| File: | build/gcc/tree-vect-stmts.cc |
| Warning: | line 10151, column 9 Value stored to 'msq' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* Statement Analysis and Transformation for Vectorization |
| 2 | Copyright (C) 2003-2023 Free Software Foundation, Inc. |
| 3 | Contributed by Dorit Naishlos <dorit@il.ibm.com> |
| 4 | and Ira Rosen <irar@il.ibm.com> |
| 5 | |
| 6 | This file is part of GCC. |
| 7 | |
| 8 | GCC is free software; you can redistribute it and/or modify it under |
| 9 | the terms of the GNU General Public License as published by the Free |
| 10 | Software Foundation; either version 3, or (at your option) any later |
| 11 | version. |
| 12 | |
| 13 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
| 14 | WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 16 | for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with GCC; see the file COPYING3. If not see |
| 20 | <http://www.gnu.org/licenses/>. */ |
| 21 | |
| 22 | #include "config.h" |
| 23 | #include "system.h" |
| 24 | #include "coretypes.h" |
| 25 | #include "backend.h" |
| 26 | #include "target.h" |
| 27 | #include "rtl.h" |
| 28 | #include "tree.h" |
| 29 | #include "gimple.h" |
| 30 | #include "ssa.h" |
| 31 | #include "optabs-tree.h" |
| 32 | #include "insn-config.h" |
| 33 | #include "recog.h" /* FIXME: for insn_data */ |
| 34 | #include "cgraph.h" |
| 35 | #include "dumpfile.h" |
| 36 | #include "alias.h" |
| 37 | #include "fold-const.h" |
| 38 | #include "stor-layout.h" |
| 39 | #include "tree-eh.h" |
| 40 | #include "gimplify.h" |
| 41 | #include "gimple-iterator.h" |
| 42 | #include "gimplify-me.h" |
| 43 | #include "tree-cfg.h" |
| 44 | #include "tree-ssa-loop-manip.h" |
| 45 | #include "cfgloop.h" |
| 46 | #include "explow.h" |
| 47 | #include "tree-ssa-loop.h" |
| 48 | #include "tree-scalar-evolution.h" |
| 49 | #include "tree-vectorizer.h" |
| 50 | #include "builtins.h" |
| 51 | #include "internal-fn.h" |
| 52 | #include "tree-vector-builder.h" |
| 53 | #include "vec-perm-indices.h" |
| 54 | #include "tree-ssa-loop-niter.h" |
| 55 | #include "gimple-fold.h" |
| 56 | #include "regs.h" |
| 57 | #include "attribs.h" |
| 58 | |
| 59 | /* For lang_hooks.types.type_for_mode. */ |
| 60 | #include "langhooks.h" |
| 61 | |
| 62 | /* Return the vectorized type for the given statement. */ |
| 63 | |
| 64 | tree |
| 65 | stmt_vectype (class _stmt_vec_info *stmt_info) |
| 66 | { |
| 67 | return STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 68 | } |
| 69 | |
| 70 | /* Return TRUE iff the given statement is in an inner loop relative to |
| 71 | the loop being vectorized. */ |
| 72 | bool |
| 73 | stmt_in_inner_loop_p (vec_info *vinfo, class _stmt_vec_info *stmt_info) |
| 74 | { |
| 75 | gimple *stmt = STMT_VINFO_STMT (stmt_info)(stmt_info)->stmt; |
| 76 | basic_block bb = gimple_bb (stmt); |
| 77 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 78 | class loop* loop; |
| 79 | |
| 80 | if (!loop_vinfo) |
| 81 | return false; |
| 82 | |
| 83 | loop = LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop; |
| 84 | |
| 85 | return (bb->loop_father == loop->inner); |
| 86 | } |
| 87 | |
| 88 | /* Record the cost of a statement, either by directly informing the |
| 89 | target model or by saving it in a vector for later processing. |
| 90 | Return a preliminary estimate of the statement's cost. */ |
| 91 | |
| 92 | static unsigned |
| 93 | record_stmt_cost (stmt_vector_for_cost *body_cost_vec, int count, |
| 94 | enum vect_cost_for_stmt kind, |
| 95 | stmt_vec_info stmt_info, slp_tree node, |
| 96 | tree vectype, int misalign, |
| 97 | enum vect_cost_model_location where) |
| 98 | { |
| 99 | if ((kind == vector_load || kind == unaligned_load) |
| 100 | && (stmt_info && STMT_VINFO_GATHER_SCATTER_P (stmt_info)(stmt_info)->gather_scatter_p)) |
| 101 | kind = vector_gather_load; |
| 102 | if ((kind == vector_store || kind == unaligned_store) |
| 103 | && (stmt_info && STMT_VINFO_GATHER_SCATTER_P (stmt_info)(stmt_info)->gather_scatter_p)) |
| 104 | kind = vector_scatter_store; |
| 105 | |
| 106 | stmt_info_for_cost si |
| 107 | = { count, kind, where, stmt_info, node, vectype, misalign }; |
| 108 | body_cost_vec->safe_push (si); |
| 109 | |
| 110 | return (unsigned) |
| 111 | (builtin_vectorization_cost (kind, vectype, misalign) * count); |
| 112 | } |
| 113 | |
| 114 | unsigned |
| 115 | record_stmt_cost (stmt_vector_for_cost *body_cost_vec, int count, |
| 116 | enum vect_cost_for_stmt kind, stmt_vec_info stmt_info, |
| 117 | tree vectype, int misalign, |
| 118 | enum vect_cost_model_location where) |
| 119 | { |
| 120 | return record_stmt_cost (body_cost_vec, count, kind, stmt_info, NULLnullptr, |
| 121 | vectype, misalign, where); |
| 122 | } |
| 123 | |
| 124 | unsigned |
| 125 | record_stmt_cost (stmt_vector_for_cost *body_cost_vec, int count, |
| 126 | enum vect_cost_for_stmt kind, slp_tree node, |
| 127 | tree vectype, int misalign, |
| 128 | enum vect_cost_model_location where) |
| 129 | { |
| 130 | return record_stmt_cost (body_cost_vec, count, kind, NULLnullptr, node, |
| 131 | vectype, misalign, where); |
| 132 | } |
| 133 | |
| 134 | unsigned |
| 135 | record_stmt_cost (stmt_vector_for_cost *body_cost_vec, int count, |
| 136 | enum vect_cost_for_stmt kind, |
| 137 | enum vect_cost_model_location where) |
| 138 | { |
| 139 | gcc_assert (kind == cond_branch_taken || kind == cond_branch_not_taken((void)(!(kind == cond_branch_taken || kind == cond_branch_not_taken || kind == scalar_stmt) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 140, __FUNCTION__), 0 : 0)) |
| 140 | || kind == scalar_stmt)((void)(!(kind == cond_branch_taken || kind == cond_branch_not_taken || kind == scalar_stmt) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 140, __FUNCTION__), 0 : 0)); |
| 141 | return record_stmt_cost (body_cost_vec, count, kind, NULLnullptr, NULLnullptr, |
| 142 | NULL_TREE(tree) nullptr, 0, where); |
| 143 | } |
| 144 | |
| 145 | /* Return a variable of type ELEM_TYPE[NELEMS]. */ |
| 146 | |
| 147 | static tree |
| 148 | create_vector_array (tree elem_type, unsigned HOST_WIDE_INTlong nelems) |
| 149 | { |
| 150 | return create_tmp_var (build_array_type_nelts (elem_type, nelems), |
| 151 | "vect_array"); |
| 152 | } |
| 153 | |
| 154 | /* ARRAY is an array of vectors created by create_vector_array. |
| 155 | Return an SSA_NAME for the vector in index N. The reference |
| 156 | is part of the vectorization of STMT_INFO and the vector is associated |
| 157 | with scalar destination SCALAR_DEST. */ |
| 158 | |
| 159 | static tree |
| 160 | read_vector_array (vec_info *vinfo, |
| 161 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 162 | tree scalar_dest, tree array, unsigned HOST_WIDE_INTlong n) |
| 163 | { |
| 164 | tree vect_type, vect, vect_name, array_ref; |
| 165 | gimple *new_stmt; |
| 166 | |
| 167 | gcc_assert (TREE_CODE (TREE_TYPE (array)) == ARRAY_TYPE)((void)(!(((enum tree_code) (((contains_struct_check ((array) , (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 167, __FUNCTION__))->typed.type))->base.code) == ARRAY_TYPE ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 167, __FUNCTION__), 0 : 0)); |
| 168 | vect_type = TREE_TYPE (TREE_TYPE (array))((contains_struct_check ((((contains_struct_check ((array), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 168, __FUNCTION__))->typed.type)), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 168, __FUNCTION__))->typed.type); |
| 169 | vect = vect_create_destination_var (scalar_dest, vect_type); |
| 170 | array_ref = build4 (ARRAY_REF, vect_type, array, |
| 171 | build_int_cst (size_type_nodeglobal_trees[TI_SIZE_TYPE], n), |
| 172 | NULL_TREE(tree) nullptr, NULL_TREE(tree) nullptr); |
| 173 | |
| 174 | new_stmt = gimple_build_assign (vect, array_ref); |
| 175 | vect_name = make_ssa_name (vect, new_stmt); |
| 176 | gimple_assign_set_lhs (new_stmt, vect_name); |
| 177 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 178 | |
| 179 | return vect_name; |
| 180 | } |
| 181 | |
| 182 | /* ARRAY is an array of vectors created by create_vector_array. |
| 183 | Emit code to store SSA_NAME VECT in index N of the array. |
| 184 | The store is part of the vectorization of STMT_INFO. */ |
| 185 | |
| 186 | static void |
| 187 | write_vector_array (vec_info *vinfo, |
| 188 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 189 | tree vect, tree array, unsigned HOST_WIDE_INTlong n) |
| 190 | { |
| 191 | tree array_ref; |
| 192 | gimple *new_stmt; |
| 193 | |
| 194 | array_ref = build4 (ARRAY_REF, TREE_TYPE (vect)((contains_struct_check ((vect), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 194, __FUNCTION__))->typed.type), array, |
| 195 | build_int_cst (size_type_nodeglobal_trees[TI_SIZE_TYPE], n), |
| 196 | NULL_TREE(tree) nullptr, NULL_TREE(tree) nullptr); |
| 197 | |
| 198 | new_stmt = gimple_build_assign (array_ref, vect); |
| 199 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 200 | } |
| 201 | |
| 202 | /* PTR is a pointer to an array of type TYPE. Return a representation |
| 203 | of *PTR. The memory reference replaces those in FIRST_DR |
| 204 | (and its group). */ |
| 205 | |
| 206 | static tree |
| 207 | create_array_ref (tree type, tree ptr, tree alias_ptr_type) |
| 208 | { |
| 209 | tree mem_ref; |
| 210 | |
| 211 | mem_ref = build2 (MEM_REF, type, ptr, build_int_cst (alias_ptr_type, 0)); |
| 212 | /* Arrays have the same alignment as their type. */ |
| 213 | set_ptr_info_alignment (get_ptr_info (ptr), TYPE_ALIGN_UNIT (type)((((tree_class_check ((type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 213, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 213, __FUNCTION__))->type_common.align) - 1) : 0) / (8)), 0); |
| 214 | return mem_ref; |
| 215 | } |
| 216 | |
| 217 | /* Add a clobber of variable VAR to the vectorization of STMT_INFO. |
| 218 | Emit the clobber before *GSI. */ |
| 219 | |
| 220 | static void |
| 221 | vect_clobber_variable (vec_info *vinfo, stmt_vec_info stmt_info, |
| 222 | gimple_stmt_iterator *gsi, tree var) |
| 223 | { |
| 224 | tree clobber = build_clobber (TREE_TYPE (var)((contains_struct_check ((var), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 224, __FUNCTION__))->typed.type)); |
| 225 | gimple *new_stmt = gimple_build_assign (var, clobber); |
| 226 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 227 | } |
| 228 | |
| 229 | /* Utility functions used by vect_mark_stmts_to_be_vectorized. */ |
| 230 | |
| 231 | /* Function vect_mark_relevant. |
| 232 | |
| 233 | Mark STMT_INFO as "relevant for vectorization" and add it to WORKLIST. */ |
| 234 | |
| 235 | static void |
| 236 | vect_mark_relevant (vec<stmt_vec_info> *worklist, stmt_vec_info stmt_info, |
| 237 | enum vect_relevant relevant, bool live_p) |
| 238 | { |
| 239 | enum vect_relevant save_relevant = STMT_VINFO_RELEVANT (stmt_info)(stmt_info)->relevant; |
| 240 | bool save_live_p = STMT_VINFO_LIVE_P (stmt_info)(stmt_info)->live; |
| 241 | |
| 242 | if (dump_enabled_p ()) |
| 243 | dump_printf_loc (MSG_NOTE, vect_location, |
| 244 | "mark relevant %d, live %d: %G", relevant, live_p, |
| 245 | stmt_info->stmt); |
| 246 | |
| 247 | /* If this stmt is an original stmt in a pattern, we might need to mark its |
| 248 | related pattern stmt instead of the original stmt. However, such stmts |
| 249 | may have their own uses that are not in any pattern, in such cases the |
| 250 | stmt itself should be marked. */ |
| 251 | if (STMT_VINFO_IN_PATTERN_P (stmt_info)(stmt_info)->in_pattern_p) |
| 252 | { |
| 253 | /* This is the last stmt in a sequence that was detected as a |
| 254 | pattern that can potentially be vectorized. Don't mark the stmt |
| 255 | as relevant/live because it's not going to be vectorized. |
| 256 | Instead mark the pattern-stmt that replaces it. */ |
| 257 | |
| 258 | if (dump_enabled_p ()) |
| 259 | dump_printf_loc (MSG_NOTE, vect_location, |
| 260 | "last stmt in pattern. don't mark" |
| 261 | " relevant/live.\n"); |
| 262 | stmt_vec_info old_stmt_info = stmt_info; |
| 263 | stmt_info = STMT_VINFO_RELATED_STMT (stmt_info)(stmt_info)->related_stmt; |
| 264 | gcc_assert (STMT_VINFO_RELATED_STMT (stmt_info) == old_stmt_info)((void)(!((stmt_info)->related_stmt == old_stmt_info) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 264, __FUNCTION__), 0 : 0)); |
| 265 | save_relevant = STMT_VINFO_RELEVANT (stmt_info)(stmt_info)->relevant; |
| 266 | save_live_p = STMT_VINFO_LIVE_P (stmt_info)(stmt_info)->live; |
| 267 | } |
| 268 | |
| 269 | STMT_VINFO_LIVE_P (stmt_info)(stmt_info)->live |= live_p; |
| 270 | if (relevant > STMT_VINFO_RELEVANT (stmt_info)(stmt_info)->relevant) |
| 271 | STMT_VINFO_RELEVANT (stmt_info)(stmt_info)->relevant = relevant; |
| 272 | |
| 273 | if (STMT_VINFO_RELEVANT (stmt_info)(stmt_info)->relevant == save_relevant |
| 274 | && STMT_VINFO_LIVE_P (stmt_info)(stmt_info)->live == save_live_p) |
| 275 | { |
| 276 | if (dump_enabled_p ()) |
| 277 | dump_printf_loc (MSG_NOTE, vect_location, |
| 278 | "already marked relevant/live.\n"); |
| 279 | return; |
| 280 | } |
| 281 | |
| 282 | worklist->safe_push (stmt_info); |
| 283 | } |
| 284 | |
| 285 | |
| 286 | /* Function is_simple_and_all_uses_invariant |
| 287 | |
| 288 | Return true if STMT_INFO is simple and all uses of it are invariant. */ |
| 289 | |
| 290 | bool |
| 291 | is_simple_and_all_uses_invariant (stmt_vec_info stmt_info, |
| 292 | loop_vec_info loop_vinfo) |
| 293 | { |
| 294 | tree op; |
| 295 | ssa_op_iter iter; |
| 296 | |
| 297 | gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt); |
| 298 | if (!stmt) |
| 299 | return false; |
| 300 | |
| 301 | FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)for (op = op_iter_init_tree (&(iter), stmt, 0x01); !op_iter_done (&(iter)); (void) (op = op_iter_next_tree (&(iter))) ) |
| 302 | { |
| 303 | enum vect_def_type dt = vect_uninitialized_def; |
| 304 | |
| 305 | if (!vect_is_simple_use (op, loop_vinfo, &dt)) |
| 306 | { |
| 307 | if (dump_enabled_p ()) |
| 308 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 309 | "use not simple.\n"); |
| 310 | return false; |
| 311 | } |
| 312 | |
| 313 | if (dt != vect_external_def && dt != vect_constant_def) |
| 314 | return false; |
| 315 | } |
| 316 | return true; |
| 317 | } |
| 318 | |
| 319 | /* Function vect_stmt_relevant_p. |
| 320 | |
| 321 | Return true if STMT_INFO, in the loop that is represented by LOOP_VINFO, |
| 322 | is "relevant for vectorization". |
| 323 | |
| 324 | A stmt is considered "relevant for vectorization" if: |
| 325 | - it has uses outside the loop. |
| 326 | - it has vdefs (it alters memory). |
| 327 | - control stmts in the loop (except for the exit condition). |
| 328 | |
| 329 | CHECKME: what other side effects would the vectorizer allow? */ |
| 330 | |
| 331 | static bool |
| 332 | vect_stmt_relevant_p (stmt_vec_info stmt_info, loop_vec_info loop_vinfo, |
| 333 | enum vect_relevant *relevant, bool *live_p) |
| 334 | { |
| 335 | class loop *loop = LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop; |
| 336 | ssa_op_iter op_iter; |
| 337 | imm_use_iterator imm_iter; |
| 338 | use_operand_p use_p; |
| 339 | def_operand_p def_p; |
| 340 | |
| 341 | *relevant = vect_unused_in_scope; |
| 342 | *live_p = false; |
| 343 | |
| 344 | /* cond stmt other than loop exit cond. */ |
| 345 | if (is_ctrl_stmt (stmt_info->stmt) |
| 346 | && STMT_VINFO_TYPE (stmt_info)(stmt_info)->type != loop_exit_ctrl_vec_info_type) |
| 347 | *relevant = vect_used_in_scope; |
| 348 | |
| 349 | /* changing memory. */ |
| 350 | if (gimple_code (stmt_info->stmt) != GIMPLE_PHI) |
| 351 | if (gimple_vdef (stmt_info->stmt) |
| 352 | && !gimple_clobber_p (stmt_info->stmt)) |
| 353 | { |
| 354 | if (dump_enabled_p ()) |
| 355 | dump_printf_loc (MSG_NOTE, vect_location, |
| 356 | "vec_stmt_relevant_p: stmt has vdefs.\n"); |
| 357 | *relevant = vect_used_in_scope; |
| 358 | } |
| 359 | |
| 360 | /* uses outside the loop. */ |
| 361 | FOR_EACH_PHI_OR_STMT_DEF (def_p, stmt_info->stmt, op_iter, SSA_OP_DEF)for ((def_p) = (gimple_code (stmt_info->stmt) == GIMPLE_PHI ? op_iter_init_phidef (&(op_iter), as_a <gphi *> ( stmt_info->stmt), 0x02) : op_iter_init_def (&(op_iter) , stmt_info->stmt, 0x02)); !op_iter_done (&(op_iter)); (def_p) = op_iter_next_def (&(op_iter))) |
| 362 | { |
| 363 | FOR_EACH_IMM_USE_FAST (use_p, imm_iter, DEF_FROM_PTR (def_p))for ((use_p) = first_readonly_imm_use (&(imm_iter), (get_def_from_ptr (def_p))); !end_readonly_imm_use_p (&(imm_iter)); (void) ((use_p) = next_readonly_imm_use (&(imm_iter)))) |
| 364 | { |
| 365 | basic_block bb = gimple_bb (USE_STMT (use_p)(use_p)->loc.stmt); |
| 366 | if (!flow_bb_inside_loop_p (loop, bb)) |
| 367 | { |
| 368 | if (is_gimple_debug (USE_STMT (use_p)(use_p)->loc.stmt)) |
| 369 | continue; |
| 370 | |
| 371 | if (dump_enabled_p ()) |
| 372 | dump_printf_loc (MSG_NOTE, vect_location, |
| 373 | "vec_stmt_relevant_p: used out of loop.\n"); |
| 374 | |
| 375 | /* We expect all such uses to be in the loop exit phis |
| 376 | (because of loop closed form) */ |
| 377 | gcc_assert (gimple_code (USE_STMT (use_p)) == GIMPLE_PHI)((void)(!(gimple_code ((use_p)->loc.stmt) == GIMPLE_PHI) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 377, __FUNCTION__), 0 : 0)); |
| 378 | gcc_assert (bb == single_exit (loop)->dest)((void)(!(bb == single_exit (loop)->dest) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 378, __FUNCTION__), 0 : 0)); |
| 379 | |
| 380 | *live_p = true; |
| 381 | } |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | if (*live_p && *relevant == vect_unused_in_scope |
| 386 | && !is_simple_and_all_uses_invariant (stmt_info, loop_vinfo)) |
| 387 | { |
| 388 | if (dump_enabled_p ()) |
| 389 | dump_printf_loc (MSG_NOTE, vect_location, |
| 390 | "vec_stmt_relevant_p: stmt live but not relevant.\n"); |
| 391 | *relevant = vect_used_only_live; |
| 392 | } |
| 393 | |
| 394 | return (*live_p || *relevant); |
| 395 | } |
| 396 | |
| 397 | |
| 398 | /* Function exist_non_indexing_operands_for_use_p |
| 399 | |
| 400 | USE is one of the uses attached to STMT_INFO. Check if USE is |
| 401 | used in STMT_INFO for anything other than indexing an array. */ |
| 402 | |
| 403 | static bool |
| 404 | exist_non_indexing_operands_for_use_p (tree use, stmt_vec_info stmt_info) |
| 405 | { |
| 406 | tree operand; |
| 407 | |
| 408 | /* USE corresponds to some operand in STMT. If there is no data |
| 409 | reference in STMT, then any operand that corresponds to USE |
| 410 | is not indexing an array. */ |
| 411 | if (!STMT_VINFO_DATA_REF (stmt_info)((stmt_info)->dr_aux.dr + 0)) |
| 412 | return true; |
| 413 | |
| 414 | /* STMT has a data_ref. FORNOW this means that its of one of |
| 415 | the following forms: |
| 416 | -1- ARRAY_REF = var |
| 417 | -2- var = ARRAY_REF |
| 418 | (This should have been verified in analyze_data_refs). |
| 419 | |
| 420 | 'var' in the second case corresponds to a def, not a use, |
| 421 | so USE cannot correspond to any operands that are not used |
| 422 | for array indexing. |
| 423 | |
| 424 | Therefore, all we need to check is if STMT falls into the |
| 425 | first case, and whether var corresponds to USE. */ |
| 426 | |
| 427 | gassign *assign = dyn_cast <gassign *> (stmt_info->stmt); |
| 428 | if (!assign || !gimple_assign_copy_p (assign)) |
| 429 | { |
| 430 | gcall *call = dyn_cast <gcall *> (stmt_info->stmt); |
| 431 | if (call && gimple_call_internal_p (call)) |
| 432 | { |
| 433 | internal_fn ifn = gimple_call_internal_fn (call); |
| 434 | int mask_index = internal_fn_mask_index (ifn); |
| 435 | if (mask_index >= 0 |
| 436 | && use == gimple_call_arg (call, mask_index)) |
| 437 | return true; |
| 438 | int stored_value_index = internal_fn_stored_value_index (ifn); |
| 439 | if (stored_value_index >= 0 |
| 440 | && use == gimple_call_arg (call, stored_value_index)) |
| 441 | return true; |
| 442 | if (internal_gather_scatter_fn_p (ifn) |
| 443 | && use == gimple_call_arg (call, 1)) |
| 444 | return true; |
| 445 | } |
| 446 | return false; |
| 447 | } |
| 448 | |
| 449 | if (TREE_CODE (gimple_assign_lhs (assign))((enum tree_code) (gimple_assign_lhs (assign))->base.code) == SSA_NAME) |
| 450 | return false; |
| 451 | operand = gimple_assign_rhs1 (assign); |
| 452 | if (TREE_CODE (operand)((enum tree_code) (operand)->base.code) != SSA_NAME) |
| 453 | return false; |
| 454 | |
| 455 | if (operand == use) |
| 456 | return true; |
| 457 | |
| 458 | return false; |
| 459 | } |
| 460 | |
| 461 | |
| 462 | /* |
| 463 | Function process_use. |
| 464 | |
| 465 | Inputs: |
| 466 | - a USE in STMT_VINFO in a loop represented by LOOP_VINFO |
| 467 | - RELEVANT - enum value to be set in the STMT_VINFO of the stmt |
| 468 | that defined USE. This is done by calling mark_relevant and passing it |
| 469 | the WORKLIST (to add DEF_STMT to the WORKLIST in case it is relevant). |
| 470 | - FORCE is true if exist_non_indexing_operands_for_use_p check shouldn't |
| 471 | be performed. |
| 472 | |
| 473 | Outputs: |
| 474 | Generally, LIVE_P and RELEVANT are used to define the liveness and |
| 475 | relevance info of the DEF_STMT of this USE: |
| 476 | STMT_VINFO_LIVE_P (DEF_stmt_vinfo) <-- live_p |
| 477 | STMT_VINFO_RELEVANT (DEF_stmt_vinfo) <-- relevant |
| 478 | Exceptions: |
| 479 | - case 1: If USE is used only for address computations (e.g. array indexing), |
| 480 | which does not need to be directly vectorized, then the liveness/relevance |
| 481 | of the respective DEF_STMT is left unchanged. |
| 482 | - case 2: If STMT_VINFO is a reduction phi and DEF_STMT is a reduction stmt, |
| 483 | we skip DEF_STMT cause it had already been processed. |
| 484 | - case 3: If DEF_STMT and STMT_VINFO are in different nests, then |
| 485 | "relevant" will be modified accordingly. |
| 486 | |
| 487 | Return true if everything is as expected. Return false otherwise. */ |
| 488 | |
| 489 | static opt_result |
| 490 | process_use (stmt_vec_info stmt_vinfo, tree use, loop_vec_info loop_vinfo, |
| 491 | enum vect_relevant relevant, vec<stmt_vec_info> *worklist, |
| 492 | bool force) |
| 493 | { |
| 494 | stmt_vec_info dstmt_vinfo; |
| 495 | enum vect_def_type dt; |
| 496 | |
| 497 | /* case 1: we are only interested in uses that need to be vectorized. Uses |
| 498 | that are used for address computation are not considered relevant. */ |
| 499 | if (!force && !exist_non_indexing_operands_for_use_p (use, stmt_vinfo)) |
| 500 | return opt_result::success (); |
| 501 | |
| 502 | if (!vect_is_simple_use (use, loop_vinfo, &dt, &dstmt_vinfo)) |
| 503 | return opt_result::failure_at (stmt_vinfo->stmt, |
| 504 | "not vectorized:" |
| 505 | " unsupported use in stmt.\n"); |
| 506 | |
| 507 | if (!dstmt_vinfo) |
| 508 | return opt_result::success (); |
| 509 | |
| 510 | basic_block def_bb = gimple_bb (dstmt_vinfo->stmt); |
| 511 | basic_block bb = gimple_bb (stmt_vinfo->stmt); |
| 512 | |
| 513 | /* case 2: A reduction phi (STMT) defined by a reduction stmt (DSTMT_VINFO). |
| 514 | We have to force the stmt live since the epilogue loop needs it to |
| 515 | continue computing the reduction. */ |
| 516 | if (gimple_code (stmt_vinfo->stmt) == GIMPLE_PHI |
| 517 | && STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type == vect_reduction_def |
| 518 | && gimple_code (dstmt_vinfo->stmt) != GIMPLE_PHI |
| 519 | && STMT_VINFO_DEF_TYPE (dstmt_vinfo)(dstmt_vinfo)->def_type == vect_reduction_def |
| 520 | && bb->loop_father == def_bb->loop_father) |
| 521 | { |
| 522 | if (dump_enabled_p ()) |
| 523 | dump_printf_loc (MSG_NOTE, vect_location, |
| 524 | "reduc-stmt defining reduc-phi in the same nest.\n"); |
| 525 | vect_mark_relevant (worklist, dstmt_vinfo, relevant, true); |
| 526 | return opt_result::success (); |
| 527 | } |
| 528 | |
| 529 | /* case 3a: outer-loop stmt defining an inner-loop stmt: |
| 530 | outer-loop-header-bb: |
| 531 | d = dstmt_vinfo |
| 532 | inner-loop: |
| 533 | stmt # use (d) |
| 534 | outer-loop-tail-bb: |
| 535 | ... */ |
| 536 | if (flow_loop_nested_p (def_bb->loop_father, bb->loop_father)) |
| 537 | { |
| 538 | if (dump_enabled_p ()) |
| 539 | dump_printf_loc (MSG_NOTE, vect_location, |
| 540 | "outer-loop def-stmt defining inner-loop stmt.\n"); |
| 541 | |
| 542 | switch (relevant) |
| 543 | { |
| 544 | case vect_unused_in_scope: |
| 545 | relevant = (STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type == vect_nested_cycle) ? |
| 546 | vect_used_in_scope : vect_unused_in_scope; |
| 547 | break; |
| 548 | |
| 549 | case vect_used_in_outer_by_reduction: |
| 550 | gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)((void)(!((stmt_vinfo)->def_type != vect_reduction_def) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 550, __FUNCTION__), 0 : 0)); |
| 551 | relevant = vect_used_by_reduction; |
| 552 | break; |
| 553 | |
| 554 | case vect_used_in_outer: |
| 555 | gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)((void)(!((stmt_vinfo)->def_type != vect_reduction_def) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 555, __FUNCTION__), 0 : 0)); |
| 556 | relevant = vect_used_in_scope; |
| 557 | break; |
| 558 | |
| 559 | case vect_used_in_scope: |
| 560 | break; |
| 561 | |
| 562 | default: |
| 563 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 563, __FUNCTION__)); |
| 564 | } |
| 565 | } |
| 566 | |
| 567 | /* case 3b: inner-loop stmt defining an outer-loop stmt: |
| 568 | outer-loop-header-bb: |
| 569 | ... |
| 570 | inner-loop: |
| 571 | d = dstmt_vinfo |
| 572 | outer-loop-tail-bb (or outer-loop-exit-bb in double reduction): |
| 573 | stmt # use (d) */ |
| 574 | else if (flow_loop_nested_p (bb->loop_father, def_bb->loop_father)) |
| 575 | { |
| 576 | if (dump_enabled_p ()) |
| 577 | dump_printf_loc (MSG_NOTE, vect_location, |
| 578 | "inner-loop def-stmt defining outer-loop stmt.\n"); |
| 579 | |
| 580 | switch (relevant) |
| 581 | { |
| 582 | case vect_unused_in_scope: |
| 583 | relevant = (STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type == vect_reduction_def |
| 584 | || STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type == vect_double_reduction_def) ? |
| 585 | vect_used_in_outer_by_reduction : vect_unused_in_scope; |
| 586 | break; |
| 587 | |
| 588 | case vect_used_by_reduction: |
| 589 | case vect_used_only_live: |
| 590 | relevant = vect_used_in_outer_by_reduction; |
| 591 | break; |
| 592 | |
| 593 | case vect_used_in_scope: |
| 594 | relevant = vect_used_in_outer; |
| 595 | break; |
| 596 | |
| 597 | default: |
| 598 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 598, __FUNCTION__)); |
| 599 | } |
| 600 | } |
| 601 | /* We are also not interested in uses on loop PHI backedges that are |
| 602 | inductions. Otherwise we'll needlessly vectorize the IV increment |
| 603 | and cause hybrid SLP for SLP inductions. Unless the PHI is live |
| 604 | of course. */ |
| 605 | else if (gimple_code (stmt_vinfo->stmt) == GIMPLE_PHI |
| 606 | && STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type == vect_induction_def |
| 607 | && ! STMT_VINFO_LIVE_P (stmt_vinfo)(stmt_vinfo)->live |
| 608 | && (PHI_ARG_DEF_FROM_EDGE (stmt_vinfo->stmt,gimple_phi_arg_def (((stmt_vinfo->stmt)), ((loop_latch_edge (bb->loop_father))->dest_idx)) |
| 609 | loop_latch_edge (bb->loop_father))gimple_phi_arg_def (((stmt_vinfo->stmt)), ((loop_latch_edge (bb->loop_father))->dest_idx)) |
| 610 | == use)) |
| 611 | { |
| 612 | if (dump_enabled_p ()) |
| 613 | dump_printf_loc (MSG_NOTE, vect_location, |
| 614 | "induction value on backedge.\n"); |
| 615 | return opt_result::success (); |
| 616 | } |
| 617 | |
| 618 | |
| 619 | vect_mark_relevant (worklist, dstmt_vinfo, relevant, false); |
| 620 | return opt_result::success (); |
| 621 | } |
| 622 | |
| 623 | |
| 624 | /* Function vect_mark_stmts_to_be_vectorized. |
| 625 | |
| 626 | Not all stmts in the loop need to be vectorized. For example: |
| 627 | |
| 628 | for i... |
| 629 | for j... |
| 630 | 1. T0 = i + j |
| 631 | 2. T1 = a[T0] |
| 632 | |
| 633 | 3. j = j + 1 |
| 634 | |
| 635 | Stmt 1 and 3 do not need to be vectorized, because loop control and |
| 636 | addressing of vectorized data-refs are handled differently. |
| 637 | |
| 638 | This pass detects such stmts. */ |
| 639 | |
| 640 | opt_result |
| 641 | vect_mark_stmts_to_be_vectorized (loop_vec_info loop_vinfo, bool *fatal) |
| 642 | { |
| 643 | class loop *loop = LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop; |
| 644 | basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo)(loop_vinfo)->bbs; |
| 645 | unsigned int nbbs = loop->num_nodes; |
| 646 | gimple_stmt_iterator si; |
| 647 | unsigned int i; |
| 648 | basic_block bb; |
| 649 | bool live_p; |
| 650 | enum vect_relevant relevant; |
| 651 | |
| 652 | DUMP_VECT_SCOPE ("vect_mark_stmts_to_be_vectorized")auto_dump_scope scope ("vect_mark_stmts_to_be_vectorized", vect_location ); |
| 653 | |
| 654 | auto_vec<stmt_vec_info, 64> worklist; |
| 655 | |
| 656 | /* 1. Init worklist. */ |
| 657 | for (i = 0; i < nbbs; i++) |
| 658 | { |
| 659 | bb = bbs[i]; |
| 660 | for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) |
| 661 | { |
| 662 | stmt_vec_info phi_info = loop_vinfo->lookup_stmt (gsi_stmt (si)); |
| 663 | if (dump_enabled_p ()) |
| 664 | dump_printf_loc (MSG_NOTE, vect_location, "init: phi relevant? %G", |
| 665 | phi_info->stmt); |
| 666 | |
| 667 | if (vect_stmt_relevant_p (phi_info, loop_vinfo, &relevant, &live_p)) |
| 668 | vect_mark_relevant (&worklist, phi_info, relevant, live_p); |
| 669 | } |
| 670 | for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) |
| 671 | { |
| 672 | if (is_gimple_debug (gsi_stmt (si))) |
| 673 | continue; |
| 674 | stmt_vec_info stmt_info = loop_vinfo->lookup_stmt (gsi_stmt (si)); |
| 675 | if (dump_enabled_p ()) |
| 676 | dump_printf_loc (MSG_NOTE, vect_location, |
| 677 | "init: stmt relevant? %G", stmt_info->stmt); |
| 678 | |
| 679 | if (vect_stmt_relevant_p (stmt_info, loop_vinfo, &relevant, &live_p)) |
| 680 | vect_mark_relevant (&worklist, stmt_info, relevant, live_p); |
| 681 | } |
| 682 | } |
| 683 | |
| 684 | /* 2. Process_worklist */ |
| 685 | while (worklist.length () > 0) |
| 686 | { |
| 687 | use_operand_p use_p; |
| 688 | ssa_op_iter iter; |
| 689 | |
| 690 | stmt_vec_info stmt_vinfo = worklist.pop (); |
| 691 | if (dump_enabled_p ()) |
| 692 | dump_printf_loc (MSG_NOTE, vect_location, |
| 693 | "worklist: examine stmt: %G", stmt_vinfo->stmt); |
| 694 | |
| 695 | /* Examine the USEs of STMT. For each USE, mark the stmt that defines it |
| 696 | (DEF_STMT) as relevant/irrelevant according to the relevance property |
| 697 | of STMT. */ |
| 698 | relevant = STMT_VINFO_RELEVANT (stmt_vinfo)(stmt_vinfo)->relevant; |
| 699 | |
| 700 | /* Generally, the relevance property of STMT (in STMT_VINFO_RELEVANT) is |
| 701 | propagated as is to the DEF_STMTs of its USEs. |
| 702 | |
| 703 | One exception is when STMT has been identified as defining a reduction |
| 704 | variable; in this case we set the relevance to vect_used_by_reduction. |
| 705 | This is because we distinguish between two kinds of relevant stmts - |
| 706 | those that are used by a reduction computation, and those that are |
| 707 | (also) used by a regular computation. This allows us later on to |
| 708 | identify stmts that are used solely by a reduction, and therefore the |
| 709 | order of the results that they produce does not have to be kept. */ |
| 710 | |
| 711 | switch (STMT_VINFO_DEF_TYPE (stmt_vinfo)(stmt_vinfo)->def_type) |
| 712 | { |
| 713 | case vect_reduction_def: |
| 714 | gcc_assert (relevant != vect_unused_in_scope)((void)(!(relevant != vect_unused_in_scope) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 714, __FUNCTION__), 0 : 0)); |
| 715 | if (relevant != vect_unused_in_scope |
| 716 | && relevant != vect_used_in_scope |
| 717 | && relevant != vect_used_by_reduction |
| 718 | && relevant != vect_used_only_live) |
| 719 | return opt_result::failure_at |
| 720 | (stmt_vinfo->stmt, "unsupported use of reduction.\n"); |
| 721 | break; |
| 722 | |
| 723 | case vect_nested_cycle: |
| 724 | if (relevant != vect_unused_in_scope |
| 725 | && relevant != vect_used_in_outer_by_reduction |
| 726 | && relevant != vect_used_in_outer) |
| 727 | return opt_result::failure_at |
| 728 | (stmt_vinfo->stmt, "unsupported use of nested cycle.\n"); |
| 729 | break; |
| 730 | |
| 731 | case vect_double_reduction_def: |
| 732 | if (relevant != vect_unused_in_scope |
| 733 | && relevant != vect_used_by_reduction |
| 734 | && relevant != vect_used_only_live) |
| 735 | return opt_result::failure_at |
| 736 | (stmt_vinfo->stmt, "unsupported use of double reduction.\n"); |
| 737 | break; |
| 738 | |
| 739 | default: |
| 740 | break; |
| 741 | } |
| 742 | |
| 743 | if (is_pattern_stmt_p (stmt_vinfo)) |
| 744 | { |
| 745 | /* Pattern statements are not inserted into the code, so |
| 746 | FOR_EACH_PHI_OR_STMT_USE optimizes their operands out, and we |
| 747 | have to scan the RHS or function arguments instead. */ |
| 748 | if (gassign *assign = dyn_cast <gassign *> (stmt_vinfo->stmt)) |
| 749 | { |
| 750 | enum tree_code rhs_code = gimple_assign_rhs_code (assign); |
| 751 | tree op = gimple_assign_rhs1 (assign); |
| 752 | |
| 753 | i = 1; |
| 754 | if (rhs_code == COND_EXPR && COMPARISON_CLASS_P (op)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (op)->base.code))] == tcc_comparison)) |
| 755 | { |
| 756 | opt_result res |
| 757 | = process_use (stmt_vinfo, TREE_OPERAND (op, 0)(*((const_cast<tree*> (tree_operand_check ((op), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 757, __FUNCTION__))))), |
| 758 | loop_vinfo, relevant, &worklist, false); |
| 759 | if (!res) |
| 760 | return res; |
| 761 | res = process_use (stmt_vinfo, TREE_OPERAND (op, 1)(*((const_cast<tree*> (tree_operand_check ((op), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 761, __FUNCTION__))))), |
| 762 | loop_vinfo, relevant, &worklist, false); |
| 763 | if (!res) |
| 764 | return res; |
| 765 | i = 2; |
| 766 | } |
| 767 | for (; i < gimple_num_ops (assign); i++) |
| 768 | { |
| 769 | op = gimple_op (assign, i); |
| 770 | if (TREE_CODE (op)((enum tree_code) (op)->base.code) == SSA_NAME) |
| 771 | { |
| 772 | opt_result res |
| 773 | = process_use (stmt_vinfo, op, loop_vinfo, relevant, |
| 774 | &worklist, false); |
| 775 | if (!res) |
| 776 | return res; |
| 777 | } |
| 778 | } |
| 779 | } |
| 780 | else if (gcall *call = dyn_cast <gcall *> (stmt_vinfo->stmt)) |
| 781 | { |
| 782 | for (i = 0; i < gimple_call_num_args (call); i++) |
| 783 | { |
| 784 | tree arg = gimple_call_arg (call, i); |
| 785 | opt_result res |
| 786 | = process_use (stmt_vinfo, arg, loop_vinfo, relevant, |
| 787 | &worklist, false); |
| 788 | if (!res) |
| 789 | return res; |
| 790 | } |
| 791 | } |
| 792 | } |
| 793 | else |
| 794 | FOR_EACH_PHI_OR_STMT_USE (use_p, stmt_vinfo->stmt, iter, SSA_OP_USE)for ((use_p) = (gimple_code (stmt_vinfo->stmt) == GIMPLE_PHI ? op_iter_init_phiuse (&(iter), as_a <gphi *> (stmt_vinfo ->stmt), 0x01) : op_iter_init_use (&(iter), stmt_vinfo ->stmt, 0x01)); !op_iter_done (&(iter)); (use_p) = op_iter_next_use (&(iter))) |
| 795 | { |
| 796 | tree op = USE_FROM_PTR (use_p)get_use_from_ptr (use_p); |
| 797 | opt_result res |
| 798 | = process_use (stmt_vinfo, op, loop_vinfo, relevant, |
| 799 | &worklist, false); |
| 800 | if (!res) |
| 801 | return res; |
| 802 | } |
| 803 | |
| 804 | if (STMT_VINFO_GATHER_SCATTER_P (stmt_vinfo)(stmt_vinfo)->gather_scatter_p) |
| 805 | { |
| 806 | gather_scatter_info gs_info; |
| 807 | if (!vect_check_gather_scatter (stmt_vinfo, loop_vinfo, &gs_info)) |
| 808 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 808, __FUNCTION__)); |
| 809 | opt_result res |
| 810 | = process_use (stmt_vinfo, gs_info.offset, loop_vinfo, relevant, |
| 811 | &worklist, true); |
| 812 | if (!res) |
| 813 | { |
| 814 | if (fatal) |
| 815 | *fatal = false; |
| 816 | return res; |
| 817 | } |
