1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4 (vectorizable_mask_load_store): Delete.
5 (vectorizable_call): Return false for masked loads and stores.
6 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
7 instead of gimple_assign_rhs1.
8 (vectorizable_load): Handle IFN_MASK_LOAD.
9 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
11 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
13 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
15 (vectorizable_mask_load_store): ...here.
16 (vectorizable_load): ...and here.
18 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
20 * tree-vect-stmts.c (vect_build_all_ones_mask)
21 (vect_build_zero_merge_argument): New functions, split out from...
22 (vectorizable_load): ...here.
24 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
26 * tree-vect-stmts.c (vect_check_store_rhs): New function,
28 (vectorizable_mask_load_store): ...here.
29 (vectorizable_store): ...and here.
31 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
33 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
35 (vectorizable_mask_load_store): ...here.
37 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
39 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
40 (vect_model_store_cost): Take a vec_load_store_type instead of a
42 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
43 (vect_model_store_cost): Take a vec_load_store_type instead of a
45 (vectorizable_mask_load_store): Update accordingly.
46 (vectorizable_store): Likewise.
47 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
49 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
51 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
52 IFN_MASK_LOAD calls here rather than...
53 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
55 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
56 Alan Hayward <alan.hayward@arm.com>
57 David Sherwood <david.sherwood@arm.com>
59 * expmed.c (extract_bit_field_1): For vector extracts,
60 fall back to extract_bit_field_as_subreg if vec_extract
63 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
64 Alan Hayward <alan.hayward@arm.com>
65 David Sherwood <david.sherwood@arm.com>
67 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
68 they are variable or constant sized.
69 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
70 slots for constant-sized data.
72 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
73 Alan Hayward <alan.hayward@arm.com>
74 David Sherwood <david.sherwood@arm.com>
76 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
77 handling COND_EXPRs with boolean comparisons, try to find a better
78 basis for the mask type than the boolean itself.
80 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
82 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
83 is calculated and how it can be overridden.
84 * genmodes.c (max_bitsize_mode_any_mode): New variable.
85 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
87 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
90 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
91 Alan Hayward <alan.hayward@arm.com>
92 David Sherwood <david.sherwood@arm.com>
94 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
95 Remove the mode argument.
96 (aarch64_simd_valid_immediate): Remove the mode and inverse
98 * config/aarch64/iterators.md (bitsize): New iterator.
99 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
100 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
101 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
102 aarch64_simd_valid_immediate.
103 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
104 (aarch64_reg_or_bic_imm): Likewise.
105 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
106 with an insn_type enum and msl with a modifier_type enum.
107 Replace element_width with a scalar_mode. Change the shift
108 to unsigned int. Add constructors for scalar_float_mode and
109 scalar_int_mode elements.
110 (aarch64_vect_float_const_representable_p): Delete.
111 (aarch64_can_const_movi_rtx_p)
112 (aarch64_simd_scalar_immediate_valid_for_move)
113 (aarch64_simd_make_constant): Update call to
114 aarch64_simd_valid_immediate.
115 (aarch64_advsimd_valid_immediate_hs): New function.
116 (aarch64_advsimd_valid_immediate): Likewise.
117 (aarch64_simd_valid_immediate): Remove mode and inverse
118 arguments. Rewrite to use the above. Use const_vec_duplicate_p
119 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
120 and aarch64_float_const_representable_p on the result.
121 (aarch64_output_simd_mov_immediate): Remove mode argument.
122 Update call to aarch64_simd_valid_immediate and use of
124 (aarch64_output_scalar_simd_mov_immediate): Update call
127 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
128 Alan Hayward <alan.hayward@arm.com>
129 David Sherwood <david.sherwood@arm.com>
131 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
132 (mode_nunits): Likewise CONST_MODE_NUNITS.
133 * machmode.def (ADJUST_NUNITS): Document.
134 * genmodes.c (mode_data::need_nunits_adj): New field.
135 (blank_mode): Update accordingly.
136 (adj_nunits): New variable.
137 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
139 (emit_mode_size_inline): Set need_bytesize_adj for all modes
140 listed in adj_nunits.
141 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
142 listed in adj_nunits. Don't emit case statements for such modes.
143 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
144 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
145 nothing if adj_nunits is nonnull.
146 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
147 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
148 (emit_mode_fbit): Update use of print_maybe_const_decl.
149 (emit_move_size): Likewise. Treat the array as non-const
151 (emit_mode_adjustments): Handle adj_nunits.
153 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
155 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
156 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
157 (VECTOR_MODES): Use it.
158 (make_vector_modes): Take the prefix as an argument.
160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
161 Alan Hayward <alan.hayward@arm.com>
162 David Sherwood <david.sherwood@arm.com>
164 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
165 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
166 for MODE_VECTOR_BOOL.
167 * machmode.def (VECTOR_BOOL_MODE): Document.
168 * genmodes.c (VECTOR_BOOL_MODE): New macro.
169 (make_vector_bool_mode): New function.
170 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
172 * lto-streamer-in.c (lto_input_mode_table): Likewise.
173 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
175 * stor-layout.c (int_mode_for_mode): Likewise.
176 * tree.c (build_vector_type_for_mode): Likewise.
177 * varasm.c (output_constant_pool_2): Likewise.
178 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
179 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
180 for MODE_VECTOR_BOOL.
181 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
182 of mode class checks.
183 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
184 instead of a list of mode class checks.
185 (expand_vector_scalar_condition): Likewise.
186 (type_for_widest_vector_mode): Handle BImode as an inner mode.
188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
189 Alan Hayward <alan.hayward@arm.com>
190 David Sherwood <david.sherwood@arm.com>
192 * machmode.h (mode_size): Change from unsigned short to
194 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
195 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
196 or if measurement_type is not polynomial.
197 (fixed_size_mode::includes_p): Check for constant-sized modes.
198 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
199 return a poly_uint16 rather than an unsigned short.
200 (emit_mode_size): Change the type of mode_size from unsigned short
201 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
202 (emit_mode_adjustments): Cope with polynomial vector sizes.
203 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
205 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
207 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
208 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
209 * caller-save.c (setup_save_areas): Likewise.
210 (replace_reg_with_saved_mem): Likewise.
211 * calls.c (emit_library_call_value_1): Likewise.
212 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
213 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
214 (gen_lowpart_for_combine): Likewise.
215 * convert.c (convert_to_integer_1): Likewise.
216 * cse.c (equiv_constant, cse_insn): Likewise.
217 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
218 (cselib_subst_to_values): Likewise.
