https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66286
Bug ID: 66286
Summary: Inconsistent handling of array sections
Product: gcc
Version: 5.1.1
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
CC: ienkovich at gcc dot gnu.org, izamyatin at gmail dot com,
kyukhin at gcc dot gnu.org
Target Milestone: ---
int
main ()
{
int a[10], i;
a[0:10] = 10; a[0:2] *= 2; a[5:2] *= 3;
for (i = 0; i < 10; i++)
if (a[i] != (i < 2 ? 20 : (i - 5U <= 1U ? 30 : 10)))
__builtin_abort ();
return 0;
}
passes in C++, but not in C, and passes with icc (both C and C++).
https://www.cilkplus.org/tutorial-array-notation says that omitted stride
defaults to 1 and that is what the C++ FE does:
if (!stride)
stride = build_one_cst (ptrdiff_type_node);
in build_array_notation_ref. But the C FE does something different:
if (!stride)
{
if (TREE_CONSTANT (start_index) && TREE_CONSTANT (length)
&& tree_int_cst_lt (length, start_index))
stride = build_int_cst (TREE_TYPE (start_index), -1);
else
stride = build_int_cst (TREE_TYPE (start_index), 1);
}
Not to mention that TREE_CONSTANT doesn't necessarily imply INTEGER_CST that
can be passed to tree_int_cst_lt...
IMHO we should just replace that hunk with
if (!stride)
stride = build_int_cst (TREE_TYPE (start_index), 1);
but I'd like to understand why this has been added there at all.
Seems the C++ FE had similar code too, but that got removed in r200554 without
mentioning it as a wrong-code bugfix.