| 818 | } |
| 819 | } /* while worklist */ |
| 820 | |
| 821 | return opt_result::success (); |
| 822 | } |
| 823 | |
| 824 | /* Function vect_model_simple_cost. |
| 825 | |
| 826 | Models cost for simple operations, i.e. those that only emit ncopies of a |
| 827 | single op. Right now, this does not account for multiple insns that could |
| 828 | be generated for the single vector op. We will handle that shortly. */ |
| 829 | |
| 830 | static void |
| 831 | vect_model_simple_cost (vec_info *, |
| 832 | stmt_vec_info stmt_info, int ncopies, |
| 833 | enum vect_def_type *dt, |
| 834 | int ndts, |
| 835 | slp_tree node, |
| 836 | stmt_vector_for_cost *cost_vec, |
| 837 | vect_cost_for_stmt kind = vector_stmt) |
| 838 | { |
| 839 | int inside_cost = 0, prologue_cost = 0; |
| 840 | |
| 841 | gcc_assert (cost_vec != NULL)((void)(!(cost_vec != nullptr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 841, __FUNCTION__), 0 : 0)); |
| 842 | |
| 843 | /* ??? Somehow we need to fix this at the callers. */ |
| 844 | if (node) |
| 845 | ncopies = SLP_TREE_NUMBER_OF_VEC_STMTS (node)(node)->vec_stmts_size; |
| 846 | |
| 847 | if (!node) |
| 848 | /* Cost the "broadcast" of a scalar operand in to a vector operand. |
| 849 | Use scalar_to_vec to cost the broadcast, as elsewhere in the vector |
| 850 | cost model. */ |
| 851 | for (int i = 0; i < ndts; i++) |
| 852 | if (dt[i] == vect_constant_def || dt[i] == vect_external_def) |
| 853 | prologue_cost += record_stmt_cost (cost_vec, 1, scalar_to_vec, |
| 854 | stmt_info, 0, vect_prologue); |
| 855 | |
| 856 | /* Pass the inside-of-loop statements to the target-specific cost model. */ |
| 857 | inside_cost += record_stmt_cost (cost_vec, ncopies, kind, |
| 858 | stmt_info, 0, vect_body); |
| 859 | |
| 860 | if (dump_enabled_p ()) |
| 861 | dump_printf_loc (MSG_NOTE, vect_location, |
| 862 | "vect_model_simple_cost: inside_cost = %d, " |
| 863 | "prologue_cost = %d .\n", inside_cost, prologue_cost); |
| 864 | } |
| 865 | |
| 866 | |
| 867 | /* Model cost for type demotion and promotion operations. PWR is |
| 868 | normally zero for single-step promotions and demotions. It will be |
| 869 | one if two-step promotion/demotion is required, and so on. NCOPIES |
| 870 | is the number of vector results (and thus number of instructions) |
| 871 | for the narrowest end of the operation chain. Each additional |
| 872 | step doubles the number of instructions required. If WIDEN_ARITH |
| 873 | is true the stmt is doing widening arithmetic. */ |
| 874 | |
| 875 | static void |
| 876 | vect_model_promotion_demotion_cost (stmt_vec_info stmt_info, |
| 877 | enum vect_def_type *dt, |
| 878 | unsigned int ncopies, int pwr, |
| 879 | stmt_vector_for_cost *cost_vec, |
| 880 | bool widen_arith) |
| 881 | { |
| 882 | int i; |
| 883 | int inside_cost = 0, prologue_cost = 0; |
| 884 | |
| 885 | for (i = 0; i < pwr + 1; i++) |
| 886 | { |
| 887 | inside_cost += record_stmt_cost (cost_vec, ncopies, |
| 888 | widen_arith |
| 889 | ? vector_stmt : vec_promote_demote, |
| 890 | stmt_info, 0, vect_body); |
| 891 | ncopies *= 2; |
| 892 | } |
| 893 | |
| 894 | /* FORNOW: Assuming maximum 2 args per stmts. */ |
| 895 | for (i = 0; i < 2; i++) |
| 896 | if (dt[i] == vect_constant_def || dt[i] == vect_external_def) |
| 897 | prologue_cost += record_stmt_cost (cost_vec, 1, vector_stmt, |
| 898 | stmt_info, 0, vect_prologue); |
| 899 | |
| 900 | if (dump_enabled_p ()) |
| 901 | dump_printf_loc (MSG_NOTE, vect_location, |
| 902 | "vect_model_promotion_demotion_cost: inside_cost = %d, " |
| 903 | "prologue_cost = %d .\n", inside_cost, prologue_cost); |
| 904 | } |
| 905 | |
| 906 | /* Returns true if the current function returns DECL. */ |
| 907 | |
| 908 | static bool |
| 909 | cfun_returns (tree decl) |
| 910 | { |
| 911 | edge_iterator ei; |
| 912 | edge e; |
| 913 | FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)for ((ei) = ei_start_1 (&(((((cfun + 0))->cfg->x_exit_block_ptr )->preds))); ei_cond ((ei), &(e)); ei_next (&(ei)) ) |
| 914 | { |
| 915 | greturn *ret = safe_dyn_cast <greturn *> (last_stmt (e->src)); |
| 916 | if (!ret) |
| 917 | continue; |
| 918 | if (gimple_return_retval (ret) == decl) |
| 919 | return true; |
| 920 | /* We often end up with an aggregate copy to the result decl, |
| 921 | handle that case as well. First skip intermediate clobbers |
| 922 | though. */ |
| 923 | gimple *def = ret; |
| 924 | do |
| 925 | { |
| 926 | def = SSA_NAME_DEF_STMT (gimple_vuse (def))(tree_check ((gimple_vuse (def)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 926, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 927 | } |
| 928 | while (gimple_clobber_p (def)); |
| 929 | if (is_a <gassign *> (def) |
| 930 | && gimple_assign_lhs (def) == gimple_return_retval (ret) |
| 931 | && gimple_assign_rhs1 (def) == decl) |
| 932 | return true; |
| 933 | } |
| 934 | return false; |
| 935 | } |
| 936 | |
| 937 | /* Function vect_model_store_cost |
| 938 | |
| 939 | Models cost for stores. In the case of grouped accesses, one access |
| 940 | has the overhead of the grouped access attributed to it. */ |
| 941 | |
| 942 | static void |
| 943 | vect_model_store_cost (vec_info *vinfo, stmt_vec_info stmt_info, int ncopies, |
| 944 | vect_memory_access_type memory_access_type, |
| 945 | dr_alignment_support alignment_support_scheme, |
| 946 | int misalignment, |
| 947 | vec_load_store_type vls_type, slp_tree slp_node, |
| 948 | stmt_vector_for_cost *cost_vec) |
| 949 | { |
| 950 | unsigned int inside_cost = 0, prologue_cost = 0; |
| 951 | stmt_vec_info first_stmt_info = stmt_info; |
| 952 | bool grouped_access_p = STMT_VINFO_GROUPED_ACCESS (stmt_info)((stmt_info)->dr_aux.dr && (((void)(!((stmt_info)-> dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 952, __FUNCTION__), 0 : 0)), (stmt_info)->first_element) ); |
| 953 | |
| 954 | /* ??? Somehow we need to fix this at the callers. */ |
| 955 | if (slp_node) |
| 956 | ncopies = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size; |
| 957 | |
| 958 | if (vls_type == VLS_STORE_INVARIANT) |
| 959 | { |
| 960 | if (!slp_node) |
| 961 | prologue_cost += record_stmt_cost (cost_vec, 1, scalar_to_vec, |
| 962 | stmt_info, 0, vect_prologue); |
| 963 | } |
| 964 | |
| 965 | /* Grouped stores update all elements in the group at once, |
| 966 | so we want the DR for the first statement. */ |
| 967 | if (!slp_node && grouped_access_p) |
| 968 | first_stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 968, __FUNCTION__), 0 : 0)), (stmt_info)->first_element); |
| 969 | |
| 970 | /* True if we should include any once-per-group costs as well as |
| 971 | the cost of the statement itself. For SLP we only get called |
| 972 | once per group anyhow. */ |
| 973 | bool first_stmt_p = (first_stmt_info == stmt_info); |
| 974 | |
| 975 | /* We assume that the cost of a single store-lanes instruction is |
| 976 | equivalent to the cost of DR_GROUP_SIZE separate stores. If a grouped |
| 977 | access is instead being provided by a permute-and-store operation, |
| 978 | include the cost of the permutes. */ |
| 979 | if (first_stmt_p |
| 980 | && memory_access_type == VMAT_CONTIGUOUS_PERMUTE) |
| 981 | { |
| 982 | /* Uses a high and low interleave or shuffle operations for each |
| 983 | needed permute. */ |
| 984 | int group_size = DR_GROUP_SIZE (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 984, __FUNCTION__), 0 : 0)), (first_stmt_info)->size); |
| 985 | int nstmts = ncopies * ceil_log2 (group_size) * group_size; |
| 986 | inside_cost = record_stmt_cost (cost_vec, nstmts, vec_perm, |
| 987 | stmt_info, 0, vect_body); |
| 988 | |
| 989 | if (dump_enabled_p ()) |
| 990 | dump_printf_loc (MSG_NOTE, vect_location, |
| 991 | "vect_model_store_cost: strided group_size = %d .\n", |
| 992 | group_size); |
| 993 | } |
| 994 | |
| 995 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 996 | /* Costs of the stores. */ |
| 997 | if (memory_access_type == VMAT_ELEMENTWISE |
| 998 | || memory_access_type == VMAT_GATHER_SCATTER) |
| 999 | { |
| 1000 | /* N scalar stores plus extracting the elements. */ |
| 1001 | unsigned int assumed_nunits = vect_nunits_for_cost (vectype); |
| 1002 | inside_cost += record_stmt_cost (cost_vec, |
| 1003 | ncopies * assumed_nunits, |
| 1004 | scalar_store, stmt_info, 0, vect_body); |
| 1005 | } |
| 1006 | else |
| 1007 | vect_get_store_cost (vinfo, stmt_info, ncopies, alignment_support_scheme, |
| 1008 | misalignment, &inside_cost, cost_vec); |
| 1009 | |
| 1010 | if (memory_access_type == VMAT_ELEMENTWISE |
| 1011 | || memory_access_type == VMAT_STRIDED_SLP) |
| 1012 | { |
| 1013 | /* N scalar stores plus extracting the elements. */ |
| 1014 | unsigned int assumed_nunits = vect_nunits_for_cost (vectype); |
| 1015 | inside_cost += record_stmt_cost (cost_vec, |
| 1016 | ncopies * assumed_nunits, |
| 1017 | vec_to_scalar, stmt_info, 0, vect_body); |
| 1018 | } |
| 1019 | |
| 1020 | /* When vectorizing a store into the function result assign |
| 1021 | a penalty if the function returns in a multi-register location. |
| 1022 | In this case we assume we'll end up with having to spill the |
| 1023 | vector result and do piecewise loads as a conservative estimate. */ |
| 1024 | tree base = get_base_address (STMT_VINFO_DATA_REF (stmt_info)((stmt_info)->dr_aux.dr + 0)->ref); |
| 1025 | if (base |
| 1026 | && (TREE_CODE (base)((enum tree_code) (base)->base.code) == RESULT_DECL |
| 1027 | || (DECL_P (base)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (base)->base.code))] == tcc_declaration) && cfun_returns (base))) |
| 1028 | && !aggregate_value_p (base, cfun(cfun + 0)->decl)) |
| 1029 | { |
| 1030 | rtx reg = hard_function_value (TREE_TYPE (base)((contains_struct_check ((base), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1030, __FUNCTION__))->typed.type), cfun(cfun + 0)->decl, 0, 1); |
| 1031 | /* ??? Handle PARALLEL in some way. */ |
| 1032 | if (REG_P (reg)(((enum rtx_code) (reg)->code) == REG)) |
| 1033 | { |
| 1034 | int nregs = hard_regno_nregs (REGNO (reg)(rhs_regno(reg)), GET_MODE (reg)((machine_mode) (reg)->mode)); |
| 1035 | /* Assume that a single reg-reg move is possible and cheap, |
| 1036 | do not account for vector to gp register move cost. */ |
| 1037 | if (nregs > 1) |
| 1038 | { |
| 1039 | /* Spill. */ |
| 1040 | prologue_cost += record_stmt_cost (cost_vec, ncopies, |
| 1041 | vector_store, |
| 1042 | stmt_info, 0, vect_epilogue); |
| 1043 | /* Loads. */ |
| 1044 | prologue_cost += record_stmt_cost (cost_vec, ncopies * nregs, |
| 1045 | scalar_load, |
| 1046 | stmt_info, 0, vect_epilogue); |
| 1047 | } |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | if (dump_enabled_p ()) |
| 1052 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1053 | "vect_model_store_cost: inside_cost = %d, " |
| 1054 | "prologue_cost = %d .\n", inside_cost, prologue_cost); |
| 1055 | } |
| 1056 | |
| 1057 | |
| 1058 | /* Calculate cost of DR's memory access. */ |
| 1059 | void |
| 1060 | vect_get_store_cost (vec_info *, stmt_vec_info stmt_info, int ncopies, |
| 1061 | dr_alignment_support alignment_support_scheme, |
| 1062 | int misalignment, |
| 1063 | unsigned int *inside_cost, |
| 1064 | stmt_vector_for_cost *body_cost_vec) |
| 1065 | { |
| 1066 | switch (alignment_support_scheme) |
| 1067 | { |
| 1068 | case dr_aligned: |
| 1069 | { |
| 1070 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, |
| 1071 | vector_store, stmt_info, 0, |
| 1072 | vect_body); |
| 1073 | |
| 1074 | if (dump_enabled_p ()) |
| 1075 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1076 | "vect_model_store_cost: aligned.\n"); |
| 1077 | break; |
| 1078 | } |
| 1079 | |
| 1080 | case dr_unaligned_supported: |
| 1081 | { |
| 1082 | /* Here, we assign an additional cost for the unaligned store. */ |
| 1083 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, |
| 1084 | unaligned_store, stmt_info, |
| 1085 | misalignment, vect_body); |
| 1086 | if (dump_enabled_p ()) |
| 1087 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1088 | "vect_model_store_cost: unaligned supported by " |
| 1089 | "hardware.\n"); |
| 1090 | break; |
| 1091 | } |
| 1092 | |
| 1093 | case dr_unaligned_unsupported: |
| 1094 | { |
| 1095 | *inside_cost = VECT_MAX_COST1000; |
| 1096 | |
| 1097 | if (dump_enabled_p ()) |
| 1098 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1099 | "vect_model_store_cost: unsupported access.\n"); |
| 1100 | break; |
| 1101 | } |
| 1102 | |
| 1103 | default: |
| 1104 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1104, __FUNCTION__)); |
| 1105 | } |
| 1106 | } |
| 1107 | |
| 1108 | |
| 1109 | /* Function vect_model_load_cost |
| 1110 | |
| 1111 | Models cost for loads. In the case of grouped accesses, one access has |
| 1112 | the overhead of the grouped access attributed to it. Since unaligned |
| 1113 | accesses are supported for loads, we also account for the costs of the |
| 1114 | access scheme chosen. */ |
| 1115 | |
| 1116 | static void |
| 1117 | vect_model_load_cost (vec_info *vinfo, |
| 1118 | stmt_vec_info stmt_info, unsigned ncopies, poly_uint64 vf, |
| 1119 | vect_memory_access_type memory_access_type, |
| 1120 | dr_alignment_support alignment_support_scheme, |
| 1121 | int misalignment, |
| 1122 | gather_scatter_info *gs_info, |
| 1123 | slp_tree slp_node, |
| 1124 | stmt_vector_for_cost *cost_vec) |
| 1125 | { |
| 1126 | unsigned int inside_cost = 0, prologue_cost = 0; |
| 1127 | bool grouped_access_p = STMT_VINFO_GROUPED_ACCESS (stmt_info)((stmt_info)->dr_aux.dr && (((void)(!((stmt_info)-> dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1127, __FUNCTION__), 0 : 0)), (stmt_info)->first_element )); |
| 1128 | |
| 1129 | gcc_assert (cost_vec)((void)(!(cost_vec) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1129, __FUNCTION__), 0 : 0)); |
| 1130 | |
| 1131 | /* ??? Somehow we need to fix this at the callers. */ |
| 1132 | if (slp_node) |
| 1133 | ncopies = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size; |
| 1134 | |
| 1135 | if (slp_node && SLP_TREE_LOAD_PERMUTATION (slp_node)(slp_node)->load_permutation.exists ()) |
| 1136 | { |
| 1137 | /* If the load is permuted then the alignment is determined by |
| 1138 | the first group element not by the first scalar stmt DR. */ |
| 1139 | stmt_vec_info first_stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1139, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ); |
| 1140 | /* Record the cost for the permutation. */ |
| 1141 | unsigned n_perms, n_loads; |
| 1142 | vect_transform_slp_perm_load (vinfo, slp_node, vNULL, NULLnullptr, |
| 1143 | vf, true, &n_perms, &n_loads); |
| 1144 | inside_cost += record_stmt_cost (cost_vec, n_perms, vec_perm, |
| 1145 | first_stmt_info, 0, vect_body); |
| 1146 | |
| 1147 | /* And adjust the number of loads performed. This handles |
| 1148 | redundancies as well as loads that are later dead. */ |
| 1149 | ncopies = n_loads; |
| 1150 | } |
| 1151 | |
| 1152 | /* Grouped loads read all elements in the group at once, |
| 1153 | so we want the DR for the first statement. */ |
| 1154 | stmt_vec_info first_stmt_info = stmt_info; |
| 1155 | if (!slp_node && grouped_access_p) |
| 1156 | first_stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1156, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ); |
| 1157 | |
| 1158 | /* True if we should include any once-per-group costs as well as |
| 1159 | the cost of the statement itself. For SLP we only get called |
| 1160 | once per group anyhow. */ |
| 1161 | bool first_stmt_p = (first_stmt_info == stmt_info); |
| 1162 | |
| 1163 | /* An IFN_LOAD_LANES will load all its vector results, regardless of which |
| 1164 | ones we actually need. Account for the cost of unused results. */ |
| 1165 | if (first_stmt_p && !slp_node && memory_access_type == VMAT_LOAD_STORE_LANES) |
| 1166 | { |
| 1167 | unsigned int gaps = DR_GROUP_SIZE (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1167, __FUNCTION__), 0 : 0)), (first_stmt_info)->size); |
| 1168 | stmt_vec_info next_stmt_info = first_stmt_info; |
| 1169 | do |
| 1170 | { |
| 1171 | gaps -= 1; |
| 1172 | next_stmt_info = DR_GROUP_NEXT_ELEMENT (next_stmt_info)(((void)(!((next_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1172, __FUNCTION__), 0 : 0)), (next_stmt_info)->next_element ); |
| 1173 | } |
| 1174 | while (next_stmt_info); |
| 1175 | if (gaps) |
| 1176 | { |
| 1177 | if (dump_enabled_p ()) |
| 1178 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1179 | "vect_model_load_cost: %d unused vectors.\n", |
| 1180 | gaps); |
| 1181 | vect_get_load_cost (vinfo, stmt_info, ncopies * gaps, |
| 1182 | alignment_support_scheme, misalignment, false, |
| 1183 | &inside_cost, &prologue_cost, |
| 1184 | cost_vec, cost_vec, true); |
| 1185 | } |
| 1186 | } |
| 1187 | |
| 1188 | /* We assume that the cost of a single load-lanes instruction is |
| 1189 | equivalent to the cost of DR_GROUP_SIZE separate loads. If a grouped |
| 1190 | access is instead being provided by a load-and-permute operation, |
| 1191 | include the cost of the permutes. */ |
| 1192 | if (first_stmt_p |
| 1193 | && memory_access_type == VMAT_CONTIGUOUS_PERMUTE) |
| 1194 | { |
| 1195 | /* Uses an even and odd extract operations or shuffle operations |
| 1196 | for each needed permute. */ |
| 1197 | int group_size = DR_GROUP_SIZE (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1197, __FUNCTION__), 0 : 0)), (first_stmt_info)->size); |
| 1198 | int nstmts = ncopies * ceil_log2 (group_size) * group_size; |
| 1199 | inside_cost += record_stmt_cost (cost_vec, nstmts, vec_perm, |
| 1200 | stmt_info, 0, vect_body); |
| 1201 | |
| 1202 | if (dump_enabled_p ()) |
| 1203 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1204 | "vect_model_load_cost: strided group_size = %d .\n", |
| 1205 | group_size); |
| 1206 | } |
| 1207 | |
| 1208 | /* The loads themselves. */ |
| 1209 | if (memory_access_type == VMAT_ELEMENTWISE |
| 1210 | || memory_access_type == VMAT_GATHER_SCATTER) |
| 1211 | { |
| 1212 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 1213 | unsigned int assumed_nunits = vect_nunits_for_cost (vectype); |
| 1214 | if (memory_access_type == VMAT_GATHER_SCATTER |
| 1215 | && gs_info->ifn == IFN_LAST && !gs_info->decl) |
| 1216 | /* For emulated gathers N offset vector element extracts |
| 1217 | (we assume the scalar scaling and ptr + offset add is consumed by |
| 1218 | the load). */ |
| 1219 | inside_cost += record_stmt_cost (cost_vec, ncopies * assumed_nunits, |
| 1220 | vec_to_scalar, stmt_info, 0, |
| 1221 | vect_body); |
| 1222 | /* N scalar loads plus gathering them into a vector. */ |
| 1223 | inside_cost += record_stmt_cost (cost_vec, |
| 1224 | ncopies * assumed_nunits, |
| 1225 | scalar_load, stmt_info, 0, vect_body); |
| 1226 | } |
| 1227 | else if (memory_access_type == VMAT_INVARIANT) |
| 1228 | { |
| 1229 | /* Invariant loads will ideally be hoisted and splat to a vector. */ |
| 1230 | prologue_cost += record_stmt_cost (cost_vec, 1, |
| 1231 | scalar_load, stmt_info, 0, |
| 1232 | vect_prologue); |
| 1233 | prologue_cost += record_stmt_cost (cost_vec, 1, |
| 1234 | scalar_to_vec, stmt_info, 0, |
| 1235 | vect_prologue); |
| 1236 | } |
| 1237 | else |
| 1238 | vect_get_load_cost (vinfo, stmt_info, ncopies, |
| 1239 | alignment_support_scheme, misalignment, first_stmt_p, |
| 1240 | &inside_cost, &prologue_cost, |
| 1241 | cost_vec, cost_vec, true); |
| 1242 | if (memory_access_type == VMAT_ELEMENTWISE |
| 1243 | || memory_access_type == VMAT_STRIDED_SLP |
| 1244 | || (memory_access_type == VMAT_GATHER_SCATTER |
| 1245 | && gs_info->ifn == IFN_LAST && !gs_info->decl)) |
| 1246 | inside_cost += record_stmt_cost (cost_vec, ncopies, vec_construct, |
| 1247 | stmt_info, 0, vect_body); |
| 1248 | |
| 1249 | if (dump_enabled_p ()) |
| 1250 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1251 | "vect_model_load_cost: inside_cost = %d, " |
| 1252 | "prologue_cost = %d .\n", inside_cost, prologue_cost); |
| 1253 | } |
| 1254 | |
| 1255 | |
| 1256 | /* Calculate cost of DR's memory access. */ |
| 1257 | void |
| 1258 | vect_get_load_cost (vec_info *, stmt_vec_info stmt_info, int ncopies, |
| 1259 | dr_alignment_support alignment_support_scheme, |
| 1260 | int misalignment, |
| 1261 | bool add_realign_cost, unsigned int *inside_cost, |
| 1262 | unsigned int *prologue_cost, |
| 1263 | stmt_vector_for_cost *prologue_cost_vec, |
| 1264 | stmt_vector_for_cost *body_cost_vec, |
| 1265 | bool record_prologue_costs) |
| 1266 | { |
| 1267 | switch (alignment_support_scheme) |
| 1268 | { |
| 1269 | case dr_aligned: |
| 1270 | { |
| 1271 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, vector_load, |
| 1272 | stmt_info, 0, vect_body); |
| 1273 | |
| 1274 | if (dump_enabled_p ()) |
| 1275 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1276 | "vect_model_load_cost: aligned.\n"); |
| 1277 | |
| 1278 | break; |
| 1279 | } |
| 1280 | case dr_unaligned_supported: |
| 1281 | { |
| 1282 | /* Here, we assign an additional cost for the unaligned load. */ |
| 1283 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, |
| 1284 | unaligned_load, stmt_info, |
| 1285 | misalignment, vect_body); |
| 1286 | |
| 1287 | if (dump_enabled_p ()) |
| 1288 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1289 | "vect_model_load_cost: unaligned supported by " |
| 1290 | "hardware.\n"); |
| 1291 | |
| 1292 | break; |
| 1293 | } |
| 1294 | case dr_explicit_realign: |
| 1295 | { |
| 1296 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies * 2, |
| 1297 | vector_load, stmt_info, 0, vect_body); |
| 1298 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, |
| 1299 | vec_perm, stmt_info, 0, vect_body); |
| 1300 | |
| 1301 | /* FIXME: If the misalignment remains fixed across the iterations of |
| 1302 | the containing loop, the following cost should be added to the |
| 1303 | prologue costs. */ |
| 1304 | if (targetm.vectorize.builtin_mask_for_load) |
| 1305 | *inside_cost += record_stmt_cost (body_cost_vec, 1, vector_stmt, |
| 1306 | stmt_info, 0, vect_body); |
| 1307 | |
| 1308 | if (dump_enabled_p ()) |
| 1309 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1310 | "vect_model_load_cost: explicit realign\n"); |
| 1311 | |
| 1312 | break; |
| 1313 | } |
| 1314 | case dr_explicit_realign_optimized: |
| 1315 | { |
| 1316 | if (dump_enabled_p ()) |
| 1317 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1318 | "vect_model_load_cost: unaligned software " |
| 1319 | "pipelined.\n"); |
| 1320 | |
| 1321 | /* Unaligned software pipeline has a load of an address, an initial |
| 1322 | load, and possibly a mask operation to "prime" the loop. However, |
| 1323 | if this is an access in a group of loads, which provide grouped |
| 1324 | access, then the above cost should only be considered for one |
| 1325 | access in the group. Inside the loop, there is a load op |
| 1326 | and a realignment op. */ |
| 1327 | |
| 1328 | if (add_realign_cost && record_prologue_costs) |
| 1329 | { |
| 1330 | *prologue_cost += record_stmt_cost (prologue_cost_vec, 2, |
| 1331 | vector_stmt, stmt_info, |
| 1332 | 0, vect_prologue); |
| 1333 | if (targetm.vectorize.builtin_mask_for_load) |
| 1334 | *prologue_cost += record_stmt_cost (prologue_cost_vec, 1, |
| 1335 | vector_stmt, stmt_info, |
| 1336 | 0, vect_prologue); |
| 1337 | } |
| 1338 | |
| 1339 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, vector_load, |
| 1340 | stmt_info, 0, vect_body); |
| 1341 | *inside_cost += record_stmt_cost (body_cost_vec, ncopies, vec_perm, |
| 1342 | stmt_info, 0, vect_body); |
| 1343 | |
| 1344 | if (dump_enabled_p ()) |
| 1345 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1346 | "vect_model_load_cost: explicit realign optimized" |
| 1347 | "\n"); |
| 1348 | |
| 1349 | break; |
| 1350 | } |
| 1351 | |
| 1352 | case dr_unaligned_unsupported: |
| 1353 | { |
| 1354 | *inside_cost = VECT_MAX_COST1000; |
| 1355 | |
| 1356 | if (dump_enabled_p ()) |
| 1357 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1358 | "vect_model_load_cost: unsupported access.\n"); |
| 1359 | break; |
| 1360 | } |
| 1361 | |
| 1362 | default: |
| 1363 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1363, __FUNCTION__)); |
| 1364 | } |
| 1365 | } |
| 1366 | |
| 1367 | /* Insert the new stmt NEW_STMT at *GSI or at the appropriate place in |
| 1368 | the loop preheader for the vectorized stmt STMT_VINFO. */ |
| 1369 | |
| 1370 | static void |
| 1371 | vect_init_vector_1 (vec_info *vinfo, stmt_vec_info stmt_vinfo, gimple *new_stmt, |
| 1372 | gimple_stmt_iterator *gsi) |
| 1373 | { |
| 1374 | if (gsi) |
| 1375 | vect_finish_stmt_generation (vinfo, stmt_vinfo, new_stmt, gsi); |
| 1376 | else |
| 1377 | vinfo->insert_on_entry (stmt_vinfo, new_stmt); |
| 1378 | |
| 1379 | if (dump_enabled_p ()) |
| 1380 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1381 | "created new init_stmt: %G", new_stmt); |
| 1382 | } |
| 1383 | |
| 1384 | /* Function vect_init_vector. |
| 1385 | |
| 1386 | Insert a new stmt (INIT_STMT) that initializes a new variable of type |
| 1387 | TYPE with the value VAL. If TYPE is a vector type and VAL does not have |
| 1388 | vector type a vector with all elements equal to VAL is created first. |
| 1389 | Place the initialization at GSI if it is not NULL. Otherwise, place the |
| 1390 | initialization at the loop preheader. |
| 1391 | Return the DEF of INIT_STMT. |
| 1392 | It will be used in the vectorization of STMT_INFO. */ |
| 1393 | |
| 1394 | tree |
| 1395 | vect_init_vector (vec_info *vinfo, stmt_vec_info stmt_info, tree val, tree type, |
| 1396 | gimple_stmt_iterator *gsi) |
| 1397 | { |
| 1398 | gimple *init_stmt; |
| 1399 | tree new_temp; |
| 1400 | |
| 1401 | /* We abuse this function to push sth to a SSA name with initial 'val'. */ |
| 1402 | if (! useless_type_conversion_p (type, TREE_TYPE (val)((contains_struct_check ((val), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1402, __FUNCTION__))->typed.type))) |
| 1403 | { |
| 1404 | gcc_assert (TREE_CODE (type) == VECTOR_TYPE)((void)(!(((enum tree_code) (type)->base.code) == VECTOR_TYPE ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1404, __FUNCTION__), 0 : 0)); |
| 1405 | if (! types_compatible_p (TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1405, __FUNCTION__))->typed.type), TREE_TYPE (val)((contains_struct_check ((val), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1405, __FUNCTION__))->typed.type))) |
| 1406 | { |
| 1407 | /* Scalar boolean value should be transformed into |
| 1408 | all zeros or all ones value before building a vector. */ |
| 1409 | if (VECTOR_BOOLEAN_TYPE_P (type)(((enum tree_code) (type)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((type), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1409, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 1410 | { |
| 1411 | tree true_val = build_all_ones_cst (TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1411, __FUNCTION__))->typed.type)); |
| 1412 | tree false_val = build_zero_cst (TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1412, __FUNCTION__))->typed.type)); |
| 1413 | |
| 1414 | if (CONSTANT_CLASS_P (val)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (val)->base.code))] == tcc_constant)) |
| 1415 | val = integer_zerop (val) ? false_val : true_val; |
| 1416 | else |
| 1417 | { |
| 1418 | new_temp = make_ssa_name (TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1418, __FUNCTION__))->typed.type)); |
| 1419 | init_stmt = gimple_build_assign (new_temp, COND_EXPR, |
| 1420 | val, true_val, false_val); |
| 1421 | vect_init_vector_1 (vinfo, stmt_info, init_stmt, gsi); |
| 1422 | val = new_temp; |
| 1423 | } |
| 1424 | } |
| 1425 | else |
| 1426 | { |
| 1427 | gimple_seq stmts = NULLnullptr; |
| 1428 | if (! INTEGRAL_TYPE_P (TREE_TYPE (val))(((enum tree_code) (((contains_struct_check ((val), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1428, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE || ((enum tree_code) (((contains_struct_check ((val), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1428, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((enum tree_code) (((contains_struct_check ((val), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1428, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE )) |
| 1429 | val = gimple_build (&stmts, VIEW_CONVERT_EXPR, |
| 1430 | TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1430, __FUNCTION__))->typed.type), val); |
| 1431 | else |
| 1432 | /* ??? Condition vectorization expects us to do |
| 1433 | promotion of invariant/external defs. */ |
| 1434 | val = gimple_convert (&stmts, TREE_TYPE (type)((contains_struct_check ((type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1434, __FUNCTION__))->typed.type), val); |
| 1435 | for (gimple_stmt_iterator gsi2 = gsi_start (stmts); |
| 1436 | !gsi_end_p (gsi2); ) |
| 1437 | { |
| 1438 | init_stmt = gsi_stmt (gsi2); |
| 1439 | gsi_remove (&gsi2, false); |
| 1440 | vect_init_vector_1 (vinfo, stmt_info, init_stmt, gsi); |
| 1441 | } |
| 1442 | } |
| 1443 | } |
| 1444 | val = build_vector_from_val (type, val); |
| 1445 | } |
| 1446 | |
| 1447 | new_temp = vect_get_new_ssa_name (type, vect_simple_var, "cst_"); |
| 1448 | init_stmt = gimple_build_assign (new_temp, val); |
| 1449 | vect_init_vector_1 (vinfo, stmt_info, init_stmt, gsi); |
| 1450 | return new_temp; |
| 1451 | } |
| 1452 | |
| 1453 | |
| 1454 | /* Function vect_get_vec_defs_for_operand. |
| 1455 | |
| 1456 | OP is an operand in STMT_VINFO. This function returns a vector of |
| 1457 | NCOPIES defs that will be used in the vectorized stmts for STMT_VINFO. |
| 1458 | |
| 1459 | In the case that OP is an SSA_NAME which is defined in the loop, then |
| 1460 | STMT_VINFO_VEC_STMTS of the defining stmt holds the relevant defs. |
| 1461 | |
| 1462 | In case OP is an invariant or constant, a new stmt that creates a vector def |
| 1463 | needs to be introduced. VECTYPE may be used to specify a required type for |
| 1464 | vector invariant. */ |
| 1465 | |
| 1466 | void |
| 1467 | vect_get_vec_defs_for_operand (vec_info *vinfo, stmt_vec_info stmt_vinfo, |
| 1468 | unsigned ncopies, |
| 1469 | tree op, vec<tree> *vec_oprnds, tree vectype) |
| 1470 | { |
| 1471 | gimple *def_stmt; |
| 1472 | enum vect_def_type dt; |
| 1473 | bool is_simple_use; |
| 1474 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 1475 | |
| 1476 | if (dump_enabled_p ()) |
| 1477 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1478 | "vect_get_vec_defs_for_operand: %T\n", op); |
| 1479 | |
| 1480 | stmt_vec_info def_stmt_info; |
| 1481 | is_simple_use = vect_is_simple_use (op, loop_vinfo, &dt, |
| 1482 | &def_stmt_info, &def_stmt); |
| 1483 | gcc_assert (is_simple_use)((void)(!(is_simple_use) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1483, __FUNCTION__), 0 : 0)); |
| 1484 | if (def_stmt && dump_enabled_p ()) |
| 1485 | dump_printf_loc (MSG_NOTE, vect_location, " def_stmt = %G", def_stmt); |
| 1486 | |
| 1487 | vec_oprnds->create (ncopies); |
| 1488 | if (dt == vect_constant_def || dt == vect_external_def) |
| 1489 | { |
| 1490 | tree stmt_vectype = STMT_VINFO_VECTYPE (stmt_vinfo)(stmt_vinfo)->vectype; |
| 1491 | tree vector_type; |
| 1492 | |
| 1493 | if (vectype) |
| 1494 | vector_type = vectype; |
| 1495 | else if (VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE ) && ((tree_class_check ((((contains_struct_check ((op ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->type_common.precision) == 1 && ((tree_class_check ((((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1495, __FUNCTION__))->base.u.bits.unsigned_flag))) |
| 1496 | && VECTOR_BOOLEAN_TYPE_P (stmt_vectype)(((enum tree_code) (stmt_vectype)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((stmt_vectype ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1496, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 1497 | vector_type = truth_type_for (stmt_vectype); |
| 1498 | else |
| 1499 | vector_type = get_vectype_for_scalar_type (loop_vinfo, TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1499, __FUNCTION__))->typed.type)); |
| 1500 | |
| 1501 | gcc_assert (vector_type)((void)(!(vector_type) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1501, __FUNCTION__), 0 : 0)); |
| 1502 | tree vop = vect_init_vector (vinfo, stmt_vinfo, op, vector_type, NULLnullptr); |
| 1503 | while (ncopies--) |
| 1504 | vec_oprnds->quick_push (vop); |
| 1505 | } |
| 1506 | else |
| 1507 | { |
| 1508 | def_stmt_info = vect_stmt_to_vectorize (def_stmt_info); |
| 1509 | gcc_assert (STMT_VINFO_VEC_STMTS (def_stmt_info).length () == ncopies)((void)(!((def_stmt_info)->vec_stmts.length () == ncopies) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1509, __FUNCTION__), 0 : 0)); |
| 1510 | for (unsigned i = 0; i < ncopies; ++i) |
| 1511 | vec_oprnds->quick_push (gimple_get_lhs |
| 1512 | (STMT_VINFO_VEC_STMTS (def_stmt_info)(def_stmt_info)->vec_stmts[i])); |
| 1513 | } |
| 1514 | } |
| 1515 | |
| 1516 | |
| 1517 | /* Get vectorized definitions for OP0 and OP1. */ |
| 1518 | |
| 1519 | void |
| 1520 | vect_get_vec_defs (vec_info *vinfo, stmt_vec_info stmt_info, slp_tree slp_node, |
| 1521 | unsigned ncopies, |
| 1522 | tree op0, vec<tree> *vec_oprnds0, tree vectype0, |
| 1523 | tree op1, vec<tree> *vec_oprnds1, tree vectype1, |
| 1524 | tree op2, vec<tree> *vec_oprnds2, tree vectype2, |
| 1525 | tree op3, vec<tree> *vec_oprnds3, tree vectype3) |
| 1526 | { |
| 1527 | if (slp_node) |
| 1528 | { |
| 1529 | if (op0) |
| 1530 | vect_get_slp_defs (SLP_TREE_CHILDREN (slp_node)(slp_node)->children[0], vec_oprnds0); |
| 1531 | if (op1) |
| 1532 | vect_get_slp_defs (SLP_TREE_CHILDREN (slp_node)(slp_node)->children[1], vec_oprnds1); |
| 1533 | if (op2) |
| 1534 | vect_get_slp_defs (SLP_TREE_CHILDREN (slp_node)(slp_node)->children[2], vec_oprnds2); |
| 1535 | if (op3) |
| 1536 | vect_get_slp_defs (SLP_TREE_CHILDREN (slp_node)(slp_node)->children[3], vec_oprnds3); |
| 1537 | } |
| 1538 | else |
| 1539 | { |
| 1540 | if (op0) |
| 1541 | vect_get_vec_defs_for_operand (vinfo, stmt_info, ncopies, |
| 1542 | op0, vec_oprnds0, vectype0); |
| 1543 | if (op1) |
| 1544 | vect_get_vec_defs_for_operand (vinfo, stmt_info, ncopies, |
| 1545 | op1, vec_oprnds1, vectype1); |
| 1546 | if (op2) |
| 1547 | vect_get_vec_defs_for_operand (vinfo, stmt_info, ncopies, |
| 1548 | op2, vec_oprnds2, vectype2); |
| 1549 | if (op3) |
| 1550 | vect_get_vec_defs_for_operand (vinfo, stmt_info, ncopies, |
| 1551 | op3, vec_oprnds3, vectype3); |
| 1552 | } |
| 1553 | } |
| 1554 | |
| 1555 | void |
| 1556 | vect_get_vec_defs (vec_info *vinfo, stmt_vec_info stmt_info, slp_tree slp_node, |
| 1557 | unsigned ncopies, |
| 1558 | tree op0, vec<tree> *vec_oprnds0, |
| 1559 | tree op1, vec<tree> *vec_oprnds1, |
| 1560 | tree op2, vec<tree> *vec_oprnds2, |
| 1561 | tree op3, vec<tree> *vec_oprnds3) |
| 1562 | { |
| 1563 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies, |
| 1564 | op0, vec_oprnds0, NULL_TREE(tree) nullptr, |
| 1565 | op1, vec_oprnds1, NULL_TREE(tree) nullptr, |
| 1566 | op2, vec_oprnds2, NULL_TREE(tree) nullptr, |
| 1567 | op3, vec_oprnds3, NULL_TREE(tree) nullptr); |
| 1568 | } |
| 1569 | |
| 1570 | /* Helper function called by vect_finish_replace_stmt and |
| 1571 | vect_finish_stmt_generation. Set the location of the new |
| 1572 | statement and create and return a stmt_vec_info for it. */ |
| 1573 | |
| 1574 | static void |
| 1575 | vect_finish_stmt_generation_1 (vec_info *, |
| 1576 | stmt_vec_info stmt_info, gimple *vec_stmt) |
| 1577 | { |
| 1578 | if (dump_enabled_p ()) |
| 1579 | dump_printf_loc (MSG_NOTE, vect_location, "add new stmt: %G", vec_stmt); |
| 1580 | |
| 1581 | if (stmt_info) |
| 1582 | { |
| 1583 | gimple_set_location (vec_stmt, gimple_location (stmt_info->stmt)); |
| 1584 | |
| 1585 | /* While EH edges will generally prevent vectorization, stmt might |