219 * dce.c (word_dce_process_block): Likewise.
220 * df-problems.c (df_word_lr_mark_ref): Likewise.
221 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
222 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
223 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
224 (rtl_for_decl_location): Likewise.
225 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
226 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
227 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
228 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
229 (expand_expr_real_1): Likewise.
230 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
231 (pad_below): Likewise.
232 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
233 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
234 * ira.c (get_subreg_tracking_sizes): Likewise.
235 * ira-build.c (ira_create_allocno_objects): Likewise.
236 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
237 (ira_sort_regnos_for_alter_reg): Likewise.
238 * ira-costs.c (record_operand_costs): Likewise.
239 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
240 (resolve_simple_move): Likewise.
241 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
242 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
243 (lra_constraints): Likewise.
244 (CONST_POOL_OK_P): Reject variable-sized modes.
245 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
246 (add_pseudo_to_slot, lra_spill): Likewise.
247 * omp-low.c (omp_clause_aligned_alignment): Likewise.
248 * optabs-query.c (get_best_extraction_insn): Likewise.
249 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
250 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
251 (expand_mult_highpart, valid_multiword_target_p): Likewise.
252 * recog.c (offsettable_address_addr_space_p): Likewise.
253 * regcprop.c (maybe_mode_change): Likewise.
254 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
255 * regrename.c (build_def_use): Likewise.
256 * regstat.c (dump_reg_info): Likewise.
257 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
258 (find_reloads, find_reloads_subreg_address): Likewise.
259 * reload1.c (eliminate_regs_1): Likewise.
260 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
261 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
262 (simplify_binary_operation_1, simplify_subreg): Likewise.
263 * targhooks.c (default_function_arg_padding): Likewise.
264 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
265 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
266 (verify_gimple_assign_ternary): Likewise.
267 * tree-inline.c (estimate_move_cost): Likewise.
268 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
269 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
270 (get_address_cost_ainc): Likewise.
271 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
272 (vect_supportable_dr_alignment): Likewise.
273 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
274 (vectorizable_reduction): Likewise.
275 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
276 (vectorizable_operation, vectorizable_load): Likewise.
277 * tree.c (build_same_sized_truth_vector_type): Likewise.
278 * valtrack.c (cleanup_auto_inc_dec): Likewise.
279 * var-tracking.c (emit_note_insn_var_location): Likewise.
280 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
281 (ADDR_VEC_ALIGN): Likewise.
283 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
284 Alan Hayward <alan.hayward@arm.com>
285 David Sherwood <david.sherwood@arm.com>
287 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
289 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
290 or if measurement_type is polynomial.
291 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
292 * combine.c (make_extraction): Likewise.
293 * dse.c (find_shift_sequence): Likewise.
294 * dwarf2out.c (mem_loc_descriptor): Likewise.
295 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
296 (extract_bit_field, extract_low_bits): Likewise.
297 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
298 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
299 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
300 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
301 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
302 * reload.c (find_reloads): Likewise.
303 * reload1.c (alter_reg): Likewise.
304 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
305 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
306 * tree-if-conv.c (predicate_mem_writes): Likewise.
307 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
308 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
309 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
310 * valtrack.c (dead_debug_insert_temp): Likewise.
311 * varasm.c (mergeable_constant_section): Likewise.
312 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
314 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
315 Alan Hayward <alan.hayward@arm.com>
316 David Sherwood <david.sherwood@arm.com>
318 * expr.c (expand_assignment): Cope with polynomial mode sizes
319 when assigning to a CONCAT.
321 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
322 Alan Hayward <alan.hayward@arm.com>
323 David Sherwood <david.sherwood@arm.com>
325 * machmode.h (mode_precision): Change from unsigned short to
327 (mode_to_precision): Return a poly_uint16 rather than an unsigned
329 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
330 or if measurement_type is not polynomial.
331 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
332 in which the mode is already known to be a scalar_int_mode.
333 * genmodes.c (emit_mode_precision): Change the type of mode_precision
334 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
336 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
337 for GET_MODE_PRECISION.
338 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
339 for GET_MODE_PRECISION.
340 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
342 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
343 (expand_field_assignment, make_extraction): Likewise.
344 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
345 (get_last_value): Likewise.
346 * convert.c (convert_to_integer_1): Likewise.
347 * cse.c (cse_insn): Likewise.
348 * expr.c (expand_expr_real_1): Likewise.
349 * lra-constraints.c (simplify_operand_subreg): Likewise.
350 * optabs-query.c (can_atomic_load_p): Likewise.
351 * optabs.c (expand_atomic_load): Likewise.
352 (expand_atomic_store): Likewise.
353 * ree.c (combine_reaching_defs): Likewise.
354 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
355 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
356 * tree.h (type_has_mode_precision_p): Likewise.
357 * ubsan.c (instrument_si_overflow): Likewise.
359 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
360 Alan Hayward <alan.hayward@arm.com>
361 David Sherwood <david.sherwood@arm.com>
363 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
364 polynomial numbers of units.
365 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
366 (valid_vector_subparts_p): New function.
367 (build_vector_type): Remove temporary shim and take the number
368 of units as a poly_uint64 rather than an int.
369 (build_opaque_vector_type): Take the number of units as a
370 poly_uint64 rather than an int.
371 * tree.c (build_vector_from_ctor): Handle polynomial
372 TYPE_VECTOR_SUBPARTS.
373 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
374 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
375 (build_vector_from_val): If the number of units is variable,
376 use build_vec_duplicate_cst for constant operands and
377 VEC_DUPLICATE_EXPR otherwise.
378 (make_vector_type): Remove temporary is_constant ().
379 (build_vector_type, build_opaque_vector_type): Take the number of
380 units as a poly_uint64 rather than an int.
381 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
383 * cfgexpand.c (expand_debug_expr): Likewise.
384 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
385 (store_constructor, expand_expr_real_1): Likewise.
386 (const_scalar_mask_from_tree): Likewise.
387 * fold-const-call.c (fold_const_reduction): Likewise.
388 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
389 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
390 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
391 (fold_relational_const): Likewise.
392 (native_interpret_vector): Likewise. Change the size from an
393 int to an unsigned int.
394 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
395 TYPE_VECTOR_SUBPARTS.
396 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
397 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
398 duplicating a non-constant operand into a variable-length vector.
399 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
400 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
401 * ipa-icf.c (sem_variable::equals): Likewise.
402 * match.pd: Likewise.
403 * omp-simd-clone.c (simd_clone_subparts): Likewise.
