https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116956
Bug ID: 116956
Summary: ICE when building PALM with gfortran: in
vect_analyze_loop_1, at tree-vect-loop.cc:3510
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: soumyaa at nvidia dot com
CC: tamar.christina at arm dot com
Target Milestone: ---
When building the PALM Model <https://palm.muk.uni-hannover.de/trac>, gfortran
emits an ICE.
The reduced test case to reproduce this bug is:
SUBROUTINE nesting_offl_init(u, v, mask)
IMPLICIT NONE
real :: u(:)
real :: v(:)
integer :: mask(:)
u = MERGE( u, v, BTEST (mask, 1) )
END SUBROUTINE nesting_offl_init
and can be checked when running with -O3 -mcpu=neoverse-v2.
ICEing after the following commit:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=0c5c0c959c2e592b84739f19ca771fa69eb8dfee
The assert at tree-vect-loop.cc:3510 seems to be checking if the main loop is
empty when processing the epilogue loop. Removing this assert fixes the ICE.
Backtrace:
during GIMPLE pass: vect
palm.f90:1:28:
1 | SUBROUTINE nesting_offl_init(u, v, mask)
| ^
internal compiler error: in vect_analyze_loop_1, at tree-vect-loop.cc:3510
0x22953db internal_error(char const*, ...)
../../gcc_src/gcc/diagnostic-global-context.cc:517
0x7e616b fancy_abort(char const*, int, char const*)
../../gcc_src/gcc/diagnostic.cc:1572
0x134fd43 vect_analyze_loop_1
../../gcc_src/gcc/tree-vect-loop.cc:3510
0x135062b vect_analyze_loop(loop*, vec_info_shared*)
../../gcc_src/gcc/tree-vect-loop.cc:3711
0x13997cb try_vectorize_loop_1
../../gcc_src/gcc/tree-vectorizer.cc:1070
0x13997cb try_vectorize_loop
../../gcc_src/gcc/tree-vectorizer.cc:1186
0x1399d83 execute
../../gcc_src/gcc/tree-vectorizer.cc:1302