| 1586 | e.g. be in a must-not-throw region. Ensure newly created stmts |
| 1587 | that could throw are part of the same region. */ |
| 1588 | int lp_nr = lookup_stmt_eh_lp (stmt_info->stmt); |
| 1589 | if (lp_nr != 0 && stmt_could_throw_p (cfun(cfun + 0), vec_stmt)) |
| 1590 | add_stmt_to_eh_lp (vec_stmt, lp_nr); |
| 1591 | } |
| 1592 | else |
| 1593 | gcc_assert (!stmt_could_throw_p (cfun, vec_stmt))((void)(!(!stmt_could_throw_p ((cfun + 0), vec_stmt)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1593, __FUNCTION__), 0 : 0)); |
| 1594 | } |
| 1595 | |
| 1596 | /* Replace the scalar statement STMT_INFO with a new vector statement VEC_STMT, |
| 1597 | which sets the same scalar result as STMT_INFO did. Create and return a |
| 1598 | stmt_vec_info for VEC_STMT. */ |
| 1599 | |
| 1600 | void |
| 1601 | vect_finish_replace_stmt (vec_info *vinfo, |
| 1602 | stmt_vec_info stmt_info, gimple *vec_stmt) |
| 1603 | { |
| 1604 | gimple *scalar_stmt = vect_orig_stmt (stmt_info)->stmt; |
| 1605 | gcc_assert (gimple_get_lhs (scalar_stmt) == gimple_get_lhs (vec_stmt))((void)(!(gimple_get_lhs (scalar_stmt) == gimple_get_lhs (vec_stmt )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1605, __FUNCTION__), 0 : 0)); |
| 1606 | |
| 1607 | gimple_stmt_iterator gsi = gsi_for_stmt (scalar_stmt); |
| 1608 | gsi_replace (&gsi, vec_stmt, true); |
| 1609 | |
| 1610 | vect_finish_stmt_generation_1 (vinfo, stmt_info, vec_stmt); |
| 1611 | } |
| 1612 | |
| 1613 | /* Add VEC_STMT to the vectorized implementation of STMT_INFO and insert it |
| 1614 | before *GSI. Create and return a stmt_vec_info for VEC_STMT. */ |
| 1615 | |
| 1616 | void |
| 1617 | vect_finish_stmt_generation (vec_info *vinfo, |
| 1618 | stmt_vec_info stmt_info, gimple *vec_stmt, |
| 1619 | gimple_stmt_iterator *gsi) |
| 1620 | { |
| 1621 | gcc_assert (!stmt_info || gimple_code (stmt_info->stmt) != GIMPLE_LABEL)((void)(!(!stmt_info || gimple_code (stmt_info->stmt) != GIMPLE_LABEL ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1621, __FUNCTION__), 0 : 0)); |
| 1622 | |
| 1623 | if (!gsi_end_p (*gsi) |
| 1624 | && gimple_has_mem_ops (vec_stmt)) |
| 1625 | { |
| 1626 | gimple *at_stmt = gsi_stmt (*gsi); |
| 1627 | tree vuse = gimple_vuse (at_stmt); |
| 1628 | if (vuse && TREE_CODE (vuse)((enum tree_code) (vuse)->base.code) == SSA_NAME) |
| 1629 | { |
| 1630 | tree vdef = gimple_vdef (at_stmt); |
| 1631 | gimple_set_vuse (vec_stmt, gimple_vuse (at_stmt)); |
| 1632 | gimple_set_modified (vec_stmt, true); |
| 1633 | /* If we have an SSA vuse and insert a store, update virtual |
| 1634 | SSA form to avoid triggering the renamer. Do so only |
| 1635 | if we can easily see all uses - which is what almost always |
| 1636 | happens with the way vectorized stmts are inserted. */ |
| 1637 | if ((vdef && TREE_CODE (vdef)((enum tree_code) (vdef)->base.code) == SSA_NAME) |
| 1638 | && ((is_gimple_assign (vec_stmt) |
| 1639 | && !is_gimple_reg (gimple_assign_lhs (vec_stmt))) |
| 1640 | || (is_gimple_call (vec_stmt) |
| 1641 | && (!(gimple_call_flags (vec_stmt) |
| 1642 | & (ECF_CONST(1 << 0)|ECF_PURE(1 << 1)|ECF_NOVOPS(1 << 9))) |
| 1643 | || (gimple_call_lhs (vec_stmt) |
| 1644 | && !is_gimple_reg (gimple_call_lhs (vec_stmt))))))) |
| 1645 | { |
| 1646 | tree new_vdef = copy_ssa_name (vuse, vec_stmt); |
| 1647 | gimple_set_vdef (vec_stmt, new_vdef); |
| 1648 | SET_USE (gimple_vuse_op (at_stmt), new_vdef)set_ssa_use_from_ptr (gimple_vuse_op (at_stmt), new_vdef); |
| 1649 | } |
| 1650 | } |
| 1651 | } |
| 1652 | gsi_insert_before (gsi, vec_stmt, GSI_SAME_STMT); |
| 1653 | vect_finish_stmt_generation_1 (vinfo, stmt_info, vec_stmt); |
| 1654 | } |
| 1655 | |
| 1656 | /* We want to vectorize a call to combined function CFN with function |
| 1657 | decl FNDECL, using VECTYPE_OUT as the type of the output and VECTYPE_IN |
| 1658 | as the types of all inputs. Check whether this is possible using |
| 1659 | an internal function, returning its code if so or IFN_LAST if not. */ |
| 1660 | |
| 1661 | static internal_fn |
| 1662 | vectorizable_internal_function (combined_fn cfn, tree fndecl, |
| 1663 | tree vectype_out, tree vectype_in) |
| 1664 | { |
| 1665 | internal_fn ifn; |
| 1666 | if (internal_fn_p (cfn)) |
| 1667 | ifn = as_internal_fn (cfn); |
| 1668 | else |
| 1669 | ifn = associated_internal_fn (fndecl); |
| 1670 | if (ifn != IFN_LAST && direct_internal_fn_p (ifn)) |
| 1671 | { |
| 1672 | const direct_internal_fn_info &info = direct_internal_fn (ifn); |
| 1673 | if (info.vectorizable) |
| 1674 | { |
| 1675 | tree type0 = (info.type0 < 0 ? vectype_out : vectype_in); |
| 1676 | tree type1 = (info.type1 < 0 ? vectype_out : vectype_in); |
| 1677 | if (direct_internal_fn_supported_p (ifn, tree_pair (type0, type1), |
| 1678 | OPTIMIZE_FOR_SPEED)) |
| 1679 | return ifn; |
| 1680 | } |
| 1681 | } |
| 1682 | return IFN_LAST; |
| 1683 | } |
| 1684 | |
| 1685 | |
| 1686 | static tree permute_vec_elements (vec_info *, tree, tree, tree, stmt_vec_info, |
| 1687 | gimple_stmt_iterator *); |
| 1688 | |
| 1689 | /* Check whether a load or store statement in the loop described by |
| 1690 | LOOP_VINFO is possible in a loop using partial vectors. This is |
| 1691 | testing whether the vectorizer pass has the appropriate support, |
| 1692 | as well as whether the target does. |
| 1693 | |
| 1694 | VLS_TYPE says whether the statement is a load or store and VECTYPE |
| 1695 | is the type of the vector being loaded or stored. SLP_NODE is the SLP |
| 1696 | node that contains the statement, or null if none. MEMORY_ACCESS_TYPE |
| 1697 | says how the load or store is going to be implemented and GROUP_SIZE |
| 1698 | is the number of load or store statements in the containing group. |
| 1699 | If the access is a gather load or scatter store, GS_INFO describes |
| 1700 | its arguments. If the load or store is conditional, SCALAR_MASK is the |
| 1701 | condition under which it occurs. |
| 1702 | |
| 1703 | Clear LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P if a loop using partial |
| 1704 | vectors is not supported, otherwise record the required rgroup control |
| 1705 | types. */ |
| 1706 | |
| 1707 | static void |
| 1708 | check_load_store_for_partial_vectors (loop_vec_info loop_vinfo, tree vectype, |
| 1709 | slp_tree slp_node, |
| 1710 | vec_load_store_type vls_type, |
| 1711 | int group_size, |
| 1712 | vect_memory_access_type |
| 1713 | memory_access_type, |
| 1714 | gather_scatter_info *gs_info, |
| 1715 | tree scalar_mask) |
| 1716 | { |
| 1717 | /* Invariant loads need no special support. */ |
| 1718 | if (memory_access_type == VMAT_INVARIANT) |
| 1719 | return; |
| 1720 | |
| 1721 | unsigned int nvectors; |
| 1722 | if (slp_node) |
| 1723 | nvectors = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size; |
| 1724 | else |
| 1725 | nvectors = vect_get_num_copies (loop_vinfo, vectype); |
| 1726 | |
| 1727 | vec_loop_masks *masks = &LOOP_VINFO_MASKS (loop_vinfo)(loop_vinfo)->masks; |
| 1728 | machine_mode vecmode = TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1728, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode); |
| 1729 | bool is_load = (vls_type == VLS_LOAD); |
| 1730 | if (memory_access_type == VMAT_LOAD_STORE_LANES) |
| 1731 | { |
| 1732 | if (is_load |
| 1733 | ? !vect_load_lanes_supported (vectype, group_size, true) |
| 1734 | : !vect_store_lanes_supported (vectype, group_size, true)) |
| 1735 | { |
| 1736 | if (dump_enabled_p ()) |
| 1737 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1738 | "can't operate on partial vectors because" |
| 1739 | " the target doesn't have an appropriate" |
| 1740 | " load/store-lanes instruction.\n"); |
| 1741 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 1742 | return; |
| 1743 | } |
| 1744 | vect_record_loop_mask (loop_vinfo, masks, nvectors, vectype, |
| 1745 | scalar_mask); |
| 1746 | return; |
| 1747 | } |
| 1748 | |
| 1749 | if (memory_access_type == VMAT_GATHER_SCATTER) |
| 1750 | { |
| 1751 | internal_fn ifn = (is_load |
| 1752 | ? IFN_MASK_GATHER_LOAD |
| 1753 | : IFN_MASK_SCATTER_STORE); |
| 1754 | if (!internal_gather_scatter_fn_supported_p (ifn, vectype, |
| 1755 | gs_info->memory_type, |
| 1756 | gs_info->offset_vectype, |
| 1757 | gs_info->scale)) |
| 1758 | { |
| 1759 | if (dump_enabled_p ()) |
| 1760 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1761 | "can't operate on partial vectors because" |
| 1762 | " the target doesn't have an appropriate" |
| 1763 | " gather load or scatter store instruction.\n"); |
| 1764 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 1765 | return; |
| 1766 | } |
| 1767 | vect_record_loop_mask (loop_vinfo, masks, nvectors, vectype, |
| 1768 | scalar_mask); |
| 1769 | return; |
| 1770 | } |
| 1771 | |
| 1772 | if (memory_access_type != VMAT_CONTIGUOUS |
| 1773 | && memory_access_type != VMAT_CONTIGUOUS_PERMUTE) |
| 1774 | { |
| 1775 | /* Element X of the data must come from iteration i * VF + X of the |
| 1776 | scalar loop. We need more work to support other mappings. */ |
| 1777 | if (dump_enabled_p ()) |
| 1778 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1779 | "can't operate on partial vectors because an" |
| 1780 | " access isn't contiguous.\n"); |
| 1781 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 1782 | return; |
| 1783 | } |
| 1784 | |
| 1785 | if (!VECTOR_MODE_P (vecmode)(((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_BOOL || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_INT || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_FLOAT || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_FRACT || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_UFRACT || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_ACCUM || ((enum mode_class) mode_class[vecmode]) == MODE_VECTOR_UACCUM )) |
| 1786 | { |
| 1787 | if (dump_enabled_p ()) |
| 1788 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1789 | "can't operate on partial vectors when emulating" |
| 1790 | " vector operations.\n"); |
| 1791 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 1792 | return; |
| 1793 | } |
| 1794 | |
| 1795 | /* We might load more scalars than we need for permuting SLP loads. |
| 1796 | We checked in get_group_load_store_type that the extra elements |
| 1797 | don't leak into a new vector. */ |
| 1798 | auto group_memory_nvectors = [](poly_uint64 size, poly_uint64 nunits) |
| 1799 | { |
| 1800 | unsigned int nvectors; |
| 1801 | if (can_div_away_from_zero_p (size, nunits, &nvectors)) |
| 1802 | return nvectors; |
| 1803 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1803, __FUNCTION__)); |
| 1804 | }; |
| 1805 | |
| 1806 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 1807 | poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo)(loop_vinfo)->vectorization_factor; |
| 1808 | machine_mode mask_mode; |
| 1809 | bool using_partial_vectors_p = false; |
| 1810 | if (targetm.vectorize.get_mask_mode (vecmode).exists (&mask_mode) |
| 1811 | && can_vec_mask_load_store_p (vecmode, mask_mode, is_load)) |
| 1812 | { |
| 1813 | nvectors = group_memory_nvectors (group_size * vf, nunits); |
| 1814 | vect_record_loop_mask (loop_vinfo, masks, nvectors, vectype, scalar_mask); |
| 1815 | using_partial_vectors_p = true; |
| 1816 | } |
| 1817 | |
| 1818 | machine_mode vmode; |
| 1819 | if (get_len_load_store_mode (vecmode, is_load).exists (&vmode)) |
| 1820 | { |
| 1821 | nvectors = group_memory_nvectors (group_size * vf, nunits); |
| 1822 | vec_loop_lens *lens = &LOOP_VINFO_LENS (loop_vinfo)(loop_vinfo)->lens; |
| 1823 | unsigned factor = (vecmode == vmode) ? 1 : GET_MODE_UNIT_SIZE (vecmode)mode_to_unit_size (vecmode); |
| 1824 | vect_record_loop_len (loop_vinfo, lens, nvectors, vectype, factor); |
| 1825 | using_partial_vectors_p = true; |
| 1826 | } |
| 1827 | |
| 1828 | if (!using_partial_vectors_p) |
| 1829 | { |
| 1830 | if (dump_enabled_p ()) |
| 1831 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 1832 | "can't operate on partial vectors because the" |
| 1833 | " target doesn't have the appropriate partial" |
| 1834 | " vectorization load or store.\n"); |
| 1835 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 1836 | } |
| 1837 | } |
| 1838 | |
| 1839 | /* Return the mask input to a masked load or store. VEC_MASK is the vectorized |
| 1840 | form of the scalar mask condition and LOOP_MASK, if nonnull, is the mask |
| 1841 | that needs to be applied to all loads and stores in a vectorized loop. |
| 1842 | Return VEC_MASK if LOOP_MASK is null or if VEC_MASK is already masked, |
| 1843 | otherwise return VEC_MASK & LOOP_MASK. |
| 1844 | |
| 1845 | MASK_TYPE is the type of both masks. If new statements are needed, |
| 1846 | insert them before GSI. */ |
| 1847 | |
| 1848 | static tree |
| 1849 | prepare_vec_mask (loop_vec_info loop_vinfo, tree mask_type, tree loop_mask, |
| 1850 | tree vec_mask, gimple_stmt_iterator *gsi) |
| 1851 | { |
| 1852 | gcc_assert (useless_type_conversion_p (mask_type, TREE_TYPE (vec_mask)))((void)(!(useless_type_conversion_p (mask_type, ((contains_struct_check ((vec_mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1852, __FUNCTION__))->typed.type))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1852, __FUNCTION__), 0 : 0)); |
| 1853 | if (!loop_mask) |
| 1854 | return vec_mask; |
| 1855 | |
| 1856 | gcc_assert (TREE_TYPE (loop_mask) == mask_type)((void)(!(((contains_struct_check ((loop_mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1856, __FUNCTION__))->typed.type) == mask_type) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1856, __FUNCTION__), 0 : 0)); |
| 1857 | |
| 1858 | if (loop_vinfo->vec_cond_masked_set.contains ({ vec_mask, loop_mask })) |
| 1859 | return vec_mask; |
| 1860 | |
| 1861 | tree and_res = make_temp_ssa_name (mask_type, NULLnullptr, "vec_mask_and"); |
| 1862 | gimple *and_stmt = gimple_build_assign (and_res, BIT_AND_EXPR, |
| 1863 | vec_mask, loop_mask); |
| 1864 | |
| 1865 | gsi_insert_before (gsi, and_stmt, GSI_SAME_STMT); |
| 1866 | return and_res; |
| 1867 | } |
| 1868 | |
| 1869 | /* Determine whether we can use a gather load or scatter store to vectorize |
| 1870 | strided load or store STMT_INFO by truncating the current offset to a |
| 1871 | smaller width. We need to be able to construct an offset vector: |
| 1872 | |
| 1873 | { 0, X, X*2, X*3, ... } |
| 1874 | |
| 1875 | without loss of precision, where X is STMT_INFO's DR_STEP. |
| 1876 | |
| 1877 | Return true if this is possible, describing the gather load or scatter |
| 1878 | store in GS_INFO. MASKED_P is true if the load or store is conditional. */ |
| 1879 | |
| 1880 | static bool |
| 1881 | vect_truncate_gather_scatter_offset (stmt_vec_info stmt_info, |
| 1882 | loop_vec_info loop_vinfo, bool masked_p, |
| 1883 | gather_scatter_info *gs_info) |
| 1884 | { |
| 1885 | dr_vec_info *dr_info = STMT_VINFO_DR_INFO (stmt_info)(((void)(!((stmt_info)->dr_aux.stmt == (stmt_info)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1885, __FUNCTION__), 0 : 0)), &(stmt_info)->dr_aux); |
| 1886 | data_reference *dr = dr_info->dr; |
| 1887 | tree step = DR_STEP (dr)(dr)->innermost.step; |
| 1888 | if (TREE_CODE (step)((enum tree_code) (step)->base.code) != INTEGER_CST) |
| 1889 | { |
| 1890 | /* ??? Perhaps we could use range information here? */ |
| 1891 | if (dump_enabled_p ()) |
| 1892 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1893 | "cannot truncate variable step.\n"); |
| 1894 | return false; |
| 1895 | } |
| 1896 | |
| 1897 | /* Get the number of bits in an element. */ |
| 1898 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 1899 | scalar_mode element_mode = SCALAR_TYPE_MODE (TREE_TYPE (vectype))(as_a <scalar_mode> ((tree_class_check ((((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1899, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1899, __FUNCTION__))->type_common.mode)); |
| 1900 | unsigned int element_bits = GET_MODE_BITSIZE (element_mode); |
| 1901 | |
| 1902 | /* Set COUNT to the upper limit on the number of elements - 1. |
| 1903 | Start with the maximum vectorization factor. */ |
| 1904 | unsigned HOST_WIDE_INTlong count = vect_max_vf (loop_vinfo) - 1; |
| 1905 | |
| 1906 | /* Try lowering COUNT to the number of scalar latch iterations. */ |
| 1907 | class loop *loop = LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop; |
| 1908 | widest_int max_iters; |
| 1909 | if (max_loop_iterations (loop, &max_iters) |
| 1910 | && max_iters < count) |
| 1911 | count = max_iters.to_shwi (); |
| 1912 | |
| 1913 | /* Try scales of 1 and the element size. */ |
| 1914 | int scales[] = { 1, vect_get_scalar_dr_size (dr_info) }; |
| 1915 | wi::overflow_type overflow = wi::OVF_NONE; |
| 1916 | for (int i = 0; i < 2; ++i) |
| 1917 | { |
| 1918 | int scale = scales[i]; |
| 1919 | widest_int factor; |
| 1920 | if (!wi::multiple_of_p (wi::to_widest (step), scale, SIGNED, &factor)) |
| 1921 | continue; |
| 1922 | |
| 1923 | /* Determine the minimum precision of (COUNT - 1) * STEP / SCALE. */ |
| 1924 | widest_int range = wi::mul (count, factor, SIGNED, &overflow); |
| 1925 | if (overflow) |
| 1926 | continue; |
| 1927 | signop sign = range >= 0 ? UNSIGNED : SIGNED; |
| 1928 | unsigned int min_offset_bits = wi::min_precision (range, sign); |
| 1929 | |
| 1930 | /* Find the narrowest viable offset type. */ |
| 1931 | unsigned int offset_bits = 1U << ceil_log2 (min_offset_bits); |
| 1932 | tree offset_type = build_nonstandard_integer_type (offset_bits, |
| 1933 | sign == UNSIGNED); |
| 1934 | |
| 1935 | /* See whether the target supports the operation with an offset |
| 1936 | no narrower than OFFSET_TYPE. */ |
| 1937 | tree memory_type = TREE_TYPE (DR_REF (dr))((contains_struct_check (((dr)->ref), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1937, __FUNCTION__))->typed.type); |
| 1938 | if (!vect_gather_scatter_fn_p (loop_vinfo, DR_IS_READ (dr)(dr)->is_read, masked_p, |
| 1939 | vectype, memory_type, offset_type, scale, |
| 1940 | &gs_info->ifn, &gs_info->offset_vectype) |
| 1941 | || gs_info->ifn == IFN_LAST) |
| 1942 | continue; |
| 1943 | |
| 1944 | gs_info->decl = NULL_TREE(tree) nullptr; |
| 1945 | /* Logically the sum of DR_BASE_ADDRESS, DR_INIT and DR_OFFSET, |
| 1946 | but we don't need to store that here. */ |
| 1947 | gs_info->base = NULL_TREE(tree) nullptr; |
| 1948 | gs_info->element_type = TREE_TYPE (vectype)((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1948, __FUNCTION__))->typed.type); |
| 1949 | gs_info->offset = fold_convert (offset_type, step)fold_convert_loc (((location_t) 0), offset_type, step); |
| 1950 | gs_info->offset_dt = vect_constant_def; |
| 1951 | gs_info->scale = scale; |
| 1952 | gs_info->memory_type = memory_type; |
| 1953 | return true; |
| 1954 | } |
| 1955 | |
| 1956 | if (overflow && dump_enabled_p ()) |
| 1957 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1958 | "truncating gather/scatter offset to %d bits" |
| 1959 | " might change its value.\n", element_bits); |
| 1960 | |
| 1961 | return false; |
| 1962 | } |
| 1963 | |
| 1964 | /* Return true if we can use gather/scatter internal functions to |
| 1965 | vectorize STMT_INFO, which is a grouped or strided load or store. |
| 1966 | MASKED_P is true if load or store is conditional. When returning |
| 1967 | true, fill in GS_INFO with the information required to perform the |
| 1968 | operation. */ |
| 1969 | |
| 1970 | static bool |
| 1971 | vect_use_strided_gather_scatters_p (stmt_vec_info stmt_info, |
| 1972 | loop_vec_info loop_vinfo, bool masked_p, |
| 1973 | gather_scatter_info *gs_info) |
| 1974 | { |
| 1975 | if (!vect_check_gather_scatter (stmt_info, loop_vinfo, gs_info) |
| 1976 | || gs_info->ifn == IFN_LAST) |
| 1977 | return vect_truncate_gather_scatter_offset (stmt_info, loop_vinfo, |
| 1978 | masked_p, gs_info); |
| 1979 | |
| 1980 | tree old_offset_type = TREE_TYPE (gs_info->offset)((contains_struct_check ((gs_info->offset), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1980, __FUNCTION__))->typed.type); |
| 1981 | tree new_offset_type = TREE_TYPE (gs_info->offset_vectype)((contains_struct_check ((gs_info->offset_vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1981, __FUNCTION__))->typed.type); |
| 1982 | |
| 1983 | gcc_assert (TYPE_PRECISION (new_offset_type)((void)(!(((tree_class_check ((new_offset_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1983, __FUNCTION__))->type_common.precision) >= ((tree_class_check ((old_offset_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1984, __FUNCTION__))->type_common.precision)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1984, __FUNCTION__), 0 : 0)) |
| 1984 | >= TYPE_PRECISION (old_offset_type))((void)(!(((tree_class_check ((new_offset_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1983, __FUNCTION__))->type_common.precision) >= ((tree_class_check ((old_offset_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1984, __FUNCTION__))->type_common.precision)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 1984, __FUNCTION__), 0 : 0)); |
| 1985 | gs_info->offset = fold_convert (new_offset_type, gs_info->offset)fold_convert_loc (((location_t) 0), new_offset_type, gs_info-> offset); |
| 1986 | |
| 1987 | if (dump_enabled_p ()) |
| 1988 | dump_printf_loc (MSG_NOTE, vect_location, |
| 1989 | "using gather/scatter for strided/grouped access," |
| 1990 | " scale = %d\n", gs_info->scale); |
| 1991 | |
| 1992 | return true; |
| 1993 | } |
| 1994 | |
| 1995 | /* STMT_INFO is a non-strided load or store, meaning that it accesses |
| 1996 | elements with a known constant step. Return -1 if that step |
| 1997 | is negative, 0 if it is zero, and 1 if it is greater than zero. */ |
| 1998 | |
| 1999 | static int |
| 2000 | compare_step_with_zero (vec_info *vinfo, stmt_vec_info stmt_info) |
| 2001 | { |
| 2002 | dr_vec_info *dr_info = STMT_VINFO_DR_INFO (stmt_info)(((void)(!((stmt_info)->dr_aux.stmt == (stmt_info)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2002, __FUNCTION__), 0 : 0)), &(stmt_info)->dr_aux); |
| 2003 | return tree_int_cst_compare (vect_dr_behavior (vinfo, dr_info)->step, |
| 2004 | size_zero_nodeglobal_trees[TI_SIZE_ZERO]); |
| 2005 | } |
| 2006 | |
| 2007 | /* If the target supports a permute mask that reverses the elements in |
| 2008 | a vector of type VECTYPE, return that mask, otherwise return null. */ |
| 2009 | |
| 2010 | static tree |
| 2011 | perm_mask_for_reverse (tree vectype) |
| 2012 | { |
| 2013 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 2014 | |
| 2015 | /* The encoding has a single stepped pattern. */ |
| 2016 | vec_perm_builder sel (nunits, 1, 3); |
| 2017 | for (int i = 0; i < 3; ++i) |
| 2018 | sel.quick_push (nunits - 1 - i); |
| 2019 | |
| 2020 | vec_perm_indices indices (sel, 1, nunits); |
| 2021 | if (!can_vec_perm_const_p (TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2021, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode), TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2021, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode), |
| 2022 | indices)) |
| 2023 | return NULL_TREE(tree) nullptr; |
| 2024 | return vect_gen_perm_mask_checked (vectype, indices); |
| 2025 | } |
| 2026 | |
| 2027 | /* A subroutine of get_load_store_type, with a subset of the same |
| 2028 | arguments. Handle the case where STMT_INFO is a load or store that |
| 2029 | accesses consecutive elements with a negative step. Sets *POFFSET |
| 2030 | to the offset to be applied to the DR for the first access. */ |
| 2031 | |
| 2032 | static vect_memory_access_type |
| 2033 | get_negative_load_store_type (vec_info *vinfo, |
| 2034 | stmt_vec_info stmt_info, tree vectype, |
| 2035 | vec_load_store_type vls_type, |
| 2036 | unsigned int ncopies, poly_int64 *poffset) |
| 2037 | { |
| 2038 | dr_vec_info *dr_info = STMT_VINFO_DR_INFO (stmt_info)(((void)(!((stmt_info)->dr_aux.stmt == (stmt_info)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2038, __FUNCTION__), 0 : 0)), &(stmt_info)->dr_aux); |
| 2039 | dr_alignment_support alignment_support_scheme; |
| 2040 | |
| 2041 | if (ncopies > 1) |
| 2042 | { |
| 2043 | if (dump_enabled_p ()) |
| 2044 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2045 | "multiple types with negative step.\n"); |
| 2046 | return VMAT_ELEMENTWISE; |
| 2047 | } |
| 2048 | |
| 2049 | /* For backward running DRs the first access in vectype actually is |
| 2050 | N-1 elements before the address of the DR. */ |
| 2051 | *poffset = ((-TYPE_VECTOR_SUBPARTS (vectype) + 1) |
| 2052 | * TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (vectype)))((unsigned long) (*tree_int_cst_elt_check ((((tree_class_check ((((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2052, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2052, __FUNCTION__))->type_common.size_unit)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2052, __FUNCTION__)))); |
| 2053 | |
| 2054 | int misalignment = dr_misalignment (dr_info, vectype, *poffset); |
| 2055 | alignment_support_scheme |
| 2056 | = vect_supportable_dr_alignment (vinfo, dr_info, vectype, misalignment); |
| 2057 | if (alignment_support_scheme != dr_aligned |
| 2058 | && alignment_support_scheme != dr_unaligned_supported) |
| 2059 | { |
| 2060 | if (dump_enabled_p ()) |
| 2061 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2062 | "negative step but alignment required.\n"); |
| 2063 | *poffset = 0; |
| 2064 | return VMAT_ELEMENTWISE; |
| 2065 | } |
| 2066 | |
| 2067 | if (vls_type == VLS_STORE_INVARIANT) |
| 2068 | { |
| 2069 | if (dump_enabled_p ()) |
| 2070 | dump_printf_loc (MSG_NOTE, vect_location, |
| 2071 | "negative step with invariant source;" |
| 2072 | " no permute needed.\n"); |
| 2073 | return VMAT_CONTIGUOUS_DOWN; |
| 2074 | } |
| 2075 | |
| 2076 | if (!perm_mask_for_reverse (vectype)) |
| 2077 | { |
| 2078 | if (dump_enabled_p ()) |
| 2079 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2080 | "negative step and reversing not supported.\n"); |
| 2081 | *poffset = 0; |
| 2082 | return VMAT_ELEMENTWISE; |
| 2083 | } |
| 2084 | |
| 2085 | return VMAT_CONTIGUOUS_REVERSE; |
| 2086 | } |
| 2087 | |
| 2088 | /* STMT_INFO is either a masked or unconditional store. Return the value |
| 2089 | being stored. */ |
| 2090 | |
| 2091 | tree |
| 2092 | vect_get_store_rhs (stmt_vec_info stmt_info) |
| 2093 | { |
| 2094 | if (gassign *assign = dyn_cast <gassign *> (stmt_info->stmt)) |
| 2095 | { |
| 2096 | gcc_assert (gimple_assign_single_p (assign))((void)(!(gimple_assign_single_p (assign)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2096, __FUNCTION__), 0 : 0)); |
| 2097 | return gimple_assign_rhs1 (assign); |
| 2098 | } |
| 2099 | if (gcall *call = dyn_cast <gcall *> (stmt_info->stmt)) |
| 2100 | { |
| 2101 | internal_fn ifn = gimple_call_internal_fn (call); |
| 2102 | int index = internal_fn_stored_value_index (ifn); |
| 2103 | gcc_assert (index >= 0)((void)(!(index >= 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2103, __FUNCTION__), 0 : 0)); |
| 2104 | return gimple_call_arg (call, index); |
| 2105 | } |
| 2106 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2106, __FUNCTION__)); |
| 2107 | } |
| 2108 | |
| 2109 | /* Function VECTOR_VECTOR_COMPOSITION_TYPE |
| 2110 | |
| 2111 | This function returns a vector type which can be composed with NETLS pieces, |
| 2112 | whose type is recorded in PTYPE. VTYPE should be a vector type, and has the |
| 2113 | same vector size as the return vector. It checks target whether supports |
| 2114 | pieces-size vector mode for construction firstly, if target fails to, check |
| 2115 | pieces-size scalar mode for construction further. It returns NULL_TREE if |
| 2116 | fails to find the available composition. |
| 2117 | |
| 2118 | For example, for (vtype=V16QI, nelts=4), we can probably get: |
| 2119 | - V16QI with PTYPE V4QI. |
| 2120 | - V4SI with PTYPE SI. |
| 2121 | - NULL_TREE. */ |
| 2122 | |
| 2123 | static tree |
| 2124 | vector_vector_composition_type (tree vtype, poly_uint64 nelts, tree *ptype) |
| 2125 | { |
| 2126 | gcc_assert (VECTOR_TYPE_P (vtype))((void)(!((((enum tree_code) (vtype)->base.code) == VECTOR_TYPE )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2126, __FUNCTION__), 0 : 0)); |
| 2127 | gcc_assert (known_gt (nelts, 0U))((void)(!((!maybe_le (nelts, 0U))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2127, __FUNCTION__), 0 : 0)); |
| 2128 | |
| 2129 | machine_mode vmode = TYPE_MODE (vtype)((((enum tree_code) ((tree_class_check ((vtype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2129, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vtype) : (vtype)->type_common.mode); |
| 2130 | if (!VECTOR_MODE_P (vmode)(((enum mode_class) mode_class[vmode]) == MODE_VECTOR_BOOL || ((enum mode_class) mode_class[vmode]) == MODE_VECTOR_INT || ( (enum mode_class) mode_class[vmode]) == MODE_VECTOR_FLOAT || ( (enum mode_class) mode_class[vmode]) == MODE_VECTOR_FRACT || ( (enum mode_class) mode_class[vmode]) == MODE_VECTOR_UFRACT || ((enum mode_class) mode_class[vmode]) == MODE_VECTOR_ACCUM || ((enum mode_class) mode_class[vmode]) == MODE_VECTOR_UACCUM)) |
| 2131 | return NULL_TREE(tree) nullptr; |
| 2132 | |
| 2133 | poly_uint64 vbsize = GET_MODE_BITSIZE (vmode); |
| 2134 | unsigned int pbsize; |
| 2135 | if (constant_multiple_p (vbsize, nelts, &pbsize)) |
| 2136 | { |
| 2137 | /* First check if vec_init optab supports construction from |
| 2138 | vector pieces directly. */ |
| 2139 | scalar_mode elmode = SCALAR_TYPE_MODE (TREE_TYPE (vtype))(as_a <scalar_mode> ((tree_class_check ((((contains_struct_check ((vtype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2139, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2139, __FUNCTION__))->type_common.mode)); |
| 2140 | poly_uint64 inelts = pbsize / GET_MODE_BITSIZE (elmode); |
| 2141 | machine_mode rmode; |
| 2142 | if (related_vector_mode (vmode, elmode, inelts).exists (&rmode) |
| 2143 | && (convert_optab_handler (vec_init_optab, vmode, rmode) |
| 2144 | != CODE_FOR_nothing)) |
| 2145 | { |
| 2146 | *ptype = build_vector_type (TREE_TYPE (vtype)((contains_struct_check ((vtype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2146, __FUNCTION__))->typed.type), inelts); |
| 2147 | return vtype; |
| 2148 | } |
| 2149 | |
| 2150 | /* Otherwise check if exists an integer type of the same piece size and |
| 2151 | if vec_init optab supports construction from it directly. */ |
| 2152 | if (int_mode_for_size (pbsize, 0).exists (&elmode) |
| 2153 | && related_vector_mode (vmode, elmode, nelts).exists (&rmode) |
| 2154 | && (convert_optab_handler (vec_init_optab, rmode, elmode) |
| 2155 | != CODE_FOR_nothing)) |
| 2156 | { |
| 2157 | *ptype = build_nonstandard_integer_type (pbsize, 1); |
| 2158 | return build_vector_type (*ptype, nelts); |
| 2159 | } |
| 2160 | } |
| 2161 | |
| 2162 | return NULL_TREE(tree) nullptr; |
| 2163 | } |
| 2164 | |
| 2165 | /* A subroutine of get_load_store_type, with a subset of the same |
| 2166 | arguments. Handle the case where STMT_INFO is part of a grouped load |
| 2167 | or store. |
| 2168 | |
| 2169 | For stores, the statements in the group are all consecutive |
| 2170 | and there is no gap at the end. For loads, the statements in the |
| 2171 | group might not be consecutive; there can be gaps between statements |
| 2172 | as well as at the end. */ |
| 2173 | |
| 2174 | static bool |
| 2175 | get_group_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, |
| 2176 | tree vectype, slp_tree slp_node, |
| 2177 | bool masked_p, vec_load_store_type vls_type, |
| 2178 | vect_memory_access_type *memory_access_type, |
| 2179 | poly_int64 *poffset, |
| 2180 | dr_alignment_support *alignment_support_scheme, |
| 2181 | int *misalignment, |
| 2182 | gather_scatter_info *gs_info) |
| 2183 | { |
| 2184 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 2185 | class loop *loop = loop_vinfo ? LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop : NULLnullptr; |
| 2186 | stmt_vec_info first_stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2186, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ); |
| 2187 | dr_vec_info *first_dr_info = STMT_VINFO_DR_INFO (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.stmt == (first_stmt_info )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2187, __FUNCTION__), 0 : 0)), &(first_stmt_info)->dr_aux ); |
| 2188 | unsigned int group_size = DR_GROUP_SIZE (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2188, __FUNCTION__), 0 : 0)), (first_stmt_info)->size); |
| 2189 | bool single_element_p = (stmt_info == first_stmt_info |
| 2190 | && !DR_GROUP_NEXT_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2190, __FUNCTION__), 0 : 0)), (stmt_info)->next_element)); |
| 2191 | unsigned HOST_WIDE_INTlong gap = DR_GROUP_GAP (first_stmt_info)(((void)(!((first_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2191, __FUNCTION__), 0 : 0)), (first_stmt_info)->gap); |
| 2192 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 2193 | |
| 2194 | /* True if the vectorized statements would access beyond the last |
| 2195 | statement in the group. */ |
| 2196 | bool overrun_p = false; |
| 2197 | |
| 2198 | /* True if we can cope with such overrun by peeling for gaps, so that |
| 2199 | there is at least one final scalar iteration after the vector loop. */ |
| 2200 | bool can_overrun_p = (!masked_p |
| 2201 | && vls_type == VLS_LOAD |
| 2202 | && loop_vinfo |
| 2203 | && !loop->inner); |
| 2204 | |
| 2205 | /* There can only be a gap at the end of the group if the stride is |
| 2206 | known at compile time. */ |
| 2207 | gcc_assert (!STMT_VINFO_STRIDED_P (first_stmt_info) || gap == 0)((void)(!(!(first_stmt_info)->strided_p || gap == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2207, __FUNCTION__), 0 : 0)); |
| 2208 | |
| 2209 | /* Stores can't yet have gaps. */ |
| 2210 | gcc_assert (slp_node || vls_type == VLS_LOAD || gap == 0)((void)(!(slp_node || vls_type == VLS_LOAD || gap == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2210, __FUNCTION__), 0 : 0)); |
| 2211 | |
| 2212 | if (slp_node) |
| 2213 | { |
| 2214 | /* For SLP vectorization we directly vectorize a subchain |
| 2215 | without permutation. */ |
| 2216 | if (! SLP_TREE_LOAD_PERMUTATION (slp_node)(slp_node)->load_permutation.exists ()) |
| 2217 | first_dr_info |
| 2218 | = STMT_VINFO_DR_INFO (SLP_TREE_SCALAR_STMTS (slp_node)[0])(((void)(!(((slp_node)->stmts[0])->dr_aux.stmt == ((slp_node )->stmts[0])) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2218, __FUNCTION__), 0 : 0)), &((slp_node)->stmts[0] )->dr_aux); |
| 2219 | if (STMT_VINFO_STRIDED_P (first_stmt_info)(first_stmt_info)->strided_p) |
| 2220 | { |
| 2221 | /* Try to use consecutive accesses of DR_GROUP_SIZE elements, |
| 2222 | separated by the stride, until we have a complete vector. |
| 2223 | Fall back to scalar accesses if that isn't possible. */ |
| 2224 | if (multiple_p (nunits, group_size)) |
| 2225 | *memory_access_type = VMAT_STRIDED_SLP; |
| 2226 | else |
| 2227 | *memory_access_type = VMAT_ELEMENTWISE; |
| 2228 | } |
| 2229 | else |
| 2230 | { |
| 2231 | overrun_p = loop_vinfo && gap != 0; |
| 2232 | if (overrun_p && vls_type != VLS_LOAD) |
| 2233 | { |
| 2234 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2235 | "Grouped store with gaps requires" |
| 2236 | " non-consecutive accesses\n"); |
| 2237 | return false; |
| 2238 | } |
| 2239 | /* An overrun is fine if the trailing elements are smaller |
| 2240 | than the alignment boundary B. Every vector access will |
| 2241 | be a multiple of B and so we are guaranteed to access a |
| 2242 | non-gap element in the same B-sized block. */ |
| 2243 | if (overrun_p |
| 2244 | && gap < (vect_known_alignment_in_bytes (first_dr_info, |
| 2245 | vectype) |
| 2246 | / vect_get_scalar_dr_size (first_dr_info))) |
| 2247 | overrun_p = false; |
| 2248 | |
| 2249 | /* If the gap splits the vector in half and the target |
