1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2 Alan Hayward <alan.hayward@arm.com>
3 David Sherwood <david.sherwood@arm.com>
5 * expmed.c (extract_bit_field_1): For vector extracts,
6 fall back to extract_bit_field_as_subreg if vec_extract
9 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
10 Alan Hayward <alan.hayward@arm.com>
11 David Sherwood <david.sherwood@arm.com>
13 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
14 they are variable or constant sized.
15 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
16 slots for constant-sized data.
18 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
19 Alan Hayward <alan.hayward@arm.com>
20 David Sherwood <david.sherwood@arm.com>
22 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
23 handling COND_EXPRs with boolean comparisons, try to find a better
24 basis for the mask type than the boolean itself.
26 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
28 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
29 is calculated and how it can be overridden.
30 * genmodes.c (max_bitsize_mode_any_mode): New variable.
31 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
33 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
36 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
37 Alan Hayward <alan.hayward@arm.com>
38 David Sherwood <david.sherwood@arm.com>
40 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
41 Remove the mode argument.
42 (aarch64_simd_valid_immediate): Remove the mode and inverse
44 * config/aarch64/iterators.md (bitsize): New iterator.
45 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
46 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
47 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
48 aarch64_simd_valid_immediate.
49 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
50 (aarch64_reg_or_bic_imm): Likewise.
51 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
52 with an insn_type enum and msl with a modifier_type enum.
53 Replace element_width with a scalar_mode. Change the shift
54 to unsigned int. Add constructors for scalar_float_mode and
55 scalar_int_mode elements.
56 (aarch64_vect_float_const_representable_p): Delete.
57 (aarch64_can_const_movi_rtx_p)
58 (aarch64_simd_scalar_immediate_valid_for_move)
59 (aarch64_simd_make_constant): Update call to
60 aarch64_simd_valid_immediate.
61 (aarch64_advsimd_valid_immediate_hs): New function.
62 (aarch64_advsimd_valid_immediate): Likewise.
63 (aarch64_simd_valid_immediate): Remove mode and inverse
64 arguments. Rewrite to use the above. Use const_vec_duplicate_p
65 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
66 and aarch64_float_const_representable_p on the result.
67 (aarch64_output_simd_mov_immediate): Remove mode argument.
68 Update call to aarch64_simd_valid_immediate and use of
70 (aarch64_output_scalar_simd_mov_immediate): Update call
73 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
74 Alan Hayward <alan.hayward@arm.com>
75 David Sherwood <david.sherwood@arm.com>
77 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
78 (mode_nunits): Likewise CONST_MODE_NUNITS.
79 * machmode.def (ADJUST_NUNITS): Document.
80 * genmodes.c (mode_data::need_nunits_adj): New field.
81 (blank_mode): Update accordingly.
82 (adj_nunits): New variable.
83 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
85 (emit_mode_size_inline): Set need_bytesize_adj for all modes
87 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
88 listed in adj_nunits. Don't emit case statements for such modes.
89 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
90 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
91 nothing if adj_nunits is nonnull.
92 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
93 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
94 (emit_mode_fbit): Update use of print_maybe_const_decl.
95 (emit_move_size): Likewise. Treat the array as non-const
97 (emit_mode_adjustments): Handle adj_nunits.
99 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
101 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
102 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
103 (VECTOR_MODES): Use it.
104 (make_vector_modes): Take the prefix as an argument.
106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
107 Alan Hayward <alan.hayward@arm.com>
108 David Sherwood <david.sherwood@arm.com>
110 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
111 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
112 for MODE_VECTOR_BOOL.
113 * machmode.def (VECTOR_BOOL_MODE): Document.
114 * genmodes.c (VECTOR_BOOL_MODE): New macro.
115 (make_vector_bool_mode): New function.
116 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
118 * lto-streamer-in.c (lto_input_mode_table): Likewise.
119 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
121 * stor-layout.c (int_mode_for_mode): Likewise.
122 * tree.c (build_vector_type_for_mode): Likewise.
123 * varasm.c (output_constant_pool_2): Likewise.
124 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
125 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
126 for MODE_VECTOR_BOOL.
127 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
128 of mode class checks.
129 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
130 instead of a list of mode class checks.
131 (expand_vector_scalar_condition): Likewise.
132 (type_for_widest_vector_mode): Handle BImode as an inner mode.
134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
135 Alan Hayward <alan.hayward@arm.com>
136 David Sherwood <david.sherwood@arm.com>
138 * machmode.h (mode_size): Change from unsigned short to
140 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
141 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
142 or if measurement_type is not polynomial.
143 (fixed_size_mode::includes_p): Check for constant-sized modes.
144 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
145 return a poly_uint16 rather than an unsigned short.
146 (emit_mode_size): Change the type of mode_size from unsigned short
147 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
148 (emit_mode_adjustments): Cope with polynomial vector sizes.
