https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125863
Bug ID: 125863
Summary: False positive -Warray-bounds in context of -O3 +
__builtin_unreachable()
Product: gcc
Version: 16.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: nicula at nicula dot xyz
Target Milestone: ---
Created attachment 64761
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=64761&action=edit
false positive -Warray-bounds
Here is some code that does a `std::vector::resize()` to a size of 1, while
trying to get rid of the branch that may need to reallocate (i.e. the
`current_size < new_size` branch).
#include <vector>
void foo(std::vector<int> &vec) {
if (vec.empty())
__builtin_unreachable();
vec.resize(1);
}
GCC 16.1 result with `-O3`:
"foo(std::vector<int, std::allocator<int>>&)":
mov rdx, QWORD PTR [rdi+8]
mov rax, QWORD PTR [rdi]
mov rcx, rdx
sub rcx, rax
cmp rcx, 4
jle .L1
add rax, 4
cmp rdx, rax
je .L1
mov QWORD PTR [rdi+8], rax
.L1:
ret
The problem is that you get a false positive `-Warray-bounds` when compiling
with `-O3 -Wall`. See the logs.txt file for the warning.
Note: if you do the same thing, but with `std::vector<Foo>` instead, where
`Foo` is `struct Foo { int x; }` you also get the warning. But if you change it
to `struct Foo { int x{}; }`, the warning goes away.
Also, changing the `__builtin_unreachable()`'s condition to `vec.size() < 1`
instead of `vec.empty()` achieves the same intended optimization but you don't
get the false positive `-Warray-bounds`.