| 2250 | can do half-vector operations avoid the epilogue peeling |
| 2251 | by simply loading half of the vector only. Usually |
| 2252 | the construction with an upper zero half will be elided. */ |
| 2253 | dr_alignment_support alss; |
| 2254 | int misalign = dr_misalignment (first_dr_info, vectype); |
| 2255 | tree half_vtype; |
| 2256 | if (overrun_p |
| 2257 | && !masked_p |
| 2258 | && (((alss = vect_supportable_dr_alignment (vinfo, first_dr_info, |
| 2259 | vectype, misalign))) |
| 2260 | == dr_aligned |
| 2261 | || alss == dr_unaligned_supported) |
| 2262 | && known_eq (nunits, (group_size - gap) * 2)(!maybe_ne (nunits, (group_size - gap) * 2)) |
| 2263 | && known_eq (nunits, group_size)(!maybe_ne (nunits, group_size)) |
| 2264 | && (vector_vector_composition_type (vectype, 2, &half_vtype) |
| 2265 | != NULL_TREE(tree) nullptr)) |
| 2266 | overrun_p = false; |
| 2267 | |
| 2268 | if (overrun_p && !can_overrun_p) |
| 2269 | { |
| 2270 | if (dump_enabled_p ()) |
| 2271 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2272 | "Peeling for outer loop is not supported\n"); |
| 2273 | return false; |
| 2274 | } |
| 2275 | int cmp = compare_step_with_zero (vinfo, stmt_info); |
| 2276 | if (cmp < 0) |
| 2277 | { |
| 2278 | if (single_element_p) |
| 2279 | /* ??? The VMAT_CONTIGUOUS_REVERSE code generation is |
| 2280 | only correct for single element "interleaving" SLP. */ |
| 2281 | *memory_access_type = get_negative_load_store_type |
| 2282 | (vinfo, stmt_info, vectype, vls_type, 1, poffset); |
| 2283 | else |
| 2284 | { |
| 2285 | /* Try to use consecutive accesses of DR_GROUP_SIZE elements, |
| 2286 | separated by the stride, until we have a complete vector. |
| 2287 | Fall back to scalar accesses if that isn't possible. */ |
| 2288 | if (multiple_p (nunits, group_size)) |
| 2289 | *memory_access_type = VMAT_STRIDED_SLP; |
| 2290 | else |
| 2291 | *memory_access_type = VMAT_ELEMENTWISE; |
| 2292 | } |
| 2293 | } |
| 2294 | else |
| 2295 | { |
| 2296 | gcc_assert (!loop_vinfo || cmp > 0)((void)(!(!loop_vinfo || cmp > 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2296, __FUNCTION__), 0 : 0)); |
| 2297 | *memory_access_type = VMAT_CONTIGUOUS; |
| 2298 | } |
| 2299 | |
| 2300 | /* When we have a contiguous access across loop iterations |
| 2301 | but the access in the loop doesn't cover the full vector |
| 2302 | we can end up with no gap recorded but still excess |
| 2303 | elements accessed, see PR103116. Make sure we peel for |
| 2304 | gaps if necessary and sufficient and give up if not. */ |
| 2305 | if (loop_vinfo |
| 2306 | && *memory_access_type == VMAT_CONTIGUOUS |
| 2307 | && SLP_TREE_LOAD_PERMUTATION (slp_node)(slp_node)->load_permutation.exists () |
| 2308 | && !multiple_p (group_size * LOOP_VINFO_VECT_FACTOR (loop_vinfo)(loop_vinfo)->vectorization_factor, |
| 2309 | nunits)) |
| 2310 | { |
| 2311 | unsigned HOST_WIDE_INTlong cnunits, cvf; |
| 2312 | if (!can_overrun_p |
| 2313 | || !nunits.is_constant (&cnunits) |
| 2314 | || !LOOP_VINFO_VECT_FACTOR (loop_vinfo)(loop_vinfo)->vectorization_factor.is_constant (&cvf) |
| 2315 | /* Peeling for gaps assumes that a single scalar iteration |
| 2316 | is enough to make sure the last vector iteration doesn't |
| 2317 | access excess elements. |
| 2318 | ??? Enhancements include peeling multiple iterations |
| 2319 | or using masked loads with a static mask. */ |
| 2320 | || (group_size * cvf) % cnunits + group_size < cnunits) |
| 2321 | { |
| 2322 | if (dump_enabled_p ()) |
| 2323 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2324 | "peeling for gaps insufficient for " |
| 2325 | "access\n"); |
| 2326 | return false; |
| 2327 | } |
| 2328 | overrun_p = true; |
| 2329 | } |
| 2330 | } |
| 2331 | } |
| 2332 | else |
| 2333 | { |
| 2334 | /* We can always handle this case using elementwise accesses, |
| 2335 | but see if something more efficient is available. */ |
| 2336 | *memory_access_type = VMAT_ELEMENTWISE; |
| 2337 | |
| 2338 | /* If there is a gap at the end of the group then these optimizations |
| 2339 | would access excess elements in the last iteration. */ |
| 2340 | bool would_overrun_p = (gap != 0); |
| 2341 | /* An overrun is fine if the trailing elements are smaller than the |
| 2342 | alignment boundary B. Every vector access will be a multiple of B |
| 2343 | and so we are guaranteed to access a non-gap element in the |
| 2344 | same B-sized block. */ |
| 2345 | if (would_overrun_p |
| 2346 | && !masked_p |
| 2347 | && gap < (vect_known_alignment_in_bytes (first_dr_info, vectype) |
| 2348 | / vect_get_scalar_dr_size (first_dr_info))) |
| 2349 | would_overrun_p = false; |
| 2350 | |
| 2351 | if (!STMT_VINFO_STRIDED_P (first_stmt_info)(first_stmt_info)->strided_p |
| 2352 | && (can_overrun_p || !would_overrun_p) |
| 2353 | && compare_step_with_zero (vinfo, stmt_info) > 0) |
| 2354 | { |
| 2355 | /* First cope with the degenerate case of a single-element |
| 2356 | vector. */ |
| 2357 | if (known_eq (TYPE_VECTOR_SUBPARTS (vectype), 1U)(!maybe_ne (TYPE_VECTOR_SUBPARTS (vectype), 1U))) |
| 2358 | ; |
| 2359 | |
| 2360 | /* Otherwise try using LOAD/STORE_LANES. */ |
| 2361 | else if (vls_type == VLS_LOAD |
| 2362 | ? vect_load_lanes_supported (vectype, group_size, masked_p) |
| 2363 | : vect_store_lanes_supported (vectype, group_size, |
| 2364 | masked_p)) |
| 2365 | { |
| 2366 | *memory_access_type = VMAT_LOAD_STORE_LANES; |
| 2367 | overrun_p = would_overrun_p; |
| 2368 | } |
| 2369 | |
| 2370 | /* If that fails, try using permuting loads. */ |
| 2371 | else if (vls_type == VLS_LOAD |
| 2372 | ? vect_grouped_load_supported (vectype, single_element_p, |
| 2373 | group_size) |
| 2374 | : vect_grouped_store_supported (vectype, group_size)) |
| 2375 | { |
| 2376 | *memory_access_type = VMAT_CONTIGUOUS_PERMUTE; |
| 2377 | overrun_p = would_overrun_p; |
| 2378 | } |
| 2379 | } |
| 2380 | |
| 2381 | /* As a last resort, trying using a gather load or scatter store. |
| 2382 | |
| 2383 | ??? Although the code can handle all group sizes correctly, |
| 2384 | it probably isn't a win to use separate strided accesses based |
| 2385 | on nearby locations. Or, even if it's a win over scalar code, |
| 2386 | it might not be a win over vectorizing at a lower VF, if that |
| 2387 | allows us to use contiguous accesses. */ |
| 2388 | if (*memory_access_type == VMAT_ELEMENTWISE |
| 2389 | && single_element_p |
| 2390 | && loop_vinfo |
| 2391 | && vect_use_strided_gather_scatters_p (stmt_info, loop_vinfo, |
| 2392 | masked_p, gs_info)) |
| 2393 | *memory_access_type = VMAT_GATHER_SCATTER; |
| 2394 | } |
| 2395 | |
| 2396 | if (*memory_access_type == VMAT_GATHER_SCATTER |
| 2397 | || *memory_access_type == VMAT_ELEMENTWISE) |
| 2398 | { |
| 2399 | *alignment_support_scheme = dr_unaligned_supported; |
| 2400 | *misalignment = DR_MISALIGNMENT_UNKNOWN(-1); |
| 2401 | } |
| 2402 | else |
| 2403 | { |
| 2404 | *misalignment = dr_misalignment (first_dr_info, vectype, *poffset); |
| 2405 | *alignment_support_scheme |
| 2406 | = vect_supportable_dr_alignment (vinfo, first_dr_info, vectype, |
| 2407 | *misalignment); |
| 2408 | } |
| 2409 | |
| 2410 | if (vls_type != VLS_LOAD && first_stmt_info == stmt_info) |
| 2411 | { |
| 2412 | /* STMT is the leader of the group. Check the operands of all the |
| 2413 | stmts of the group. */ |
| 2414 | stmt_vec_info next_stmt_info = DR_GROUP_NEXT_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2414, __FUNCTION__), 0 : 0)), (stmt_info)->next_element); |
| 2415 | while (next_stmt_info) |
| 2416 | { |
| 2417 | tree op = vect_get_store_rhs (next_stmt_info); |
| 2418 | enum vect_def_type dt; |
| 2419 | if (!vect_is_simple_use (op, vinfo, &dt)) |
| 2420 | { |
| 2421 | if (dump_enabled_p ()) |
| 2422 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2423 | "use not simple.\n"); |
| 2424 | return false; |
| 2425 | } |
| 2426 | next_stmt_info = DR_GROUP_NEXT_ELEMENT (next_stmt_info)(((void)(!((next_stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2426, __FUNCTION__), 0 : 0)), (next_stmt_info)->next_element ); |
| 2427 | } |
| 2428 | } |
| 2429 | |
| 2430 | if (overrun_p) |
| 2431 | { |
| 2432 | gcc_assert (can_overrun_p)((void)(!(can_overrun_p) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2432, __FUNCTION__), 0 : 0)); |
| 2433 | if (dump_enabled_p ()) |
| 2434 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2435 | "Data access with gaps requires scalar " |
| 2436 | "epilogue loop\n"); |
| 2437 | LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo)(loop_vinfo)->peeling_for_gaps = true; |
| 2438 | } |
| 2439 | |
| 2440 | return true; |
| 2441 | } |
| 2442 | |
| 2443 | /* Analyze load or store statement STMT_INFO of type VLS_TYPE. Return true |
| 2444 | if there is a memory access type that the vectorized form can use, |
| 2445 | storing it in *MEMORY_ACCESS_TYPE if so. If we decide to use gathers |
| 2446 | or scatters, fill in GS_INFO accordingly. In addition |
| 2447 | *ALIGNMENT_SUPPORT_SCHEME is filled out and false is returned if |
| 2448 | the target does not support the alignment scheme. *MISALIGNMENT |
| 2449 | is set according to the alignment of the access (including |
| 2450 | DR_MISALIGNMENT_UNKNOWN when it is unknown). |
| 2451 | |
| 2452 | SLP says whether we're performing SLP rather than loop vectorization. |
| 2453 | MASKED_P is true if the statement is conditional on a vectorized mask. |
| 2454 | VECTYPE is the vector type that the vectorized statements will use. |
| 2455 | NCOPIES is the number of vector statements that will be needed. */ |
| 2456 | |
| 2457 | static bool |
| 2458 | get_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, |
| 2459 | tree vectype, slp_tree slp_node, |
| 2460 | bool masked_p, vec_load_store_type vls_type, |
| 2461 | unsigned int ncopies, |
| 2462 | vect_memory_access_type *memory_access_type, |
| 2463 | poly_int64 *poffset, |
| 2464 | dr_alignment_support *alignment_support_scheme, |
| 2465 | int *misalignment, |
| 2466 | gather_scatter_info *gs_info) |
| 2467 | { |
| 2468 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 2469 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 2470 | *misalignment = DR_MISALIGNMENT_UNKNOWN(-1); |
| 2471 | *poffset = 0; |
| 2472 | if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)(stmt_info)->gather_scatter_p) |
| 2473 | { |
| 2474 | *memory_access_type = VMAT_GATHER_SCATTER; |
| 2475 | if (!vect_check_gather_scatter (stmt_info, loop_vinfo, gs_info)) |
| 2476 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2476, __FUNCTION__)); |
| 2477 | /* When using internal functions, we rely on pattern recognition |
| 2478 | to convert the type of the offset to the type that the target |
| 2479 | requires, with the result being a call to an internal function. |
| 2480 | If that failed for some reason (e.g. because another pattern |
| 2481 | took priority), just handle cases in which the offset already |
| 2482 | has the right type. */ |
| 2483 | else if (gs_info->ifn != IFN_LAST |
| 2484 | && !is_gimple_call (stmt_info->stmt) |
| 2485 | && !tree_nop_conversion_p (TREE_TYPE (gs_info->offset)((contains_struct_check ((gs_info->offset), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2485, __FUNCTION__))->typed.type), |
| 2486 | TREE_TYPE (gs_info->offset_vectype)((contains_struct_check ((gs_info->offset_vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2486, __FUNCTION__))->typed.type))) |
| 2487 | { |
| 2488 | if (dump_enabled_p ()) |
| 2489 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2490 | "%s offset requires a conversion\n", |
| 2491 | vls_type == VLS_LOAD ? "gather" : "scatter"); |
| 2492 | return false; |
| 2493 | } |
| 2494 | else if (!vect_is_simple_use (gs_info->offset, vinfo, |
| 2495 | &gs_info->offset_dt, |
| 2496 | &gs_info->offset_vectype)) |
| 2497 | { |
| 2498 | if (dump_enabled_p ()) |
| 2499 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2500 | "%s index use not simple.\n", |
| 2501 | vls_type == VLS_LOAD ? "gather" : "scatter"); |
| 2502 | return false; |
| 2503 | } |
| 2504 | else if (gs_info->ifn == IFN_LAST && !gs_info->decl) |
| 2505 | { |
| 2506 | if (vls_type != VLS_LOAD) |
| 2507 | { |
| 2508 | if (dump_enabled_p ()) |
| 2509 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2510 | "unsupported emulated scatter.\n"); |
| 2511 | return false; |
| 2512 | } |
| 2513 | else if (!TYPE_VECTOR_SUBPARTS (vectype).is_constant () |
| 2514 | || !TYPE_VECTOR_SUBPARTS |
| 2515 | (gs_info->offset_vectype).is_constant () |
| 2516 | || !constant_multiple_p (TYPE_VECTOR_SUBPARTS |
| 2517 | (gs_info->offset_vectype), |
| 2518 | TYPE_VECTOR_SUBPARTS (vectype))) |
| 2519 | { |
| 2520 | if (dump_enabled_p ()) |
| 2521 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2522 | "unsupported vector types for emulated " |
| 2523 | "gather.\n"); |
| 2524 | return false; |
| 2525 | } |
| 2526 | } |
| 2527 | /* Gather-scatter accesses perform only component accesses, alignment |
| 2528 | is irrelevant for them. */ |
| 2529 | *alignment_support_scheme = dr_unaligned_supported; |
| 2530 | } |
| 2531 | else if (STMT_VINFO_GROUPED_ACCESS (stmt_info)((stmt_info)->dr_aux.dr && (((void)(!((stmt_info)-> dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2531, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ))) |
| 2532 | { |
| 2533 | if (!get_group_load_store_type (vinfo, stmt_info, vectype, slp_node, |
| 2534 | masked_p, |
| 2535 | vls_type, memory_access_type, poffset, |
| 2536 | alignment_support_scheme, |
| 2537 | misalignment, gs_info)) |
| 2538 | return false; |
| 2539 | } |
| 2540 | else if (STMT_VINFO_STRIDED_P (stmt_info)(stmt_info)->strided_p) |
| 2541 | { |
| 2542 | gcc_assert (!slp_node)((void)(!(!slp_node) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2542, __FUNCTION__), 0 : 0)); |
| 2543 | if (loop_vinfo |
| 2544 | && vect_use_strided_gather_scatters_p (stmt_info, loop_vinfo, |
| 2545 | masked_p, gs_info)) |
| 2546 | *memory_access_type = VMAT_GATHER_SCATTER; |
| 2547 | else |
| 2548 | *memory_access_type = VMAT_ELEMENTWISE; |
| 2549 | /* Alignment is irrelevant here. */ |
| 2550 | *alignment_support_scheme = dr_unaligned_supported; |
| 2551 | } |
| 2552 | else |
| 2553 | { |
| 2554 | int cmp = compare_step_with_zero (vinfo, stmt_info); |
| 2555 | if (cmp == 0) |
| 2556 | { |
| 2557 | gcc_assert (vls_type == VLS_LOAD)((void)(!(vls_type == VLS_LOAD) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2557, __FUNCTION__), 0 : 0)); |
| 2558 | *memory_access_type = VMAT_INVARIANT; |
| 2559 | /* Invariant accesses perform only component accesses, alignment |
| 2560 | is irrelevant for them. */ |
| 2561 | *alignment_support_scheme = dr_unaligned_supported; |
| 2562 | } |
| 2563 | else |
| 2564 | { |
| 2565 | if (cmp < 0) |
| 2566 | *memory_access_type = get_negative_load_store_type |
| 2567 | (vinfo, stmt_info, vectype, vls_type, ncopies, poffset); |
| 2568 | else |
| 2569 | *memory_access_type = VMAT_CONTIGUOUS; |
| 2570 | *misalignment = dr_misalignment (STMT_VINFO_DR_INFO (stmt_info)(((void)(!((stmt_info)->dr_aux.stmt == (stmt_info)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2570, __FUNCTION__), 0 : 0)), &(stmt_info)->dr_aux), |
| 2571 | vectype, *poffset); |
| 2572 | *alignment_support_scheme |
| 2573 | = vect_supportable_dr_alignment (vinfo, |
| 2574 | STMT_VINFO_DR_INFO (stmt_info)(((void)(!((stmt_info)->dr_aux.stmt == (stmt_info)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2574, __FUNCTION__), 0 : 0)), &(stmt_info)->dr_aux), |
| 2575 | vectype, *misalignment); |
| 2576 | } |
| 2577 | } |
| 2578 | |
| 2579 | if ((*memory_access_type == VMAT_ELEMENTWISE |
| 2580 | || *memory_access_type == VMAT_STRIDED_SLP) |
| 2581 | && !nunits.is_constant ()) |
| 2582 | { |
| 2583 | if (dump_enabled_p ()) |
| 2584 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2585 | "Not using elementwise accesses due to variable " |
| 2586 | "vectorization factor.\n"); |
| 2587 | return false; |
| 2588 | } |
| 2589 | |
| 2590 | if (*alignment_support_scheme == dr_unaligned_unsupported) |
| 2591 | { |
| 2592 | if (dump_enabled_p ()) |
| 2593 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2594 | "unsupported unaligned access\n"); |
| 2595 | return false; |
| 2596 | } |
| 2597 | |
| 2598 | /* FIXME: At the moment the cost model seems to underestimate the |
| 2599 | cost of using elementwise accesses. This check preserves the |
| 2600 | traditional behavior until that can be fixed. */ |
| 2601 | stmt_vec_info first_stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2601, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ); |
| 2602 | if (!first_stmt_info) |
| 2603 | first_stmt_info = stmt_info; |
| 2604 | if (*memory_access_type == VMAT_ELEMENTWISE |
| 2605 | && !STMT_VINFO_STRIDED_P (first_stmt_info)(first_stmt_info)->strided_p |
| 2606 | && !(stmt_info == DR_GROUP_FIRST_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2606, __FUNCTION__), 0 : 0)), (stmt_info)->first_element ) |
| 2607 | && !DR_GROUP_NEXT_ELEMENT (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2607, __FUNCTION__), 0 : 0)), (stmt_info)->next_element) |
| 2608 | && !pow2p_hwi (DR_GROUP_SIZE (stmt_info)(((void)(!((stmt_info)->dr_aux.dr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2608, __FUNCTION__), 0 : 0)), (stmt_info)->size)))) |
| 2609 | { |
| 2610 | if (dump_enabled_p ()) |
| 2611 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2612 | "not falling back to elementwise accesses\n"); |
| 2613 | return false; |
| 2614 | } |
| 2615 | return true; |
| 2616 | } |
| 2617 | |
| 2618 | /* Return true if boolean argument at MASK_INDEX is suitable for vectorizing |
| 2619 | conditional operation STMT_INFO. When returning true, store the mask |
| 2620 | in *MASK, the type of its definition in *MASK_DT_OUT, the type of the |
| 2621 | vectorized mask in *MASK_VECTYPE_OUT and the SLP node corresponding |
| 2622 | to the mask in *MASK_NODE if MASK_NODE is not NULL. */ |
| 2623 | |
| 2624 | static bool |
| 2625 | vect_check_scalar_mask (vec_info *vinfo, stmt_vec_info stmt_info, |
| 2626 | slp_tree slp_node, unsigned mask_index, |
| 2627 | tree *mask, slp_tree *mask_node, |
| 2628 | vect_def_type *mask_dt_out, tree *mask_vectype_out) |
| 2629 | { |
| 2630 | enum vect_def_type mask_dt; |
| 2631 | tree mask_vectype; |
| 2632 | slp_tree mask_node_1; |
| 2633 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, mask_index, |
| 2634 | mask, &mask_node_1, &mask_dt, &mask_vectype)) |
| 2635 | { |
| 2636 | if (dump_enabled_p ()) |
| 2637 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2638 | "mask use not simple.\n"); |
| 2639 | return false; |
| 2640 | } |
| 2641 | |
| 2642 | if (!VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (*mask))(((enum tree_code) (((contains_struct_check ((*mask), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((((enum tree_code) (((contains_struct_check ((*mask), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE || ((enum tree_code) (((contains_struct_check ((*mask), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE ) && ((tree_class_check ((((contains_struct_check ((* mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->type_common.precision) == 1 && ((tree_class_check ((((contains_struct_check ((*mask), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2642, __FUNCTION__))->base.u.bits.unsigned_flag)))) |
| 2643 | { |
| 2644 | if (dump_enabled_p ()) |
| 2645 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2646 | "mask argument is not a boolean.\n"); |
| 2647 | return false; |
| 2648 | } |
| 2649 | |
| 2650 | /* If the caller is not prepared for adjusting an external/constant |
| 2651 | SLP mask vector type fail. */ |
| 2652 | if (slp_node |
| 2653 | && !mask_node |
| 2654 | && SLP_TREE_DEF_TYPE (mask_node_1)(mask_node_1)->def_type != vect_internal_def) |
| 2655 | { |
| 2656 | if (dump_enabled_p ()) |
| 2657 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2658 | "SLP mask argument is not vectorized.\n"); |
| 2659 | return false; |
| 2660 | } |
| 2661 | |
| 2662 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 2663 | if (!mask_vectype) |
| 2664 | mask_vectype = get_mask_type_for_scalar_type (vinfo, TREE_TYPE (vectype)((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2664, __FUNCTION__))->typed.type)); |
| 2665 | |
| 2666 | if (!mask_vectype || !VECTOR_BOOLEAN_TYPE_P (mask_vectype)(((enum tree_code) (mask_vectype)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((mask_vectype ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2666, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 2667 | { |
| 2668 | if (dump_enabled_p ()) |
| 2669 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2670 | "could not find an appropriate vector mask type.\n"); |
| 2671 | return false; |
| 2672 | } |
| 2673 | |
| 2674 | if (maybe_ne (TYPE_VECTOR_SUBPARTS (mask_vectype), |
| 2675 | TYPE_VECTOR_SUBPARTS (vectype))) |
| 2676 | { |
| 2677 | if (dump_enabled_p ()) |
| 2678 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2679 | "vector mask type %T" |
| 2680 | " does not match vector data type %T.\n", |
| 2681 | mask_vectype, vectype); |
| 2682 | |
| 2683 | return false; |
| 2684 | } |
| 2685 | |
| 2686 | *mask_dt_out = mask_dt; |
| 2687 | *mask_vectype_out = mask_vectype; |
| 2688 | if (mask_node) |
| 2689 | *mask_node = mask_node_1; |
| 2690 | return true; |
| 2691 | } |
| 2692 | |
| 2693 | /* Return true if stored value RHS is suitable for vectorizing store |
| 2694 | statement STMT_INFO. When returning true, store the type of the |
| 2695 | definition in *RHS_DT_OUT, the type of the vectorized store value in |
| 2696 | *RHS_VECTYPE_OUT and the type of the store in *VLS_TYPE_OUT. */ |
| 2697 | |
| 2698 | static bool |
| 2699 | vect_check_store_rhs (vec_info *vinfo, stmt_vec_info stmt_info, |
| 2700 | slp_tree slp_node, tree rhs, |
| 2701 | vect_def_type *rhs_dt_out, tree *rhs_vectype_out, |
| 2702 | vec_load_store_type *vls_type_out) |
| 2703 | { |
| 2704 | /* In the case this is a store from a constant make sure |
| 2705 | native_encode_expr can handle it. */ |
| 2706 | if (CONSTANT_CLASS_P (rhs)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (rhs)->base.code))] == tcc_constant) && native_encode_expr (rhs, NULLnullptr, 64) == 0) |
| 2707 | { |
| 2708 | if (dump_enabled_p ()) |
| 2709 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2710 | "cannot encode constant as a byte sequence.\n"); |
| 2711 | return false; |
| 2712 | } |
| 2713 | |
| 2714 | unsigned op_no = 0; |
| 2715 | if (gcall *call = dyn_cast <gcall *> (stmt_info->stmt)) |
| 2716 | { |
| 2717 | if (gimple_call_internal_p (call) |
| 2718 | && internal_store_fn_p (gimple_call_internal_fn (call))) |
| 2719 | op_no = internal_fn_stored_value_index (gimple_call_internal_fn (call)); |
| 2720 | } |
| 2721 | |
| 2722 | enum vect_def_type rhs_dt; |
| 2723 | tree rhs_vectype; |
| 2724 | slp_tree slp_op; |
| 2725 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, op_no, |
| 2726 | &rhs, &slp_op, &rhs_dt, &rhs_vectype)) |
| 2727 | { |
| 2728 | if (dump_enabled_p ()) |
| 2729 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2730 | "use not simple.\n"); |
| 2731 | return false; |
| 2732 | } |
| 2733 | |
| 2734 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 2735 | if (rhs_vectype && !useless_type_conversion_p (vectype, rhs_vectype)) |
| 2736 | { |
| 2737 | if (dump_enabled_p ()) |
| 2738 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 2739 | "incompatible vector types.\n"); |
| 2740 | return false; |
| 2741 | } |
| 2742 | |
| 2743 | *rhs_dt_out = rhs_dt; |
| 2744 | *rhs_vectype_out = rhs_vectype; |
| 2745 | if (rhs_dt == vect_constant_def || rhs_dt == vect_external_def) |
| 2746 | *vls_type_out = VLS_STORE_INVARIANT; |
| 2747 | else |
| 2748 | *vls_type_out = VLS_STORE; |
| 2749 | return true; |
| 2750 | } |
| 2751 | |
| 2752 | /* Build an all-ones vector mask of type MASKTYPE while vectorizing STMT_INFO. |
| 2753 | Note that we support masks with floating-point type, in which case the |
| 2754 | floats are interpreted as a bitmask. */ |
| 2755 | |
| 2756 | static tree |
| 2757 | vect_build_all_ones_mask (vec_info *vinfo, |
| 2758 | stmt_vec_info stmt_info, tree masktype) |
| 2759 | { |
| 2760 | if (TREE_CODE (masktype)((enum tree_code) (masktype)->base.code) == INTEGER_TYPE) |
| 2761 | return build_int_cst (masktype, -1); |
| 2762 | else if (TREE_CODE (TREE_TYPE (masktype))((enum tree_code) (((contains_struct_check ((masktype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2762, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE) |
| 2763 | { |
| 2764 | tree mask = build_int_cst (TREE_TYPE (masktype)((contains_struct_check ((masktype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2764, __FUNCTION__))->typed.type), -1); |
| 2765 | mask = build_vector_from_val (masktype, mask); |
| 2766 | return vect_init_vector (vinfo, stmt_info, mask, masktype, NULLnullptr); |
| 2767 | } |
| 2768 | else if (SCALAR_FLOAT_TYPE_P (TREE_TYPE (masktype))(((enum tree_code) (((contains_struct_check ((masktype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2768, __FUNCTION__))->typed.type))->base.code) == REAL_TYPE )) |
| 2769 | { |
| 2770 | REAL_VALUE_TYPEstruct real_value r; |
| 2771 | long tmp[6]; |
| 2772 | for (int j = 0; j < 6; ++j) |
| 2773 | tmp[j] = -1; |
| 2774 | real_from_target (&r, tmp, TYPE_MODE (TREE_TYPE (masktype))((((enum tree_code) ((tree_class_check ((((contains_struct_check ((masktype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2774, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2774, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (((contains_struct_check ((masktype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2774, __FUNCTION__))->typed.type)) : (((contains_struct_check ((masktype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2774, __FUNCTION__))->typed.type))->type_common.mode)); |
| 2775 | tree mask = build_real (TREE_TYPE (masktype)((contains_struct_check ((masktype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2775, __FUNCTION__))->typed.type), r); |
| 2776 | mask = build_vector_from_val (masktype, mask); |
| 2777 | return vect_init_vector (vinfo, stmt_info, mask, masktype, NULLnullptr); |
| 2778 | } |
| 2779 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2779, __FUNCTION__)); |
| 2780 | } |
| 2781 | |
| 2782 | /* Build an all-zero merge value of type VECTYPE while vectorizing |
| 2783 | STMT_INFO as a gather load. */ |
| 2784 | |
| 2785 | static tree |
| 2786 | vect_build_zero_merge_argument (vec_info *vinfo, |
| 2787 | stmt_vec_info stmt_info, tree vectype) |
| 2788 | { |
| 2789 | tree merge; |
| 2790 | if (TREE_CODE (TREE_TYPE (vectype))((enum tree_code) (((contains_struct_check ((vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2790, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE) |
| 2791 | merge = build_int_cst (TREE_TYPE (vectype)((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2791, __FUNCTION__))->typed.type), 0); |
| 2792 | else if (SCALAR_FLOAT_TYPE_P (TREE_TYPE (vectype))(((enum tree_code) (((contains_struct_check ((vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2792, __FUNCTION__))->typed.type))->base.code) == REAL_TYPE )) |
| 2793 | { |
| 2794 | REAL_VALUE_TYPEstruct real_value r; |
| 2795 | long tmp[6]; |
| 2796 | for (int j = 0; j < 6; ++j) |
| 2797 | tmp[j] = 0; |
| 2798 | real_from_target (&r, tmp, TYPE_MODE (TREE_TYPE (vectype))((((enum tree_code) ((tree_class_check ((((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2798, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2798, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2798, __FUNCTION__))->typed.type)) : (((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2798, __FUNCTION__))->typed.type))->type_common.mode)); |
| 2799 | merge = build_real (TREE_TYPE (vectype)((contains_struct_check ((vectype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2799, __FUNCTION__))->typed.type), r); |
| 2800 | } |
| 2801 | else |
| 2802 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2802, __FUNCTION__)); |
| 2803 | merge = build_vector_from_val (vectype, merge); |
| 2804 | return vect_init_vector (vinfo, stmt_info, merge, vectype, NULLnullptr); |
| 2805 | } |
| 2806 | |
| 2807 | /* Build a gather load call while vectorizing STMT_INFO. Insert new |
| 2808 | instructions before GSI and add them to VEC_STMT. GS_INFO describes |
| 2809 | the gather load operation. If the load is conditional, MASK is the |
| 2810 | unvectorized condition and MASK_DT is its definition type, otherwise |
| 2811 | MASK is null. */ |
| 2812 | |
| 2813 | static void |
| 2814 | vect_build_gather_load_calls (vec_info *vinfo, stmt_vec_info stmt_info, |
| 2815 | gimple_stmt_iterator *gsi, |
| 2816 | gimple **vec_stmt, |
| 2817 | gather_scatter_info *gs_info, |
| 2818 | tree mask) |
| 2819 | { |
| 2820 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 2821 | class loop *loop = LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop; |
| 2822 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 2823 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 2824 | int ncopies = vect_get_num_copies (loop_vinfo, vectype); |
| 2825 | edge pe = loop_preheader_edge (loop); |
| 2826 | enum { NARROW, NONE, WIDEN } modifier; |
| 2827 | poly_uint64 gather_off_nunits |
| 2828 | = TYPE_VECTOR_SUBPARTS (gs_info->offset_vectype); |
| 2829 | |
| 2830 | tree arglist = TYPE_ARG_TYPES (TREE_TYPE (gs_info->decl))((tree_check2 ((((contains_struct_check ((gs_info->decl), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2830, __FUNCTION__))->typed.type)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2830, __FUNCTION__, (FUNCTION_TYPE), (METHOD_TYPE)))->type_non_common .values); |
| 2831 | tree rettype = TREE_TYPE (TREE_TYPE (gs_info->decl))((contains_struct_check ((((contains_struct_check ((gs_info-> decl), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2831, __FUNCTION__))->typed.type)), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2831, __FUNCTION__))->typed.type); |
| 2832 | tree srctype = TREE_VALUE (arglist)((tree_check ((arglist), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2832, __FUNCTION__, (TREE_LIST)))->list.value); arglist = TREE_CHAIN (arglist)((contains_struct_check ((arglist), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2832, __FUNCTION__))->common.chain); |
| 2833 | tree ptrtype = TREE_VALUE (arglist)((tree_check ((arglist), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2833, __FUNCTION__, (TREE_LIST)))->list.value); arglist = TREE_CHAIN (arglist)((contains_struct_check ((arglist), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2833, __FUNCTION__))->common.chain); |
| 2834 | tree idxtype = TREE_VALUE (arglist)((tree_check ((arglist), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2834, __FUNCTION__, (TREE_LIST)))->list.value); arglist = TREE_CHAIN (arglist)((contains_struct_check ((arglist), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2834, __FUNCTION__))->common.chain); |
| 2835 | tree masktype = TREE_VALUE (arglist)((tree_check ((arglist), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2835, __FUNCTION__, (TREE_LIST)))->list.value); arglist = TREE_CHAIN (arglist)((contains_struct_check ((arglist), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2835, __FUNCTION__))->common.chain); |
| 2836 | tree scaletype = TREE_VALUE (arglist)((tree_check ((arglist), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2836, __FUNCTION__, (TREE_LIST)))->list.value); |
| 2837 | tree real_masktype = masktype; |
| 2838 | gcc_checking_assert (types_compatible_p (srctype, rettype)((void)(!(types_compatible_p (srctype, rettype) && (! mask || ((enum tree_code) (masktype)->base.code) == INTEGER_TYPE || types_compatible_p (srctype, masktype))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2841, __FUNCTION__), 0 : 0)) |
| 2839 | && (!mask((void)(!(types_compatible_p (srctype, rettype) && (! mask || ((enum tree_code) (masktype)->base.code) == INTEGER_TYPE || types_compatible_p (srctype, masktype))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2841, __FUNCTION__), 0 : 0)) |
| 2840 | || TREE_CODE (masktype) == INTEGER_TYPE((void)(!(types_compatible_p (srctype, rettype) && (! mask || ((enum tree_code) (masktype)->base.code) == INTEGER_TYPE || types_compatible_p (srctype, masktype))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2841, __FUNCTION__), 0 : 0)) |
| 2841 | || types_compatible_p (srctype, masktype)))((void)(!(types_compatible_p (srctype, rettype) && (! mask || ((enum tree_code) (masktype)->base.code) == INTEGER_TYPE || types_compatible_p (srctype, masktype))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2841, __FUNCTION__), 0 : 0)); |
| 2842 | if (mask) |
| 2843 | masktype = truth_type_for (srctype); |
| 2844 | |
| 2845 | tree mask_halftype = masktype; |
| 2846 | tree perm_mask = NULL_TREE(tree) nullptr; |
| 2847 | tree mask_perm_mask = NULL_TREE(tree) nullptr; |
| 2848 | if (known_eq (nunits, gather_off_nunits)(!maybe_ne (nunits, gather_off_nunits))) |
| 2849 | modifier = NONE; |
| 2850 | else if (known_eq (nunits * 2, gather_off_nunits)(!maybe_ne (nunits * 2, gather_off_nunits))) |
| 2851 | { |
| 2852 | modifier = WIDEN; |
| 2853 | |
| 2854 | /* Currently widening gathers and scatters are only supported for |
| 2855 | fixed-length vectors. */ |
| 2856 | int count = gather_off_nunits.to_constant (); |
| 2857 | vec_perm_builder sel (count, count, 1); |
| 2858 | for (int i = 0; i < count; ++i) |
| 2859 | sel.quick_push (i | (count / 2)); |
| 2860 | |
| 2861 | vec_perm_indices indices (sel, 1, count); |
| 2862 | perm_mask = vect_gen_perm_mask_checked (gs_info->offset_vectype, |
| 2863 | indices); |
| 2864 | } |
| 2865 | else if (known_eq (nunits, gather_off_nunits * 2)(!maybe_ne (nunits, gather_off_nunits * 2))) |
| 2866 | { |
| 2867 | modifier = NARROW; |
| 2868 | |
| 2869 | /* Currently narrowing gathers and scatters are only supported for |
| 2870 | fixed-length vectors. */ |
| 2871 | int count = nunits.to_constant (); |
| 2872 | vec_perm_builder sel (count, count, 1); |
| 2873 | sel.quick_grow (count); |
| 2874 | for (int i = 0; i < count; ++i) |
| 2875 | sel[i] = i < count / 2 ? i : i + count / 2; |
| 2876 | vec_perm_indices indices (sel, 2, count); |
| 2877 | perm_mask = vect_gen_perm_mask_checked (vectype, indices); |
| 2878 | |
| 2879 | ncopies *= 2; |
| 2880 | |
| 2881 | if (mask && VECTOR_TYPE_P (real_masktype)(((enum tree_code) (real_masktype)->base.code) == VECTOR_TYPE )) |
| 2882 | { |
| 2883 | for (int i = 0; i < count; ++i) |
| 2884 | sel[i] = i | (count / 2); |
| 2885 | indices.new_vector (sel, 2, count); |
| 2886 | mask_perm_mask = vect_gen_perm_mask_checked (masktype, indices); |
| 2887 | } |
| 2888 | else if (mask) |
| 2889 | mask_halftype = truth_type_for (gs_info->offset_vectype); |
| 2890 | } |
| 2891 | else |
| 2892 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2892, __FUNCTION__)); |
| 2893 | |
| 2894 | tree scalar_dest = gimple_get_lhs (stmt_info->stmt); |
| 2895 | tree vec_dest = vect_create_destination_var (scalar_dest, vectype); |
| 2896 | |
| 2897 | tree ptr = fold_convert (ptrtype, gs_info->base)fold_convert_loc (((location_t) 0), ptrtype, gs_info->base ); |
| 2898 | if (!is_gimple_min_invariant (ptr)) |
| 2899 | { |
| 2900 | gimple_seq seq; |
| 2901 | ptr = force_gimple_operand (ptr, &seq, true, NULL_TREE(tree) nullptr); |
| 2902 | basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, seq); |
| 2903 | gcc_assert (!new_bb)((void)(!(!new_bb) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2903, __FUNCTION__), 0 : 0)); |
| 2904 | } |
| 2905 | |
| 2906 | tree scale = build_int_cst (scaletype, gs_info->scale); |
| 2907 | |
| 2908 | tree vec_oprnd0 = NULL_TREE(tree) nullptr; |
| 2909 | tree vec_mask = NULL_TREE(tree) nullptr; |
| 2910 | tree src_op = NULL_TREE(tree) nullptr; |
