https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121649
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Priority|P3 |P1
CC| |jakub at gcc dot gnu.org,
| |ktkachov at gcc dot gnu.org
Summary|[16 Regression] wrong code |[16 Regression] wrong code
|at -O2 with large number of |with _BitInt() conversions
|arguments |(?) at -O2 since r16-2382
Keywords|needs-bisection |
Component|target |tree-optimization
Last reconfirmed|2025-08-24 00:00:00 |
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r16-2382-gf87b632f98dc6d5cf8ed75d57d15adb8d7ffea6e
Before that change, expand dump had
(insn 947 946 948 (set (reg:V8QI 578)
(const_vector:V8QI [
(const_int 3 [0x3])
(const_int -4 [0xfffffffffffffffc])
(const_int 0 [0]) repeated x6
])) "pr121649.c":63:3 -1
(nil))
...
(insn 1174 1173 1175 (set (reg:V8QI 33 v1)
(reg:V8QI 578)) "pr121649.c":63:3 -1
(nil))
but in r16-2382 it has instead:
(insn 1173 1172 1174 (set (reg:V8QI 33 v1)
(const_vector:V8QI [
(const_int 3 [0x3])
(const_int -4 [0xfffffffffffffffc])
(const_int 0 [0]) repeated x6
])) "pr121649.c":63:3 -1
(nil))
The problem seems to happen during cse2. We have
(insn 1173 1172 1174 2 (set (reg:V8QI 33 v1)
(const_vector:V8QI [
(const_int 3 [0x3])
(const_int -4 [0xfffffffffffffffc])
(const_int 0 [0]) repeated x6
])) "pr121649.c":63:3 1325 {*aarch64_simd_movv8qi}
(nil))
...
(insn 1178 1177 1179 2 (set (reg:QI 4 x4)
(vec_select:QI (reg:V8QI 33 v1)
(parallel [
(const_int 1 [0x1])
]))) "pr121649.c":63:3 2976 {aarch64_get_lanev8qi}
(expr_list:REG_EQUAL (const_int -4 [0xfffffffffffffffc])
(nil)))
but cse2 turns the latter into:
(insn 1178 1177 1179 2 (set (reg:QI 4 x4)
(const_int 0 [0])) "pr121649.c":63:3 107 {*movqi_aarch64}
(expr_list:REG_EQUAL (const_int -4 [0xfffffffffffffffc])
(nil)))
