diff options
Diffstat (limited to 'gcc/tree-vect-stmts.cc')
| -rw-r--r-- | gcc/tree-vect-stmts.cc | 93 |
1 files changed, 40 insertions, 53 deletions
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc index 9edc4a8..4cf6c3b 100644 --- a/gcc/tree-vect-stmts.cc +++ b/gcc/tree-vect-stmts.cc @@ -1676,7 +1676,6 @@ vect_truncate_gather_scatter_offset (stmt_vec_info stmt_info, tree vectype, get_object_alignment (DR_REF (dr))); gs_info->element_type = TREE_TYPE (vectype); gs_info->offset = fold_convert (offset_type, step); - gs_info->offset_dt = vect_constant_def; gs_info->scale = scale; gs_info->memory_type = memory_type; return true; @@ -2229,22 +2228,23 @@ get_group_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, *memory_access_type = VMAT_GATHER_SCATTER; else if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)) { - tree offset; - slp_tree offset_node; *memory_access_type = VMAT_GATHER_SCATTER; if (!vect_check_gather_scatter (stmt_info, loop_vinfo, gs_info, elsvals)) gcc_unreachable (); + slp_tree offset_node = SLP_TREE_CHILDREN (slp_node)[0]; + tree offset_vectype = SLP_TREE_VECTYPE (offset_node); + gs_info->offset_vectype = offset_vectype; /* When using internal functions, we rely on pattern recognition to convert the type of the offset to the type that the target requires, with the result being a call to an internal function. If that failed for some reason (e.g. because another pattern took priority), just handle cases in which the offset already has the right type. */ - else if (GATHER_SCATTER_IFN_P (*gs_info) - && !is_gimple_call (stmt_info->stmt) - && !tree_nop_conversion_p (TREE_TYPE (gs_info->offset), - TREE_TYPE (gs_info->offset_vectype))) + if (GATHER_SCATTER_IFN_P (*gs_info) + && !is_gimple_call (stmt_info->stmt) + && !tree_nop_conversion_p (TREE_TYPE (gs_info->offset), + offset_vectype)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, @@ -2252,23 +2252,12 @@ get_group_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, vls_type == VLS_LOAD ? "gather" : "scatter"); return false; } - else if (!vect_is_simple_use (vinfo, slp_node, 0, &offset, &offset_node, - &gs_info->offset_dt, - &gs_info->offset_vectype)) - { - if (dump_enabled_p ()) - dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, - "%s index use not simple.\n", - vls_type == VLS_LOAD ? "gather" : "scatter"); - return false; - } else if (GATHER_SCATTER_EMULATED_P (*gs_info)) { if (!TYPE_VECTOR_SUBPARTS (vectype).is_constant () - || !TYPE_VECTOR_SUBPARTS (gs_info->offset_vectype).is_constant () - || VECTOR_BOOLEAN_TYPE_P (gs_info->offset_vectype) - || !constant_multiple_p (TYPE_VECTOR_SUBPARTS - (gs_info->offset_vectype), + || !TYPE_VECTOR_SUBPARTS (offset_vectype).is_constant () + || VECTOR_BOOLEAN_TYPE_P (offset_vectype) + || !constant_multiple_p (TYPE_VECTOR_SUBPARTS (offset_vectype), TYPE_VECTOR_SUBPARTS (vectype))) { if (dump_enabled_p ()) @@ -10211,29 +10200,6 @@ vectorizable_load (vec_info *vinfo, tree bump; tree vec_offset = NULL_TREE; - if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)) - { - aggr_type = NULL_TREE; - bump = NULL_TREE; - } - else if (memory_access_type == VMAT_GATHER_SCATTER) - { - aggr_type = elem_type; - if (!costing_p) - vect_get_strided_load_store_ops (stmt_info, vectype, loop_vinfo, - gsi, &gs_info, - &bump, &vec_offset, loop_lens); - } - else - { - if (memory_access_type == VMAT_LOAD_STORE_LANES) - aggr_type = build_array_type_nelts (elem_type, group_size * nunits); - else - aggr_type = vectype; - if (!costing_p) - bump = vect_get_data_ptr_increment (vinfo, gsi, dr_info, aggr_type, - memory_access_type, loop_lens); - } auto_vec<tree> vec_offsets; auto_vec<tree> vec_masks; @@ -10248,6 +10214,11 @@ vectorizable_load (vec_info *vinfo, gcc_assert (alignment_support_scheme == dr_aligned || alignment_support_scheme == dr_unaligned_supported); + aggr_type = build_array_type_nelts (elem_type, group_size * nunits); + if (!costing_p) + bump = vect_get_data_ptr_increment (vinfo, gsi, dr_info, aggr_type, + memory_access_type, loop_lens); + unsigned int inside_cost = 0, prologue_cost = 0; /* For costing some adjacent vector loads, we'd like to cost with the total number of them once instead of cost each one by one. */ @@ -10409,20 +10380,31 @@ vectorizable_load (vec_info *vinfo, { gcc_assert (!grouped_load && !slp_perm); - unsigned int inside_cost = 0, prologue_cost = 0; - /* 1. Create the vector or array pointer update chain. */ - if (!costing_p) + if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)) { - if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)) + aggr_type = NULL_TREE; + bump = NULL_TREE; + if (!costing_p) vect_get_gather_scatter_ops (loop, slp_node, &gs_info, &dataref_ptr, &vec_offsets); - else - dataref_ptr - = vect_create_data_ref_ptr (vinfo, first_stmt_info, aggr_type, - at_loop, offset, &dummy, gsi, - &ptr_incr, false, bump); } + else + { + aggr_type = elem_type; + if (!costing_p) + { + vect_get_strided_load_store_ops (stmt_info, vectype, loop_vinfo, + gsi, &gs_info, + &bump, &vec_offset, loop_lens); + dataref_ptr + = vect_create_data_ref_ptr (vinfo, first_stmt_info, aggr_type, + at_loop, offset, &dummy, gsi, + &ptr_incr, false, bump); + } + } + + unsigned int inside_cost = 0, prologue_cost = 0; gimple *new_stmt = NULL; for (i = 0; i < vec_num; i++) @@ -10744,6 +10726,11 @@ vectorizable_load (vec_info *vinfo, return true; } + aggr_type = vectype; + if (!costing_p) + bump = vect_get_data_ptr_increment (vinfo, gsi, dr_info, aggr_type, + memory_access_type, loop_lens); + poly_uint64 group_elt = 0; unsigned int inside_cost = 0, prologue_cost = 0; /* For costing some adjacent vector loads, we'd like to cost with |