The reported regression of libgomp loop-14.C shows that there isn't generally a good reliable place to insert the permute upfront so the following simply restricts recurrence vectorization to the cases where the latch value isn't defined by a PHI.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed. * tree-vect-loop.cc (vect_phi_first_order_recurrence_p): Disallow latch PHI defs. (vectorizable_recurr): Revert previous change. --- gcc/tree-vect-loop.cc | 11 +++-------- 1 file changed, 3 insertions(+), 8 deletions(-) diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc index 92790bd8095..d5c2bff80be 100644 --- a/gcc/tree-vect-loop.cc +++ b/gcc/tree-vect-loop.cc @@ -543,6 +543,7 @@ vect_phi_first_order_recurrence_p (loop_vec_info loop_vinfo, class loop *loop, tree ldef = PHI_ARG_DEF_FROM_EDGE (phi, latch); if (TREE_CODE (ldef) != SSA_NAME || SSA_NAME_IS_DEFAULT_DEF (ldef) + || is_a <gphi *> (SSA_NAME_DEF_STMT (ldef)) || !flow_bb_inside_loop_p (loop, gimple_bb (SSA_NAME_DEF_STMT (ldef)))) return false; @@ -8486,14 +8487,8 @@ vectorizable_recurr (loop_vec_info loop_vinfo, stmt_vec_info stmt_info, vectorized the latch definition. */ edge le = loop_latch_edge (LOOP_VINFO_LOOP (loop_vinfo)); gimple *latch_def = SSA_NAME_DEF_STMT (PHI_ARG_DEF_FROM_EDGE (phi, le)); - gimple_stmt_iterator gsi2; - if (is_a <gphi *> (latch_def)) - gsi2 = gsi_after_labels (gimple_bb (latch_def)); - else - { - gsi2 = gsi_for_stmt (latch_def); - gsi_next (&gsi2); - } + gimple_stmt_iterator gsi2 = gsi_for_stmt (latch_def); + gsi_next (&gsi2); for (unsigned i = 0; i < ncopies; ++i) { -- 2.35.3