https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97983
Bug ID: 97983
Summary: [11 Regression] Bootstrap failure on s390x-linux
related to vec-perm-indices.c
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: bootstrap
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
Target Milestone: ---
Non-profiled bootstrap on s390x fails when compiling vec-perm-indices.c.
Can be reproduced in a cross-compiler with
./cc1plus -quiet -nostdinc -mtune=z13 -march=zEC12 -m64 -O2 -fchecking=1
-fno-exceptions -fno-rtti vec-perm-indices.ii
../../gcc/vec-perm-indices.c: In member function ‘void
vec_perm_indices::new_vector(const vec_perm_builder&, unsigned int,
poly_uint64)’:
../../gcc/vec-perm-indices.c:78:1: error: insn 308 outside of basic blocks has
non-NULL bb field
78 | }
| ^
../../gcc/vec-perm-indices.c:78:1: error: verify_flow_info: Incorrect fallthru
6->7
../../gcc/vec-perm-indices.c:78:1: error: wrong insn in the fallthru edge
(insn 308 150 68 7 (set (reg:SI 1 %r1 [orig:157 i ] [157])
(reg:SI 1 %r1 [orig:157 i ] [157])) 1467 {*movsi_zarch}
(nil))
during RTL pass: reload
../../gcc/vec-perm-indices.c:78:1: internal compiler error: in
rtl_verify_fallthru, at cfgrtl.c:2978
0x141e2c9 _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
../../gcc/rtl-error.c:108
0xe6637c rtl_verify_fallthru
../../gcc/cfgrtl.c:2978
0xe665f8 rtl_verify_flow_info
../../gcc/cfgrtl.c:3078
0xe4bf2e verify_flow_info()
../../gcc/cfghooks.c:267
0x1fe0b0d checking_verify_flow_info
../../gcc/cfghooks.h:212
0x1fe897f try_optimize_cfg
../../gcc/cfgcleanup.c:3009
0x1fe8de9 cleanup_cfg(int)
../../gcc/cfgcleanup.c:3174
0x120ebc4 do_reload
../../gcc/ira.c:5851
0x120ef38 execute
../../gcc/ira.c:5988
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
profiledbootstrap fails too, but rather when building libdecnumber, but the ICE
is in vec-perm-indices.c new_vector and it looks really weird (e.g. the
function has .5MB stack frame).
