https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97294
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org
Status|NEW |ASSIGNED
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 49323
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49323&action=edit
gcc11-pr97294.patch
Untested fix.
Self-contained testcase with -O0 -fsanitize=address -fopenmp could be e.g.
__attribute__((noipa)) void
foo (int *p, int n)
{
int i;
#pragma omp parallel for num_threads(2) reduction(+:p[:n])
for (i = 0; i < 10; i++)
{
p[0]++;
p[n - 1] += 2;
}
}
__attribute__((noipa)) void
bar (void)
{
unsigned char buf[1024];
int i;
asm volatile ("" : : "r" (&buf[0]) : "memory");
for (i = 0; i < 1024; i++)
buf[i] = i;
asm volatile ("" : : "r" (&buf[0]) : "memory");
}
int
main ()
{
int p[50], i;
for (i = 0; i < 50; i++)
p[i] = 0;
foo (p, 50);
bar ();
if (p[0] != 10 || p[49] != 20)
__builtin_abort ();
return 0;
}
The problem was that nothing set cfun->calls_alloca flag in the child omp
function and thus the asan code wouldn't add __asan_allocas_unpoison call at
the end of the function. Normally when optimizing, DCE clears those flags and
recomputes them again, so this was only problematic with -O0.
