https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117161
Bug ID: 117161
Summary: internal compiler error: in
vect_get_vector_types_for_stmt, at
tree-vect-stmts.c:12252 on gcc 10.2 for aarch64+sve
Product: gcc
Version: 10.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: Robert.Hardwick at arm dot com
Target Milestone: ---
We are seeing an internal compiler error when O3 optimization is used on
aarch64+sve ( specifically neoverse-v1 )
compiler flags where the error is seen
-O3 -march=armv8-a+sve -Wall -std=c++17
workaround
Adding the compiler flag -fno-tree-vectorize fixes the problem
Error:
"""
during GIMPLE pass: slp
test.cpp: In function ‘int main()’:
test.cpp:3:5: internal compiler error: in vect_get_vector_types_for_stmt, at
tree-vect-stmts.c:12252
3 | int main()
| ^~~~
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://bugzilla.redhat.com/bugzilla> for instructions.
Preprocessed source stored into /tmp/cc1Tnw7l.out file, please attach this to
your bugreport.
"""
Reproducer:
"""
#include <iostream>
int main()
{
int64_t values[8];
for (int i = 0; i < 8; i++) {
values[i] = 1.0f + i;
}
for (int i = 0; i < 8; i++) {
if (values[i] <= 15) {
std::cout << values[i] << std::endl;
}
}
}
"""
The std::cout is there simply to prevent the compiler from optimizing values
away since it is unused.
This appears to be a GCC/G++ 10 issue as we don't see this in GCC 10 and above.