149 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
151 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
153 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
154 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
155 * caller-save.c (setup_save_areas): Likewise.
156 (replace_reg_with_saved_mem): Likewise.
157 * calls.c (emit_library_call_value_1): Likewise.
158 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
159 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
160 (gen_lowpart_for_combine): Likewise.
161 * convert.c (convert_to_integer_1): Likewise.
162 * cse.c (equiv_constant, cse_insn): Likewise.
163 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
164 (cselib_subst_to_values): Likewise.
165 * dce.c (word_dce_process_block): Likewise.
166 * df-problems.c (df_word_lr_mark_ref): Likewise.
167 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
168 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
169 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
170 (rtl_for_decl_location): Likewise.
171 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
172 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
173 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
174 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
175 (expand_expr_real_1): Likewise.
176 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
177 (pad_below): Likewise.
178 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
179 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
180 * ira.c (get_subreg_tracking_sizes): Likewise.
181 * ira-build.c (ira_create_allocno_objects): Likewise.
182 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
183 (ira_sort_regnos_for_alter_reg): Likewise.
184 * ira-costs.c (record_operand_costs): Likewise.
185 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
186 (resolve_simple_move): Likewise.
187 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
188 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
189 (lra_constraints): Likewise.
190 (CONST_POOL_OK_P): Reject variable-sized modes.
191 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
192 (add_pseudo_to_slot, lra_spill): Likewise.
193 * omp-low.c (omp_clause_aligned_alignment): Likewise.
194 * optabs-query.c (get_best_extraction_insn): Likewise.
195 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
196 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
197 (expand_mult_highpart, valid_multiword_target_p): Likewise.
198 * recog.c (offsettable_address_addr_space_p): Likewise.
199 * regcprop.c (maybe_mode_change): Likewise.
200 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
201 * regrename.c (build_def_use): Likewise.
202 * regstat.c (dump_reg_info): Likewise.
203 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
204 (find_reloads, find_reloads_subreg_address): Likewise.
205 * reload1.c (eliminate_regs_1): Likewise.
206 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
207 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
208 (simplify_binary_operation_1, simplify_subreg): Likewise.
209 * targhooks.c (default_function_arg_padding): Likewise.
210 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
211 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
212 (verify_gimple_assign_ternary): Likewise.
213 * tree-inline.c (estimate_move_cost): Likewise.
214 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
215 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
216 (get_address_cost_ainc): Likewise.
217 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
218 (vect_supportable_dr_alignment): Likewise.
219 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
220 (vectorizable_reduction): Likewise.
221 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
222 (vectorizable_operation, vectorizable_load): Likewise.
223 * tree.c (build_same_sized_truth_vector_type): Likewise.
224 * valtrack.c (cleanup_auto_inc_dec): Likewise.
225 * var-tracking.c (emit_note_insn_var_location): Likewise.
226 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
227 (ADDR_VEC_ALIGN): Likewise.
229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
230 Alan Hayward <alan.hayward@arm.com>
231 David Sherwood <david.sherwood@arm.com>
233 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
235 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
236 or if measurement_type is polynomial.
237 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
238 * combine.c (make_extraction): Likewise.
239 * dse.c (find_shift_sequence): Likewise.
240 * dwarf2out.c (mem_loc_descriptor): Likewise.
241 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
242 (extract_bit_field, extract_low_bits): Likewise.
243 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
244 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
245 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
246 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
247 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
248 * reload.c (find_reloads): Likewise.
249 * reload1.c (alter_reg): Likewise.
250 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
251 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
252 * tree-if-conv.c (predicate_mem_writes): Likewise.
253 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
254 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
255 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
256 * valtrack.c (dead_debug_insert_temp): Likewise.
257 * varasm.c (mergeable_constant_section): Likewise.
258 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
260 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
261 Alan Hayward <alan.hayward@arm.com>
262 David Sherwood <david.sherwood@arm.com>
264 * expr.c (expand_assignment): Cope with polynomial mode sizes
265 when assigning to a CONCAT.
267 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
268 Alan Hayward <alan.hayward@arm.com>
269 David Sherwood <david.sherwood@arm.com>
271 * machmode.h (mode_precision): Change from unsigned short to
273 (mode_to_precision): Return a poly_uint16 rather than an unsigned
275 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
276 or if measurement_type is not polynomial.
277 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
278 in which the mode is already known to be a scalar_int_mode.
279 * genmodes.c (emit_mode_precision): Change the type of mode_precision
280 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
282 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
283 for GET_MODE_PRECISION.
284 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
285 for GET_MODE_PRECISION.
286 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
288 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
289 (expand_field_assignment, make_extraction): Likewise.
290 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
291 (get_last_value): Likewise.
292 * convert.c (convert_to_integer_1): Likewise.
293 * cse.c (cse_insn): Likewise.
