https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98837
Bug ID: 98837
Summary: SLP discovery does not consider all lane permutes
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rguenth at gcc dot gnu.org
Target Milestone: ---
While we SLP vectorize
int a[1024], b[1024], c[1024];
void foo ()
{
for (int i = 0; i < 1024; i += 4)
{
c[i] = a[i] + b[i];
c[i+1] = a[i+1] + b[i+1];
c[i+2] = a[i+2] * b[i+2];
c[i+3] = a[i+3] * b[i+3];
}
}
by splitting the SLP group into two the very similar
int a[1024], b[1024], c[1024];
void foo ()
{
for (int i = 0; i < 1024; i += 4)
{
c[i] = a[i] + b[i];
c[i+1] = a[i+1] * b[i+1];
c[i+2] = a[i+2] + b[i+2];
c[i+3] = a[i+3] * b[i+3];
}
}
is not SLPed because we do not consider splitting the group into
non-adjacent sets. The same applies to basic-block SLP when
you make the data type double (so we don't need unrolling),
of course we simply fall back to a scalar build then.