https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117556
Bug ID: 117556
Summary: gcc.dg/vect/vect-live-slp-3.c fails to SLP with
variable-length vectors
Product: gcc
Version: 15.0
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: ---
gcc.dg/vect/vect-live-slp-3.c:31:1: missed: Cannot determine which vector
holds the final result.
gcc.dg/vect/vect-live-slp-3.c:31:1: missed: not vectorized: live stmt not
supported: n0_29 = *_4;
gcc.dg/vect/vect-live-slp-3.c:31:1: note: unsupported SLP instance starting
from: *_14 = _15;
on aarch64 with SVE we fall back to AdvSIMD, with RISC-V we fall back to
fixed length modes.
With --param vect-force-slp=0 we can use VLA vectors.
The above is
bool
vectorizable_live_operation (vec_info *vinfo, stmt_vec_info stmt_info,
slp_tree slp_node, slp_instance slp_node_instance,
int slp_index, bool vec_stmt_p,
stmt_vector_for_cost *cost_vec)
{
...
/* Calculate which vector contains the result, and which lane of
that vector we need. */
if (!can_div_trunc_p (pos, nunits, &vec_entry, &vec_index))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Cannot determine which vector holds the"
" final result.\n");
return false;
}
I suppose for single-lane SLP this can possibly be relaxed.