404 * print-tree.c (print_node): Likewise.
405 * stor-layout.c (layout_type): Likewise.
406 * targhooks.c (default_builtin_vectorization_cost): Likewise.
407 * tree-cfg.c (verify_gimple_comparison): Likewise.
408 (verify_gimple_assign_binary): Likewise.
409 (verify_gimple_assign_ternary): Likewise.
410 (verify_gimple_assign_single): Likewise.
411 * tree-pretty-print.c (dump_generic_node): Likewise.
412 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
413 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
414 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
415 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
416 (vect_shift_permute_load_chain): Likewise.
417 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
418 (expand_vector_condition, optimize_vector_constructor): Likewise.
419 (lower_vec_perm, get_compute_type): Likewise.
420 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
421 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
422 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
423 (vect_recog_mask_conversion_pattern): Likewise.
424 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
425 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
426 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
427 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
428 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
429 (vectorizable_shift, vectorizable_operation, vectorizable_store)
430 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
431 (supportable_widening_operation): Likewise.
432 (supportable_narrowing_operation): Likewise.
433 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
435 * varasm.c (output_constant): Likewise.
437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
438 Alan Hayward <alan.hayward@arm.com>
439 David Sherwood <david.sherwood@arm.com>
441 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
442 so that both the length == 3 and length != 3 cases set up their
443 own permute vectors. Add comments explaining why we know the
444 number of elements is constant.
445 (vect_permute_load_chain): Likewise.
447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
448 Alan Hayward <alan.hayward@arm.com>
449 David Sherwood <david.sherwood@arm.com>
451 * machmode.h (mode_nunits): Change from unsigned char to
453 (ONLY_FIXED_SIZE_MODES): New macro.
454 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
455 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
456 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
458 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
459 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
460 or if measurement_type is not polynomial.
461 * genmodes.c (ZERO_COEFFS): New macro.
462 (emit_mode_nunits_inline): Make mode_nunits_inline return a
464 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
465 Use ZERO_COEFFS when emitting initializers.
466 * data-streamer.h (bp_pack_poly_value): New function.
467 (bp_unpack_poly_value): Likewise.
468 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
470 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
472 * tree.c (make_vector_type): Remove temporary shim and make
473 the real function take the number of units as a poly_uint64
475 (build_vector_type_for_mode): Handle polynomial nunits.
476 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
477 * emit-rtl.c (const_vec_series_p_1): Likewise.
478 (gen_rtx_CONST_VECTOR): Likewise.
479 * fold-const.c (test_vec_duplicate_folding): Likewise.
480 * genrecog.c (validate_pattern): Likewise.
481 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
482 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
483 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
484 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
485 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
486 * rtlanal.c (subreg_get_info): Likewise.
487 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
488 (vect_grouped_load_supported): Likewise.
489 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
490 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
491 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
492 (simplify_const_unary_operation, simplify_binary_operation_1)
493 (simplify_const_binary_operation, simplify_ternary_operation)
494 (test_vector_ops_duplicate, test_vector_ops): Likewise.
495 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
496 instead of CONST_VECTOR_NUNITS.
497 * varasm.c (output_constant_pool_2): Likewise.
498 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
499 explicit-encoded elements in the XVEC for variable-length vectors.
501 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
503 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
505 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
506 Alan Hayward <alan.hayward@arm.com>
507 David Sherwood <david.sherwood@arm.com>
509 * coretypes.h (fixed_size_mode): Declare.
510 (fixed_size_mode_pod): New typedef.
511 * builtins.h (target_builtins::x_apply_args_mode)
512 (target_builtins::x_apply_result_mode): Change type to
514 * builtins.c (apply_args_size, apply_result_size, result_vector)
515 (expand_builtin_apply_args_1, expand_builtin_apply)
516 (expand_builtin_return): Update accordingly.
518 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
520 * cse.c (hash_rtx_cb): Hash only the encoded elements.
521 * cselib.c (cselib_hash_rtx): Likewise.
522 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
523 CONST_VECTOR encoding.
525 2017-01-03 Jakub Jelinek <jakub@redhat.com>
526 Jeff Law <law@redhat.com>
529 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
530 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
531 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
532 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
535 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
536 explicitly probe *sp in a noreturn function if there were any callee
537 register saves or frame pointer is needed.
539 2018-01-03 Jakub Jelinek <jakub@redhat.com>
542 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
543 BLKmode for ternary, binary or unary expressions.
546 * var-tracking.c (delete_vta_debug_insn): New inline function.
547 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
548 insns from get_insns () to NULL instead of each bb separately.
549 Use delete_vta_debug_insn. No longer static.
550 (vt_debug_insns_local, variable_tracking_main_1): Adjust
551 delete_vta_debug_insns callers.
552 * rtl.h (delete_vta_debug_insns): Declare.
553 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
554 instead of variable_tracking_main.
556 2018-01-03 Martin Sebor <msebor@redhat.com>
558 PR tree-optimization/83603
559 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
560 arguments past the endof the argument list in functions declared
562 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
563 Avoid checking when arguments are null.
565 2018-01-03 Martin Sebor <msebor@redhat.com>
568 * doc/extend.texi (attribute const): Fix a typo.
569 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
570 issuing -Wsuggest-attribute for void functions.
572 2018-01-03 Martin Sebor <msebor@redhat.com>
574 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
575 offset_int::from instead of wide_int::to_shwi.
576 (maybe_diag_overlap): Remove assertion.
577 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
578 * gimple-ssa-sprintf.c (format_directive): Same.
579 (parse_directive): Same.
580 (sprintf_dom_walker::compute_format_length): Same.
581 (try_substitute_return_value): Same.
583 2017-01-03 Jeff Law <law@redhat.com>
586 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
587 non-constant residual for zero at runtime and avoid probing in
588 that case. Reorganize code for trailing problem to mirror handling
591 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
593 PR tree-optimization/83501
594 * tree-ssa-strlen.c (get_string_cst): New.
595 (handle_char_store): Call get_string_cst.
597 2018-01-03 Martin Liska <mliska@suse.cz>
599 PR tree-optimization/83593
600 * tree-ssa-strlen.c: Include tree-cfg.h.
601 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
602 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
603 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
605 (strlen_dom_walker::before_dom_children): Call
606 gimple_purge_dead_eh_edges. Dump tranformation with details
608 (strlen_dom_walker::before_dom_children): Update call by adding
609 new argument cleanup_eh.