| 2911 | tree mask_op = NULL_TREE(tree) nullptr; |
| 2912 | tree prev_res = NULL_TREE(tree) nullptr; |
| 2913 | |
| 2914 | if (!mask) |
| 2915 | { |
| 2916 | src_op = vect_build_zero_merge_argument (vinfo, stmt_info, rettype); |
| 2917 | mask_op = vect_build_all_ones_mask (vinfo, stmt_info, masktype); |
| 2918 | } |
| 2919 | |
| 2920 | auto_vec<tree> vec_oprnds0; |
| 2921 | auto_vec<tree> vec_masks; |
| 2922 | vect_get_vec_defs_for_operand (vinfo, stmt_info, |
| 2923 | modifier == WIDEN ? ncopies / 2 : ncopies, |
| 2924 | gs_info->offset, &vec_oprnds0); |
| 2925 | if (mask) |
| 2926 | vect_get_vec_defs_for_operand (vinfo, stmt_info, |
| 2927 | modifier == NARROW ? ncopies / 2 : ncopies, |
| 2928 | mask, &vec_masks, masktype); |
| 2929 | for (int j = 0; j < ncopies; ++j) |
| 2930 | { |
| 2931 | tree op, var; |
| 2932 | if (modifier == WIDEN && (j & 1)) |
| 2933 | op = permute_vec_elements (vinfo, vec_oprnd0, vec_oprnd0, |
| 2934 | perm_mask, stmt_info, gsi); |
| 2935 | else |
| 2936 | op = vec_oprnd0 = vec_oprnds0[modifier == WIDEN ? j / 2 : j]; |
| 2937 | |
| 2938 | if (!useless_type_conversion_p (idxtype, TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2938, __FUNCTION__))->typed.type))) |
| 2939 | { |
| 2940 | gcc_assert (known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (op)),((void)(!((!maybe_ne (TYPE_VECTOR_SUBPARTS (((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2940, __FUNCTION__))->typed.type)), TYPE_VECTOR_SUBPARTS (idxtype)))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2941, __FUNCTION__), 0 : 0)) |
| 2941 | TYPE_VECTOR_SUBPARTS (idxtype)))((void)(!((!maybe_ne (TYPE_VECTOR_SUBPARTS (((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2940, __FUNCTION__))->typed.type)), TYPE_VECTOR_SUBPARTS (idxtype)))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2941, __FUNCTION__), 0 : 0)); |
| 2942 | var = vect_get_new_ssa_name (idxtype, vect_simple_var); |
| 2943 | op = build1 (VIEW_CONVERT_EXPR, idxtype, op); |
| 2944 | gassign *new_stmt = gimple_build_assign (var, VIEW_CONVERT_EXPR, op); |
| 2945 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 2946 | op = var; |
| 2947 | } |
| 2948 | |
| 2949 | if (mask) |
| 2950 | { |
| 2951 | if (mask_perm_mask && (j & 1)) |
| 2952 | mask_op = permute_vec_elements (vinfo, mask_op, mask_op, |
| 2953 | mask_perm_mask, stmt_info, gsi); |
| 2954 | else |
| 2955 | { |
| 2956 | if (modifier == NARROW) |
| 2957 | { |
| 2958 | if ((j & 1) == 0) |
| 2959 | vec_mask = vec_masks[j / 2]; |
| 2960 | } |
| 2961 | else |
| 2962 | vec_mask = vec_masks[j]; |
| 2963 | |
| 2964 | mask_op = vec_mask; |
| 2965 | if (!useless_type_conversion_p (masktype, TREE_TYPE (vec_mask)((contains_struct_check ((vec_mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2965, __FUNCTION__))->typed.type))) |
| 2966 | { |
| 2967 | poly_uint64 sub1 = TYPE_VECTOR_SUBPARTS (TREE_TYPE (mask_op)((contains_struct_check ((mask_op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2967, __FUNCTION__))->typed.type)); |
| 2968 | poly_uint64 sub2 = TYPE_VECTOR_SUBPARTS (masktype); |
| 2969 | gcc_assert (known_eq (sub1, sub2))((void)(!((!maybe_ne (sub1, sub2))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2969, __FUNCTION__), 0 : 0)); |
| 2970 | var = vect_get_new_ssa_name (masktype, vect_simple_var); |
| 2971 | mask_op = build1 (VIEW_CONVERT_EXPR, masktype, mask_op); |
| 2972 | gassign *new_stmt |
| 2973 | = gimple_build_assign (var, VIEW_CONVERT_EXPR, mask_op); |
| 2974 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 2975 | mask_op = var; |
| 2976 | } |
| 2977 | } |
| 2978 | if (modifier == NARROW && !VECTOR_TYPE_P (real_masktype)(((enum tree_code) (real_masktype)->base.code) == VECTOR_TYPE )) |
| 2979 | { |
| 2980 | var = vect_get_new_ssa_name (mask_halftype, vect_simple_var); |
| 2981 | gassign *new_stmt |
| 2982 | = gimple_build_assign (var, (j & 1) ? VEC_UNPACK_HI_EXPR |
| 2983 | : VEC_UNPACK_LO_EXPR, |
| 2984 | mask_op); |
| 2985 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 2986 | mask_op = var; |
| 2987 | } |
| 2988 | src_op = mask_op; |
| 2989 | } |
| 2990 | |
| 2991 | tree mask_arg = mask_op; |
| 2992 | if (masktype != real_masktype) |
| 2993 | { |
| 2994 | tree utype, optype = TREE_TYPE (mask_op)((contains_struct_check ((mask_op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2994, __FUNCTION__))->typed.type); |
| 2995 | if (VECTOR_TYPE_P (real_masktype)(((enum tree_code) (real_masktype)->base.code) == VECTOR_TYPE ) |
| 2996 | || TYPE_MODE (real_masktype)((((enum tree_code) ((tree_class_check ((real_masktype), (tcc_type ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2996, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (real_masktype) : (real_masktype)->type_common.mode) == TYPE_MODE (optype)((((enum tree_code) ((tree_class_check ((optype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2996, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (optype) : (optype)->type_common.mode)) |
| 2997 | utype = real_masktype; |
| 2998 | else |
| 2999 | utype = lang_hooks.types.type_for_mode (TYPE_MODE (optype)((((enum tree_code) ((tree_class_check ((optype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 2999, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (optype) : (optype)->type_common.mode), 1); |
| 3000 | var = vect_get_new_ssa_name (utype, vect_scalar_var); |
| 3001 | mask_arg = build1 (VIEW_CONVERT_EXPR, utype, mask_op); |
| 3002 | gassign *new_stmt |
| 3003 | = gimple_build_assign (var, VIEW_CONVERT_EXPR, mask_arg); |
| 3004 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3005 | mask_arg = var; |
| 3006 | if (!useless_type_conversion_p (real_masktype, utype)) |
| 3007 | { |
| 3008 | gcc_assert (TYPE_PRECISION (utype)((void)(!(((tree_class_check ((utype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3008, __FUNCTION__))->type_common.precision) <= ((tree_class_check ((real_masktype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3009, __FUNCTION__))->type_common.precision)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3009, __FUNCTION__), 0 : 0)) |
| 3009 | <= TYPE_PRECISION (real_masktype))((void)(!(((tree_class_check ((utype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3008, __FUNCTION__))->type_common.precision) <= ((tree_class_check ((real_masktype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3009, __FUNCTION__))->type_common.precision)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3009, __FUNCTION__), 0 : 0)); |
| 3010 | var = vect_get_new_ssa_name (real_masktype, vect_scalar_var); |
| 3011 | new_stmt = gimple_build_assign (var, NOP_EXPR, mask_arg); |
| 3012 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3013 | mask_arg = var; |
| 3014 | } |
| 3015 | src_op = build_zero_cst (srctype); |
| 3016 | } |
| 3017 | gimple *new_stmt = gimple_build_call (gs_info->decl, 5, src_op, ptr, op, |
| 3018 | mask_arg, scale); |
| 3019 | |
| 3020 | if (!useless_type_conversion_p (vectype, rettype)) |
| 3021 | { |
| 3022 | gcc_assert (known_eq (TYPE_VECTOR_SUBPARTS (vectype),((void)(!((!maybe_ne (TYPE_VECTOR_SUBPARTS (vectype), TYPE_VECTOR_SUBPARTS (rettype)))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3023, __FUNCTION__), 0 : 0)) |
| 3023 | TYPE_VECTOR_SUBPARTS (rettype)))((void)(!((!maybe_ne (TYPE_VECTOR_SUBPARTS (vectype), TYPE_VECTOR_SUBPARTS (rettype)))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3023, __FUNCTION__), 0 : 0)); |
| 3024 | op = vect_get_new_ssa_name (rettype, vect_simple_var); |
| 3025 | gimple_call_set_lhs (new_stmt, op); |
| 3026 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3027 | var = make_ssa_name (vec_dest); |
| 3028 | op = build1 (VIEW_CONVERT_EXPR, vectype, op); |
| 3029 | new_stmt = gimple_build_assign (var, VIEW_CONVERT_EXPR, op); |
| 3030 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3031 | } |
| 3032 | else |
| 3033 | { |
| 3034 | var = make_ssa_name (vec_dest, new_stmt); |
| 3035 | gimple_call_set_lhs (new_stmt, var); |
| 3036 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3037 | } |
| 3038 | |
| 3039 | if (modifier == NARROW) |
| 3040 | { |
| 3041 | if ((j & 1) == 0) |
| 3042 | { |
| 3043 | prev_res = var; |
| 3044 | continue; |
| 3045 | } |
| 3046 | var = permute_vec_elements (vinfo, prev_res, var, perm_mask, |
| 3047 | stmt_info, gsi); |
| 3048 | new_stmt = SSA_NAME_DEF_STMT (var)(tree_check ((var), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3048, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 3049 | } |
| 3050 | |
| 3051 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 3052 | } |
| 3053 | *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts[0]; |
| 3054 | } |
| 3055 | |
| 3056 | /* Prepare the base and offset in GS_INFO for vectorization. |
| 3057 | Set *DATAREF_PTR to the loop-invariant base address and *VEC_OFFSET |
| 3058 | to the vectorized offset argument for the first copy of STMT_INFO. |
| 3059 | STMT_INFO is the statement described by GS_INFO and LOOP is the |
| 3060 | containing loop. */ |
| 3061 | |
| 3062 | static void |
| 3063 | vect_get_gather_scatter_ops (loop_vec_info loop_vinfo, |
| 3064 | class loop *loop, stmt_vec_info stmt_info, |
| 3065 | slp_tree slp_node, gather_scatter_info *gs_info, |
| 3066 | tree *dataref_ptr, vec<tree> *vec_offset) |
| 3067 | { |
| 3068 | gimple_seq stmts = NULLnullptr; |
| 3069 | *dataref_ptr = force_gimple_operand (gs_info->base, &stmts, true, NULL_TREE(tree) nullptr); |
| 3070 | if (stmts != NULLnullptr) |
| 3071 | { |
| 3072 | basic_block new_bb; |
| 3073 | edge pe = loop_preheader_edge (loop); |
| 3074 | new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); |
| 3075 | gcc_assert (!new_bb)((void)(!(!new_bb) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3075, __FUNCTION__), 0 : 0)); |
| 3076 | } |
| 3077 | if (slp_node) |
| 3078 | vect_get_slp_defs (SLP_TREE_CHILDREN (slp_node)(slp_node)->children[0], vec_offset); |
| 3079 | else |
| 3080 | { |
| 3081 | unsigned ncopies |
| 3082 | = vect_get_num_copies (loop_vinfo, gs_info->offset_vectype); |
| 3083 | vect_get_vec_defs_for_operand (loop_vinfo, stmt_info, ncopies, |
| 3084 | gs_info->offset, vec_offset, |
| 3085 | gs_info->offset_vectype); |
| 3086 | } |
| 3087 | } |
| 3088 | |
| 3089 | /* Prepare to implement a grouped or strided load or store using |
| 3090 | the gather load or scatter store operation described by GS_INFO. |
| 3091 | STMT_INFO is the load or store statement. |
| 3092 | |
| 3093 | Set *DATAREF_BUMP to the amount that should be added to the base |
| 3094 | address after each copy of the vectorized statement. Set *VEC_OFFSET |
| 3095 | to an invariant offset vector in which element I has the value |
| 3096 | I * DR_STEP / SCALE. */ |
| 3097 | |
| 3098 | static void |
| 3099 | vect_get_strided_load_store_ops (stmt_vec_info stmt_info, |
| 3100 | loop_vec_info loop_vinfo, |
| 3101 | gather_scatter_info *gs_info, |
| 3102 | tree *dataref_bump, tree *vec_offset) |
| 3103 | { |
| 3104 | struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info)((stmt_info)->dr_aux.dr + 0); |
| 3105 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 3106 | |
| 3107 | tree bump = size_binop (MULT_EXPR,size_binop_loc (((location_t) 0), MULT_EXPR, fold_convert_loc (((location_t) 0), sizetype_tab[(int) stk_sizetype], unshare_expr ((dr)->innermost.step)), size_int_kind (TYPE_VECTOR_SUBPARTS (vectype), stk_sizetype)) |
| 3108 | fold_convert (sizetype, unshare_expr (DR_STEP (dr))),size_binop_loc (((location_t) 0), MULT_EXPR, fold_convert_loc (((location_t) 0), sizetype_tab[(int) stk_sizetype], unshare_expr ((dr)->innermost.step)), size_int_kind (TYPE_VECTOR_SUBPARTS (vectype), stk_sizetype)) |
| 3109 | size_int (TYPE_VECTOR_SUBPARTS (vectype)))size_binop_loc (((location_t) 0), MULT_EXPR, fold_convert_loc (((location_t) 0), sizetype_tab[(int) stk_sizetype], unshare_expr ((dr)->innermost.step)), size_int_kind (TYPE_VECTOR_SUBPARTS (vectype), stk_sizetype)); |
| 3110 | *dataref_bump = cse_and_gimplify_to_preheader (loop_vinfo, bump); |
| 3111 | |
| 3112 | /* The offset given in GS_INFO can have pointer type, so use the element |
| 3113 | type of the vector instead. */ |
| 3114 | tree offset_type = TREE_TYPE (gs_info->offset_vectype)((contains_struct_check ((gs_info->offset_vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3114, __FUNCTION__))->typed.type); |
| 3115 | |
| 3116 | /* Calculate X = DR_STEP / SCALE and convert it to the appropriate type. */ |
| 3117 | tree step = size_binop (EXACT_DIV_EXPR, unshare_expr (DR_STEP (dr)),size_binop_loc (((location_t) 0), EXACT_DIV_EXPR, unshare_expr ((dr)->innermost.step), size_int_kind (gs_info->scale, stk_ssizetype)) |
| 3118 | ssize_int (gs_info->scale))size_binop_loc (((location_t) 0), EXACT_DIV_EXPR, unshare_expr ((dr)->innermost.step), size_int_kind (gs_info->scale, stk_ssizetype)); |
| 3119 | step = fold_convert (offset_type, step)fold_convert_loc (((location_t) 0), offset_type, step); |
| 3120 | |
| 3121 | /* Create {0, X, X*2, X*3, ...}. */ |
| 3122 | tree offset = fold_build2 (VEC_SERIES_EXPR, gs_info->offset_vectype,fold_build2_loc (((location_t) 0), VEC_SERIES_EXPR, gs_info-> offset_vectype, build_zero_cst (offset_type), step ) |
| 3123 | build_zero_cst (offset_type), step)fold_build2_loc (((location_t) 0), VEC_SERIES_EXPR, gs_info-> offset_vectype, build_zero_cst (offset_type), step ); |
| 3124 | *vec_offset = cse_and_gimplify_to_preheader (loop_vinfo, offset); |
| 3125 | } |
| 3126 | |
| 3127 | /* Return the amount that should be added to a vector pointer to move |
| 3128 | to the next or previous copy of AGGR_TYPE. DR_INFO is the data reference |
| 3129 | being vectorized and MEMORY_ACCESS_TYPE describes the type of |
| 3130 | vectorization. */ |
| 3131 | |
| 3132 | static tree |
| 3133 | vect_get_data_ptr_increment (vec_info *vinfo, |
| 3134 | dr_vec_info *dr_info, tree aggr_type, |
| 3135 | vect_memory_access_type memory_access_type) |
| 3136 | { |
| 3137 | if (memory_access_type == VMAT_INVARIANT) |
| 3138 | return size_zero_nodeglobal_trees[TI_SIZE_ZERO]; |
| 3139 | |
| 3140 | tree iv_step = TYPE_SIZE_UNIT (aggr_type)((tree_class_check ((aggr_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3140, __FUNCTION__))->type_common.size_unit); |
| 3141 | tree step = vect_dr_behavior (vinfo, dr_info)->step; |
| 3142 | if (tree_int_cst_sgn (step) == -1) |
| 3143 | iv_step = fold_build1 (NEGATE_EXPR, TREE_TYPE (iv_step), iv_step)fold_build1_loc (((location_t) 0), NEGATE_EXPR, ((contains_struct_check ((iv_step), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3143, __FUNCTION__))->typed.type), iv_step ); |
| 3144 | return iv_step; |
| 3145 | } |
| 3146 | |
| 3147 | /* Check and perform vectorization of BUILT_IN_BSWAP{16,32,64,128}. */ |
| 3148 | |
| 3149 | static bool |
| 3150 | vectorizable_bswap (vec_info *vinfo, |
| 3151 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 3152 | gimple **vec_stmt, slp_tree slp_node, |
| 3153 | slp_tree *slp_op, |
| 3154 | tree vectype_in, stmt_vector_for_cost *cost_vec) |
| 3155 | { |
| 3156 | tree op, vectype; |
| 3157 | gcall *stmt = as_a <gcall *> (stmt_info->stmt); |
| 3158 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 3159 | unsigned ncopies; |
| 3160 | |
| 3161 | op = gimple_call_arg (stmt, 0); |
| 3162 | vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 3163 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 3164 | |
| 3165 | /* Multiple types in SLP are handled by creating the appropriate number of |
| 3166 | vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in |
| 3167 | case of SLP. */ |
| 3168 | if (slp_node) |
| 3169 | ncopies = 1; |
| 3170 | else |
| 3171 | ncopies = vect_get_num_copies (loop_vinfo, vectype); |
| 3172 | |
| 3173 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3173, __FUNCTION__), 0 : 0)); |
| 3174 | |
| 3175 | tree char_vectype = get_same_sized_vectype (char_type_nodeinteger_types[itk_char], vectype_in); |
| 3176 | if (! char_vectype) |
| 3177 | return false; |
| 3178 | |
| 3179 | poly_uint64 num_bytes = TYPE_VECTOR_SUBPARTS (char_vectype); |
| 3180 | unsigned word_bytes; |
| 3181 | if (!constant_multiple_p (num_bytes, nunits, &word_bytes)) |
| 3182 | return false; |
| 3183 | |
| 3184 | /* The encoding uses one stepped pattern for each byte in the word. */ |
| 3185 | vec_perm_builder elts (num_bytes, word_bytes, 3); |
| 3186 | for (unsigned i = 0; i < 3; ++i) |
| 3187 | for (unsigned j = 0; j < word_bytes; ++j) |
| 3188 | elts.quick_push ((i + 1) * word_bytes - j - 1); |
| 3189 | |
| 3190 | vec_perm_indices indices (elts, 1, num_bytes); |
| 3191 | machine_mode vmode = TYPE_MODE (char_vectype)((((enum tree_code) ((tree_class_check ((char_vectype), (tcc_type ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3191, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (char_vectype) : (char_vectype)->type_common.mode); |
| 3192 | if (!can_vec_perm_const_p (vmode, vmode, indices)) |
| 3193 | return false; |
| 3194 | |
| 3195 | if (! vec_stmt) |
| 3196 | { |
| 3197 | if (slp_node |
| 3198 | && !vect_maybe_update_slp_op_vectype (slp_op[0], vectype_in)) |
| 3199 | { |
| 3200 | if (dump_enabled_p ()) |
| 3201 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3202 | "incompatible vector types for invariants\n"); |
| 3203 | return false; |
| 3204 | } |
| 3205 | |
| 3206 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = call_vec_info_type; |
| 3207 | DUMP_VECT_SCOPE ("vectorizable_bswap")auto_dump_scope scope ("vectorizable_bswap", vect_location); |
| 3208 | record_stmt_cost (cost_vec, |
| 3209 | 1, vector_stmt, stmt_info, 0, vect_prologue); |
| 3210 | record_stmt_cost (cost_vec, |
| 3211 | slp_node |
| 3212 | ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size : ncopies, |
| 3213 | vec_perm, stmt_info, 0, vect_body); |
| 3214 | return true; |
| 3215 | } |
| 3216 | |
| 3217 | tree bswap_vconst = vec_perm_indices_to_tree (char_vectype, indices); |
| 3218 | |
| 3219 | /* Transform. */ |
| 3220 | vec<tree> vec_oprnds = vNULL; |
| 3221 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies, |
| 3222 | op, &vec_oprnds); |
| 3223 | /* Arguments are ready. create the new vector stmt. */ |
| 3224 | unsigned i; |
| 3225 | tree vop; |
| 3226 | FOR_EACH_VEC_ELT (vec_oprnds, i, vop)for (i = 0; (vec_oprnds).iterate ((i), &(vop)); ++(i)) |
| 3227 | { |
| 3228 | gimple *new_stmt; |
| 3229 | tree tem = make_ssa_name (char_vectype); |
| 3230 | new_stmt = gimple_build_assign (tem, build1 (VIEW_CONVERT_EXPR, |
| 3231 | char_vectype, vop)); |
| 3232 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3233 | tree tem2 = make_ssa_name (char_vectype); |
| 3234 | new_stmt = gimple_build_assign (tem2, VEC_PERM_EXPR, |
| 3235 | tem, tem, bswap_vconst); |
| 3236 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3237 | tem = make_ssa_name (vectype); |
| 3238 | new_stmt = gimple_build_assign (tem, build1 (VIEW_CONVERT_EXPR, |
| 3239 | vectype, tem2)); |
| 3240 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3241 | if (slp_node) |
| 3242 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 3243 | else |
| 3244 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 3245 | } |
| 3246 | |
| 3247 | if (!slp_node) |
| 3248 | *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts[0]; |
| 3249 | |
| 3250 | vec_oprnds.release (); |
| 3251 | return true; |
| 3252 | } |
| 3253 | |
| 3254 | /* Return true if vector types VECTYPE_IN and VECTYPE_OUT have |
| 3255 | integer elements and if we can narrow VECTYPE_IN to VECTYPE_OUT |
| 3256 | in a single step. On success, store the binary pack code in |
| 3257 | *CONVERT_CODE. */ |
| 3258 | |
| 3259 | static bool |
| 3260 | simple_integer_narrowing (tree vectype_out, tree vectype_in, |
| 3261 | tree_code *convert_code) |
| 3262 | { |
| 3263 | if (!INTEGRAL_TYPE_P (TREE_TYPE (vectype_out))(((enum tree_code) (((contains_struct_check ((vectype_out), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3263, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE || ((enum tree_code) (((contains_struct_check ((vectype_out) , (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3263, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((enum tree_code) (((contains_struct_check ((vectype_out) , (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3263, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE ) |
| 3264 | || !INTEGRAL_TYPE_P (TREE_TYPE (vectype_in))(((enum tree_code) (((contains_struct_check ((vectype_in), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3264, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE || ((enum tree_code) (((contains_struct_check ((vectype_in), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3264, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((enum tree_code) (((contains_struct_check ((vectype_in), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3264, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE )) |
| 3265 | return false; |
| 3266 | |
| 3267 | tree_code code; |
| 3268 | int multi_step_cvt = 0; |
| 3269 | auto_vec <tree, 8> interm_types; |
| 3270 | if (!supportable_narrowing_operation (NOP_EXPR, vectype_out, vectype_in, |
| 3271 | &code, &multi_step_cvt, &interm_types) |
| 3272 | || multi_step_cvt) |
| 3273 | return false; |
| 3274 | |
| 3275 | *convert_code = code; |
| 3276 | return true; |
| 3277 | } |
| 3278 | |
| 3279 | /* Function vectorizable_call. |
| 3280 | |
| 3281 | Check if STMT_INFO performs a function call that can be vectorized. |
| 3282 | If VEC_STMT is also passed, vectorize STMT_INFO: create a vectorized |
| 3283 | stmt to replace it, put it in VEC_STMT, and insert it at GSI. |
| 3284 | Return true if STMT_INFO is vectorizable in this way. */ |
| 3285 | |
| 3286 | static bool |
| 3287 | vectorizable_call (vec_info *vinfo, |
| 3288 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 3289 | gimple **vec_stmt, slp_tree slp_node, |
| 3290 | stmt_vector_for_cost *cost_vec) |
| 3291 | { |
| 3292 | gcall *stmt; |
| 3293 | tree vec_dest; |
| 3294 | tree scalar_dest; |
| 3295 | tree op; |
| 3296 | tree vec_oprnd0 = NULL_TREE(tree) nullptr, vec_oprnd1 = NULL_TREE(tree) nullptr; |
| 3297 | tree vectype_out, vectype_in; |
| 3298 | poly_uint64 nunits_in; |
| 3299 | poly_uint64 nunits_out; |
| 3300 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 3301 | bb_vec_info bb_vinfo = dyn_cast <bb_vec_info> (vinfo); |
| 3302 | tree fndecl, new_temp, rhs_type; |
| 3303 | enum vect_def_type dt[4] |
| 3304 | = { vect_unknown_def_type, vect_unknown_def_type, vect_unknown_def_type, |
| 3305 | vect_unknown_def_type }; |
| 3306 | tree vectypes[ARRAY_SIZE (dt)(sizeof (dt) / sizeof ((dt)[0]))] = {}; |
| 3307 | slp_tree slp_op[ARRAY_SIZE (dt)(sizeof (dt) / sizeof ((dt)[0]))] = {}; |
| 3308 | int ndts = ARRAY_SIZE (dt)(sizeof (dt) / sizeof ((dt)[0])); |
| 3309 | int ncopies, j; |
| 3310 | auto_vec<tree, 8> vargs; |
| 3311 | enum { NARROW, NONE, WIDEN } modifier; |
| 3312 | size_t i, nargs; |
| 3313 | tree lhs; |
| 3314 | |
| 3315 | if (!STMT_VINFO_RELEVANT_P (stmt_info)((stmt_info)->relevant != vect_unused_in_scope) && !bb_vinfo) |
| 3316 | return false; |
| 3317 | |
| 3318 | if (STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_internal_def |
| 3319 | && ! vec_stmt) |
| 3320 | return false; |
| 3321 | |
| 3322 | /* Is STMT_INFO a vectorizable call? */ |
| 3323 | stmt = dyn_cast <gcall *> (stmt_info->stmt); |
| 3324 | if (!stmt) |
| 3325 | return false; |
| 3326 | |
| 3327 | if (gimple_call_internal_p (stmt) |
| 3328 | && (internal_load_fn_p (gimple_call_internal_fn (stmt)) |
| 3329 | || internal_store_fn_p (gimple_call_internal_fn (stmt)))) |
| 3330 | /* Handled by vectorizable_load and vectorizable_store. */ |
| 3331 | return false; |
| 3332 | |
| 3333 | if (gimple_call_lhs (stmt) == NULL_TREE(tree) nullptr |
| 3334 | || TREE_CODE (gimple_call_lhs (stmt))((enum tree_code) (gimple_call_lhs (stmt))->base.code) != SSA_NAME) |
| 3335 | return false; |
| 3336 | |
| 3337 | gcc_checking_assert (!stmt_can_throw_internal (cfun, stmt))((void)(!(!stmt_can_throw_internal ((cfun + 0), stmt)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3337, __FUNCTION__), 0 : 0)); |
| 3338 | |
| 3339 | vectype_out = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 3340 | |
| 3341 | /* Process function arguments. */ |
| 3342 | rhs_type = NULL_TREE(tree) nullptr; |
| 3343 | vectype_in = NULL_TREE(tree) nullptr; |
| 3344 | nargs = gimple_call_num_args (stmt); |
| 3345 | |
| 3346 | /* Bail out if the function has more than four arguments, we do not have |
| 3347 | interesting builtin functions to vectorize with more than two arguments |
| 3348 | except for fma. No arguments is also not good. */ |
| 3349 | if (nargs == 0 || nargs > 4) |
| 3350 | return false; |
| 3351 | |
| 3352 | /* Ignore the arguments of IFN_GOMP_SIMD_LANE, they are magic. */ |
| 3353 | combined_fn cfn = gimple_call_combined_fn (stmt); |
| 3354 | if (cfn == CFN_GOMP_SIMD_LANE) |
| 3355 | { |
| 3356 | nargs = 0; |
| 3357 | rhs_type = unsigned_type_nodeinteger_types[itk_unsigned_int]; |
| 3358 | } |
| 3359 | |
| 3360 | int mask_opno = -1; |
| 3361 | if (internal_fn_p (cfn)) |
| 3362 | mask_opno = internal_fn_mask_index (as_internal_fn (cfn)); |
| 3363 | |
| 3364 | for (i = 0; i < nargs; i++) |
| 3365 | { |
| 3366 | if ((int) i == mask_opno) |
| 3367 | { |
| 3368 | if (!vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_opno, |
| 3369 | &op, &slp_op[i], &dt[i], &vectypes[i])) |
| 3370 | return false; |
| 3371 | continue; |
| 3372 | } |
| 3373 | |
| 3374 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, |
| 3375 | i, &op, &slp_op[i], &dt[i], &vectypes[i])) |
| 3376 | { |
| 3377 | if (dump_enabled_p ()) |
| 3378 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3379 | "use not simple.\n"); |
| 3380 | return false; |
| 3381 | } |
| 3382 | |
| 3383 | /* We can only handle calls with arguments of the same type. */ |
| 3384 | if (rhs_type |
| 3385 | && !types_compatible_p (rhs_type, TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3385, __FUNCTION__))->typed.type))) |
| 3386 | { |
| 3387 | if (dump_enabled_p ()) |
| 3388 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3389 | "argument types differ.\n"); |
| 3390 | return false; |
| 3391 | } |
| 3392 | if (!rhs_type) |
| 3393 | rhs_type = TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3393, __FUNCTION__))->typed.type); |
| 3394 | |
| 3395 | if (!vectype_in) |
| 3396 | vectype_in = vectypes[i]; |
| 3397 | else if (vectypes[i] |
| 3398 | && !types_compatible_p (vectypes[i], vectype_in)) |
| 3399 | { |
| 3400 | if (dump_enabled_p ()) |
| 3401 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3402 | "argument vector types differ.\n"); |
| 3403 | return false; |
| 3404 | } |
| 3405 | } |
| 3406 | /* If all arguments are external or constant defs, infer the vector type |
| 3407 | from the scalar type. */ |
| 3408 | if (!vectype_in) |
| 3409 | vectype_in = get_vectype_for_scalar_type (vinfo, rhs_type, slp_node); |
| 3410 | if (vec_stmt) |
| 3411 | gcc_assert (vectype_in)((void)(!(vectype_in) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3411, __FUNCTION__), 0 : 0)); |
| 3412 | if (!vectype_in) |
| 3413 | { |
| 3414 | if (dump_enabled_p ()) |
| 3415 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3416 | "no vectype for scalar type %T\n", rhs_type); |
| 3417 | |
| 3418 | return false; |
| 3419 | } |
| 3420 | /* FORNOW: we don't yet support mixtures of vector sizes for calls, |
| 3421 | just mixtures of nunits. E.g. DI->SI versions of __builtin_ctz* |
| 3422 | are traditionally vectorized as two VnDI->VnDI IFN_CTZs followed |
| 3423 | by a pack of the two vectors into an SI vector. We would need |
| 3424 | separate code to handle direct VnDI->VnSI IFN_CTZs. */ |
| 3425 | if (TYPE_SIZE (vectype_in)((tree_class_check ((vectype_in), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3425, __FUNCTION__))->type_common.size) != TYPE_SIZE (vectype_out)((tree_class_check ((vectype_out), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3425, __FUNCTION__))->type_common.size)) |
| 3426 | { |
| 3427 | if (dump_enabled_p ()) |
| 3428 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3429 | "mismatched vector sizes %T and %T\n", |
| 3430 | vectype_in, vectype_out); |
| 3431 | return false; |
| 3432 | } |
| 3433 | |
| 3434 | if (VECTOR_BOOLEAN_TYPE_P (vectype_out)(((enum tree_code) (vectype_out)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_out ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3434, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE ) |
| 3435 | != VECTOR_BOOLEAN_TYPE_P (vectype_in)(((enum tree_code) (vectype_in)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_in ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3435, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 3436 | { |
| 3437 | if (dump_enabled_p ()) |
| 3438 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3439 | "mixed mask and nonmask vector types\n"); |
| 3440 | return false; |
| 3441 | } |
| 3442 | |
| 3443 | if (vect_emulated_vector_p (vectype_in) || vect_emulated_vector_p (vectype_out)) |
| 3444 | { |
| 3445 | if (dump_enabled_p ()) |
| 3446 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3447 | "use emulated vector type for call\n"); |
| 3448 | return false; |
| 3449 | } |
| 3450 | |
| 3451 | /* FORNOW */ |
| 3452 | nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); |
| 3453 | nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); |
| 3454 | if (known_eq (nunits_in * 2, nunits_out)(!maybe_ne (nunits_in * 2, nunits_out))) |
| 3455 | modifier = NARROW; |
| 3456 | else if (known_eq (nunits_out, nunits_in)(!maybe_ne (nunits_out, nunits_in))) |
| 3457 | modifier = NONE; |
| 3458 | else if (known_eq (nunits_out * 2, nunits_in)(!maybe_ne (nunits_out * 2, nunits_in))) |
| 3459 | modifier = WIDEN; |
| 3460 | else |
| 3461 | return false; |
| 3462 | |
| 3463 | /* We only handle functions that do not read or clobber memory. */ |
| 3464 | if (gimple_vuse (stmt)) |
| 3465 | { |
| 3466 | if (dump_enabled_p ()) |
| 3467 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3468 | "function reads from or writes to memory.\n"); |
| 3469 | return false; |
| 3470 | } |
| 3471 | |
| 3472 | /* For now, we only vectorize functions if a target specific builtin |
| 3473 | is available. TODO -- in some cases, it might be profitable to |
| 3474 | insert the calls for pieces of the vector, in order to be able |
| 3475 | to vectorize other operations in the loop. */ |
| 3476 | fndecl = NULL_TREE(tree) nullptr; |
| 3477 | internal_fn ifn = IFN_LAST; |
| 3478 | tree callee = gimple_call_fndecl (stmt); |
| 3479 | |
| 3480 | /* First try using an internal function. */ |
| 3481 | tree_code convert_code = ERROR_MARK; |
| 3482 | if (cfn != CFN_LAST |
| 3483 | && (modifier == NONE |
| 3484 | || (modifier == NARROW |
| 3485 | && simple_integer_narrowing (vectype_out, vectype_in, |
| 3486 | &convert_code)))) |
| 3487 | ifn = vectorizable_internal_function (cfn, callee, vectype_out, |
| 3488 | vectype_in); |
| 3489 | |
| 3490 | /* If that fails, try asking for a target-specific built-in function. */ |
| 3491 | if (ifn == IFN_LAST) |
| 3492 | { |
| 3493 | if (cfn != CFN_LAST) |
| 3494 | fndecl = targetm.vectorize.builtin_vectorized_function |
| 3495 | (cfn, vectype_out, vectype_in); |
| 3496 | else if (callee && fndecl_built_in_p (callee, BUILT_IN_MD)) |
| 3497 | fndecl = targetm.vectorize.builtin_md_vectorized_function |
| 3498 | (callee, vectype_out, vectype_in); |
| 3499 | } |
| 3500 | |
| 3501 | if (ifn == IFN_LAST && !fndecl) |
| 3502 | { |
| 3503 | if (cfn == CFN_GOMP_SIMD_LANE |
| 3504 | && !slp_node |
| 3505 | && loop_vinfo |
| 3506 | && LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop->simduid |
| 3507 | && TREE_CODE (gimple_call_arg (stmt, 0))((enum tree_code) (gimple_call_arg (stmt, 0))->base.code) == SSA_NAME |
| 3508 | && LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop->simduid |
| 3509 | == SSA_NAME_VAR (gimple_call_arg (stmt, 0))((tree_check ((gimple_call_arg (stmt, 0)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3509, __FUNCTION__, (SSA_NAME)))->ssa_name.var == (tree) nullptr || ((enum tree_code) ((gimple_call_arg (stmt, 0))-> ssa_name.var)->base.code) == IDENTIFIER_NODE ? (tree) nullptr : (gimple_call_arg (stmt, 0))->ssa_name.var)) |
| 3510 | { |
| 3511 | /* We can handle IFN_GOMP_SIMD_LANE by returning a |
| 3512 | { 0, 1, 2, ... vf - 1 } vector. */ |
| 3513 | gcc_assert (nargs == 0)((void)(!(nargs == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3513, __FUNCTION__), 0 : 0)); |
| 3514 | } |
| 3515 | else if (modifier == NONE |
| 3516 | && (gimple_call_builtin_p (stmt, BUILT_IN_BSWAP16) |
| 3517 | || gimple_call_builtin_p (stmt, BUILT_IN_BSWAP32) |
| 3518 | || gimple_call_builtin_p (stmt, BUILT_IN_BSWAP64) |
| 3519 | || gimple_call_builtin_p (stmt, BUILT_IN_BSWAP128))) |
| 3520 | return vectorizable_bswap (vinfo, stmt_info, gsi, vec_stmt, slp_node, |
| 3521 | slp_op, vectype_in, cost_vec); |
| 3522 | else |
| 3523 | { |
| 3524 | if (dump_enabled_p ()) |
| 3525 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3526 | "function is not vectorizable.\n"); |
| 3527 | return false; |
| 3528 | } |
| 3529 | } |
| 3530 | |
| 3531 | if (slp_node) |
| 3532 | ncopies = 1; |
| 3533 | else if (modifier == NARROW && ifn == IFN_LAST) |
| 3534 | ncopies = vect_get_num_copies (loop_vinfo, vectype_out); |
| 3535 | else |
| 3536 | ncopies = vect_get_num_copies (loop_vinfo, vectype_in); |
| 3537 | |
| 3538 | /* Sanity check: make sure that at least one copy of the vectorized stmt |
| 3539 | needs to be generated. */ |
| 3540 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3540, __FUNCTION__), 0 : 0)); |
| 3541 | |
| 3542 | int reduc_idx = STMT_VINFO_REDUC_IDX (stmt_info)(stmt_info)->reduc_idx; |
| 3543 | internal_fn cond_fn = get_conditional_internal_fn (ifn); |
| 3544 | vec_loop_masks *masks = (loop_vinfo ? &LOOP_VINFO_MASKS (loop_vinfo)(loop_vinfo)->masks : NULLnullptr); |
| 3545 | if (!vec_stmt) /* transformation not required. */ |
| 3546 | { |
| 3547 | if (slp_node) |
| 3548 | for (i = 0; i < nargs; ++i) |
| 3549 | if (!vect_maybe_update_slp_op_vectype (slp_op[i], |
| 3550 | vectypes[i] |
| 3551 | ? vectypes[i] : vectype_in)) |
| 3552 | { |
| 3553 | if (dump_enabled_p ()) |
| 3554 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3555 | "incompatible vector types for invariants\n"); |
| 3556 | return false; |
| 3557 | } |
| 3558 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = call_vec_info_type; |
| 3559 | DUMP_VECT_SCOPE ("vectorizable_call")auto_dump_scope scope ("vectorizable_call", vect_location); |
| 3560 | vect_model_simple_cost (vinfo, stmt_info, |
| 3561 | ncopies, dt, ndts, slp_node, cost_vec); |
| 3562 | if (ifn != IFN_LAST && modifier == NARROW && !slp_node) |
| 3563 | record_stmt_cost (cost_vec, ncopies / 2, |
| 3564 | vec_promote_demote, stmt_info, 0, vect_body); |
| 3565 | |
| 3566 | if (loop_vinfo |
| 3567 | && LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p |
| 3568 | && (reduc_idx >= 0 || mask_opno >= 0)) |
| 3569 | { |
| 3570 | if (reduc_idx >= 0 |
| 3571 | && (cond_fn == IFN_LAST |
| 3572 | || !direct_internal_fn_supported_p (cond_fn, vectype_out, |
| 3573 | OPTIMIZE_FOR_SPEED))) |
| 3574 | { |
| 3575 | if (dump_enabled_p ()) |
| 3576 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 3577 | "can't use a fully-masked loop because no" |
| 3578 | " conditional operation is available.\n"); |
| 3579 | LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)(loop_vinfo)->can_use_partial_vectors_p = false; |
| 3580 | } |
| 3581 | else |
| 3582 | { |
| 3583 | unsigned int nvectors |
| 3584 | = (slp_node |
| 3585 | ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size |
| 3586 | : ncopies); |
| 3587 | tree scalar_mask = NULL_TREE(tree) nullptr; |
| 3588 | if (mask_opno >= 0) |
| 3589 | scalar_mask = gimple_call_arg (stmt_info->stmt, mask_opno); |
