https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122473
Bug ID: 122473
Summary: False-positive -Wstringop-overflow warning when using
`-march=x86-64-v3 -O3`
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: abbeyj+gcc at gmail dot com
Target Milestone: ---
When compiling the following with `-march=x86-64-v3 -O3`, a -Wstringop-overflow
warning is produced:
```
struct my_descriptor {
int length;
char *pointer;
};
void sink(const char *);
void print(const my_descriptor *desc)
{
char output[9];
char *p = output;
for (int i = desc->length; i >= 0; i--) {
p[i] = desc->pointer[i];
}
sink(output);
}
```
Output:
> $ g++ -march=x86-64-v3 -O3 -c repro.cpp
> repro.cpp: In function ‘void print(const my_descriptor*)’:
> repro.cpp:13:10: warning: writing 16 bytes into a region of size 9
> [-Wstringop-overflow=]
> 13 | p[i] = desc->pointer[i];
> | ~~~~~^~~~~~~~~~~~~~~~~~
> repro.cpp:10:8: note: destination object ‘output’ of size 9
> 10 | char output[9];
> | ^~~~~~
Instead of `-O3`, one could also use `-O2 -fvect-cost-model=dynamic` or `-O1
-foptimize-strlen -ftree-loop-vectorize -fvect-cost-model=dynamic`.
This looks like it started in GCC 12.1 and is still reproducible on the current
trunk version available on Godbolt: https://godbolt.org/z/foMd1Pq4s
I thought this might be a duplicate of Bug 106020, but the testcase there no
longer reproduces on GCC 15 or later while this testcase does. Maybe also Bug
109852 but that has a lot of warnings and no minimized reproducer so it is hard
to tell.