610 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
612 2018-01-03 Martin Liska <mliska@suse.cz>
615 * cif-code.def (VARIADIC_THUNK): New enum value.
616 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
619 2018-01-03 Jan Beulich <jbeulich@suse.com>
621 * sse.md (mov<mode>_internal): Tighten condition for when to use
622 vmovdqu<ssescalarsize> for TI and OI modes.
624 2018-01-03 Jakub Jelinek <jakub@redhat.com>
626 Update copyright years.
628 2018-01-03 Martin Liska <mliska@suse.cz>
631 * ipa-visibility.c (function_and_variable_visibility): Skip
632 functions with noipa attribure.
634 2018-01-03 Jakub Jelinek <jakub@redhat.com>
636 * gcc.c (process_command): Update copyright notice dates.
637 * gcov-dump.c (print_version): Ditto.
638 * gcov.c (print_version): Ditto.
639 * gcov-tool.c (print_version): Ditto.
640 * gengtype.c (create_file): Ditto.
641 * doc/cpp.texi: Bump @copying's copyright year.
642 * doc/cppinternals.texi: Ditto.
643 * doc/gcc.texi: Ditto.
644 * doc/gccint.texi: Ditto.
645 * doc/gcov.texi: Ditto.
646 * doc/install.texi: Ditto.
647 * doc/invoke.texi: Ditto.
649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
651 * vector-builder.h (vector_builder::m_full_nelts): Change from
652 unsigned int to poly_uint64.
653 (vector_builder::full_nelts): Update prototype accordingly.
654 (vector_builder::new_vector): Likewise.
655 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
656 (vector_builder::operator ==): Likewise.
657 (vector_builder::finalize): Likewise.
658 * int-vector-builder.h (int_vector_builder::int_vector_builder):
659 Take the number of elements as a poly_uint64 rather than an
661 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
662 from unsigned int to poly_uint64.
663 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
664 (vec_perm_indices::new_vector): Likewise.
665 (vec_perm_indices::length): Likewise.
666 (vec_perm_indices::nelts_per_input): Likewise.
667 (vec_perm_indices::input_nelts): Likewise.
668 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
669 number of elements per input as a poly_uint64 rather than an
670 unsigned int. Use the original encoding for variable-length
671 vectors, rather than clamping each individual element.
672 For the second and subsequent elements in each pattern,
673 clamp the step and base before clamping their sum.
674 (vec_perm_indices::series_p): Handle polynomial element counts.
675 (vec_perm_indices::all_in_range_p): Likewise.
676 (vec_perm_indices_to_tree): Likewise.
677 (vec_perm_indices_to_rtx): Likewise.
678 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
679 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
680 (tree_vector_builder::new_binary_operation): Handle polynomial
681 element counts. Return false if we need to know the number
682 of elements at compile time.
683 * fold-const.c (fold_vec_perm): Punt if the number of elements
684 isn't known at compile time.
686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
688 * vec-perm-indices.h (vec_perm_builder): Change element type
689 from HOST_WIDE_INT to poly_int64.
690 (vec_perm_indices::element_type): Update accordingly.
691 (vec_perm_indices::clamp): Handle polynomial element_types.
692 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
693 (vec_perm_indices::all_in_range_p): Likewise.
694 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
696 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
697 polynomial vec_perm_indices element types.
698 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
699 * fold-const.c (fold_vec_perm): Likewise.
700 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
701 * tree-vect-generic.c (lower_vec_perm): Likewise.
702 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
703 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
704 element type to HOST_WIDE_INT.
706 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
707 Alan Hayward <alan.hayward@arm.com>
708 David Sherwood <david.sherwood@arm.com>
710 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
711 rather than an int. Use plus_constant.
712 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
713 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
716 Alan Hayward <alan.hayward@arm.com>
717 David Sherwood <david.sherwood@arm.com>
719 * calls.c (emit_call_1, expand_call): Change struct_value_size from
720 a HOST_WIDE_INT to a poly_int64.
722 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
723 Alan Hayward <alan.hayward@arm.com>
724 David Sherwood <david.sherwood@arm.com>
726 * calls.c (load_register_parameters): Cope with polynomial
727 mode sizes. Require a constant size for BLKmode parameters
728 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
729 forces a parameter to be padded at the lsb end in order to
730 fill a complete number of words, require the parameter size
731 to be ordered wrt UNITS_PER_WORD.
733 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
734 Alan Hayward <alan.hayward@arm.com>
735 David Sherwood <david.sherwood@arm.com>
737 * reload1.c (spill_stack_slot_width): Change element type
738 from unsigned int to poly_uint64_pod.
739 (alter_reg): Treat mode sizes as polynomial.
741 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
742 Alan Hayward <alan.hayward@arm.com>
743 David Sherwood <david.sherwood@arm.com>
745 * reload.c (complex_word_subreg_p): New function.
746 (reload_inner_reg_of_subreg, push_reload): Use it.
748 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
749 Alan Hayward <alan.hayward@arm.com>
750 David Sherwood <david.sherwood@arm.com>
752 * lra-constraints.c (process_alt_operands): Reject matched
753 operands whose sizes aren't ordered.
754 (match_reload): Refer to this check here.
756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
757 Alan Hayward <alan.hayward@arm.com>
758 David Sherwood <david.sherwood@arm.com>
760 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
761 that the mode size is in the set {1, 2, 4, 8, 16}.
763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
764 Alan Hayward <alan.hayward@arm.com>
765 David Sherwood <david.sherwood@arm.com>
767 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
768 Use plus_constant instead of gen_rtx_PLUS.
770 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
771 Alan Hayward <alan.hayward@arm.com>
772 David Sherwood <david.sherwood@arm.com>
774 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
775 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
776 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
777 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
778 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
779 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
780 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
781 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
782 * config/i386/i386.c (ix86_push_rounding): ...this new function.
783 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
785 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
786 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
787 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
788 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
789 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
790 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
791 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
792 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
793 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
794 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
796 * expr.c (emit_move_resolve_push): Treat the input and result
797 of PUSH_ROUNDING as a poly_int64.
798 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
799 (emit_push_insn): Likewise.
800 * lra-eliminations.c (mark_not_eliminable): Likewise.
801 * recog.c (push_operand): Likewise.
802 * reload1.c (elimination_effects): Likewise.
803 * rtlanal.c (nonzero_bits1): Likewise.
804 * calls.c (store_one_arg): Likewise. Require the padding to be
805 known at compile time.
