https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103761
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rsandifo at gcc dot
gnu.org
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed btw,
#1 0x0000000000f8b9b2 in exact_div<2u, unsigned long, unsigned long> (a=...,
b=...)
at /home/rguenther/src/trunk/gcc/poly-int.h:2239
2239 gcc_checking_assert (a.coeffs[0] % b.coeffs[0] == 0);
(gdb) p a
$3 = (const poly_int_pod<2, unsigned long> &) @0x7fffffffada0: {coeffs = {4,
4}}
(gdb) p b
$4 = (const poly_int_pod<2, unsigned long> &) @0x3862a00: {coeffs = {8, 8}}
8916 if (loop_vinfo
8917 && LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo))
8918 check_load_store_for_partial_vectors (loop_vinfo, vectype,
VLS_LOAD,
8919 group_size,
memory_access_type,
8920 ncopies, &gs_info, mask);
passes down ncopies == 1 but with SLP this is always one. From what
vect_record_loop_mask does it looks like it would do OK with just
SLP_TREE_NUMBER_OF_VEC_STMTS. The following works for the testcase but
the APIs are somewhat iffy (I guess I need to think on how to merge the
SLP vs. non-SLP idea of "ncopies").
Richard, can you take over from this? Test coverage on x86-64 is zero here.
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 5c9e8cfefa5..3304df26842 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -8917,7 +8917,10 @@ vectorizable_load (vec_info *vinfo,
&& LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo))
check_load_store_for_partial_vectors (loop_vinfo, vectype, VLS_LOAD,
group_size, memory_access_type,
- ncopies, &gs_info, mask);
+ slp_node
+ ? SLP_TREE_NUMBER_OF_VEC_STMTS
+ (slp_node) : ncopies,
+ &gs_info, mask);
if (dump_enabled_p ()
&& memory_access_type != VMAT_ELEMENTWISE