| 3590 | vect_record_loop_mask (loop_vinfo, masks, nvectors, |
| 3591 | vectype_out, scalar_mask); |
| 3592 | } |
| 3593 | } |
| 3594 | return true; |
| 3595 | } |
| 3596 | |
| 3597 | /* Transform. */ |
| 3598 | |
| 3599 | if (dump_enabled_p ()) |
| 3600 | dump_printf_loc (MSG_NOTE, vect_location, "transform call.\n"); |
| 3601 | |
| 3602 | /* Handle def. */ |
| 3603 | scalar_dest = gimple_call_lhs (stmt); |
| 3604 | vec_dest = vect_create_destination_var (scalar_dest, vectype_out); |
| 3605 | |
| 3606 | bool masked_loop_p = loop_vinfo && LOOP_VINFO_FULLY_MASKED_P (loop_vinfo)((loop_vinfo)->using_partial_vectors_p && !(loop_vinfo )->masks.is_empty ()); |
| 3607 | unsigned int vect_nargs = nargs; |
| 3608 | if (masked_loop_p && reduc_idx >= 0) |
| 3609 | { |
| 3610 | ifn = cond_fn; |
| 3611 | vect_nargs += 2; |
| 3612 | } |
| 3613 | |
| 3614 | if (modifier == NONE || ifn != IFN_LAST) |
| 3615 | { |
| 3616 | tree prev_res = NULL_TREE(tree) nullptr; |
| 3617 | vargs.safe_grow (vect_nargs, true); |
| 3618 | auto_vec<vec<tree> > vec_defs (nargs); |
| 3619 | for (j = 0; j < ncopies; ++j) |
| 3620 | { |
| 3621 | /* Build argument list for the vectorized call. */ |
| 3622 | if (slp_node) |
| 3623 | { |
| 3624 | vec<tree> vec_oprnds0; |
| 3625 | |
| 3626 | vect_get_slp_defs (vinfo, slp_node, &vec_defs); |
| 3627 | vec_oprnds0 = vec_defs[0]; |
| 3628 | |
| 3629 | /* Arguments are ready. Create the new vector stmt. */ |
| 3630 | FOR_EACH_VEC_ELT (vec_oprnds0, i, vec_oprnd0)for (i = 0; (vec_oprnds0).iterate ((i), &(vec_oprnd0)); ++ (i)) |
| 3631 | { |
| 3632 | int varg = 0; |
| 3633 | if (masked_loop_p && reduc_idx >= 0) |
| 3634 | { |
| 3635 | unsigned int vec_num = vec_oprnds0.length (); |
| 3636 | /* Always true for SLP. */ |
| 3637 | gcc_assert (ncopies == 1)((void)(!(ncopies == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3637, __FUNCTION__), 0 : 0)); |
| 3638 | vargs[varg++] = vect_get_loop_mask (gsi, masks, vec_num, |
| 3639 | vectype_out, i); |
| 3640 | } |
| 3641 | size_t k; |
| 3642 | for (k = 0; k < nargs; k++) |
| 3643 | { |
| 3644 | vec<tree> vec_oprndsk = vec_defs[k]; |
| 3645 | vargs[varg++] = vec_oprndsk[i]; |
| 3646 | } |
| 3647 | if (masked_loop_p && reduc_idx >= 0) |
| 3648 | vargs[varg++] = vargs[reduc_idx + 1]; |
| 3649 | gimple *new_stmt; |
| 3650 | if (modifier == NARROW) |
| 3651 | { |
| 3652 | /* We don't define any narrowing conditional functions |
| 3653 | at present. */ |
| 3654 | gcc_assert (mask_opno < 0)((void)(!(mask_opno < 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3654, __FUNCTION__), 0 : 0)); |
| 3655 | tree half_res = make_ssa_name (vectype_in); |
| 3656 | gcall *call |
| 3657 | = gimple_build_call_internal_vec (ifn, vargs); |
| 3658 | gimple_call_set_lhs (call, half_res); |
| 3659 | gimple_call_set_nothrow (call, true); |
| 3660 | vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); |
| 3661 | if ((i & 1) == 0) |
| 3662 | { |
| 3663 | prev_res = half_res; |
| 3664 | continue; |
| 3665 | } |
| 3666 | new_temp = make_ssa_name (vec_dest); |
| 3667 | new_stmt = gimple_build_assign (new_temp, convert_code, |
| 3668 | prev_res, half_res); |
| 3669 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 3670 | new_stmt, gsi); |
| 3671 | } |
| 3672 | else |
| 3673 | { |
| 3674 | if (mask_opno >= 0 && masked_loop_p) |
| 3675 | { |
| 3676 | unsigned int vec_num = vec_oprnds0.length (); |
| 3677 | /* Always true for SLP. */ |
| 3678 | gcc_assert (ncopies == 1)((void)(!(ncopies == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3678, __FUNCTION__), 0 : 0)); |
| 3679 | tree mask = vect_get_loop_mask (gsi, masks, vec_num, |
| 3680 | vectype_out, i); |
| 3681 | vargs[mask_opno] = prepare_vec_mask |
| 3682 | (loop_vinfo, TREE_TYPE (mask)((contains_struct_check ((mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3682, __FUNCTION__))->typed.type), mask, |
| 3683 | vargs[mask_opno], gsi); |
| 3684 | } |
| 3685 | |
| 3686 | gcall *call; |
| 3687 | if (ifn != IFN_LAST) |
| 3688 | call = gimple_build_call_internal_vec (ifn, vargs); |
| 3689 | else |
| 3690 | call = gimple_build_call_vec (fndecl, vargs); |
| 3691 | new_temp = make_ssa_name (vec_dest, call); |
| 3692 | gimple_call_set_lhs (call, new_temp); |
| 3693 | gimple_call_set_nothrow (call, true); |
| 3694 | vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); |
| 3695 | new_stmt = call; |
| 3696 | } |
| 3697 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 3698 | } |
| 3699 | continue; |
| 3700 | } |
| 3701 | |
| 3702 | int varg = 0; |
| 3703 | if (masked_loop_p && reduc_idx >= 0) |
| 3704 | vargs[varg++] = vect_get_loop_mask (gsi, masks, ncopies, |
| 3705 | vectype_out, j); |
| 3706 | for (i = 0; i < nargs; i++) |
| 3707 | { |
| 3708 | op = gimple_call_arg (stmt, i); |
| 3709 | if (j == 0) |
| 3710 | { |
| 3711 | vec_defs.quick_push (vNULL); |
| 3712 | vect_get_vec_defs_for_operand (vinfo, stmt_info, ncopies, |
| 3713 | op, &vec_defs[i], |
| 3714 | vectypes[i]); |
| 3715 | } |
| 3716 | vargs[varg++] = vec_defs[i][j]; |
| 3717 | } |
| 3718 | if (masked_loop_p && reduc_idx >= 0) |
| 3719 | vargs[varg++] = vargs[reduc_idx + 1]; |
| 3720 | |
| 3721 | if (mask_opno >= 0 && masked_loop_p) |
| 3722 | { |
| 3723 | tree mask = vect_get_loop_mask (gsi, masks, ncopies, |
| 3724 | vectype_out, j); |
| 3725 | vargs[mask_opno] |
| 3726 | = prepare_vec_mask (loop_vinfo, TREE_TYPE (mask)((contains_struct_check ((mask), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3726, __FUNCTION__))->typed.type), mask, |
| 3727 | vargs[mask_opno], gsi); |
| 3728 | } |
| 3729 | |
| 3730 | gimple *new_stmt; |
| 3731 | if (cfn == CFN_GOMP_SIMD_LANE) |
| 3732 | { |
| 3733 | tree cst = build_index_vector (vectype_out, j * nunits_out, 1); |
| 3734 | tree new_var |
| 3735 | = vect_get_new_ssa_name (vectype_out, vect_simple_var, "cst_"); |
| 3736 | gimple *init_stmt = gimple_build_assign (new_var, cst); |
| 3737 | vect_init_vector_1 (vinfo, stmt_info, init_stmt, NULLnullptr); |
| 3738 | new_temp = make_ssa_name (vec_dest); |
| 3739 | new_stmt = gimple_build_assign (new_temp, new_var); |
| 3740 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3741 | } |
| 3742 | else if (modifier == NARROW) |
| 3743 | { |
| 3744 | /* We don't define any narrowing conditional functions at |
| 3745 | present. */ |
| 3746 | gcc_assert (mask_opno < 0)((void)(!(mask_opno < 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3746, __FUNCTION__), 0 : 0)); |
| 3747 | tree half_res = make_ssa_name (vectype_in); |
| 3748 | gcall *call = gimple_build_call_internal_vec (ifn, vargs); |
| 3749 | gimple_call_set_lhs (call, half_res); |
| 3750 | gimple_call_set_nothrow (call, true); |
| 3751 | vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); |
| 3752 | if ((j & 1) == 0) |
| 3753 | { |
| 3754 | prev_res = half_res; |
| 3755 | continue; |
| 3756 | } |
| 3757 | new_temp = make_ssa_name (vec_dest); |
| 3758 | new_stmt = gimple_build_assign (new_temp, convert_code, |
| 3759 | prev_res, half_res); |
| 3760 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3761 | } |
| 3762 | else |
| 3763 | { |
| 3764 | gcall *call; |
| 3765 | if (ifn != IFN_LAST) |
| 3766 | call = gimple_build_call_internal_vec (ifn, vargs); |
| 3767 | else |
| 3768 | call = gimple_build_call_vec (fndecl, vargs); |
| 3769 | new_temp = make_ssa_name (vec_dest, call); |
| 3770 | gimple_call_set_lhs (call, new_temp); |
| 3771 | gimple_call_set_nothrow (call, true); |
| 3772 | vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); |
| 3773 | new_stmt = call; |
| 3774 | } |
| 3775 | |
| 3776 | if (j == (modifier == NARROW ? 1 : 0)) |
| 3777 | *vec_stmt = new_stmt; |
| 3778 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 3779 | } |
| 3780 | for (i = 0; i < nargs; i++) |
| 3781 | { |
| 3782 | vec<tree> vec_oprndsi = vec_defs[i]; |
| 3783 | vec_oprndsi.release (); |
| 3784 | } |
| 3785 | } |
| 3786 | else if (modifier == NARROW) |
| 3787 | { |
| 3788 | auto_vec<vec<tree> > vec_defs (nargs); |
| 3789 | /* We don't define any narrowing conditional functions at present. */ |
| 3790 | gcc_assert (mask_opno < 0)((void)(!(mask_opno < 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3790, __FUNCTION__), 0 : 0)); |
| 3791 | for (j = 0; j < ncopies; ++j) |
| 3792 | { |
| 3793 | /* Build argument list for the vectorized call. */ |
| 3794 | if (j == 0) |
| 3795 | vargs.create (nargs * 2); |
| 3796 | else |
| 3797 | vargs.truncate (0); |
| 3798 | |
| 3799 | if (slp_node) |
| 3800 | { |
| 3801 | vec<tree> vec_oprnds0; |
| 3802 | |
| 3803 | vect_get_slp_defs (vinfo, slp_node, &vec_defs); |
| 3804 | vec_oprnds0 = vec_defs[0]; |
| 3805 | |
| 3806 | /* Arguments are ready. Create the new vector stmt. */ |
| 3807 | for (i = 0; vec_oprnds0.iterate (i, &vec_oprnd0); i += 2) |
| 3808 | { |
| 3809 | size_t k; |
| 3810 | vargs.truncate (0); |
| 3811 | for (k = 0; k < nargs; k++) |
| 3812 | { |
| 3813 | vec<tree> vec_oprndsk = vec_defs[k]; |
| 3814 | vargs.quick_push (vec_oprndsk[i]); |
| 3815 | vargs.quick_push (vec_oprndsk[i + 1]); |
| 3816 | } |
| 3817 | gcall *call; |
| 3818 | if (ifn != IFN_LAST) |
| 3819 | call = gimple_build_call_internal_vec (ifn, vargs); |
| 3820 | else |
| 3821 | call = gimple_build_call_vec (fndecl, vargs); |
| 3822 | new_temp = make_ssa_name (vec_dest, call); |
| 3823 | gimple_call_set_lhs (call, new_temp); |
| 3824 | gimple_call_set_nothrow (call, true); |
| 3825 | vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); |
| 3826 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (call); |
| 3827 | } |
| 3828 | continue; |
| 3829 | } |
| 3830 | |
| 3831 | for (i = 0; i < nargs; i++) |
| 3832 | { |
| 3833 | op = gimple_call_arg (stmt, i); |
| 3834 | if (j == 0) |
| 3835 | { |
| 3836 | vec_defs.quick_push (vNULL); |
| 3837 | vect_get_vec_defs_for_operand (vinfo, stmt_info, 2 * ncopies, |
| 3838 | op, &vec_defs[i], vectypes[i]); |
| 3839 | } |
| 3840 | vec_oprnd0 = vec_defs[i][2*j]; |
| 3841 | vec_oprnd1 = vec_defs[i][2*j+1]; |
| 3842 | |
| 3843 | vargs.quick_push (vec_oprnd0); |
| 3844 | vargs.quick_push (vec_oprnd1); |
| 3845 | } |
| 3846 | |
| 3847 | gcall *new_stmt = gimple_build_call_vec (fndecl, vargs); |
| 3848 | new_temp = make_ssa_name (vec_dest, new_stmt); |
| 3849 | gimple_call_set_lhs (new_stmt, new_temp); |
| 3850 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 3851 | |
| 3852 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 3853 | } |
| 3854 | |
| 3855 | if (!slp_node) |
| 3856 | *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts[0]; |
| 3857 | |
| 3858 | for (i = 0; i < nargs; i++) |
| 3859 | { |
| 3860 | vec<tree> vec_oprndsi = vec_defs[i]; |
| 3861 | vec_oprndsi.release (); |
| 3862 | } |
| 3863 | } |
| 3864 | else |
| 3865 | /* No current target implements this case. */ |
| 3866 | return false; |
| 3867 | |
| 3868 | vargs.release (); |
| 3869 | |
| 3870 | /* The call in STMT might prevent it from being removed in dce. |
| 3871 | We however cannot remove it here, due to the way the ssa name |
| 3872 | it defines is mapped to the new definition. So just replace |
| 3873 | rhs of the statement with something harmless. */ |
| 3874 | |
| 3875 | if (slp_node) |
| 3876 | return true; |
| 3877 | |
| 3878 | stmt_info = vect_orig_stmt (stmt_info); |
| 3879 | lhs = gimple_get_lhs (stmt_info->stmt); |
| 3880 | |
| 3881 | gassign *new_stmt |
| 3882 | = gimple_build_assign (lhs, build_zero_cst (TREE_TYPE (lhs)((contains_struct_check ((lhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3882, __FUNCTION__))->typed.type))); |
| 3883 | vinfo->replace_stmt (gsi, stmt_info, new_stmt); |
| 3884 | |
| 3885 | return true; |
| 3886 | } |
| 3887 | |
| 3888 | |
| 3889 | struct simd_call_arg_info |
| 3890 | { |
| 3891 | tree vectype; |
| 3892 | tree op; |
| 3893 | HOST_WIDE_INTlong linear_step; |
| 3894 | enum vect_def_type dt; |
| 3895 | unsigned int align; |
| 3896 | bool simd_lane_linear; |
| 3897 | }; |
| 3898 | |
| 3899 | /* Helper function of vectorizable_simd_clone_call. If OP, an SSA_NAME, |
| 3900 | is linear within simd lane (but not within whole loop), note it in |
| 3901 | *ARGINFO. */ |
| 3902 | |
| 3903 | static void |
| 3904 | vect_simd_lane_linear (tree op, class loop *loop, |
| 3905 | struct simd_call_arg_info *arginfo) |
| 3906 | { |
| 3907 | gimple *def_stmt = SSA_NAME_DEF_STMT (op)(tree_check ((op), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3907, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 3908 | |
| 3909 | if (!is_gimple_assign (def_stmt) |
| 3910 | || gimple_assign_rhs_code (def_stmt) != POINTER_PLUS_EXPR |
| 3911 | || !is_gimple_min_invariant (gimple_assign_rhs1 (def_stmt))) |
| 3912 | return; |
| 3913 | |
| 3914 | tree base = gimple_assign_rhs1 (def_stmt); |
| 3915 | HOST_WIDE_INTlong linear_step = 0; |
| 3916 | tree v = gimple_assign_rhs2 (def_stmt); |
| 3917 | while (TREE_CODE (v)((enum tree_code) (v)->base.code) == SSA_NAME) |
| 3918 | { |
| 3919 | tree t; |
| 3920 | def_stmt = SSA_NAME_DEF_STMT (v)(tree_check ((v), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3920, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 3921 | if (is_gimple_assign (def_stmt)) |
| 3922 | switch (gimple_assign_rhs_code (def_stmt)) |
| 3923 | { |
| 3924 | case PLUS_EXPR: |
| 3925 | t = gimple_assign_rhs2 (def_stmt); |
| 3926 | if (linear_step || TREE_CODE (t)((enum tree_code) (t)->base.code) != INTEGER_CST) |
| 3927 | return; |
| 3928 | base = fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (base), base, t)fold_build2_loc (((location_t) 0), POINTER_PLUS_EXPR, ((contains_struct_check ((base), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3928, __FUNCTION__))->typed.type), base, t ); |
| 3929 | v = gimple_assign_rhs1 (def_stmt); |
| 3930 | continue; |
| 3931 | case MULT_EXPR: |
| 3932 | t = gimple_assign_rhs2 (def_stmt); |
| 3933 | if (linear_step || !tree_fits_shwi_p (t) || integer_zerop (t)) |
| 3934 | return; |
| 3935 | linear_step = tree_to_shwi (t); |
| 3936 | v = gimple_assign_rhs1 (def_stmt); |
| 3937 | continue; |
| 3938 | CASE_CONVERTcase NOP_EXPR: case CONVERT_EXPR: |
| 3939 | t = gimple_assign_rhs1 (def_stmt); |
| 3940 | if (TREE_CODE (TREE_TYPE (t))((enum tree_code) (((contains_struct_check ((t), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3940, __FUNCTION__))->typed.type))->base.code) != INTEGER_TYPE |
| 3941 | || (TYPE_PRECISION (TREE_TYPE (v))((tree_class_check ((((contains_struct_check ((v), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3941, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3941, __FUNCTION__))->type_common.precision) |
| 3942 | < TYPE_PRECISION (TREE_TYPE (t))((tree_class_check ((((contains_struct_check ((t), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3942, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3942, __FUNCTION__))->type_common.precision))) |
| 3943 | return; |
| 3944 | if (!linear_step) |
| 3945 | linear_step = 1; |
| 3946 | v = t; |
| 3947 | continue; |
| 3948 | default: |
| 3949 | return; |
| 3950 | } |
| 3951 | else if (gimple_call_internal_p (def_stmt, IFN_GOMP_SIMD_LANE) |
| 3952 | && loop->simduid |
| 3953 | && TREE_CODE (gimple_call_arg (def_stmt, 0))((enum tree_code) (gimple_call_arg (def_stmt, 0))->base.code ) == SSA_NAME |
| 3954 | && (SSA_NAME_VAR (gimple_call_arg (def_stmt, 0))((tree_check ((gimple_call_arg (def_stmt, 0)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 3954, __FUNCTION__, (SSA_NAME)))->ssa_name.var == (tree) nullptr || ((enum tree_code) ((gimple_call_arg (def_stmt, 0) )->ssa_name.var)->base.code) == IDENTIFIER_NODE ? (tree ) nullptr : (gimple_call_arg (def_stmt, 0))->ssa_name.var) |
| 3955 | == loop->simduid)) |
| 3956 | { |
| 3957 | if (!linear_step) |
| 3958 | linear_step = 1; |
| 3959 | arginfo->linear_step = linear_step; |
| 3960 | arginfo->op = base; |
| 3961 | arginfo->simd_lane_linear = true; |
| 3962 | return; |
| 3963 | } |
| 3964 | } |
| 3965 | } |
| 3966 | |
| 3967 | /* Return the number of elements in vector type VECTYPE, which is associated |
| 3968 | with a SIMD clone. At present these vectors always have a constant |
| 3969 | length. */ |
| 3970 | |
| 3971 | static unsigned HOST_WIDE_INTlong |
| 3972 | simd_clone_subparts (tree vectype) |
| 3973 | { |
| 3974 | return TYPE_VECTOR_SUBPARTS (vectype).to_constant (); |
| 3975 | } |
| 3976 | |
| 3977 | /* Function vectorizable_simd_clone_call. |
| 3978 | |
| 3979 | Check if STMT_INFO performs a function call that can be vectorized |
| 3980 | by calling a simd clone of the function. |
| 3981 | If VEC_STMT is also passed, vectorize STMT_INFO: create a vectorized |
| 3982 | stmt to replace it, put it in VEC_STMT, and insert it at GSI. |
| 3983 | Return true if STMT_INFO is vectorizable in this way. */ |
| 3984 | |
| 3985 | static bool |
| 3986 | vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info, |
| 3987 | gimple_stmt_iterator *gsi, |
| 3988 | gimple **vec_stmt, slp_tree slp_node, |
| 3989 | stmt_vector_for_cost *) |
| 3990 | { |
| 3991 | tree vec_dest; |
| 3992 | tree scalar_dest; |
| 3993 | tree op, type; |
| 3994 | tree vec_oprnd0 = NULL_TREE(tree) nullptr; |
| 3995 | tree vectype; |
| 3996 | poly_uint64 nunits; |
| 3997 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 3998 | bb_vec_info bb_vinfo = dyn_cast <bb_vec_info> (vinfo); |
| 3999 | class loop *loop = loop_vinfo ? LOOP_VINFO_LOOP (loop_vinfo)(loop_vinfo)->loop : NULLnullptr; |
| 4000 | tree fndecl, new_temp; |
| 4001 | int ncopies, j; |
| 4002 | auto_vec<simd_call_arg_info> arginfo; |
| 4003 | vec<tree> vargs = vNULL; |
| 4004 | size_t i, nargs; |
| 4005 | tree lhs, rtype, ratype; |
| 4006 | vec<constructor_elt, va_gc> *ret_ctor_elts = NULLnullptr; |
| 4007 | int arg_offset = 0; |
| 4008 | |
| 4009 | /* Is STMT a vectorizable call? */ |
| 4010 | gcall *stmt = dyn_cast <gcall *> (stmt_info->stmt); |
| 4011 | if (!stmt) |
| 4012 | return false; |
| 4013 | |
| 4014 | fndecl = gimple_call_fndecl (stmt); |
| 4015 | if (fndecl == NULL_TREE(tree) nullptr |
| 4016 | && gimple_call_internal_p (stmt, IFN_MASK_CALL)) |
| 4017 | { |
| 4018 | fndecl = gimple_call_arg (stmt, 0); |
| 4019 | gcc_checking_assert (TREE_CODE (fndecl) == ADDR_EXPR)((void)(!(((enum tree_code) (fndecl)->base.code) == ADDR_EXPR ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4019, __FUNCTION__), 0 : 0)); |
| 4020 | fndecl = TREE_OPERAND (fndecl, 0)(*((const_cast<tree*> (tree_operand_check ((fndecl), (0 ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4020, __FUNCTION__))))); |
| 4021 | gcc_checking_assert (TREE_CODE (fndecl) == FUNCTION_DECL)((void)(!(((enum tree_code) (fndecl)->base.code) == FUNCTION_DECL ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4021, __FUNCTION__), 0 : 0)); |
| 4022 | arg_offset = 1; |
| 4023 | } |
| 4024 | if (fndecl == NULL_TREE(tree) nullptr) |
| 4025 | return false; |
| 4026 | |
| 4027 | struct cgraph_node *node = cgraph_node::get (fndecl); |
| 4028 | if (node == NULLnullptr || node->simd_clones == NULLnullptr) |
| 4029 | return false; |
| 4030 | |
| 4031 | if (!STMT_VINFO_RELEVANT_P (stmt_info)((stmt_info)->relevant != vect_unused_in_scope) && !bb_vinfo) |
| 4032 | return false; |
| 4033 | |
| 4034 | if (STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_internal_def |
| 4035 | && ! vec_stmt) |
| 4036 | return false; |
| 4037 | |
| 4038 | if (gimple_call_lhs (stmt) |
| 4039 | && TREE_CODE (gimple_call_lhs (stmt))((enum tree_code) (gimple_call_lhs (stmt))->base.code) != SSA_NAME) |
| 4040 | return false; |
| 4041 | |
| 4042 | gcc_checking_assert (!stmt_can_throw_internal (cfun, stmt))((void)(!(!stmt_can_throw_internal ((cfun + 0), stmt)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4042, __FUNCTION__), 0 : 0)); |
| 4043 | |
| 4044 | vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 4045 | |
| 4046 | if (loop_vinfo && nested_in_vect_loop_p (loop, stmt_info)) |
| 4047 | return false; |
| 4048 | |
| 4049 | /* FORNOW */ |
| 4050 | if (slp_node) |
| 4051 | return false; |
| 4052 | |
| 4053 | /* Process function arguments. */ |
| 4054 | nargs = gimple_call_num_args (stmt) - arg_offset; |
| 4055 | |
| 4056 | /* Bail out if the function has zero arguments. */ |
| 4057 | if (nargs == 0) |
| 4058 | return false; |
| 4059 | |
| 4060 | arginfo.reserve (nargs, true); |
| 4061 | |
| 4062 | for (i = 0; i < nargs; i++) |
| 4063 | { |
| 4064 | simd_call_arg_info thisarginfo; |
| 4065 | affine_iv iv; |
| 4066 | |
| 4067 | thisarginfo.linear_step = 0; |
| 4068 | thisarginfo.align = 0; |
| 4069 | thisarginfo.op = NULL_TREE(tree) nullptr; |
| 4070 | thisarginfo.simd_lane_linear = false; |
| 4071 | |
| 4072 | op = gimple_call_arg (stmt, i + arg_offset); |
| 4073 | if (!vect_is_simple_use (op, vinfo, &thisarginfo.dt, |
| 4074 | &thisarginfo.vectype) |
| 4075 | || thisarginfo.dt == vect_uninitialized_def) |
| 4076 | { |
| 4077 | if (dump_enabled_p ()) |
| 4078 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 4079 | "use not simple.\n"); |
| 4080 | return false; |
| 4081 | } |
| 4082 | |
| 4083 | if (thisarginfo.dt == vect_constant_def |
| 4084 | || thisarginfo.dt == vect_external_def) |
| 4085 | gcc_assert (thisarginfo.vectype == NULL_TREE)((void)(!(thisarginfo.vectype == (tree) nullptr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4085, __FUNCTION__), 0 : 0)); |
| 4086 | else |
| 4087 | gcc_assert (thisarginfo.vectype != NULL_TREE)((void)(!(thisarginfo.vectype != (tree) nullptr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4087, __FUNCTION__), 0 : 0)); |
| 4088 | |
| 4089 | /* For linear arguments, the analyze phase should have saved |
| 4090 | the base and step in STMT_VINFO_SIMD_CLONE_INFO. */ |
| 4091 | if (i * 3 + 4 <= STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.length () |
| 4092 | && STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[i * 3 + 2]) |
| 4093 | { |
| 4094 | gcc_assert (vec_stmt)((void)(!(vec_stmt) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4094, __FUNCTION__), 0 : 0)); |
| 4095 | thisarginfo.linear_step |
| 4096 | = tree_to_shwi (STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[i * 3 + 2]); |
| 4097 | thisarginfo.op |
| 4098 | = STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[i * 3 + 1]; |
| 4099 | thisarginfo.simd_lane_linear |
| 4100 | = (STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[i * 3 + 3] |
| 4101 | == boolean_true_nodeglobal_trees[TI_BOOLEAN_TRUE]); |
| 4102 | /* If loop has been peeled for alignment, we need to adjust it. */ |
| 4103 | tree n1 = LOOP_VINFO_NITERS_UNCHANGED (loop_vinfo)(loop_vinfo)->num_iters_unchanged; |
| 4104 | tree n2 = LOOP_VINFO_NITERS (loop_vinfo)(loop_vinfo)->num_iters; |
| 4105 | if (n1 != n2 && !thisarginfo.simd_lane_linear) |
| 4106 | { |
| 4107 | tree bias = fold_build2 (MINUS_EXPR, TREE_TYPE (n1), n1, n2)fold_build2_loc (((location_t) 0), MINUS_EXPR, ((contains_struct_check ((n1), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4107, __FUNCTION__))->typed.type), n1, n2 ); |
| 4108 | tree step = STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[i * 3 + 2]; |
| 4109 | tree opt = TREE_TYPE (thisarginfo.op)((contains_struct_check ((thisarginfo.op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4109, __FUNCTION__))->typed.type); |
| 4110 | bias = fold_convert (TREE_TYPE (step), bias)fold_convert_loc (((location_t) 0), ((contains_struct_check ( (step), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4110, __FUNCTION__))->typed.type), bias); |
| 4111 | bias = fold_build2 (MULT_EXPR, TREE_TYPE (step), bias, step)fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((step), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4111, __FUNCTION__))->typed.type), bias, step ); |
| 4112 | thisarginfo.op |
| 4113 | = fold_build2 (POINTER_TYPE_P (opt)fold_build2_loc (((location_t) 0), (((enum tree_code) (opt)-> base.code) == POINTER_TYPE || ((enum tree_code) (opt)->base .code) == REFERENCE_TYPE) ? POINTER_PLUS_EXPR : PLUS_EXPR, opt , thisarginfo.op, bias ) |
| 4114 | ? POINTER_PLUS_EXPR : PLUS_EXPR, opt,fold_build2_loc (((location_t) 0), (((enum tree_code) (opt)-> base.code) == POINTER_TYPE || ((enum tree_code) (opt)->base .code) == REFERENCE_TYPE) ? POINTER_PLUS_EXPR : PLUS_EXPR, opt , thisarginfo.op, bias ) |
| 4115 | thisarginfo.op, bias)fold_build2_loc (((location_t) 0), (((enum tree_code) (opt)-> base.code) == POINTER_TYPE || ((enum tree_code) (opt)->base .code) == REFERENCE_TYPE) ? POINTER_PLUS_EXPR : PLUS_EXPR, opt , thisarginfo.op, bias ); |
| 4116 | } |
| 4117 | } |
| 4118 | else if (!vec_stmt |
| 4119 | && thisarginfo.dt != vect_constant_def |
| 4120 | && thisarginfo.dt != vect_external_def |
| 4121 | && loop_vinfo |
| 4122 | && TREE_CODE (op)((enum tree_code) (op)->base.code) == SSA_NAME |
| 4123 | && simple_iv (loop, loop_containing_stmt (stmt), op, |
| 4124 | &iv, false) |
| 4125 | && tree_fits_shwi_p (iv.step)) |
| 4126 | { |
| 4127 | thisarginfo.linear_step = tree_to_shwi (iv.step); |
| 4128 | thisarginfo.op = iv.base; |
| 4129 | } |
| 4130 | else if ((thisarginfo.dt == vect_constant_def |
| 4131 | || thisarginfo.dt == vect_external_def) |
| 4132 | && POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4132, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4132, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE )) |
| 4133 | thisarginfo.align = get_pointer_alignment (op) / BITS_PER_UNIT(8); |
| 4134 | /* Addresses of array elements indexed by GOMP_SIMD_LANE are |
| 4135 | linear too. */ |
| 4136 | if (POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4136, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4136, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4137 | && !thisarginfo.linear_step |
| 4138 | && !vec_stmt |
| 4139 | && thisarginfo.dt != vect_constant_def |
| 4140 | && thisarginfo.dt != vect_external_def |
| 4141 | && loop_vinfo |
| 4142 | && !slp_node |
| 4143 | && TREE_CODE (op)((enum tree_code) (op)->base.code) == SSA_NAME) |
| 4144 | vect_simd_lane_linear (op, loop, &thisarginfo); |
| 4145 | |
| 4146 | arginfo.quick_push (thisarginfo); |
| 4147 | } |
| 4148 | |
| 4149 | poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo)(loop_vinfo)->vectorization_factor; |
| 4150 | if (!vf.is_constant ()) |
| 4151 | { |
| 4152 | if (dump_enabled_p ()) |
| 4153 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 4154 | "not considering SIMD clones; not yet supported" |
| 4155 | " for variable-width vectors.\n"); |
| 4156 | return false; |
| 4157 | } |
| 4158 | |
| 4159 | unsigned int badness = 0; |
| 4160 | struct cgraph_node *bestn = NULLnullptr; |
| 4161 | if (STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.exists ()) |
| 4162 | bestn = cgraph_node::get (STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info[0]); |
| 4163 | else |
| 4164 | for (struct cgraph_node *n = node->simd_clones; n != NULLnullptr; |
| 4165 | n = n->simdclone->next_clone) |
| 4166 | { |
| 4167 | unsigned int this_badness = 0; |
| 4168 | unsigned int num_calls; |
| 4169 | if (!constant_multiple_p (vf, n->simdclone->simdlen, &num_calls) |
| 4170 | || n->simdclone->nargs != nargs) |
| 4171 | continue; |
| 4172 | if (num_calls != 1) |
| 4173 | this_badness += exact_log2 (num_calls) * 4096; |
| 4174 | if (n->simdclone->inbranch) |
| 4175 | this_badness += 8192; |
| 4176 | int target_badness = targetm.simd_clone.usable (n); |
| 4177 | if (target_badness < 0) |
| 4178 | continue; |
| 4179 | this_badness += target_badness * 512; |
| 4180 | for (i = 0; i < nargs; i++) |
| 4181 | { |
| 4182 | switch (n->simdclone->args[i].arg_type) |
| 4183 | { |
| 4184 | case SIMD_CLONE_ARG_TYPE_VECTOR: |
| 4185 | if (!useless_type_conversion_p |
| 4186 | (n->simdclone->args[i].orig_type, |
| 4187 | TREE_TYPE (gimple_call_arg (stmt, i + arg_offset))((contains_struct_check ((gimple_call_arg (stmt, i + arg_offset )), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4187, __FUNCTION__))->typed.type))) |
| 4188 | i = -1; |
| 4189 | else if (arginfo[i].dt == vect_constant_def |
| 4190 | || arginfo[i].dt == vect_external_def |
| 4191 | || arginfo[i].linear_step) |
| 4192 | this_badness += 64; |
| 4193 | break; |
| 4194 | case SIMD_CLONE_ARG_TYPE_UNIFORM: |
| 4195 | if (arginfo[i].dt != vect_constant_def |
| 4196 | && arginfo[i].dt != vect_external_def) |
| 4197 | i = -1; |
| 4198 | break; |
| 4199 | case SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP: |
| 4200 | case SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP: |
| 4201 | if (arginfo[i].dt == vect_constant_def |
| 4202 | || arginfo[i].dt == vect_external_def |
| 4203 | || (arginfo[i].linear_step |
| 4204 | != n->simdclone->args[i].linear_step)) |
| 4205 | i = -1; |
| 4206 | break; |
| 4207 | case SIMD_CLONE_ARG_TYPE_LINEAR_VARIABLE_STEP: |
| 4208 | case SIMD_CLONE_ARG_TYPE_LINEAR_VAL_CONSTANT_STEP: |
| 4209 | case SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_CONSTANT_STEP: |
| 4210 | case SIMD_CLONE_ARG_TYPE_LINEAR_REF_VARIABLE_STEP: |
| 4211 | case SIMD_CLONE_ARG_TYPE_LINEAR_VAL_VARIABLE_STEP: |
| 4212 | case SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_VARIABLE_STEP: |
| 4213 | /* FORNOW */ |
| 4214 | i = -1; |
| 4215 | break; |
| 4216 | case SIMD_CLONE_ARG_TYPE_MASK: |
| 4217 | break; |
| 4218 | } |
| 4219 | if (i == (size_t) -1) |
| 4220 | break; |
| 4221 | if (n->simdclone->args[i].alignment > arginfo[i].align) |
| 4222 | { |
| 4223 | i = -1; |
| 4224 | break; |
| 4225 | } |
| 4226 | if (arginfo[i].align) |
| 4227 | this_badness += (exact_log2 (arginfo[i].align) |
| 4228 | - exact_log2 (n->simdclone->args[i].alignment)); |
| 4229 | } |
| 4230 | if (i == (size_t) -1) |
| 4231 | continue; |
| 4232 | if (bestn == NULLnullptr || this_badness < badness) |
| 4233 | { |
| 4234 | bestn = n; |
| 4235 | badness = this_badness; |
| 4236 | } |
| 4237 | } |
| 4238 | |
| 4239 | if (bestn == NULLnullptr) |
| 4240 | return false; |
| 4241 | |
| 4242 | for (i = 0; i < nargs; i++) |
| 4243 | { |
| 4244 | if ((arginfo[i].dt == vect_constant_def |
| 4245 | || arginfo[i].dt == vect_external_def) |
| 4246 | && bestn->simdclone->args[i].arg_type == SIMD_CLONE_ARG_TYPE_VECTOR) |
| 4247 | { |
| 4248 | tree arg_type = TREE_TYPE (gimple_call_arg (stmt, i + arg_offset))((contains_struct_check ((gimple_call_arg (stmt, i + arg_offset )), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4248, __FUNCTION__))->typed.type); |
| 4249 | arginfo[i].vectype = get_vectype_for_scalar_type (vinfo, arg_type, |
| 4250 | slp_node); |
| 4251 | if (arginfo[i].vectype == NULLnullptr |
| 4252 | || !constant_multiple_p (bestn->simdclone->simdlen, |
| 4253 | simd_clone_subparts (arginfo[i].vectype))) |
| 4254 | return false; |
| 4255 | } |
| 4256 | |
| 4257 | if (bestn->simdclone->args[i].arg_type == SIMD_CLONE_ARG_TYPE_VECTOR |
| 4258 | && VECTOR_BOOLEAN_TYPE_P (bestn->simdclone->args[i].vector_type)(((enum tree_code) (bestn->simdclone->args[i].vector_type )->base.code) == VECTOR_TYPE && ((enum tree_code) ( ((contains_struct_check ((bestn->simdclone->args[i].vector_type ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4258, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 4259 | { |
| 4260 | if (dump_enabled_p ()) |
| 4261 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 4262 | "vector mask arguments are not supported.\n"); |
| 4263 | return false; |
| 4264 | } |
| 4265 | |
| 4266 | if (bestn->simdclone->args[i].arg_type == SIMD_CLONE_ARG_TYPE_MASK |
| 4267 | && bestn->simdclone->mask_mode == VOIDmode((void) 0, E_VOIDmode) |
| 4268 | && (simd_clone_subparts (bestn->simdclone->args[i].vector_type) |
| 4269 | != simd_clone_subparts (arginfo[i].vectype))) |
| 4270 | { |
| 4271 | /* FORNOW we only have partial support for vector-type masks that |
| 4272 | can't hold all of simdlen. */ |
| 4273 | if (dump_enabled_p ()) |
| 4274 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, |
| 4275 | vect_location, |
| 4276 | "in-branch vector clones are not yet" |
| 4277 | " supported for mismatched vector sizes.\n"); |
| 4278 | return false; |
| 4279 | } |
| 4280 | if (bestn->simdclone->args[i].arg_type == SIMD_CLONE_ARG_TYPE_MASK |
| 4281 | && bestn->simdclone->mask_mode != VOIDmode((void) 0, E_VOIDmode)) |
| 4282 | { |
| 4283 | /* FORNOW don't support integer-type masks. */ |
| 4284 | if (dump_enabled_p ()) |
| 4285 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, |
| 4286 | vect_location, |
| 4287 | "in-branch vector clones are not yet" |
| 4288 | " supported for integer mask modes.\n"); |
| 4289 | return false; |
| 4290 | } |
| 4291 | } |
| 4292 | |
| 4293 | fndecl = bestn->decl; |
| 4294 | nunits = bestn->simdclone->simdlen; |
| 4295 | ncopies = vector_unroll_factor (vf, nunits)(exact_div (vf, nunits).to_constant ()); |
| 4296 | |
| 4297 | /* If the function isn't const, only allow it in simd loops where user |
| 4298 | has asserted that at least nunits consecutive iterations can be |
| 4299 | performed using SIMD instructions. */ |
| 4300 | if ((loop == NULLnullptr || maybe_lt ((unsigned) loop->safelen, nunits)) |
| 4301 | && gimple_vuse (stmt)) |
| 4302 | return false; |
| 4303 | |
| 4304 | /* Sanity check: make sure that at least one copy of the vectorized stmt |
| 4305 | needs to be generated. */ |
| 4306 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4306, __FUNCTION__), 0 : 0)); |
| 4307 | |
| 4308 | if (!vec_stmt) /* transformation not required. */ |
| 4309 | { |
| 4310 | /* When the original call is pure or const but the SIMD ABI dictates |
| 4311 | an aggregate return we will have to use a virtual definition and |
| 4312 | in a loop eventually even need to add a virtual PHI. That's |
| 4313 | not straight-forward so allow to fix this up via renaming. */ |
| 4314 | if (gimple_call_lhs (stmt) |
| 4315 | && !gimple_vdef (stmt) |
| 4316 | && TREE_CODE (TREE_TYPE (TREE_TYPE (bestn->decl)))((enum tree_code) (((contains_struct_check ((((contains_struct_check ((bestn->decl), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4316, __FUNCTION__))->typed.type)), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4316, __FUNCTION__))->typed.type))->base.code) == ARRAY_TYPE) |
| 4317 | vinfo->any_known_not_updated_vssa = true; |
| 4318 | STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.safe_push (bestn->decl); |
| 4319 | for (i = 0; i < nargs; i++) |
| 4320 | if ((bestn->simdclone->args[i].arg_type |
| 4321 | == SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP) |
| 4322 | || (bestn->simdclone->args[i].arg_type |
| 4323 | == SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP)) |
| 4324 | { |
| 4325 | STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.safe_grow_cleared (i * 3 |
| 4326 | + 1, |
| 4327 | true); |
| 4328 | STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.safe_push (arginfo[i].op); |
| 4329 | tree lst = POINTER_TYPE_P (TREE_TYPE (arginfo[i].op))(((enum tree_code) (((contains_struct_check ((arginfo[i].op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4329, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((arginfo[i].op ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4329, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4330 | ? size_type_nodeglobal_trees[TI_SIZE_TYPE] : TREE_TYPE (arginfo[i].op)((contains_struct_check ((arginfo[i].op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4330, __FUNCTION__))->typed.type); |
| 4331 | tree ls = build_int_cst (lst, arginfo[i].linear_step); |
| 4332 | STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.safe_push (ls); |
| 4333 | tree sll = arginfo[i].simd_lane_linear |
| 4334 | ? boolean_true_nodeglobal_trees[TI_BOOLEAN_TRUE] : boolean_false_nodeglobal_trees[TI_BOOLEAN_FALSE]; |
| 4335 | STMT_VINFO_SIMD_CLONE_INFO (stmt_info)(stmt_info)->simd_clone_info.safe_push (sll); |
| 4336 | } |
| 4337 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = call_simd_clone_vec_info_type; |
| 4338 | DUMP_VECT_SCOPE ("vectorizable_simd_clone_call")auto_dump_scope scope ("vectorizable_simd_clone_call", vect_location ); |
| 4339 | /* vect_model_simple_cost (vinfo, stmt_info, ncopies, |