807 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
808 Alan Hayward <alan.hayward@arm.com>
809 David Sherwood <david.sherwood@arm.com>
811 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
812 Use plus_constant instead of gen_rtx_PLUS.
814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
815 Alan Hayward <alan.hayward@arm.com>
816 David Sherwood <david.sherwood@arm.com>
818 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
821 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
822 Alan Hayward <alan.hayward@arm.com>
823 David Sherwood <david.sherwood@arm.com>
825 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
826 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
827 via stack temporaries. Treat the mode size as polynomial too.
829 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
830 Alan Hayward <alan.hayward@arm.com>
831 David Sherwood <david.sherwood@arm.com>
833 * expr.c (expand_expr_real_2): When handling conversions involving
834 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
835 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
836 as a poly_uint64 too.
838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
839 Alan Hayward <alan.hayward@arm.com>
840 David Sherwood <david.sherwood@arm.com>
842 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
845 Alan Hayward <alan.hayward@arm.com>
846 David Sherwood <david.sherwood@arm.com>
848 * combine.c (can_change_dest_mode): Handle polynomial
849 REGMODE_NATURAL_SIZE.
850 * expmed.c (store_bit_field_1): Likewise.
851 * expr.c (store_constructor): Likewise.
852 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
853 and polynomial REGMODE_NATURAL_SIZE.
854 (gen_lowpart_common): Likewise.
855 * reginfo.c (record_subregs_of_mode): Likewise.
856 * rtlanal.c (read_modify_subreg_p): Likewise.
858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
859 Alan Hayward <alan.hayward@arm.com>
860 David Sherwood <david.sherwood@arm.com>
862 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
866 Alan Hayward <alan.hayward@arm.com>
867 David Sherwood <david.sherwood@arm.com>
869 * match.pd: Cope with polynomial numbers of vector elements.
871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
872 Alan Hayward <alan.hayward@arm.com>
873 David Sherwood <david.sherwood@arm.com>
875 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
876 in a POINTER_PLUS_EXPR.
878 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
879 Alan Hayward <alan.hayward@arm.com>
880 David Sherwood <david.sherwood@arm.com>
882 * omp-simd-clone.c (simd_clone_subparts): New function.
883 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
884 (ipa_simd_modify_function_body): Likewise.
886 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
887 Alan Hayward <alan.hayward@arm.com>
888 David Sherwood <david.sherwood@arm.com>
890 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
891 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
892 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
893 (expand_vector_condition, vector_element): Likewise.
894 (subparts_gt): New function.
895 (get_compute_type): Use subparts_gt.
896 (count_type_subparts): Delete.
897 (expand_vector_operations_1): Use subparts_gt instead of
900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
901 Alan Hayward <alan.hayward@arm.com>
902 David Sherwood <david.sherwood@arm.com>
904 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
905 (vect_compile_time_alias): ...this new function. Do the calculation
906 on poly_ints rather than trees.
907 (vect_prune_runtime_alias_test_list): Update call accordingly.
909 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
910 Alan Hayward <alan.hayward@arm.com>
911 David Sherwood <david.sherwood@arm.com>
913 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
915 (vect_schedule_slp_instance): Likewise.
917 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
918 Alan Hayward <alan.hayward@arm.com>
919 David Sherwood <david.sherwood@arm.com>
921 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
922 constant and extern definitions for variable-length vectors.
923 (vect_get_constant_vectors): Note that the number of units
924 is known to be constant.
926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
927 Alan Hayward <alan.hayward@arm.com>
928 David Sherwood <david.sherwood@arm.com>
930 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
931 of units as polynomial. Choose between WIDE and NARROW based
934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
935 Alan Hayward <alan.hayward@arm.com>
936 David Sherwood <david.sherwood@arm.com>
938 * tree-vect-stmts.c (simd_clone_subparts): New function.
939 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
942 Alan Hayward <alan.hayward@arm.com>
943 David Sherwood <david.sherwood@arm.com>
945 * tree-vect-stmts.c (vectorizable_call): Treat the number of
946 vectors as polynomial. Use build_index_vector for
949 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
950 Alan Hayward <alan.hayward@arm.com>
951 David Sherwood <david.sherwood@arm.com>
953 * tree-vect-stmts.c (get_load_store_type): Treat the number of
954 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
955 for variable-length vectors.
956 (vectorizable_mask_load_store): Treat the number of units as
957 polynomial, asserting that it is constant if the condition has
958 already been enforced.
959 (vectorizable_store, vectorizable_load): Likewise.
961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
962 Alan Hayward <alan.hayward@arm.com>
963 David Sherwood <david.sherwood@arm.com>
965 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
966 of units as polynomial. Punt if we can't tell at compile time
967 which vector contains the final result.
969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
970 Alan Hayward <alan.hayward@arm.com>
971 David Sherwood <david.sherwood@arm.com>
973 * tree-vect-loop.c (vectorizable_induction): Treat the number
974 of units as polynomial. Punt on SLP inductions. Use an integer
975 VEC_SERIES_EXPR for variable-length integer reductions. Use a
976 cast of such a series for variable-length floating-point
979 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
980 Alan Hayward <alan.hayward@arm.com>
981 David Sherwood <david.sherwood@arm.com>
983 * tree.h (build_index_vector): Declare.
984 * tree.c (build_index_vector): New function.
985 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
986 of units as polynomial, forcibly converting it to a constant if
987 vectorizable_reduction has already enforced the condition.
988 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
989 to create a {1,2,3,...} vector.
990 (vectorizable_reduction): Treat the number of units as polynomial.
991 Choose vectype_in based on the largest scalar element size rather
992 than the smallest number of units. Enforce the restrictions
995 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
996 Alan Hayward <alan.hayward@arm.com>
997 David Sherwood <david.sherwood@arm.com>
999 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1000 number of units as polynomial.
1002 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1003 Alan Hayward <alan.hayward@arm.com>
1004 David Sherwood <david.sherwood@arm.com>
1006 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1007 * target.def (autovectorize_vector_sizes): Return the vector sizes
1008 by pointer, using vector_sizes rather than a bitmask.
1009 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1010 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1011 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1013 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1014 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1015 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1016 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1017 * omp-general.c (omp_max_vf): Likewise.
1018 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1019 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1020 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1021 * tree-vect-slp.c (vect_slp_bb): Likewise.
1022 * doc/tm.texi: Regenerate.
1023 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1025 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1026 the vector size as a poly_uint64 rather than an unsigned int.
1027 (current_vector_size): Change from an unsigned int to a poly_uint64.
1028 (get_vectype_for_scalar_type): Update accordingly.
