https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111754
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2023-10-10
Component|rtl-optimization |middle-end
Ever confirmed|0 |1
Target Milestone|--- |14.0
Priority|P3 |P1
Keywords| |needs-bisection
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
(gdb) p debug_tree (exp)
<vector_cst 0x7ffff6f25398
type <vector_type 0x7ffff6eebf18 F
type <real_type 0x7ffff6d672a0 float SF
size <integer_cst 0x7ffff6d61180 constant 32>
unit-size <integer_cst 0x7ffff6d61198 constant 4>
align:32 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6d672a0 precision:32
pointer_to_this <pointer_type 0x7ffff6d67930>>
sizes-gimplified V4SF
size <integer_cst 0x7ffff6d40f78 constant 128>
unit-size <integer_cst 0x7ffff6d40f90 constant 16>
align:128 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6e99b28 nunits:4>
constant npatterns:1 nelts-per-pattern:3
elt:0: <real_cst 0x7ffff6f13540 type <real_type 0x7ffff6d672a0 float>
constant 0.0>
elt:1: <real_cst 0x7ffff6f13510 type <real_type 0x7ffff6d672a0 float>
constant 9.0e+0> elt:2: <real_cst 0x7ffff6f13540 0.0>>
Confirmed. That's an odd vector mode 'F'!? And we have three elements?
Happens after inlining bar () into foo ():
F bar (F a, F b)
{
F c;
<bb 2> [local count: 1073741824]:
c_2 = VEC_PERM_EXPR <a_1(D), a_1(D), { 1, 0, 1, 2 }>;
__builtin_logbl (0.0);
return c_2;
F foo ()
{
F _3;
<bb 2> [local count: 1073741824]:
_3 = bar ({ 9.0e+0, 0.0, 0.0, 0.0 }, { 0.0, 0.0, 0.0, 0.0 });
and the constant folding of the VEC_PERM_EXPR causes this.
ICEs way earlier (during inlining) when you do -fdump-ipa-inline