| 4340 | dt, slp_node, cost_vec); */ |
| 4341 | return true; |
| 4342 | } |
| 4343 | |
| 4344 | /* Transform. */ |
| 4345 | |
| 4346 | if (dump_enabled_p ()) |
| 4347 | dump_printf_loc (MSG_NOTE, vect_location, "transform call.\n"); |
| 4348 | |
| 4349 | /* Handle def. */ |
| 4350 | scalar_dest = gimple_call_lhs (stmt); |
| 4351 | vec_dest = NULL_TREE(tree) nullptr; |
| 4352 | rtype = NULL_TREE(tree) nullptr; |
| 4353 | ratype = NULL_TREE(tree) nullptr; |
| 4354 | if (scalar_dest) |
| 4355 | { |
| 4356 | vec_dest = vect_create_destination_var (scalar_dest, vectype); |
| 4357 | rtype = TREE_TYPE (TREE_TYPE (fndecl))((contains_struct_check ((((contains_struct_check ((fndecl), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4357, __FUNCTION__))->typed.type)), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4357, __FUNCTION__))->typed.type); |
| 4358 | if (TREE_CODE (rtype)((enum tree_code) (rtype)->base.code) == ARRAY_TYPE) |
| 4359 | { |
| 4360 | ratype = rtype; |
| 4361 | rtype = TREE_TYPE (ratype)((contains_struct_check ((ratype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4361, __FUNCTION__))->typed.type); |
| 4362 | } |
| 4363 | } |
| 4364 | |
| 4365 | auto_vec<vec<tree> > vec_oprnds; |
| 4366 | auto_vec<unsigned> vec_oprnds_i; |
| 4367 | vec_oprnds.safe_grow_cleared (nargs, true); |
| 4368 | vec_oprnds_i.safe_grow_cleared (nargs, true); |
| 4369 | for (j = 0; j < ncopies; ++j) |
| 4370 | { |
| 4371 | /* Build argument list for the vectorized call. */ |
| 4372 | if (j == 0) |
| 4373 | vargs.create (nargs); |
| 4374 | else |
| 4375 | vargs.truncate (0); |
| 4376 | |
| 4377 | for (i = 0; i < nargs; i++) |
| 4378 | { |
| 4379 | unsigned int k, l, m, o; |
| 4380 | tree atype; |
| 4381 | op = gimple_call_arg (stmt, i + arg_offset); |
| 4382 | switch (bestn->simdclone->args[i].arg_type) |
| 4383 | { |
| 4384 | case SIMD_CLONE_ARG_TYPE_VECTOR: |
| 4385 | atype = bestn->simdclone->args[i].vector_type; |
| 4386 | o = vector_unroll_factor (nunits,(exact_div (nunits, simd_clone_subparts (atype)).to_constant ( )) |
| 4387 | simd_clone_subparts (atype))(exact_div (nunits, simd_clone_subparts (atype)).to_constant ( )); |
| 4388 | for (m = j * o; m < (j + 1) * o; m++) |
| 4389 | { |
| 4390 | if (simd_clone_subparts (atype) |
| 4391 | < simd_clone_subparts (arginfo[i].vectype)) |
| 4392 | { |
| 4393 | poly_uint64 prec = GET_MODE_BITSIZE (TYPE_MODE (atype)((((enum tree_code) ((tree_class_check ((atype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4393, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (atype) : (atype)->type_common.mode)); |
| 4394 | k = (simd_clone_subparts (arginfo[i].vectype) |
| 4395 | / simd_clone_subparts (atype)); |
| 4396 | gcc_assert ((k & (k - 1)) == 0)((void)(!((k & (k - 1)) == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4396, __FUNCTION__), 0 : 0)); |
| 4397 | if (m == 0) |
| 4398 | { |
| 4399 | vect_get_vec_defs_for_operand (vinfo, stmt_info, |
| 4400 | ncopies * o / k, op, |
| 4401 | &vec_oprnds[i]); |
| 4402 | vec_oprnds_i[i] = 0; |
| 4403 | vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++]; |
| 4404 | } |
| 4405 | else |
| 4406 | { |
| 4407 | vec_oprnd0 = arginfo[i].op; |
| 4408 | if ((m & (k - 1)) == 0) |
| 4409 | vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++]; |
| 4410 | } |
| 4411 | arginfo[i].op = vec_oprnd0; |
| 4412 | vec_oprnd0 |
| 4413 | = build3 (BIT_FIELD_REF, atype, vec_oprnd0, |
| 4414 | bitsize_int (prec)size_int_kind (prec, stk_bitsizetype), |
| 4415 | bitsize_int ((m & (k - 1)) * prec)size_int_kind ((m & (k - 1)) * prec, stk_bitsizetype)); |
| 4416 | gassign *new_stmt |
| 4417 | = gimple_build_assign (make_ssa_name (atype), |
| 4418 | vec_oprnd0); |
| 4419 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 4420 | new_stmt, gsi); |
| 4421 | vargs.safe_push (gimple_assign_lhs (new_stmt)); |
| 4422 | } |
| 4423 | else |
| 4424 | { |
| 4425 | k = (simd_clone_subparts (atype) |
| 4426 | / simd_clone_subparts (arginfo[i].vectype)); |
| 4427 | gcc_assert ((k & (k - 1)) == 0)((void)(!((k & (k - 1)) == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4427, __FUNCTION__), 0 : 0)); |
| 4428 | vec<constructor_elt, va_gc> *ctor_elts; |
| 4429 | if (k != 1) |
| 4430 | vec_alloc (ctor_elts, k); |
| 4431 | else |
| 4432 | ctor_elts = NULLnullptr; |
| 4433 | for (l = 0; l < k; l++) |
| 4434 | { |
| 4435 | if (m == 0 && l == 0) |
| 4436 | { |
| 4437 | vect_get_vec_defs_for_operand (vinfo, stmt_info, |
| 4438 | k * o * ncopies, |
| 4439 | op, |
| 4440 | &vec_oprnds[i]); |
| 4441 | vec_oprnds_i[i] = 0; |
| 4442 | vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++]; |
| 4443 | } |
| 4444 | else |
| 4445 | vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++]; |
| 4446 | arginfo[i].op = vec_oprnd0; |
| 4447 | if (k == 1) |
| 4448 | break; |
| 4449 | CONSTRUCTOR_APPEND_ELT (ctor_elts, NULL_TREE,do { constructor_elt _ce___ = {(tree) nullptr, vec_oprnd0}; vec_safe_push ((ctor_elts), _ce___); } while (0) |
| 4450 | vec_oprnd0)do { constructor_elt _ce___ = {(tree) nullptr, vec_oprnd0}; vec_safe_push ((ctor_elts), _ce___); } while (0); |
| 4451 | } |
| 4452 | if (k == 1) |
| 4453 | if (!useless_type_conversion_p (TREE_TYPE (vec_oprnd0)((contains_struct_check ((vec_oprnd0), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4453, __FUNCTION__))->typed.type), |
| 4454 | atype)) |
| 4455 | { |
| 4456 | vec_oprnd0 |
| 4457 | = build1 (VIEW_CONVERT_EXPR, atype, vec_oprnd0); |
| 4458 | gassign *new_stmt |
| 4459 | = gimple_build_assign (make_ssa_name (atype), |
| 4460 | vec_oprnd0); |
| 4461 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 4462 | new_stmt, gsi); |
| 4463 | vargs.safe_push (gimple_assign_lhs (new_stmt)); |
| 4464 | } |
| 4465 | else |
| 4466 | vargs.safe_push (vec_oprnd0); |
| 4467 | else |
| 4468 | { |
| 4469 | vec_oprnd0 = build_constructor (atype, ctor_elts); |
| 4470 | gassign *new_stmt |
| 4471 | = gimple_build_assign (make_ssa_name (atype), |
| 4472 | vec_oprnd0); |
| 4473 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 4474 | new_stmt, gsi); |
| 4475 | vargs.safe_push (gimple_assign_lhs (new_stmt)); |
| 4476 | } |
| 4477 | } |
| 4478 | } |
| 4479 | break; |
| 4480 | case SIMD_CLONE_ARG_TYPE_MASK: |
| 4481 | atype = bestn->simdclone->args[i].vector_type; |
| 4482 | if (bestn->simdclone->mask_mode != VOIDmode((void) 0, E_VOIDmode)) |
| 4483 | { |
| 4484 | /* FORNOW: this is disabled above. */ |
| 4485 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4485, __FUNCTION__)); |
| 4486 | } |
| 4487 | else |
| 4488 | { |
| 4489 | tree elt_type = TREE_TYPE (atype)((contains_struct_check ((atype), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4489, __FUNCTION__))->typed.type); |
| 4490 | tree one = fold_convert (elt_type, integer_one_node)fold_convert_loc (((location_t) 0), elt_type, global_trees[TI_INTEGER_ONE ]); |
| 4491 | tree zero = fold_convert (elt_type, integer_zero_node)fold_convert_loc (((location_t) 0), elt_type, global_trees[TI_INTEGER_ZERO ]); |
| 4492 | o = vector_unroll_factor (nunits,(exact_div (nunits, simd_clone_subparts (atype)).to_constant ( )) |
| 4493 | simd_clone_subparts (atype))(exact_div (nunits, simd_clone_subparts (atype)).to_constant ( )); |
| 4494 | for (m = j * o; m < (j + 1) * o; m++) |
| 4495 | { |
| 4496 | if (simd_clone_subparts (atype) |
| 4497 | < simd_clone_subparts (arginfo[i].vectype)) |
| 4498 | { |
| 4499 | /* The mask type has fewer elements than simdlen. */ |
| 4500 | |
| 4501 | /* FORNOW */ |
| 4502 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4502, __FUNCTION__)); |
| 4503 | } |
| 4504 | else if (simd_clone_subparts (atype) |
| 4505 | == simd_clone_subparts (arginfo[i].vectype)) |
| 4506 | { |
| 4507 | /* The SIMD clone function has the same number of |
| 4508 | elements as the current function. */ |
| 4509 | if (m == 0) |
| 4510 | { |
| 4511 | vect_get_vec_defs_for_operand (vinfo, stmt_info, |
| 4512 | o * ncopies, |
| 4513 | op, |
| 4514 | &vec_oprnds[i]); |
| 4515 | vec_oprnds_i[i] = 0; |
| 4516 | } |
| 4517 | vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++]; |
| 4518 | vec_oprnd0 |
| 4519 | = build3 (VEC_COND_EXPR, atype, vec_oprnd0, |
| 4520 | build_vector_from_val (atype, one), |
| 4521 | build_vector_from_val (atype, zero)); |
| 4522 | gassign *new_stmt |
| 4523 | = gimple_build_assign (make_ssa_name (atype), |
| 4524 | vec_oprnd0); |
| 4525 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 4526 | new_stmt, gsi); |
| 4527 | vargs.safe_push (gimple_assign_lhs (new_stmt)); |
| 4528 | } |
| 4529 | else |
| 4530 | { |
| 4531 | /* The mask type has more elements than simdlen. */ |
| 4532 | |
| 4533 | /* FORNOW */ |
| 4534 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4534, __FUNCTION__)); |
| 4535 | } |
| 4536 | } |
| 4537 | } |
| 4538 | break; |
| 4539 | case SIMD_CLONE_ARG_TYPE_UNIFORM: |
| 4540 | vargs.safe_push (op); |
| 4541 | break; |
| 4542 | case SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP: |
| 4543 | case SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP: |
| 4544 | if (j == 0) |
| 4545 | { |
| 4546 | gimple_seq stmts; |
| 4547 | arginfo[i].op |
| 4548 | = force_gimple_operand (unshare_expr (arginfo[i].op), |
| 4549 | &stmts, true, NULL_TREE(tree) nullptr); |
| 4550 | if (stmts != NULLnullptr) |
| 4551 | { |
| 4552 | basic_block new_bb; |
| 4553 | edge pe = loop_preheader_edge (loop); |
| 4554 | new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); |
| 4555 | gcc_assert (!new_bb)((void)(!(!new_bb) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4555, __FUNCTION__), 0 : 0)); |
| 4556 | } |
| 4557 | if (arginfo[i].simd_lane_linear) |
| 4558 | { |
| 4559 | vargs.safe_push (arginfo[i].op); |
| 4560 | break; |
| 4561 | } |
| 4562 | tree phi_res = copy_ssa_name (op); |
| 4563 | gphi *new_phi = create_phi_node (phi_res, loop->header); |
| 4564 | add_phi_arg (new_phi, arginfo[i].op, |
| 4565 | loop_preheader_edge (loop), UNKNOWN_LOCATION((location_t) 0)); |
| 4566 | enum tree_code code |
| 4567 | = POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4567, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4567, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4568 | ? POINTER_PLUS_EXPR : PLUS_EXPR; |
| 4569 | tree type = POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4569, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4569, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4570 | ? sizetypesizetype_tab[(int) stk_sizetype] : TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4570, __FUNCTION__))->typed.type); |
| 4571 | poly_widest_int cst |
| 4572 | = wi::mul (bestn->simdclone->args[i].linear_step, |
| 4573 | ncopies * nunits); |
| 4574 | tree tcst = wide_int_to_tree (type, cst); |
| 4575 | tree phi_arg = copy_ssa_name (op); |
| 4576 | gassign *new_stmt |
| 4577 | = gimple_build_assign (phi_arg, code, phi_res, tcst); |
| 4578 | gimple_stmt_iterator si = gsi_after_labels (loop->header); |
| 4579 | gsi_insert_after (&si, new_stmt, GSI_NEW_STMT); |
| 4580 | add_phi_arg (new_phi, phi_arg, loop_latch_edge (loop), |
| 4581 | UNKNOWN_LOCATION((location_t) 0)); |
| 4582 | arginfo[i].op = phi_res; |
| 4583 | vargs.safe_push (phi_res); |
| 4584 | } |
| 4585 | else |
| 4586 | { |
| 4587 | enum tree_code code |
| 4588 | = POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4588, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4588, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4589 | ? POINTER_PLUS_EXPR : PLUS_EXPR; |
| 4590 | tree type = POINTER_TYPE_P (TREE_TYPE (op))(((enum tree_code) (((contains_struct_check ((op), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4590, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4590, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE ) |
| 4591 | ? sizetypesizetype_tab[(int) stk_sizetype] : TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4591, __FUNCTION__))->typed.type); |
| 4592 | poly_widest_int cst |
| 4593 | = wi::mul (bestn->simdclone->args[i].linear_step, |
| 4594 | j * nunits); |
| 4595 | tree tcst = wide_int_to_tree (type, cst); |
| 4596 | new_temp = make_ssa_name (TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4596, __FUNCTION__))->typed.type)); |
| 4597 | gassign *new_stmt |
| 4598 | = gimple_build_assign (new_temp, code, |
| 4599 | arginfo[i].op, tcst); |
| 4600 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4601 | vargs.safe_push (new_temp); |
| 4602 | } |
| 4603 | break; |
| 4604 | case SIMD_CLONE_ARG_TYPE_LINEAR_VAL_CONSTANT_STEP: |
| 4605 | case SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_CONSTANT_STEP: |
| 4606 | case SIMD_CLONE_ARG_TYPE_LINEAR_VARIABLE_STEP: |
| 4607 | case SIMD_CLONE_ARG_TYPE_LINEAR_REF_VARIABLE_STEP: |
| 4608 | case SIMD_CLONE_ARG_TYPE_LINEAR_VAL_VARIABLE_STEP: |
| 4609 | case SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_VARIABLE_STEP: |
| 4610 | default: |
| 4611 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4611, __FUNCTION__)); |
| 4612 | } |
| 4613 | } |
| 4614 | |
| 4615 | gcall *new_call = gimple_build_call_vec (fndecl, vargs); |
| 4616 | if (vec_dest) |
| 4617 | { |
| 4618 | gcc_assert (ratype((void)(!(ratype || (!maybe_ne (simd_clone_subparts (rtype), nunits ))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4619, __FUNCTION__), 0 : 0)) |
| 4619 | || known_eq (simd_clone_subparts (rtype), nunits))((void)(!(ratype || (!maybe_ne (simd_clone_subparts (rtype), nunits ))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4619, __FUNCTION__), 0 : 0)); |
| 4620 | if (ratype) |
| 4621 | new_temp = create_tmp_var (ratype); |
| 4622 | else if (useless_type_conversion_p (vectype, rtype)) |
| 4623 | new_temp = make_ssa_name (vec_dest, new_call); |
| 4624 | else |
| 4625 | new_temp = make_ssa_name (rtype, new_call); |
| 4626 | gimple_call_set_lhs (new_call, new_temp); |
| 4627 | } |
| 4628 | vect_finish_stmt_generation (vinfo, stmt_info, new_call, gsi); |
| 4629 | gimple *new_stmt = new_call; |
| 4630 | |
| 4631 | if (vec_dest) |
| 4632 | { |
| 4633 | if (!multiple_p (simd_clone_subparts (vectype), nunits)) |
| 4634 | { |
| 4635 | unsigned int k, l; |
| 4636 | poly_uint64 prec = GET_MODE_BITSIZE (TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4636, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode)); |
| 4637 | poly_uint64 bytes = GET_MODE_SIZE (TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4637, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode)); |
| 4638 | k = vector_unroll_factor (nunits,(exact_div (nunits, simd_clone_subparts (vectype)).to_constant ()) |
| 4639 | simd_clone_subparts (vectype))(exact_div (nunits, simd_clone_subparts (vectype)).to_constant ()); |
| 4640 | gcc_assert ((k & (k - 1)) == 0)((void)(!((k & (k - 1)) == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4640, __FUNCTION__), 0 : 0)); |
| 4641 | for (l = 0; l < k; l++) |
| 4642 | { |
| 4643 | tree t; |
| 4644 | if (ratype) |
| 4645 | { |
| 4646 | t = build_fold_addr_expr (new_temp)build_fold_addr_expr_loc (((location_t) 0), (new_temp)); |
| 4647 | t = build2 (MEM_REF, vectype, t, |
| 4648 | build_int_cst (TREE_TYPE (t)((contains_struct_check ((t), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4648, __FUNCTION__))->typed.type), l * bytes)); |
| 4649 | } |
| 4650 | else |
| 4651 | t = build3 (BIT_FIELD_REF, vectype, new_temp, |
| 4652 | bitsize_int (prec)size_int_kind (prec, stk_bitsizetype), bitsize_int (l * prec)size_int_kind (l * prec, stk_bitsizetype)); |
| 4653 | new_stmt = gimple_build_assign (make_ssa_name (vectype), t); |
| 4654 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4655 | |
| 4656 | if (j == 0 && l == 0) |
| 4657 | *vec_stmt = new_stmt; |
| 4658 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 4659 | } |
| 4660 | |
| 4661 | if (ratype) |
| 4662 | vect_clobber_variable (vinfo, stmt_info, gsi, new_temp); |
| 4663 | continue; |
| 4664 | } |
| 4665 | else if (!multiple_p (nunits, simd_clone_subparts (vectype))) |
| 4666 | { |
| 4667 | unsigned int k = (simd_clone_subparts (vectype) |
| 4668 | / simd_clone_subparts (rtype)); |
| 4669 | gcc_assert ((k & (k - 1)) == 0)((void)(!((k & (k - 1)) == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4669, __FUNCTION__), 0 : 0)); |
| 4670 | if ((j & (k - 1)) == 0) |
| 4671 | vec_alloc (ret_ctor_elts, k); |
| 4672 | if (ratype) |
| 4673 | { |
| 4674 | unsigned int m, o; |
| 4675 | o = vector_unroll_factor (nunits,(exact_div (nunits, simd_clone_subparts (rtype)).to_constant ( )) |
| 4676 | simd_clone_subparts (rtype))(exact_div (nunits, simd_clone_subparts (rtype)).to_constant ( )); |
| 4677 | for (m = 0; m < o; m++) |
| 4678 | { |
| 4679 | tree tem = build4 (ARRAY_REF, rtype, new_temp, |
| 4680 | size_int (m)size_int_kind (m, stk_sizetype), NULL_TREE(tree) nullptr, NULL_TREE(tree) nullptr); |
| 4681 | new_stmt = gimple_build_assign (make_ssa_name (rtype), |
| 4682 | tem); |
| 4683 | vect_finish_stmt_generation (vinfo, stmt_info, |
| 4684 | new_stmt, gsi); |
| 4685 | CONSTRUCTOR_APPEND_ELT (ret_ctor_elts, NULL_TREE,do { constructor_elt _ce___ = {(tree) nullptr, gimple_assign_lhs (new_stmt)}; vec_safe_push ((ret_ctor_elts), _ce___); } while (0) |
| 4686 | gimple_assign_lhs (new_stmt))do { constructor_elt _ce___ = {(tree) nullptr, gimple_assign_lhs (new_stmt)}; vec_safe_push ((ret_ctor_elts), _ce___); } while (0); |
| 4687 | } |
| 4688 | vect_clobber_variable (vinfo, stmt_info, gsi, new_temp); |
| 4689 | } |
| 4690 | else |
| 4691 | CONSTRUCTOR_APPEND_ELT (ret_ctor_elts, NULL_TREE, new_temp)do { constructor_elt _ce___ = {(tree) nullptr, new_temp}; vec_safe_push ((ret_ctor_elts), _ce___); } while (0); |
| 4692 | if ((j & (k - 1)) != k - 1) |
| 4693 | continue; |
| 4694 | vec_oprnd0 = build_constructor (vectype, ret_ctor_elts); |
| 4695 | new_stmt |
| 4696 | = gimple_build_assign (make_ssa_name (vec_dest), vec_oprnd0); |
| 4697 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4698 | |
| 4699 | if ((unsigned) j == k - 1) |
| 4700 | *vec_stmt = new_stmt; |
| 4701 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 4702 | continue; |
| 4703 | } |
| 4704 | else if (ratype) |
| 4705 | { |
| 4706 | tree t = build_fold_addr_expr (new_temp)build_fold_addr_expr_loc (((location_t) 0), (new_temp)); |
| 4707 | t = build2 (MEM_REF, vectype, t, |
| 4708 | build_int_cst (TREE_TYPE (t)((contains_struct_check ((t), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4708, __FUNCTION__))->typed.type), 0)); |
| 4709 | new_stmt = gimple_build_assign (make_ssa_name (vec_dest), t); |
| 4710 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4711 | vect_clobber_variable (vinfo, stmt_info, gsi, new_temp); |
| 4712 | } |
| 4713 | else if (!useless_type_conversion_p (vectype, rtype)) |
| 4714 | { |
| 4715 | vec_oprnd0 = build1 (VIEW_CONVERT_EXPR, vectype, new_temp); |
| 4716 | new_stmt |
| 4717 | = gimple_build_assign (make_ssa_name (vec_dest), vec_oprnd0); |
| 4718 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4719 | } |
| 4720 | } |
| 4721 | |
| 4722 | if (j == 0) |
| 4723 | *vec_stmt = new_stmt; |
| 4724 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 4725 | } |
| 4726 | |
| 4727 | for (i = 0; i < nargs; ++i) |
| 4728 | { |
| 4729 | vec<tree> oprndsi = vec_oprnds[i]; |
| 4730 | oprndsi.release (); |
| 4731 | } |
| 4732 | vargs.release (); |
| 4733 | |
| 4734 | /* Mark the clone as no longer being a candidate for GC. */ |
| 4735 | bestn->gc_candidate = false; |
| 4736 | |
| 4737 | /* The call in STMT might prevent it from being removed in dce. |
| 4738 | We however cannot remove it here, due to the way the ssa name |
| 4739 | it defines is mapped to the new definition. So just replace |
| 4740 | rhs of the statement with something harmless. */ |
| 4741 | |
| 4742 | if (slp_node) |
| 4743 | return true; |
| 4744 | |
| 4745 | gimple *new_stmt; |
| 4746 | if (scalar_dest) |
| 4747 | { |
| 4748 | type = TREE_TYPE (scalar_dest)((contains_struct_check ((scalar_dest), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4748, __FUNCTION__))->typed.type); |
| 4749 | lhs = gimple_call_lhs (vect_orig_stmt (stmt_info)->stmt); |
| 4750 | new_stmt = gimple_build_assign (lhs, build_zero_cst (type)); |
| 4751 | } |
| 4752 | else |
| 4753 | new_stmt = gimple_build_nop (); |
| 4754 | vinfo->replace_stmt (gsi, vect_orig_stmt (stmt_info), new_stmt); |
| 4755 | unlink_stmt_vdef (stmt); |
| 4756 | |
| 4757 | return true; |
| 4758 | } |
| 4759 | |
| 4760 | |
| 4761 | /* Function vect_gen_widened_results_half |
| 4762 | |
| 4763 | Create a vector stmt whose code, type, number of arguments, and result |
| 4764 | variable are CODE, OP_TYPE, and VEC_DEST, and its arguments are |
| 4765 | VEC_OPRND0 and VEC_OPRND1. The new vector stmt is to be inserted at GSI. |
| 4766 | In the case that CODE is a CALL_EXPR, this means that a call to DECL |
| 4767 | needs to be created (DECL is a function-decl of a target-builtin). |
| 4768 | STMT_INFO is the original scalar stmt that we are vectorizing. */ |
| 4769 | |
| 4770 | static gimple * |
| 4771 | vect_gen_widened_results_half (vec_info *vinfo, enum tree_code code, |
| 4772 | tree vec_oprnd0, tree vec_oprnd1, int op_type, |
| 4773 | tree vec_dest, gimple_stmt_iterator *gsi, |
| 4774 | stmt_vec_info stmt_info) |
| 4775 | { |
| 4776 | gimple *new_stmt; |
| 4777 | tree new_temp; |
| 4778 | |
| 4779 | /* Generate half of the widened result: */ |
| 4780 | gcc_assert (op_type == TREE_CODE_LENGTH (code))((void)(!(op_type == tree_code_length_tmpl <0>::tree_code_length [(int) (code)]) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4780, __FUNCTION__), 0 : 0)); |
| 4781 | if (op_type != binary_op) |
| 4782 | vec_oprnd1 = NULLnullptr; |
| 4783 | new_stmt = gimple_build_assign (vec_dest, code, vec_oprnd0, vec_oprnd1); |
| 4784 | new_temp = make_ssa_name (vec_dest, new_stmt); |
| 4785 | gimple_assign_set_lhs (new_stmt, new_temp); |
| 4786 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4787 | |
| 4788 | return new_stmt; |
| 4789 | } |
| 4790 | |
| 4791 | |
| 4792 | /* Create vectorized demotion statements for vector operands from VEC_OPRNDS. |
| 4793 | For multi-step conversions store the resulting vectors and call the function |
| 4794 | recursively. */ |
| 4795 | |
| 4796 | static void |
| 4797 | vect_create_vectorized_demotion_stmts (vec_info *vinfo, vec<tree> *vec_oprnds, |
| 4798 | int multi_step_cvt, |
| 4799 | stmt_vec_info stmt_info, |
| 4800 | vec<tree> &vec_dsts, |
| 4801 | gimple_stmt_iterator *gsi, |
| 4802 | slp_tree slp_node, enum tree_code code) |
| 4803 | { |
| 4804 | unsigned int i; |
| 4805 | tree vop0, vop1, new_tmp, vec_dest; |
| 4806 | |
| 4807 | vec_dest = vec_dsts.pop (); |
| 4808 | |
| 4809 | for (i = 0; i < vec_oprnds->length (); i += 2) |
| 4810 | { |
| 4811 | /* Create demotion operation. */ |
| 4812 | vop0 = (*vec_oprnds)[i]; |
| 4813 | vop1 = (*vec_oprnds)[i + 1]; |
| 4814 | gassign *new_stmt = gimple_build_assign (vec_dest, code, vop0, vop1); |
| 4815 | new_tmp = make_ssa_name (vec_dest, new_stmt); |
| 4816 | gimple_assign_set_lhs (new_stmt, new_tmp); |
| 4817 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 4818 | |
| 4819 | if (multi_step_cvt) |
| 4820 | /* Store the resulting vector for next recursive call. */ |
| 4821 | (*vec_oprnds)[i/2] = new_tmp; |
| 4822 | else |
| 4823 | { |
| 4824 | /* This is the last step of the conversion sequence. Store the |
| 4825 | vectors in SLP_NODE or in vector info of the scalar statement |
| 4826 | (or in STMT_VINFO_RELATED_STMT chain). */ |
| 4827 | if (slp_node) |
| 4828 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 4829 | else |
| 4830 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 4831 | } |
| 4832 | } |
| 4833 | |
| 4834 | /* For multi-step demotion operations we first generate demotion operations |
| 4835 | from the source type to the intermediate types, and then combine the |
| 4836 | results (stored in VEC_OPRNDS) in demotion operation to the destination |
| 4837 | type. */ |
| 4838 | if (multi_step_cvt) |
| 4839 | { |
| 4840 | /* At each level of recursion we have half of the operands we had at the |
| 4841 | previous level. */ |
| 4842 | vec_oprnds->truncate ((i+1)/2); |
| 4843 | vect_create_vectorized_demotion_stmts (vinfo, vec_oprnds, |
| 4844 | multi_step_cvt - 1, |
| 4845 | stmt_info, vec_dsts, gsi, |
| 4846 | slp_node, VEC_PACK_TRUNC_EXPR); |
| 4847 | } |
| 4848 | |
| 4849 | vec_dsts.quick_push (vec_dest); |
| 4850 | } |
| 4851 | |
| 4852 | |
| 4853 | /* Create vectorized promotion statements for vector operands from VEC_OPRNDS0 |
| 4854 | and VEC_OPRNDS1, for a binary operation associated with scalar statement |
| 4855 | STMT_INFO. For multi-step conversions store the resulting vectors and |
| 4856 | call the function recursively. */ |
| 4857 | |
| 4858 | static void |
| 4859 | vect_create_vectorized_promotion_stmts (vec_info *vinfo, |
| 4860 | vec<tree> *vec_oprnds0, |
| 4861 | vec<tree> *vec_oprnds1, |
| 4862 | stmt_vec_info stmt_info, tree vec_dest, |
| 4863 | gimple_stmt_iterator *gsi, |
| 4864 | enum tree_code code1, |
| 4865 | enum tree_code code2, int op_type) |
| 4866 | { |
| 4867 | int i; |
| 4868 | tree vop0, vop1, new_tmp1, new_tmp2; |
| 4869 | gimple *new_stmt1, *new_stmt2; |
| 4870 | vec<tree> vec_tmp = vNULL; |
| 4871 | |
| 4872 | vec_tmp.create (vec_oprnds0->length () * 2); |
| 4873 | FOR_EACH_VEC_ELT (*vec_oprnds0, i, vop0)for (i = 0; (*vec_oprnds0).iterate ((i), &(vop0)); ++(i)) |
| 4874 | { |
| 4875 | if (op_type == binary_op) |
| 4876 | vop1 = (*vec_oprnds1)[i]; |
| 4877 | else |
| 4878 | vop1 = NULL_TREE(tree) nullptr; |
| 4879 | |
| 4880 | /* Generate the two halves of promotion operation. */ |
| 4881 | new_stmt1 = vect_gen_widened_results_half (vinfo, code1, vop0, vop1, |
| 4882 | op_type, vec_dest, gsi, |
| 4883 | stmt_info); |
| 4884 | new_stmt2 = vect_gen_widened_results_half (vinfo, code2, vop0, vop1, |
| 4885 | op_type, vec_dest, gsi, |
| 4886 | stmt_info); |
| 4887 | if (is_gimple_call (new_stmt1)) |
| 4888 | { |
| 4889 | new_tmp1 = gimple_call_lhs (new_stmt1); |
| 4890 | new_tmp2 = gimple_call_lhs (new_stmt2); |
| 4891 | } |
| 4892 | else |
| 4893 | { |
| 4894 | new_tmp1 = gimple_assign_lhs (new_stmt1); |
| 4895 | new_tmp2 = gimple_assign_lhs (new_stmt2); |
| 4896 | } |
| 4897 | |
| 4898 | /* Store the results for the next step. */ |
| 4899 | vec_tmp.quick_push (new_tmp1); |
| 4900 | vec_tmp.quick_push (new_tmp2); |
| 4901 | } |
| 4902 | |
| 4903 | vec_oprnds0->release (); |
| 4904 | *vec_oprnds0 = vec_tmp; |
| 4905 | } |
| 4906 | |
| 4907 | /* Create vectorized promotion stmts for widening stmts using only half the |
| 4908 | potential vector size for input. */ |
| 4909 | static void |
| 4910 | vect_create_half_widening_stmts (vec_info *vinfo, |
| 4911 | vec<tree> *vec_oprnds0, |
| 4912 | vec<tree> *vec_oprnds1, |
| 4913 | stmt_vec_info stmt_info, tree vec_dest, |
| 4914 | gimple_stmt_iterator *gsi, |
| 4915 | enum tree_code code1, |
| 4916 | int op_type) |
| 4917 | { |
| 4918 | int i; |
| 4919 | tree vop0, vop1; |
| 4920 | gimple *new_stmt1; |
| 4921 | gimple *new_stmt2; |
| 4922 | gimple *new_stmt3; |
| 4923 | vec<tree> vec_tmp = vNULL; |
| 4924 | |
| 4925 | vec_tmp.create (vec_oprnds0->length ()); |
| 4926 | FOR_EACH_VEC_ELT (*vec_oprnds0, i, vop0)for (i = 0; (*vec_oprnds0).iterate ((i), &(vop0)); ++(i)) |
| 4927 | { |
| 4928 | tree new_tmp1, new_tmp2, new_tmp3, out_type; |
| 4929 | |
| 4930 | gcc_assert (op_type == binary_op)((void)(!(op_type == binary_op) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4930, __FUNCTION__), 0 : 0)); |
| 4931 | vop1 = (*vec_oprnds1)[i]; |
| 4932 | |
| 4933 | /* Widen the first vector input. */ |
| 4934 | out_type = TREE_TYPE (vec_dest)((contains_struct_check ((vec_dest), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4934, __FUNCTION__))->typed.type); |
| 4935 | new_tmp1 = make_ssa_name (out_type); |
| 4936 | new_stmt1 = gimple_build_assign (new_tmp1, NOP_EXPR, vop0); |
| 4937 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt1, gsi); |
| 4938 | if (VECTOR_TYPE_P (TREE_TYPE (vop1))(((enum tree_code) (((contains_struct_check ((vop1), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 4938, __FUNCTION__))->typed.type))->base.code) == VECTOR_TYPE )) |
| 4939 | { |
| 4940 | /* Widen the second vector input. */ |
| 4941 | new_tmp2 = make_ssa_name (out_type); |
| 4942 | new_stmt2 = gimple_build_assign (new_tmp2, NOP_EXPR, vop1); |
| 4943 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt2, gsi); |
| 4944 | /* Perform the operation. With both vector inputs widened. */ |
| 4945 | new_stmt3 = gimple_build_assign (vec_dest, code1, new_tmp1, new_tmp2); |
| 4946 | } |
| 4947 | else |
| 4948 | { |
| 4949 | /* Perform the operation. With the single vector input widened. */ |
| 4950 | new_stmt3 = gimple_build_assign (vec_dest, code1, new_tmp1, vop1); |
| 4951 | } |
| 4952 | |
| 4953 | new_tmp3 = make_ssa_name (vec_dest, new_stmt3); |
| 4954 | gimple_assign_set_lhs (new_stmt3, new_tmp3); |
| 4955 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt3, gsi); |
| 4956 | |
| 4957 | /* Store the results for the next step. */ |
| 4958 | vec_tmp.quick_push (new_tmp3); |
| 4959 | } |
| 4960 | |
| 4961 | vec_oprnds0->release (); |
| 4962 | *vec_oprnds0 = vec_tmp; |
| 4963 | } |
| 4964 | |
| 4965 | |
| 4966 | /* Check if STMT_INFO performs a conversion operation that can be vectorized. |
| 4967 | If VEC_STMT is also passed, vectorize STMT_INFO: create a vectorized |
| 4968 | stmt to replace it, put it in VEC_STMT, and insert it at GSI. |
| 4969 | Return true if STMT_INFO is vectorizable in this way. */ |
| 4970 | |
| 4971 | static bool |
| 4972 | vectorizable_conversion (vec_info *vinfo, |
| 4973 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 4974 | gimple **vec_stmt, slp_tree slp_node, |
| 4975 | stmt_vector_for_cost *cost_vec) |
| 4976 | { |
| 4977 | tree vec_dest; |
| 4978 | tree scalar_dest; |
| 4979 | tree op0, op1 = NULL_TREE(tree) nullptr; |
| 4980 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 4981 | enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK; |
| 4982 | enum tree_code codecvt1 = ERROR_MARK, codecvt2 = ERROR_MARK; |
| 4983 | tree new_temp; |
| 4984 | enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; |
| 4985 | int ndts = 2; |
| 4986 | poly_uint64 nunits_in; |
| 4987 | poly_uint64 nunits_out; |
| 4988 | tree vectype_out, vectype_in; |
| 4989 | int ncopies, i; |
| 4990 | tree lhs_type, rhs_type; |
| 4991 | enum { NARROW, NONE, WIDEN } modifier; |
| 4992 | vec<tree> vec_oprnds0 = vNULL; |
| 4993 | vec<tree> vec_oprnds1 = vNULL; |
| 4994 | tree vop0; |
| 4995 | bb_vec_info bb_vinfo = dyn_cast <bb_vec_info> (vinfo); |
| 4996 | int multi_step_cvt = 0; |
| 4997 | vec<tree> interm_types = vNULL; |
| 4998 | tree intermediate_type, cvt_type = NULL_TREE(tree) nullptr; |
| 4999 | int op_type; |
| 5000 | unsigned short fltsz; |
| 5001 | |
| 5002 | /* Is STMT a vectorizable conversion? */ |
| 5003 | |
| 5004 | if (!STMT_VINFO_RELEVANT_P (stmt_info)((stmt_info)->relevant != vect_unused_in_scope) && !bb_vinfo) |
| 5005 | return false; |
| 5006 | |
| 5007 | if (STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_internal_def |
| 5008 | && ! vec_stmt) |
| 5009 | return false; |
| 5010 | |
| 5011 | gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt); |
| 5012 | if (!stmt) |
| 5013 | return false; |
| 5014 | |
| 5015 | if (TREE_CODE (gimple_assign_lhs (stmt))((enum tree_code) (gimple_assign_lhs (stmt))->base.code) != SSA_NAME) |
| 5016 | return false; |
| 5017 | |
| 5018 | code = gimple_assign_rhs_code (stmt); |
| 5019 | if (!CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR) |
| 5020 | && code != FIX_TRUNC_EXPR |
| 5021 | && code != FLOAT_EXPR |
| 5022 | && code != WIDEN_PLUS_EXPR |
| 5023 | && code != WIDEN_MINUS_EXPR |
| 5024 | && code != WIDEN_MULT_EXPR |
| 5025 | && code != WIDEN_LSHIFT_EXPR) |
| 5026 | return false; |
| 5027 | |
| 5028 | bool widen_arith = (code == WIDEN_PLUS_EXPR |
| 5029 | || code == WIDEN_MINUS_EXPR |
| 5030 | || code == WIDEN_MULT_EXPR |
| 5031 | || code == WIDEN_LSHIFT_EXPR); |
| 5032 | op_type = TREE_CODE_LENGTH (code)tree_code_length_tmpl <0>::tree_code_length[(int) (code )]; |
| 5033 | |
| 5034 | /* Check types of lhs and rhs. */ |
| 5035 | scalar_dest = gimple_assign_lhs (stmt); |
| 5036 | lhs_type = TREE_TYPE (scalar_dest)((contains_struct_check ((scalar_dest), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5036, __FUNCTION__))->typed.type); |
| 5037 | vectype_out = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 5038 | |
| 5039 | /* Check the operands of the operation. */ |
| 5040 | slp_tree slp_op0, slp_op1 = NULLnullptr; |
| 5041 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, |
| 5042 | 0, &op0, &slp_op0, &dt[0], &vectype_in)) |
| 5043 | { |
| 5044 | if (dump_enabled_p ()) |
| 5045 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5046 | "use not simple.\n"); |
| 5047 | return false; |
| 5048 | } |
| 5049 | |
| 5050 | rhs_type = TREE_TYPE (op0)((contains_struct_check ((op0), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5050, __FUNCTION__))->typed.type); |
| 5051 | if ((code != FIX_TRUNC_EXPR && code != FLOAT_EXPR) |
| 5052 | && !((INTEGRAL_TYPE_P (lhs_type)(((enum tree_code) (lhs_type)->base.code) == ENUMERAL_TYPE || ((enum tree_code) (lhs_type)->base.code) == BOOLEAN_TYPE || ((enum tree_code) (lhs_type)->base.code) == INTEGER_TYPE ) |
| 5053 | && INTEGRAL_TYPE_P (rhs_type)(((enum tree_code) (rhs_type)->base.code) == ENUMERAL_TYPE || ((enum tree_code) (rhs_type)->base.code) == BOOLEAN_TYPE || ((enum tree_code) (rhs_type)->base.code) == INTEGER_TYPE )) |
| 5054 | || (SCALAR_FLOAT_TYPE_P (lhs_type)(((enum tree_code) (lhs_type)->base.code) == REAL_TYPE) |
| 5055 | && SCALAR_FLOAT_TYPE_P (rhs_type)(((enum tree_code) (rhs_type)->base.code) == REAL_TYPE)))) |
| 5056 | return false; |
| 5057 | |
| 5058 | if (!VECTOR_BOOLEAN_TYPE_P (vectype_out)(((enum tree_code) (vectype_out)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_out ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5058, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE ) |
| 5059 | && ((INTEGRAL_TYPE_P (lhs_type)(((enum tree_code) (lhs_type)->base.code) == ENUMERAL_TYPE || ((enum tree_code) (lhs_type)->base.code) == BOOLEAN_TYPE || ((enum tree_code) (lhs_type)->base.code) == INTEGER_TYPE ) |
| 5060 | && !type_has_mode_precision_p (lhs_type)) |
| 5061 | || (INTEGRAL_TYPE_P (rhs_type)(((enum tree_code) (rhs_type)->base.code) == ENUMERAL_TYPE || ((enum tree_code) (rhs_type)->base.code) == BOOLEAN_TYPE || ((enum tree_code) (rhs_type)->base.code) == INTEGER_TYPE ) |
| 5062 | && !type_has_mode_precision_p (rhs_type)))) |
| 5063 | { |
| 5064 | if (dump_enabled_p ()) |
| 5065 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5066 | "type conversion to/from bit-precision unsupported." |
| 5067 | "\n"); |
| 5068 | return false; |
| 5069 | } |
| 5070 | |
| 5071 | if (op_type == binary_op) |
| 5072 | { |
| 5073 | gcc_assert (code == WIDEN_MULT_EXPR || code == WIDEN_LSHIFT_EXPR((void)(!(code == WIDEN_MULT_EXPR || code == WIDEN_LSHIFT_EXPR || code == WIDEN_PLUS_EXPR || code == WIDEN_MINUS_EXPR) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5074, __FUNCTION__), 0 : 0)) |
| 5074 | || code == WIDEN_PLUS_EXPR || code == WIDEN_MINUS_EXPR)((void)(!(code == WIDEN_MULT_EXPR || code == WIDEN_LSHIFT_EXPR || code == WIDEN_PLUS_EXPR || code == WIDEN_MINUS_EXPR) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5074, __FUNCTION__), 0 : 0)); |
| 5075 | |
| 5076 | op1 = gimple_assign_rhs2 (stmt); |
| 5077 | tree vectype1_in; |
| 5078 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, 1, |
| 5079 | &op1, &slp_op1, &dt[1], &vectype1_in)) |
| 5080 | { |
| 5081 | if (dump_enabled_p ()) |
| 5082 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5083 | "use not simple.\n"); |
| 5084 | return false; |
| 5085 | } |
| 5086 | /* For WIDEN_MULT_EXPR, if OP0 is a constant, use the type of |
| 5087 | OP1. */ |