1029 * tree.h (build_truth_vector_type): Take the size and number of
1030 units as a poly_uint64 rather than an unsigned int.
1031 (build_vector_type): Add a temporary overload that takes
1032 the number of units as a poly_uint64 rather than an unsigned int.
1033 * tree.c (make_vector_type): Likewise.
1034 (build_truth_vector_type): Take the number of units as a poly_uint64
1035 rather than an unsigned int.
1037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1038 Alan Hayward <alan.hayward@arm.com>
1039 David Sherwood <david.sherwood@arm.com>
1041 * target.def (get_mask_mode): Take the number of units and length
1042 as poly_uint64s rather than unsigned ints.
1043 * targhooks.h (default_get_mask_mode): Update accordingly.
1044 * targhooks.c (default_get_mask_mode): Likewise.
1045 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1046 * doc/tm.texi: Regenerate.
1048 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1049 Alan Hayward <alan.hayward@arm.com>
1050 David Sherwood <david.sherwood@arm.com>
1052 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1053 * omp-general.c (omp_max_vf): Likewise.
1054 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1055 (expand_omp_simd): Handle polynomial safelen.
1056 * omp-low.c (omplow_simd_context): Add a default constructor.
1057 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1058 (lower_rec_simd_input_clauses): Update accordingly.
1059 (lower_rec_input_clauses): Likewise.
1061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1062 Alan Hayward <alan.hayward@arm.com>
1063 David Sherwood <david.sherwood@arm.com>
1065 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1066 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1067 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1068 (vect_analyze_slp_cost): Likewise.
1069 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1070 (vect_model_load_cost): Likewise.
1072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1073 Alan Hayward <alan.hayward@arm.com>
1074 David Sherwood <david.sherwood@arm.com>
1076 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1077 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1078 from an unsigned int * to a poly_uint64_pod *.
1079 (calculate_unrolling_factor): New function.
1080 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1083 Alan Hayward <alan.hayward@arm.com>
1084 David Sherwood <david.sherwood@arm.com>
1086 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1087 from an unsigned int to a poly_uint64.
1088 (_loop_vec_info::slp_unrolling_factor): Likewise.
1089 (_loop_vec_info::vectorization_factor): Change from an int
1091 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1092 (vect_get_num_vectors): New function.
1093 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1094 (vect_get_num_copies): Use vect_get_num_vectors.
1095 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1096 to an unsigned int *.
1097 (vect_analyze_data_refs): Change min_vf from an int * to a
1099 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1100 than an unsigned HOST_WIDE_INT.
1101 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1102 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1103 to an unsigned int *.
1104 (vect_analyze_data_ref_dependences): Likewise.
1105 (vect_compute_data_ref_alignment): Handle polynomial vf.
1106 (vect_enhance_data_refs_alignment): Likewise.
1107 (vect_prune_runtime_alias_test_list): Likewise.
1108 (vect_shift_permute_load_chain): Likewise.
1109 (vect_supportable_dr_alignment): Likewise.
1110 (dependence_distance_ge_vf): Take the vectorization factor as a
1111 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1112 (vect_analyze_data_refs): Change min_vf from an int * to a
1114 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1115 vfm1 as a poly_uint64 rather than an int. Make the same change
1116 for the returned bound_scalar.
1117 (vect_gen_vector_loop_niters): Handle polynomial vf.
1118 (vect_do_peeling): Likewise. Update call to
1119 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1120 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1122 * tree-vect-loop.c (vect_determine_vectorization_factor)
1123 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1124 (vect_get_known_peeling_cost): Likewise.
1125 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1126 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1127 (vect_transform_loop): Likewise. Use the lowest possible VF when
1128 updating the upper bounds of the loop.
1129 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1131 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1132 polynomial unroll factors.
1133 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1134 (vect_make_slp_decision): Likewise.
1135 (vect_supported_load_permutation_p): Likewise, and polynomial
1137 (vect_analyze_slp_cost): Handle polynomial vf.
1138 (vect_slp_analyze_node_operations): Likewise.
1139 (vect_slp_analyze_bb_1): Likewise.
1140 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1141 than an unsigned HOST_WIDE_INT.
1142 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1143 (vectorizable_load): Handle polynomial vf.
1144 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1146 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1149 Alan Hayward <alan.hayward@arm.com>
1150 David Sherwood <david.sherwood@arm.com>
1152 * match.pd: Handle bit operations involving three constants
1153 and try to fold one pair.
1155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1157 * tree-vect-loop-manip.c: Include gimple-fold.h.
1158 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1159 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1160 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1161 Add a path that uses a step of VF instead of 1, but disable it
1163 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1164 and niters_no_overflow parameters. Update calls to
1165 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1166 Create a new SSA name if the latter choses to use a ste other
1167 than zero, and return it via niters_vector_mult_vf_var.
1168 * tree-vect-loop.c (vect_transform_loop): Update calls to
1169 vect_do_peeling, vect_gen_vector_loop_niters and
1170 slpeel_make_loop_iterate_ntimes.
1171 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1172 (vect_gen_vector_loop_niters): Update declarations after above changes.
1174 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1176 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1177 128-bit round to integer instructions.
1178 (ceil<mode>2): Likewise.
1179 (btrunc<mode>2): Likewise.
1180 (round<mode>2): Likewise.
1182 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1184 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1185 unaligned VSX load/store on P8/P9.
1186 (expand_block_clear): Allow the use of unaligned VSX
1187 load/store on P8/P9.
1189 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1191 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1193 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1194 swap associated with both a load and a store.
1196 2018-01-02 Andrew Waterman <andrew@sifive.com>
1198 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1199 * config/riscv/riscv.md (clear_cache): Use it.
1201 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1203 * web.c: Remove out-of-date comment.
1205 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1207 * expr.c (fixup_args_size_notes): Check that any existing
1208 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1209 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1210 (emit_single_push_insn): ...here.
1212 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1214 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1215 (const_vector_encoded_nelts): New function.
1216 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1217 (const_vector_int_elt, const_vector_elt): Declare.
1218 * emit-rtl.c (const_vector_int_elt_1): New function.
1219 (const_vector_elt): Likewise.
1220 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1221 of CONST_VECTOR_ELT.
1223 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1225 * expr.c: Include rtx-vector-builder.h.
1226 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1227 directly on the tree encoding.
1228 (const_vector_from_tree): Likewise.
1229 * optabs.c: Include rtx-vector-builder.h.
1230 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1231 sequence of "u" values.
1232 * vec-perm-indices.c: Include rtx-vector-builder.h.
