https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127382
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|target |tree-optimization
Priority|P3 |P2
CC| |liuhongt at gcc dot gnu.org,
| |rguenth at gcc dot gnu.org
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Target Milestone|--- |16.3
Last reconfirmed| |2026-09-14
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
#1 0x00000000015d76c0 in fold_using_range::range_of_cond_expr (
this=0x7fffffff9dff, r=..., s=0x7ffff680e820, src=...)
at /space/rguenther/src/gcc-clean/gcc/gimple-range-fold.cc:1513
1513 gcc_checking_assert (range_compatible_p (TREE_TYPE (op1), TREE_TYPE
(op2)));
confirmed.
(gdb) p debug_tree (op1)
<ssa_name 0x7ffff6605840
type <integer_type 0x7ffff681d690 unsigned int sizes-gimplified public
unsigned SI
size <integer_cst 0x7ffff681f210 constant 32>
unit-size <integer_cst 0x7ffff681f228 constant 4>
align:32 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff681d690 precision:32 min <integer_cst 0x7ffff681f240 0> max <integer_cst
0x7ffff681f1f8 4294967295>
pointer_to_this <pointer_type 0x7ffff682d738>>
visited var <var_decl 0x7ffff69ff130 i>
def_stmt i_18 = PHI <i_15(8), 0(15)>
(gdb) p debug_tree (op2)
<ssa_name 0x7ffff69d1948
type <bitint_type 0x7ffff69c9498 public unsigned SI
size <integer_cst 0x7ffff681f210 constant 32>
unit-size <integer_cst 0x7ffff681f228 constant 4>
align:32 warn_if_not_align:0 symtab:0 alias-set 1 canonical-type
0x7ffff69c9498 precision:17 min <integer_cst 0x7ffff69dc618 0> max <integer_cst
0x7ffff69dc558 131071>
pointer_to_this <pointer_type 0x7ffff69c9540>>
visited
def_stmt g1.0_1 = g1;
patt_11 = patt_38 ? i_18 : g1.0_1;
this inconsistency is introduced by vectorizer patterns. Built by
0x0000000003af9768 in vect_recog_bool_pattern (vinfo=0x57fdf50,
stmt_vinfo=0x57ff680, type_out=0x7fffffffc5d8)
at /space/rguenther/src/gcc-clean/gcc/tree-vect-patterns.cc:6209
6209 = gimple_build_assign (lhs, COND_EXPR, lhs_var,
and taken from
patt_11 = _16 ? i_18 : g1.0_1;
built by
0x0000000003ae8582 in vect_recog_cond_expr_convert_pattern (vinfo=0x57fdf50,
stmt_vinfo=0x57feb20, type_out=0x7fffffffc698)
at /space/rguenther/src/gcc-clean/gcc/tree-vect-patterns.cc:1185
1185 cond_stmt = gimple_build_assign (temp, build3 (COND_EXPR, type,
match[3],
from
_7 = _16 ? _4 : _5;
which was OK. gimple_cond_expr_convert_p has a match that ends up with
two operands not of the same precision:
/* Floatint point/integer comparison and integer->integer
or floating point -> float point conversion. */
(match (cond_expr_convert_p @0 @2 @3 @6)
(cond (simple_comparison@6 @0 @1) (convert@4 @2) (convert@5 @3))
(if ((INTEGRAL_TYPE_P (type)
|| (!flag_trapping_math && SCALAR_FLOAT_TYPE_P (type)))
&& ((INTEGRAL_TYPE_P (TREE_TYPE (@2))
&& INTEGRAL_TYPE_P (TREE_TYPE (@3)))
|| (SCALAR_FLOAT_TYPE_P (TREE_TYPE (@2))
&& types_match (TREE_TYPE (@2), TREE_TYPE (@3))))
&& !operand_equal_p (TYPE_SIZE (type), TYPE_SIZE (TREE_TYPE (@0)))
&& operand_equal_p (TYPE_SIZE (TREE_TYPE (@0)),
TYPE_SIZE (TREE_TYPE (@2)))
&& operand_equal_p (TYPE_SIZE (TREE_TYPE (@0)),
TYPE_SIZE (TREE_TYPE (@3)))
/* For vect_recog_cond_expr_convert_pattern, @2 and @3 can differ in
signess when convert is truncation, but not ok for extension since
it's sign_extend vs zero_extend. */
&& (known_gt (tree_to_poly_uint64 (TYPE_SIZE (TREE_TYPE (@0))),
tree_to_poly_uint64 (TYPE_SIZE (type)))
|| (TYPE_UNSIGNED (TREE_TYPE (@2))
== TYPE_UNSIGNED (TREE_TYPE (@3))))
&& single_use (@4)
&& single_use (@5))))
r16-605-ge3385fed3fc61a elided a (also insufficient)
- && (TYPE_PRECISION (TREE_TYPE (@0)) > TYPE_PRECISION (type)
check.
