Target: x86_64

Created attachment 32880
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=32880&action=edit
pre-processed "C" source, generates ICE at -O3 in tree-ssa

When compiling the attached pre-processed source file at -O3, the following
stack trace is generated.

$ gcc -O3 -fpreprocessed -c ice-int_cst_value-int128.i
ice-int_cst_value-int128.i: In function âtest15â:
ice-int_cst_value-int128.i:442:1: internal compiler error: in int_cst_value, at
tree.c:10492
 test15 ()
 ^
0xe3fb99 int_cst_value(tree_node const*)
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree.c:10492
0xd03709 find_bswap_or_nop_1
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:1848
0xd0354c find_bswap_or_nop_1
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:1815
0xd0354c find_bswap_or_nop_1
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:1815
0xd039a9 find_bswap_or_nop_1
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:1916
0xd03f6a find_bswap_or_nop
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:2034
0xd04ac3 execute
        /eng/upc/dev/gary/gupc-gcc-trunk/src/gcc/tree-ssa-math-opts.c:2322

The compiler is based on:

Last Changed Rev: 211071
Last Changed Date: 2014-05-29 17:17:10 -0700

The test case can likely be reduced.  The issue seems to relate to an attempt
to simplify an expression that mixes an address and an __int128 value.

The ICE triggered here:

10488     unsigned bits = TYPE_PRECISION (TREE_TYPE (x));
10489     unsigned HOST_WIDE_INT val = TREE_INT_CST_LOW (x);
10490
10491     /* Make sure the sign-extended value will fit in a HOST_WIDE_INT.  */
10492     gcc_assert (cst_and_fits_in_hwi (x));

where 'x' is:

<integer_cst 0x7ffff19ab0e0 type <integer_type 0x7ffff19be690
upcr_pshared_ptr_t> constant 0x0fffff00000000000>

and the type is a 128 bit integer:

 <integer_type 0x7ffff19be690 upcr_pshared_ptr_t sizes-gimplified public
unsigned TI
    size <integer_cst 0x7ffff189d6c0 type <integer_type 0x7ffff189f150
bitsizetype> constant 128>
    unit size <integer_cst 0x7ffff189d6d8 type <integer_type 0x7ffff189f0a8
sizetype> constant 16>
    align 128 symtab 0 alias set -1 canonical type 0x7ffff189fb28 precision 128
min <integer_cst 0x7ffff189d9a8 0> max <integer_cst 0x7ffff189c668
0xffffffffffffffffffffffffffffffff> context <translation_unit_decl
0x7ffff19c3b80 D.1942>
    pointer_to_this <pointer_type 0x7ffff19c97e0>>

The gimple statement being simplified looks like this:

_15 = array.0_8 & 18446726481523507200

(gdb) p/x 18446726481523507200
$4 = 0xfffff00000000000