1233 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1234 directly on the vec_perm_indices encoding.
1236 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1238 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1239 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1240 * rtx-vector-builder.h: New file.
1241 * rtx-vector-builder.c: Likewise.
1242 * rtl.h (rtx_def::u2): Add a const_vector field.
1243 (CONST_VECTOR_NPATTERNS): New macro.
1244 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1245 (CONST_VECTOR_DUPLICATE_P): Likewise.
1246 (CONST_VECTOR_STEPPED_P): Likewise.
1247 (CONST_VECTOR_ENCODED_ELT): Likewise.
1248 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1249 (unwrap_const_vec_duplicate): Likewise.
1250 (const_vec_series_p): Check for a non-duplicated vector encoding.
1251 Say that the function only returns true for integer vectors.
1252 * emit-rtl.c: Include rtx-vector-builder.h.
1253 (gen_const_vec_duplicate_1): Delete.
1254 (gen_const_vector): Call gen_const_vec_duplicate instead of
1255 gen_const_vec_duplicate_1.
1256 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1257 (gen_const_vec_duplicate): Use rtx_vector_builder.
1258 (gen_const_vec_series): Likewise.
1259 (gen_rtx_CONST_VECTOR): Likewise.
1260 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1261 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1262 Build a new vector rather than modifying a CONST_VECTOR in-place.
1263 (handle_special_swappables): Update call accordingly.
1264 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1265 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1266 Build a new vector rather than modifying a CONST_VECTOR in-place.
1267 (handle_special_swappables): Update call accordingly.
1269 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1271 * simplify-rtx.c (simplify_const_binary_operation): Use
1272 CONST_VECTOR_ELT instead of XVECEXP.
1274 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1276 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1277 the selector elements to be different from the data elements
1278 if the selector is a VECTOR_CST.
1279 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1280 ssizetype for the selector.
1282 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1284 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1285 before testing each element individually.
1286 * tree-vect-generic.c (lower_vec_perm): Likewise.
1288 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1290 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1291 * selftest-run-tests.c (selftest::run_tests): Call it.
1292 * vector-builder.h (vector_builder::operator ==): New function.
1293 (vector_builder::operator !=): Likewise.
1294 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1295 (vec_perm_indices::all_from_input_p): New function.
1296 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1297 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1298 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1299 instead of reading the VECTOR_CST directly. Detect whether both
1300 vector inputs are the same before constructing the vec_perm_indices,
1301 and update the number of inputs argument accordingly. Use the
1302 utility functions added above. Only construct sel2 if we need to.
1304 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1306 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1307 the broadcast of the low byte.
1308 (expand_mult_highpart): Use an explicit encoding for the permutes.
1309 * optabs-query.c (can_mult_highpart_p): Likewise.
1310 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1311 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1312 (vectorizable_bswap): Likewise.
1313 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1314 explicit encoding for the power-of-2 permutes.
1315 (vect_permute_store_chain): Likewise.
1316 (vect_grouped_load_supported): Likewise.
1317 (vect_permute_load_chain): Likewise.
1319 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1321 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1322 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1323 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1324 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1325 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1326 (vect_gen_perm_mask_any): Likewise.
1328 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1330 * int-vector-builder.h: New file.
1331 * vec-perm-indices.h: Include int-vector-builder.h.
1332 (vec_perm_indices): Redefine as an int_vector_builder.
1333 (auto_vec_perm_indices): Delete.
1334 (vec_perm_builder): Redefine as a stand-alone class.
1335 (vec_perm_indices::vec_perm_indices): New function.
1336 (vec_perm_indices::clamp): Likewise.
1337 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1338 (vec_perm_indices::new_vector): New function.
1339 (vec_perm_indices::new_expanded_vector): Update for new
1340 vec_perm_indices class.
1341 (vec_perm_indices::rotate_inputs): New function.
1342 (vec_perm_indices::all_in_range_p): Operate directly on the
1343 encoded form, without computing elided elements.
1344 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1345 encoding. Update for new vec_perm_indices class.
1346 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1347 the given vec_perm_builder.
1348 (expand_vec_perm_var): Update vec_perm_builder constructor.
1349 (expand_mult_highpart): Use vec_perm_builder instead of
1350 auto_vec_perm_indices.
1351 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1352 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1353 or double series encoding as appropriate.
1354 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1355 vec_perm_indices instead of auto_vec_perm_indices.
1356 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1357 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1358 (vect_permute_store_chain): Likewise.
1359 (vect_grouped_load_supported): Likewise.
1360 (vect_permute_load_chain): Likewise.
1361 (vect_shift_permute_load_chain): Likewise.
1362 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1363 (vect_transform_slp_perm_load): Likewise.
1364 (vect_schedule_slp_instance): Likewise.
1365 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1366 (vectorizable_mask_load_store): Likewise.
1367 (vectorizable_bswap): Likewise.
1368 (vectorizable_store): Likewise.
1369 (vectorizable_load): Likewise.
1370 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1371 vec_perm_indices instead of auto_vec_perm_indices. Use
1372 tree_to_vec_perm_builder to read the vector from a tree.
1373 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1374 vec_perm_builder instead of a vec_perm_indices.
1375 (have_whole_vector_shift): Use vec_perm_builder and
1376 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1377 truncation to calc_vec_perm_mask_for_shift.
1378 (vect_create_epilog_for_reduction): Likewise.
1379 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1380 from auto_vec_perm_indices to vec_perm_indices.
1381 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1382 instead of changing individual elements.
1383 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1384 the vector in d.perm.
1385 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1386 from auto_vec_perm_indices to vec_perm_indices.
1387 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1388 instead of changing individual elements.
1389 (arm_vectorize_vec_perm_const): Use new_vector to install
1390 the vector in d.perm.
1391 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1392 Update vec_perm_builder constructor.
1393 (rs6000_expand_interleave): Likewise.
1394 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1395 (rs6000_expand_interleave): Likewise.
1397 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1399 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1400 to qimode could truncate the indices.
1401 * optabs.c (expand_vec_perm_var): Likewise.
1403 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1405 * Makefile.in (OBJS): Add vec-perm-indices.o.
1406 * vec-perm-indices.h: New file.
1407 * vec-perm-indices.c: Likewise.
1408 * target.h (vec_perm_indices): Replace with a forward class
1410 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1411 * optabs.h: Include vec-perm-indices.h.
1412 (expand_vec_perm): Delete.
1413 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1414 (expand_vec_perm_const): Declare.
1415 * target.def (vec_perm_const_ok): Replace with...