| 5088 | if (!vectype_in) |
| 5089 | vectype_in = vectype1_in; |
| 5090 | } |
| 5091 | |
| 5092 | /* If op0 is an external or constant def, infer the vector type |
| 5093 | from the scalar type. */ |
| 5094 | if (!vectype_in) |
| 5095 | vectype_in = get_vectype_for_scalar_type (vinfo, rhs_type, slp_node); |
| 5096 | if (vec_stmt) |
| 5097 | gcc_assert (vectype_in)((void)(!(vectype_in) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5097, __FUNCTION__), 0 : 0)); |
| 5098 | if (!vectype_in) |
| 5099 | { |
| 5100 | if (dump_enabled_p ()) |
| 5101 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5102 | "no vectype for scalar type %T\n", rhs_type); |
| 5103 | |
| 5104 | return false; |
| 5105 | } |
| 5106 | |
| 5107 | if (VECTOR_BOOLEAN_TYPE_P (vectype_out)(((enum tree_code) (vectype_out)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_out ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5107, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE ) |
| 5108 | && !VECTOR_BOOLEAN_TYPE_P (vectype_in)(((enum tree_code) (vectype_in)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_in ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5108, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 5109 | { |
| 5110 | if (dump_enabled_p ()) |
| 5111 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5112 | "can't convert between boolean and non " |
| 5113 | "boolean vectors %T\n", rhs_type); |
| 5114 | |
| 5115 | return false; |
| 5116 | } |
| 5117 | |
| 5118 | nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); |
| 5119 | nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); |
| 5120 | if (known_eq (nunits_out, nunits_in)(!maybe_ne (nunits_out, nunits_in))) |
| 5121 | if (widen_arith) |
| 5122 | modifier = WIDEN; |
| 5123 | else |
| 5124 | modifier = NONE; |
| 5125 | else if (multiple_p (nunits_out, nunits_in)) |
| 5126 | modifier = NARROW; |
| 5127 | else |
| 5128 | { |
| 5129 | gcc_checking_assert (multiple_p (nunits_in, nunits_out))((void)(!(multiple_p (nunits_in, nunits_out)) ? fancy_abort ( "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5129, __FUNCTION__), 0 : 0)); |
| 5130 | modifier = WIDEN; |
| 5131 | } |
| 5132 | |
| 5133 | /* Multiple types in SLP are handled by creating the appropriate number of |
| 5134 | vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in |
| 5135 | case of SLP. */ |
| 5136 | if (slp_node) |
| 5137 | ncopies = 1; |
| 5138 | else if (modifier == NARROW) |
| 5139 | ncopies = vect_get_num_copies (loop_vinfo, vectype_out); |
| 5140 | else |
| 5141 | ncopies = vect_get_num_copies (loop_vinfo, vectype_in); |
| 5142 | |
| 5143 | /* Sanity check: make sure that at least one copy of the vectorized stmt |
| 5144 | needs to be generated. */ |
| 5145 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5145, __FUNCTION__), 0 : 0)); |
| 5146 | |
| 5147 | bool found_mode = false; |
| 5148 | scalar_mode lhs_mode = SCALAR_TYPE_MODE (lhs_type)(as_a <scalar_mode> ((tree_class_check ((lhs_type), (tcc_type ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5148, __FUNCTION__))->type_common.mode)); |
| 5149 | scalar_mode rhs_mode = SCALAR_TYPE_MODE (rhs_type)(as_a <scalar_mode> ((tree_class_check ((rhs_type), (tcc_type ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5149, __FUNCTION__))->type_common.mode)); |
| 5150 | opt_scalar_mode rhs_mode_iter; |
| 5151 | |
| 5152 | /* Supportable by target? */ |
| 5153 | switch (modifier) |
| 5154 | { |
| 5155 | case NONE: |
| 5156 | if (code != FIX_TRUNC_EXPR |
| 5157 | && code != FLOAT_EXPR |
| 5158 | && !CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR)) |
| 5159 | return false; |
| 5160 | if (supportable_convert_operation (code, vectype_out, vectype_in, &code1)) |
| 5161 | break; |
| 5162 | /* FALLTHRU */ |
| 5163 | unsupported: |
| 5164 | if (dump_enabled_p ()) |
| 5165 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5166 | "conversion not supported by target.\n"); |
| 5167 | return false; |
| 5168 | |
| 5169 | case WIDEN: |
| 5170 | if (known_eq (nunits_in, nunits_out)(!maybe_ne (nunits_in, nunits_out))) |
| 5171 | { |
| 5172 | if (!supportable_half_widening_operation (code, vectype_out, |
| 5173 | vectype_in, &code1)) |
| 5174 | goto unsupported; |
| 5175 | gcc_assert (!(multi_step_cvt && op_type == binary_op))((void)(!(!(multi_step_cvt && op_type == binary_op)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5175, __FUNCTION__), 0 : 0)); |
| 5176 | break; |
| 5177 | } |
| 5178 | if (supportable_widening_operation (vinfo, code, stmt_info, |
| 5179 | vectype_out, vectype_in, &code1, |
| 5180 | &code2, &multi_step_cvt, |
| 5181 | &interm_types)) |
| 5182 | { |
| 5183 | /* Binary widening operation can only be supported directly by the |
| 5184 | architecture. */ |
| 5185 | gcc_assert (!(multi_step_cvt && op_type == binary_op))((void)(!(!(multi_step_cvt && op_type == binary_op)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5185, __FUNCTION__), 0 : 0)); |
| 5186 | break; |
| 5187 | } |
| 5188 | |
| 5189 | if (code != FLOAT_EXPR |
| 5190 | || GET_MODE_SIZE (lhs_mode) <= GET_MODE_SIZE (rhs_mode)) |
| 5191 | goto unsupported; |
| 5192 | |
| 5193 | fltsz = GET_MODE_SIZE (lhs_mode); |
| 5194 | FOR_EACH_2XWIDER_MODE (rhs_mode_iter, rhs_mode)for ((rhs_mode_iter) = (rhs_mode), mode_iterator::get_2xwider (&(rhs_mode_iter)); mode_iterator::iterate_p (&(rhs_mode_iter )); mode_iterator::get_2xwider (&(rhs_mode_iter))) |
| 5195 | { |
| 5196 | rhs_mode = rhs_mode_iter.require (); |
| 5197 | if (GET_MODE_SIZE (rhs_mode) > fltsz) |
| 5198 | break; |
| 5199 | |
| 5200 | cvt_type |
| 5201 | = build_nonstandard_integer_type (GET_MODE_BITSIZE (rhs_mode), 0); |
| 5202 | cvt_type = get_same_sized_vectype (cvt_type, vectype_in); |
| 5203 | if (cvt_type == NULL_TREE(tree) nullptr) |
| 5204 | goto unsupported; |
| 5205 | |
| 5206 | if (GET_MODE_SIZE (rhs_mode) == fltsz) |
| 5207 | { |
| 5208 | if (!supportable_convert_operation (code, vectype_out, |
| 5209 | cvt_type, &codecvt1)) |
| 5210 | goto unsupported; |
| 5211 | } |
| 5212 | else if (!supportable_widening_operation (vinfo, code, stmt_info, |
| 5213 | vectype_out, cvt_type, |
| 5214 | &codecvt1, &codecvt2, |
| 5215 | &multi_step_cvt, |
| 5216 | &interm_types)) |
| 5217 | continue; |
| 5218 | else |
| 5219 | gcc_assert (multi_step_cvt == 0)((void)(!(multi_step_cvt == 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5219, __FUNCTION__), 0 : 0)); |
| 5220 | |
| 5221 | if (supportable_widening_operation (vinfo, NOP_EXPR, stmt_info, |
| 5222 | cvt_type, |
| 5223 | vectype_in, &code1, &code2, |
| 5224 | &multi_step_cvt, &interm_types)) |
| 5225 | { |
| 5226 | found_mode = true; |
| 5227 | break; |
| 5228 | } |
| 5229 | } |
| 5230 | |
| 5231 | if (!found_mode) |
| 5232 | goto unsupported; |
| 5233 | |
| 5234 | if (GET_MODE_SIZE (rhs_mode) == fltsz) |
| 5235 | codecvt2 = ERROR_MARK; |
| 5236 | else |
| 5237 | { |
| 5238 | multi_step_cvt++; |
| 5239 | interm_types.safe_push (cvt_type); |
| 5240 | cvt_type = NULL_TREE(tree) nullptr; |
| 5241 | } |
| 5242 | break; |
| 5243 | |
| 5244 | case NARROW: |
| 5245 | gcc_assert (op_type == unary_op)((void)(!(op_type == unary_op) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5245, __FUNCTION__), 0 : 0)); |
| 5246 | if (supportable_narrowing_operation (code, vectype_out, vectype_in, |
| 5247 | &code1, &multi_step_cvt, |
| 5248 | &interm_types)) |
| 5249 | break; |
| 5250 | |
| 5251 | if (code != FIX_TRUNC_EXPR |
| 5252 | || GET_MODE_SIZE (lhs_mode) >= GET_MODE_SIZE (rhs_mode)) |
| 5253 | goto unsupported; |
| 5254 | |
| 5255 | cvt_type |
| 5256 | = build_nonstandard_integer_type (GET_MODE_BITSIZE (rhs_mode), 0); |
| 5257 | cvt_type = get_same_sized_vectype (cvt_type, vectype_in); |
| 5258 | if (cvt_type == NULL_TREE(tree) nullptr) |
| 5259 | goto unsupported; |
| 5260 | if (!supportable_convert_operation (code, cvt_type, vectype_in, |
| 5261 | &codecvt1)) |
| 5262 | goto unsupported; |
| 5263 | if (supportable_narrowing_operation (NOP_EXPR, vectype_out, cvt_type, |
| 5264 | &code1, &multi_step_cvt, |
| 5265 | &interm_types)) |
| 5266 | break; |
| 5267 | goto unsupported; |
| 5268 | |
| 5269 | default: |
| 5270 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5270, __FUNCTION__)); |
| 5271 | } |
| 5272 | |
| 5273 | if (!vec_stmt) /* transformation not required. */ |
| 5274 | { |
| 5275 | if (slp_node |
| 5276 | && (!vect_maybe_update_slp_op_vectype (slp_op0, vectype_in) |
| 5277 | || !vect_maybe_update_slp_op_vectype (slp_op1, vectype_in))) |
| 5278 | { |
| 5279 | if (dump_enabled_p ()) |
| 5280 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5281 | "incompatible vector types for invariants\n"); |
| 5282 | return false; |
| 5283 | } |
| 5284 | DUMP_VECT_SCOPE ("vectorizable_conversion")auto_dump_scope scope ("vectorizable_conversion", vect_location ); |
| 5285 | if (modifier == NONE) |
| 5286 | { |
| 5287 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = type_conversion_vec_info_type; |
| 5288 | vect_model_simple_cost (vinfo, stmt_info, ncopies, dt, ndts, slp_node, |
| 5289 | cost_vec); |
| 5290 | } |
| 5291 | else if (modifier == NARROW) |
| 5292 | { |
| 5293 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = type_demotion_vec_info_type; |
| 5294 | /* The final packing step produces one vector result per copy. */ |
| 5295 | unsigned int nvectors |
| 5296 | = (slp_node ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size : ncopies); |
| 5297 | vect_model_promotion_demotion_cost (stmt_info, dt, nvectors, |
| 5298 | multi_step_cvt, cost_vec, |
| 5299 | widen_arith); |
| 5300 | } |
| 5301 | else |
| 5302 | { |
| 5303 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = type_promotion_vec_info_type; |
| 5304 | /* The initial unpacking step produces two vector results |
| 5305 | per copy. MULTI_STEP_CVT is 0 for a single conversion, |
| 5306 | so >> MULTI_STEP_CVT divides by 2^(number of steps - 1). */ |
| 5307 | unsigned int nvectors |
| 5308 | = (slp_node |
| 5309 | ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node)(slp_node)->vec_stmts_size >> multi_step_cvt |
| 5310 | : ncopies * 2); |
| 5311 | vect_model_promotion_demotion_cost (stmt_info, dt, nvectors, |
| 5312 | multi_step_cvt, cost_vec, |
| 5313 | widen_arith); |
| 5314 | } |
| 5315 | interm_types.release (); |
| 5316 | return true; |
| 5317 | } |
| 5318 | |
| 5319 | /* Transform. */ |
| 5320 | if (dump_enabled_p ()) |
| 5321 | dump_printf_loc (MSG_NOTE, vect_location, |
| 5322 | "transform conversion. ncopies = %d.\n", ncopies); |
| 5323 | |
| 5324 | if (op_type == binary_op) |
| 5325 | { |
| 5326 | if (CONSTANT_CLASS_P (op0)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (op0)->base.code))] == tcc_constant)) |
| 5327 | op0 = fold_convert (TREE_TYPE (op1), op0)fold_convert_loc (((location_t) 0), ((contains_struct_check ( (op1), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5327, __FUNCTION__))->typed.type), op0); |
| 5328 | else if (CONSTANT_CLASS_P (op1)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (op1)->base.code))] == tcc_constant)) |
| 5329 | op1 = fold_convert (TREE_TYPE (op0), op1)fold_convert_loc (((location_t) 0), ((contains_struct_check ( (op0), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5329, __FUNCTION__))->typed.type), op1); |
| 5330 | } |
| 5331 | |
| 5332 | /* In case of multi-step conversion, we first generate conversion operations |
| 5333 | to the intermediate types, and then from that types to the final one. |
| 5334 | We create vector destinations for the intermediate type (TYPES) received |
| 5335 | from supportable_*_operation, and store them in the correct order |
| 5336 | for future use in vect_create_vectorized_*_stmts (). */ |
| 5337 | auto_vec<tree> vec_dsts (multi_step_cvt + 1); |
| 5338 | vec_dest = vect_create_destination_var (scalar_dest, |
| 5339 | (cvt_type && modifier == WIDEN) |
| 5340 | ? cvt_type : vectype_out); |
| 5341 | vec_dsts.quick_push (vec_dest); |
| 5342 | |
| 5343 | if (multi_step_cvt) |
| 5344 | { |
| 5345 | for (i = interm_types.length () - 1; |
| 5346 | interm_types.iterate (i, &intermediate_type); i--) |
| 5347 | { |
| 5348 | vec_dest = vect_create_destination_var (scalar_dest, |
| 5349 | intermediate_type); |
| 5350 | vec_dsts.quick_push (vec_dest); |
| 5351 | } |
| 5352 | } |
| 5353 | |
| 5354 | if (cvt_type) |
| 5355 | vec_dest = vect_create_destination_var (scalar_dest, |
| 5356 | modifier == WIDEN |
| 5357 | ? vectype_out : cvt_type); |
| 5358 | |
| 5359 | int ninputs = 1; |
| 5360 | if (!slp_node) |
| 5361 | { |
| 5362 | if (modifier == WIDEN) |
| 5363 | ; |
| 5364 | else if (modifier == NARROW) |
| 5365 | { |
| 5366 | if (multi_step_cvt) |
| 5367 | ninputs = vect_pow2 (multi_step_cvt); |
| 5368 | ninputs *= 2; |
| 5369 | } |
| 5370 | } |
| 5371 | |
| 5372 | switch (modifier) |
| 5373 | { |
| 5374 | case NONE: |
| 5375 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies, |
| 5376 | op0, &vec_oprnds0); |
| 5377 | FOR_EACH_VEC_ELT (vec_oprnds0, i, vop0)for (i = 0; (vec_oprnds0).iterate ((i), &(vop0)); ++(i)) |
| 5378 | { |
| 5379 | /* Arguments are ready, create the new vector stmt. */ |
| 5380 | gcc_assert (TREE_CODE_LENGTH (code1) == unary_op)((void)(!(tree_code_length_tmpl <0>::tree_code_length[( int) (code1)] == unary_op) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5380, __FUNCTION__), 0 : 0)); |
| 5381 | gassign *new_stmt = gimple_build_assign (vec_dest, code1, vop0); |
| 5382 | new_temp = make_ssa_name (vec_dest, new_stmt); |
| 5383 | gimple_assign_set_lhs (new_stmt, new_temp); |
| 5384 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 5385 | |
| 5386 | if (slp_node) |
| 5387 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 5388 | else |
| 5389 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 5390 | } |
| 5391 | break; |
| 5392 | |
| 5393 | case WIDEN: |
| 5394 | /* In case the vectorization factor (VF) is bigger than the number |
| 5395 | of elements that we can fit in a vectype (nunits), we have to |
| 5396 | generate more than one vector stmt - i.e - we need to "unroll" |
| 5397 | the vector stmt by a factor VF/nunits. */ |
| 5398 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies * ninputs, |
| 5399 | op0, &vec_oprnds0, |
| 5400 | code == WIDEN_LSHIFT_EXPR ? NULL_TREE(tree) nullptr : op1, |
| 5401 | &vec_oprnds1); |
| 5402 | if (code == WIDEN_LSHIFT_EXPR) |
| 5403 | { |
| 5404 | int oprnds_size = vec_oprnds0.length (); |
| 5405 | vec_oprnds1.create (oprnds_size); |
| 5406 | for (i = 0; i < oprnds_size; ++i) |
| 5407 | vec_oprnds1.quick_push (op1); |
| 5408 | } |
| 5409 | /* Arguments are ready. Create the new vector stmts. */ |
| 5410 | for (i = multi_step_cvt; i >= 0; i--) |
| 5411 | { |
| 5412 | tree this_dest = vec_dsts[i]; |
| 5413 | enum tree_code c1 = code1, c2 = code2; |
| 5414 | if (i == 0 && codecvt2 != ERROR_MARK) |
| 5415 | { |
| 5416 | c1 = codecvt1; |
| 5417 | c2 = codecvt2; |
| 5418 | } |
| 5419 | if (known_eq (nunits_out, nunits_in)(!maybe_ne (nunits_out, nunits_in))) |
| 5420 | vect_create_half_widening_stmts (vinfo, &vec_oprnds0, |
| 5421 | &vec_oprnds1, stmt_info, |
| 5422 | this_dest, gsi, |
| 5423 | c1, op_type); |
| 5424 | else |
| 5425 | vect_create_vectorized_promotion_stmts (vinfo, &vec_oprnds0, |
| 5426 | &vec_oprnds1, stmt_info, |
| 5427 | this_dest, gsi, |
| 5428 | c1, c2, op_type); |
| 5429 | } |
| 5430 | |
| 5431 | FOR_EACH_VEC_ELT (vec_oprnds0, i, vop0)for (i = 0; (vec_oprnds0).iterate ((i), &(vop0)); ++(i)) |
| 5432 | { |
| 5433 | gimple *new_stmt; |
| 5434 | if (cvt_type) |
| 5435 | { |
| 5436 | gcc_assert (TREE_CODE_LENGTH (codecvt1) == unary_op)((void)(!(tree_code_length_tmpl <0>::tree_code_length[( int) (codecvt1)] == unary_op) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5436, __FUNCTION__), 0 : 0)); |
| 5437 | new_temp = make_ssa_name (vec_dest); |
| 5438 | new_stmt = gimple_build_assign (new_temp, codecvt1, vop0); |
| 5439 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 5440 | } |
| 5441 | else |
| 5442 | new_stmt = SSA_NAME_DEF_STMT (vop0)(tree_check ((vop0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5442, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 5443 | |
| 5444 | if (slp_node) |
| 5445 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 5446 | else |
| 5447 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 5448 | } |
| 5449 | break; |
| 5450 | |
| 5451 | case NARROW: |
| 5452 | /* In case the vectorization factor (VF) is bigger than the number |
| 5453 | of elements that we can fit in a vectype (nunits), we have to |
| 5454 | generate more than one vector stmt - i.e - we need to "unroll" |
| 5455 | the vector stmt by a factor VF/nunits. */ |
| 5456 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies * ninputs, |
| 5457 | op0, &vec_oprnds0); |
| 5458 | /* Arguments are ready. Create the new vector stmts. */ |
| 5459 | if (cvt_type) |
| 5460 | FOR_EACH_VEC_ELT (vec_oprnds0, i, vop0)for (i = 0; (vec_oprnds0).iterate ((i), &(vop0)); ++(i)) |
| 5461 | { |
| 5462 | gcc_assert (TREE_CODE_LENGTH (codecvt1) == unary_op)((void)(!(tree_code_length_tmpl <0>::tree_code_length[( int) (codecvt1)] == unary_op) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5462, __FUNCTION__), 0 : 0)); |
| 5463 | new_temp = make_ssa_name (vec_dest); |
| 5464 | gassign *new_stmt |
| 5465 | = gimple_build_assign (new_temp, codecvt1, vop0); |
| 5466 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 5467 | vec_oprnds0[i] = new_temp; |
| 5468 | } |
| 5469 | |
| 5470 | vect_create_vectorized_demotion_stmts (vinfo, &vec_oprnds0, |
| 5471 | multi_step_cvt, |
| 5472 | stmt_info, vec_dsts, gsi, |
| 5473 | slp_node, code1); |
| 5474 | break; |
| 5475 | } |
| 5476 | if (!slp_node) |
| 5477 | *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts[0]; |
| 5478 | |
| 5479 | vec_oprnds0.release (); |
| 5480 | vec_oprnds1.release (); |
| 5481 | interm_types.release (); |
| 5482 | |
| 5483 | return true; |
| 5484 | } |
| 5485 | |
| 5486 | /* Return true if we can assume from the scalar form of STMT_INFO that |
| 5487 | neither the scalar nor the vector forms will generate code. STMT_INFO |
| 5488 | is known not to involve a data reference. */ |
| 5489 | |
| 5490 | bool |
| 5491 | vect_nop_conversion_p (stmt_vec_info stmt_info) |
| 5492 | { |
| 5493 | gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt); |
| 5494 | if (!stmt) |
| 5495 | return false; |
| 5496 | |
| 5497 | tree lhs = gimple_assign_lhs (stmt); |
| 5498 | tree_code code = gimple_assign_rhs_code (stmt); |
| 5499 | tree rhs = gimple_assign_rhs1 (stmt); |
| 5500 | |
| 5501 | if (code == SSA_NAME || code == VIEW_CONVERT_EXPR) |
| 5502 | return true; |
| 5503 | |
| 5504 | if (CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR)) |
| 5505 | return tree_nop_conversion_p (TREE_TYPE (lhs)((contains_struct_check ((lhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5505, __FUNCTION__))->typed.type), TREE_TYPE (rhs)((contains_struct_check ((rhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5505, __FUNCTION__))->typed.type)); |
| 5506 | |
| 5507 | return false; |
| 5508 | } |
| 5509 | |
| 5510 | /* Function vectorizable_assignment. |
| 5511 | |
| 5512 | Check if STMT_INFO performs an assignment (copy) that can be vectorized. |
| 5513 | If VEC_STMT is also passed, vectorize the STMT_INFO: create a vectorized |
| 5514 | stmt to replace it, put it in VEC_STMT, and insert it at GSI. |
| 5515 | Return true if STMT_INFO is vectorizable in this way. */ |
| 5516 | |
| 5517 | static bool |
| 5518 | vectorizable_assignment (vec_info *vinfo, |
| 5519 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 5520 | gimple **vec_stmt, slp_tree slp_node, |
| 5521 | stmt_vector_for_cost *cost_vec) |
| 5522 | { |
| 5523 | tree vec_dest; |
| 5524 | tree scalar_dest; |
| 5525 | tree op; |
| 5526 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 5527 | tree new_temp; |
| 5528 | enum vect_def_type dt[1] = {vect_unknown_def_type}; |
| 5529 | int ndts = 1; |
| 5530 | int ncopies; |
| 5531 | int i; |
| 5532 | vec<tree> vec_oprnds = vNULL; |
| 5533 | tree vop; |
| 5534 | bb_vec_info bb_vinfo = dyn_cast <bb_vec_info> (vinfo); |
| 5535 | enum tree_code code; |
| 5536 | tree vectype_in; |
| 5537 | |
| 5538 | if (!STMT_VINFO_RELEVANT_P (stmt_info)((stmt_info)->relevant != vect_unused_in_scope) && !bb_vinfo) |
| 5539 | return false; |
| 5540 | |
| 5541 | if (STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_internal_def |
| 5542 | && ! vec_stmt) |
| 5543 | return false; |
| 5544 | |
| 5545 | /* Is vectorizable assignment? */ |
| 5546 | gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt); |
| 5547 | if (!stmt) |
| 5548 | return false; |
| 5549 | |
| 5550 | scalar_dest = gimple_assign_lhs (stmt); |
| 5551 | if (TREE_CODE (scalar_dest)((enum tree_code) (scalar_dest)->base.code) != SSA_NAME) |
| 5552 | return false; |
| 5553 | |
| 5554 | if (STMT_VINFO_DATA_REF (stmt_info)((stmt_info)->dr_aux.dr + 0)) |
| 5555 | return false; |
| 5556 | |
| 5557 | code = gimple_assign_rhs_code (stmt); |
| 5558 | if (!(gimple_assign_single_p (stmt) |
| 5559 | || code == PAREN_EXPR |
| 5560 | || CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR))) |
| 5561 | return false; |
| 5562 | |
| 5563 | tree vectype = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 5564 | poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype); |
| 5565 | |
| 5566 | /* Multiple types in SLP are handled by creating the appropriate number of |
| 5567 | vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in |
| 5568 | case of SLP. */ |
| 5569 | if (slp_node) |
| 5570 | ncopies = 1; |
| 5571 | else |
| 5572 | ncopies = vect_get_num_copies (loop_vinfo, vectype); |
| 5573 | |
| 5574 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5574, __FUNCTION__), 0 : 0)); |
| 5575 | |
| 5576 | slp_tree slp_op; |
| 5577 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, 0, &op, &slp_op, |
| 5578 | &dt[0], &vectype_in)) |
| 5579 | { |
| 5580 | if (dump_enabled_p ()) |
| 5581 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5582 | "use not simple.\n"); |
| 5583 | return false; |
| 5584 | } |
| 5585 | if (!vectype_in) |
| 5586 | vectype_in = get_vectype_for_scalar_type (vinfo, TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5586, __FUNCTION__))->typed.type), slp_node); |
| 5587 | |
| 5588 | /* We can handle NOP_EXPR conversions that do not change the number |
| 5589 | of elements or the vector size. */ |
| 5590 | if ((CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR) |
| 5591 | || code == VIEW_CONVERT_EXPR) |
| 5592 | && (!vectype_in |
| 5593 | || maybe_ne (TYPE_VECTOR_SUBPARTS (vectype_in), nunits) |
| 5594 | || maybe_ne (GET_MODE_SIZE (TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5594, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode)), |
| 5595 | GET_MODE_SIZE (TYPE_MODE (vectype_in)((((enum tree_code) ((tree_class_check ((vectype_in), (tcc_type ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5595, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype_in) : (vectype_in)->type_common.mode))))) |
| 5596 | return false; |
| 5597 | |
| 5598 | if (VECTOR_BOOLEAN_TYPE_P (vectype)(((enum tree_code) (vectype)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5598, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE ) |
| 5599 | && !VECTOR_BOOLEAN_TYPE_P (vectype_in)(((enum tree_code) (vectype_in)->base.code) == VECTOR_TYPE && ((enum tree_code) (((contains_struct_check ((vectype_in ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5599, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE )) |
| 5600 | { |
| 5601 | if (dump_enabled_p ()) |
| 5602 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5603 | "can't convert between boolean and non " |
| 5604 | "boolean vectors %T\n", TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5604, __FUNCTION__))->typed.type)); |
| 5605 | |
| 5606 | return false; |
| 5607 | } |
| 5608 | |
| 5609 | /* We do not handle bit-precision changes. */ |
| 5610 | if ((CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR) |
| 5611 | || code == VIEW_CONVERT_EXPR) |
| 5612 | && INTEGRAL_TYPE_P (TREE_TYPE (scalar_dest))(((enum tree_code) (((contains_struct_check ((scalar_dest), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5612, __FUNCTION__))->typed.type))->base.code) == ENUMERAL_TYPE || ((enum tree_code) (((contains_struct_check ((scalar_dest) , (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5612, __FUNCTION__))->typed.type))->base.code) == BOOLEAN_TYPE || ((enum tree_code) (((contains_struct_check ((scalar_dest) , (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5612, __FUNCTION__))->typed.type))->base.code) == INTEGER_TYPE ) |
| 5613 | && (!type_has_mode_precision_p (TREE_TYPE (scalar_dest)((contains_struct_check ((scalar_dest), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5613, __FUNCTION__))->typed.type)) |
| 5614 | || !type_has_mode_precision_p (TREE_TYPE (op)((contains_struct_check ((op), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5614, __FUNCTION__))->typed.type))) |
| 5615 | /* But a conversion that does not change the bit-pattern is ok. */ |
| 5616 | && !((TYPE_PRECISION (TREE_TYPE (scalar_dest))((tree_class_check ((((contains_struct_check ((scalar_dest), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5616, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5616, __FUNCTION__))->type_common.precision) |
| 5617 | > TYPE_PRECISION (TREE_TYPE (op))((tree_class_check ((((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5617, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5617, __FUNCTION__))->type_common.precision)) |
| 5618 | && TYPE_UNSIGNED (TREE_TYPE (op))((tree_class_check ((((contains_struct_check ((op), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5618, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5618, __FUNCTION__))->base.u.bits.unsigned_flag))) |
| 5619 | { |
| 5620 | if (dump_enabled_p ()) |
| 5621 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5622 | "type conversion to/from bit-precision " |
| 5623 | "unsupported.\n"); |
| 5624 | return false; |
| 5625 | } |
| 5626 | |
| 5627 | if (!vec_stmt) /* transformation not required. */ |
| 5628 | { |
| 5629 | if (slp_node |
| 5630 | && !vect_maybe_update_slp_op_vectype (slp_op, vectype_in)) |
| 5631 | { |
| 5632 | if (dump_enabled_p ()) |
| 5633 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5634 | "incompatible vector types for invariants\n"); |
| 5635 | return false; |
| 5636 | } |
| 5637 | STMT_VINFO_TYPE (stmt_info)(stmt_info)->type = assignment_vec_info_type; |
| 5638 | DUMP_VECT_SCOPE ("vectorizable_assignment")auto_dump_scope scope ("vectorizable_assignment", vect_location ); |
| 5639 | if (!vect_nop_conversion_p (stmt_info)) |
| 5640 | vect_model_simple_cost (vinfo, stmt_info, ncopies, dt, ndts, slp_node, |
| 5641 | cost_vec); |
| 5642 | return true; |
| 5643 | } |
| 5644 | |
| 5645 | /* Transform. */ |
| 5646 | if (dump_enabled_p ()) |
| 5647 | dump_printf_loc (MSG_NOTE, vect_location, "transform assignment.\n"); |
| 5648 | |
| 5649 | /* Handle def. */ |
| 5650 | vec_dest = vect_create_destination_var (scalar_dest, vectype); |
| 5651 | |
| 5652 | /* Handle use. */ |
| 5653 | vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies, op, &vec_oprnds); |
| 5654 | |
| 5655 | /* Arguments are ready. create the new vector stmt. */ |
| 5656 | FOR_EACH_VEC_ELT (vec_oprnds, i, vop)for (i = 0; (vec_oprnds).iterate ((i), &(vop)); ++(i)) |
| 5657 | { |
| 5658 | if (CONVERT_EXPR_CODE_P (code)((code) == NOP_EXPR || (code) == CONVERT_EXPR) |
| 5659 | || code == VIEW_CONVERT_EXPR) |
| 5660 | vop = build1 (VIEW_CONVERT_EXPR, vectype, vop); |
| 5661 | gassign *new_stmt = gimple_build_assign (vec_dest, vop); |
| 5662 | new_temp = make_ssa_name (vec_dest, new_stmt); |
| 5663 | gimple_assign_set_lhs (new_stmt, new_temp); |
| 5664 | vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi); |
| 5665 | if (slp_node) |
| 5666 | SLP_TREE_VEC_STMTS (slp_node)(slp_node)->vec_stmts.quick_push (new_stmt); |
| 5667 | else |
| 5668 | STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts.safe_push (new_stmt); |
| 5669 | } |
| 5670 | if (!slp_node) |
| 5671 | *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)(stmt_info)->vec_stmts[0]; |
| 5672 | |
| 5673 | vec_oprnds.release (); |
| 5674 | return true; |
| 5675 | } |
| 5676 | |
| 5677 | |
| 5678 | /* Return TRUE if CODE (a shift operation) is supported for SCALAR_TYPE |
| 5679 | either as shift by a scalar or by a vector. */ |
| 5680 | |
| 5681 | bool |
| 5682 | vect_supportable_shift (vec_info *vinfo, enum tree_code code, tree scalar_type) |
| 5683 | { |
| 5684 | |
| 5685 | machine_mode vec_mode; |
| 5686 | optab optab; |
| 5687 | int icode; |
| 5688 | tree vectype; |
| 5689 | |
| 5690 | vectype = get_vectype_for_scalar_type (vinfo, scalar_type); |
| 5691 | if (!vectype) |
| 5692 | return false; |
| 5693 | |
| 5694 | optab = optab_for_tree_code (code, vectype, optab_scalar); |
| 5695 | if (!optab |
| 5696 | || optab_handler (optab, TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5696, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode)) == CODE_FOR_nothing) |
| 5697 | { |
| 5698 | optab = optab_for_tree_code (code, vectype, optab_vector); |
| 5699 | if (!optab |
| 5700 | || (optab_handler (optab, TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5700, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode)) |
| 5701 | == CODE_FOR_nothing)) |
| 5702 | return false; |
| 5703 | } |
| 5704 | |
| 5705 | vec_mode = TYPE_MODE (vectype)((((enum tree_code) ((tree_class_check ((vectype), (tcc_type) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5705, __FUNCTION__)))->base.code) == VECTOR_TYPE) ? vector_type_mode (vectype) : (vectype)->type_common.mode); |
| 5706 | icode = (int) optab_handler (optab, vec_mode); |
| 5707 | if (icode == CODE_FOR_nothing) |
| 5708 | return false; |
| 5709 | |
| 5710 | return true; |
| 5711 | } |
| 5712 | |
| 5713 | |
| 5714 | /* Function vectorizable_shift. |
| 5715 | |
| 5716 | Check if STMT_INFO performs a shift operation that can be vectorized. |
| 5717 | If VEC_STMT is also passed, vectorize the STMT_INFO: create a vectorized |
| 5718 | stmt to replace it, put it in VEC_STMT, and insert it at GSI. |
| 5719 | Return true if STMT_INFO is vectorizable in this way. */ |
| 5720 | |
| 5721 | static bool |
| 5722 | vectorizable_shift (vec_info *vinfo, |
| 5723 | stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, |
| 5724 | gimple **vec_stmt, slp_tree slp_node, |
| 5725 | stmt_vector_for_cost *cost_vec) |
| 5726 | { |
| 5727 | tree vec_dest; |
| 5728 | tree scalar_dest; |
| 5729 | tree op0, op1 = NULLnullptr; |
| 5730 | tree vec_oprnd1 = NULL_TREE(tree) nullptr; |
| 5731 | tree vectype; |
| 5732 | loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo); |
| 5733 | enum tree_code code; |
| 5734 | machine_mode vec_mode; |
| 5735 | tree new_temp; |
| 5736 | optab optab; |
| 5737 | int icode; |
| 5738 | machine_mode optab_op2_mode; |
| 5739 | enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; |
| 5740 | int ndts = 2; |
| 5741 | poly_uint64 nunits_in; |
| 5742 | poly_uint64 nunits_out; |
| 5743 | tree vectype_out; |
| 5744 | tree op1_vectype; |
| 5745 | int ncopies; |
| 5746 | int i; |
| 5747 | vec<tree> vec_oprnds0 = vNULL; |
| 5748 | vec<tree> vec_oprnds1 = vNULL; |
| 5749 | tree vop0, vop1; |
| 5750 | unsigned int k; |
| 5751 | bool scalar_shift_arg = true; |
| 5752 | bb_vec_info bb_vinfo = dyn_cast <bb_vec_info> (vinfo); |
| 5753 | bool incompatible_op1_vectype_p = false; |
| 5754 | |
| 5755 | if (!STMT_VINFO_RELEVANT_P (stmt_info)((stmt_info)->relevant != vect_unused_in_scope) && !bb_vinfo) |
| 5756 | return false; |
| 5757 | |
| 5758 | if (STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_internal_def |
| 5759 | && STMT_VINFO_DEF_TYPE (stmt_info)(stmt_info)->def_type != vect_nested_cycle |
| 5760 | && ! vec_stmt) |
| 5761 | return false; |
| 5762 | |
| 5763 | /* Is STMT a vectorizable binary/unary operation? */ |
| 5764 | gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt); |
| 5765 | if (!stmt) |
| 5766 | return false; |
| 5767 | |
| 5768 | if (TREE_CODE (gimple_assign_lhs (stmt))((enum tree_code) (gimple_assign_lhs (stmt))->base.code) != SSA_NAME) |
| 5769 | return false; |
| 5770 | |
| 5771 | code = gimple_assign_rhs_code (stmt); |
| 5772 | |
| 5773 | if (!(code == LSHIFT_EXPR || code == RSHIFT_EXPR || code == LROTATE_EXPR |
| 5774 | || code == RROTATE_EXPR)) |
| 5775 | return false; |
| 5776 | |
| 5777 | scalar_dest = gimple_assign_lhs (stmt); |
| 5778 | vectype_out = STMT_VINFO_VECTYPE (stmt_info)(stmt_info)->vectype; |
| 5779 | if (!type_has_mode_precision_p (TREE_TYPE (scalar_dest)((contains_struct_check ((scalar_dest), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5779, __FUNCTION__))->typed.type))) |
| 5780 | { |
| 5781 | if (dump_enabled_p ()) |
| 5782 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5783 | "bit-precision shifts not supported.\n"); |
| 5784 | return false; |
| 5785 | } |
| 5786 | |
| 5787 | slp_tree slp_op0; |
| 5788 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, |
| 5789 | 0, &op0, &slp_op0, &dt[0], &vectype)) |
| 5790 | { |
| 5791 | if (dump_enabled_p ()) |
| 5792 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5793 | "use not simple.\n"); |
| 5794 | return false; |
| 5795 | } |
| 5796 | /* If op0 is an external or constant def, infer the vector type |
| 5797 | from the scalar type. */ |
| 5798 | if (!vectype) |
| 5799 | vectype = get_vectype_for_scalar_type (vinfo, TREE_TYPE (op0)((contains_struct_check ((op0), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5799, __FUNCTION__))->typed.type), slp_node); |
| 5800 | if (vec_stmt) |
| 5801 | gcc_assert (vectype)((void)(!(vectype) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5801, __FUNCTION__), 0 : 0)); |
| 5802 | if (!vectype) |
| 5803 | { |
| 5804 | if (dump_enabled_p ()) |
| 5805 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5806 | "no vectype for scalar type\n"); |
| 5807 | return false; |
| 5808 | } |
| 5809 | |
| 5810 | nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); |
| 5811 | nunits_in = TYPE_VECTOR_SUBPARTS (vectype); |
| 5812 | if (maybe_ne (nunits_out, nunits_in)) |
| 5813 | return false; |
| 5814 | |
| 5815 | stmt_vec_info op1_def_stmt_info; |
| 5816 | slp_tree slp_op1; |
| 5817 | if (!vect_is_simple_use (vinfo, stmt_info, slp_node, 1, &op1, &slp_op1, |
| 5818 | &dt[1], &op1_vectype, &op1_def_stmt_info)) |
| 5819 | { |
| 5820 | if (dump_enabled_p ()) |
| 5821 | dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, |
| 5822 | "use not simple.\n"); |
| 5823 | return false; |
| 5824 | } |
| 5825 | |
| 5826 | /* Multiple types in SLP are handled by creating the appropriate number of |
| 5827 | vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in |
| 5828 | case of SLP. */ |
| 5829 | if (slp_node) |
| 5830 | ncopies = 1; |
| 5831 | else |
| 5832 | ncopies = vect_get_num_copies (loop_vinfo, vectype); |
| 5833 | |
| 5834 | gcc_assert (ncopies >= 1)((void)(!(ncopies >= 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/tree-vect-stmts.cc" , 5834, __FUNCTION__), 0 : 0)); |
| 5835 | |
| 5836 | /* Determine whether the shift amount is a vector, or scalar. If the |
| 5837 | shift/rotate amount is a vector, use the vector/vector shift optabs. */ |
| 5838 | |
| 5839 | if ((dt[1] == vect_internal_def |
| 5840 | || dt[1] == vect_induction_def |
| 5841 | || dt[1] == vect_nested_cycle) |
| 5842 | && !slp_node) |
| 5843 | scalar_shift_arg = false; |
| 5844 | else if (dt[1] == vect_constant_def |
| 5845 | || dt[1] == vect_external_def |
| 5846 | || dt[1] == vect_internal_def) |
| 5847 | { |
| 5848 | /* In SLP, need to check whether the shift count is the same, |
| 5849 | in loops if it is a constant or invariant, it is always |
| 5850 | a scalar shift. */ |
| 5851 | if (slp_node) |
| 5852 | { |