https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100848
Bug ID: 100848
Summary: Cases that require -fallow-store-data-races
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rsandifo at gcc dot gnu.org
Target Milestone: ---
If a loop could be vectorised with -fallow-store-data-races, that approach
trumps IFN_MASK_LOAD/STORE even if -fallow-store-data-races is off.
This means that, in the following example, we can vectorise f2
with -O3 -march=armv8.2-a+sve, but we can't vectorise f1 without
-fallow-store-data-races:
int x[100], y[100];
void f1 (void)
{
for (int i = 0; i < 100; ++i)
if (x[i])
y[i] += 1;
}
void f2 (int *restrict x, int *restrict y)
{
for (int i = 0; i < 100; ++i)
if (x[i])
y[i] += 1;
}
Even if -fallow-store-data-races is passed, it would be better
to avoid load, select and store in the data-races version of:
int x[100], y[100];
void f1 (void)
{
for (int i = 0; i < 100; ++i)
if (x[i])
y[i] = 1;
}
when a single IFN_MASK_STORE is enough.