1416 (vec_perm_const): ...this new hook.
1417 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1418 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1419 * doc/tm.texi: Regenerate.
1420 * optabs.def (vec_perm_const): Delete.
1421 * doc/md.texi (vec_perm_const): Likewise.
1422 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1423 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1424 expand_vec_perm for constant permutation vectors. Assert that
1425 the mode of variable permutation vectors is the integer equivalent
1426 of the mode that is being permuted.
1427 * optabs-query.h (selector_fits_mode_p): Declare.
1428 * optabs-query.c: Include vec-perm-indices.h.
1429 (selector_fits_mode_p): New function.
1430 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1431 is defined, instead of checking whether the vec_perm_const_optab
1432 exists. Use targetm.vectorize.vec_perm_const instead of
1433 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1434 fit in the vector mode before using a variable permute.
1435 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1436 vec_perm_indices instead of an rtx.
1437 (expand_vec_perm): Replace with...
1438 (expand_vec_perm_const): ...this new function. Take the selector
1439 as a vec_perm_indices rather than an rtx. Also take the mode of
1440 the selector. Update call to shift_amt_for_vec_perm_mask.
1441 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1442 Use vec_perm_indices::new_expanded_vector to expand the original
1443 selector into bytes. Check whether the indices fit in the vector
1444 mode before using a variable permute.
1445 (expand_vec_perm_var): Make global.
1446 (expand_mult_highpart): Use expand_vec_perm_const.
1447 * fold-const.c: Includes vec-perm-indices.h.
1448 * tree-ssa-forwprop.c: Likewise.
1449 * tree-vect-data-refs.c: Likewise.
1450 * tree-vect-generic.c: Likewise.
1451 * tree-vect-loop.c: Likewise.
1452 * tree-vect-slp.c: Likewise.
1453 * tree-vect-stmts.c: Likewise.
1454 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1456 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1457 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1458 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1459 (aarch64_vectorize_vec_perm_const): ...this new function.
1460 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1461 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1462 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1463 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1464 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1465 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1466 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1468 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1469 check for NEON modes.
1470 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1471 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1472 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1473 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1475 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1476 the old VEC_PERM_CONST conditions.
1477 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1478 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1479 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1480 (ia64_vectorize_vec_perm_const_ok): Merge into...
1481 (ia64_vectorize_vec_perm_const): ...this new function.
1482 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1483 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1484 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1485 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1486 * config/mips/mips.c (mips_expand_vec_perm_const)
1487 (mips_vectorize_vec_perm_const_ok): Merge into...
1488 (mips_vectorize_vec_perm_const): ...this new function.
1489 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1490 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1491 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1492 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1493 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1494 (rs6000_expand_vec_perm_const): Delete.
1495 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1497 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1498 (altivec_expand_vec_perm_const_le): Take each operand individually.
1499 Operate on constant selectors rather than rtxes.
1500 (altivec_expand_vec_perm_const): Likewise. Update call to
1501 altivec_expand_vec_perm_const_le.
1502 (rs6000_expand_vec_perm_const): Delete.
1503 (rs6000_vectorize_vec_perm_const_ok): Delete.
1504 (rs6000_vectorize_vec_perm_const): New function.
1505 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1506 an element count and rtx array.
1507 (rs6000_expand_extract_even): Update call accordingly.
1508 (rs6000_expand_interleave): Likewise.
1509 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1510 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1511 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1512 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1513 (rs6000_expand_vec_perm_const): Delete.
1514 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1515 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1516 (altivec_expand_vec_perm_const_le): Take each operand individually.
1517 Operate on constant selectors rather than rtxes.
1518 (altivec_expand_vec_perm_const): Likewise. Update call to
1519 altivec_expand_vec_perm_const_le.
1520 (rs6000_expand_vec_perm_const): Delete.
1521 (rs6000_vectorize_vec_perm_const_ok): Delete.
1522 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1523 reference to the SPE evmerge intructions.
1524 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1525 an element count and rtx array.
1526 (rs6000_expand_extract_even): Update call accordingly.
1527 (rs6000_expand_interleave): Likewise.
1528 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1529 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1531 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1533 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1535 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1536 vector mode and that that mode matches the mode of the data
1538 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1539 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1540 directly using expand_vec_perm_1 when forcing selectors into
1542 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1544 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1546 * optabs-query.h (can_vec_perm_p): Delete.
1547 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1548 * optabs-query.c (can_vec_perm_p): Split into...
1549 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1550 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1551 particular selector is valid.
1552 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1553 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1554 (vect_grouped_load_supported): Likewise.
1555 (vect_shift_permute_load_chain): Likewise.
1556 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1557 (vect_transform_slp_perm_load): Likewise.
1558 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1559 (vectorizable_bswap): Likewise.
1560 (vect_gen_perm_mask_checked): Likewise.
1561 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1562 implementations of variable permutation vectors into account
1563 when deciding which selector to use.
1564 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1565 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1566 with a false third argument.
1567 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1568 to test whether the constant selector is valid and can_vec_perm_var_p
1569 to test whether a variable selector is valid.
1571 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1573 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1574 * optabs-query.c (can_vec_perm_p): Likewise.
1575 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1576 instead of vec_perm_indices.
1577 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1578 (vect_gen_perm_mask_checked): Likewise,
1579 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1580 (vect_gen_perm_mask_checked): Likewise,
1582 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1584 * optabs-query.h (qimode_for_vec_perm): Declare.
1585 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1586 (qimode_for_vec_perm): ...this new function.
1587 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1589 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1591 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1592 does not have a conditional at the top.
1594 2018-01-02 Richard Biener <rguenther@suse.de>
1596 * ipa-inline.c (big_speedup_p): Fix expression.
1598 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1601 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1604 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1608 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1609 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1610 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1611 cond_taken_branch_cost 3->4.
1613 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1615 PR tree-optimization/83581
1616 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1617 TODO_cleanup_cfg if any changes have been made.
1620 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1621 convert_modes if target mode has the right side, but different mode
1625 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1626 last argument when extracting from CONCAT. If either from_real or
1627 from_imag is NULL, use expansion through memory. If result is not
1628 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1629 the parts directly to inner mode, if even that fails, use expansion
1633 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1634 check for bswap in mode rather than HImode and use that in expand_unop
1637 Copyright (C) 2018 Free Software Foundation, Inc.
1639 Copying and distribution of this file, with or without modification,
1640 are permitted in any medium without royalty provided the copyright
1641 notice and this notice are preserved.