294 * expr.c (expand_expr_real_1): Likewise.
295 * lra-constraints.c (simplify_operand_subreg): Likewise.
296 * optabs-query.c (can_atomic_load_p): Likewise.
297 * optabs.c (expand_atomic_load): Likewise.
298 (expand_atomic_store): Likewise.
299 * ree.c (combine_reaching_defs): Likewise.
300 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
301 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
302 * tree.h (type_has_mode_precision_p): Likewise.
303 * ubsan.c (instrument_si_overflow): Likewise.
305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
306 Alan Hayward <alan.hayward@arm.com>
307 David Sherwood <david.sherwood@arm.com>
309 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
310 polynomial numbers of units.
311 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
312 (valid_vector_subparts_p): New function.
313 (build_vector_type): Remove temporary shim and take the number
314 of units as a poly_uint64 rather than an int.
315 (build_opaque_vector_type): Take the number of units as a
316 poly_uint64 rather than an int.
317 * tree.c (build_vector_from_ctor): Handle polynomial
318 TYPE_VECTOR_SUBPARTS.
319 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
320 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
321 (build_vector_from_val): If the number of units is variable,
322 use build_vec_duplicate_cst for constant operands and
323 VEC_DUPLICATE_EXPR otherwise.
324 (make_vector_type): Remove temporary is_constant ().
325 (build_vector_type, build_opaque_vector_type): Take the number of
326 units as a poly_uint64 rather than an int.
327 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
329 * cfgexpand.c (expand_debug_expr): Likewise.
330 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
331 (store_constructor, expand_expr_real_1): Likewise.
332 (const_scalar_mask_from_tree): Likewise.
333 * fold-const-call.c (fold_const_reduction): Likewise.
334 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
335 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
336 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
337 (fold_relational_const): Likewise.
338 (native_interpret_vector): Likewise. Change the size from an
339 int to an unsigned int.
340 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
341 TYPE_VECTOR_SUBPARTS.
342 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
343 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
344 duplicating a non-constant operand into a variable-length vector.
345 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
346 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
347 * ipa-icf.c (sem_variable::equals): Likewise.
348 * match.pd: Likewise.
349 * omp-simd-clone.c (simd_clone_subparts): Likewise.
350 * print-tree.c (print_node): Likewise.
351 * stor-layout.c (layout_type): Likewise.
352 * targhooks.c (default_builtin_vectorization_cost): Likewise.
353 * tree-cfg.c (verify_gimple_comparison): Likewise.
354 (verify_gimple_assign_binary): Likewise.
355 (verify_gimple_assign_ternary): Likewise.
356 (verify_gimple_assign_single): Likewise.
357 * tree-pretty-print.c (dump_generic_node): Likewise.
358 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
359 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
360 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
361 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
362 (vect_shift_permute_load_chain): Likewise.
363 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
364 (expand_vector_condition, optimize_vector_constructor): Likewise.
365 (lower_vec_perm, get_compute_type): Likewise.
366 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
367 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
368 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
369 (vect_recog_mask_conversion_pattern): Likewise.
370 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
371 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
372 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
373 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
374 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
375 (vectorizable_shift, vectorizable_operation, vectorizable_store)
376 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
377 (supportable_widening_operation): Likewise.
378 (supportable_narrowing_operation): Likewise.
379 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
381 * varasm.c (output_constant): Likewise.
383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
384 Alan Hayward <alan.hayward@arm.com>
385 David Sherwood <david.sherwood@arm.com>
387 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
388 so that both the length == 3 and length != 3 cases set up their
389 own permute vectors. Add comments explaining why we know the
390 number of elements is constant.
391 (vect_permute_load_chain): Likewise.
393 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
394 Alan Hayward <alan.hayward@arm.com>
395 David Sherwood <david.sherwood@arm.com>
397 * machmode.h (mode_nunits): Change from unsigned char to
399 (ONLY_FIXED_SIZE_MODES): New macro.
400 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
401 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
402 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
404 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
405 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
406 or if measurement_type is not polynomial.
407 * genmodes.c (ZERO_COEFFS): New macro.
408 (emit_mode_nunits_inline): Make mode_nunits_inline return a
410 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
411 Use ZERO_COEFFS when emitting initializers.
412 * data-streamer.h (bp_pack_poly_value): New function.
413 (bp_unpack_poly_value): Likewise.
414 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
416 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
418 * tree.c (make_vector_type): Remove temporary shim and make
419 the real function take the number of units as a poly_uint64
421 (build_vector_type_for_mode): Handle polynomial nunits.
422 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
423 * emit-rtl.c (const_vec_series_p_1): Likewise.
424 (gen_rtx_CONST_VECTOR): Likewise.
425 * fold-const.c (test_vec_duplicate_folding): Likewise.
426 * genrecog.c (validate_pattern): Likewise.
427 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
428 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
429 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
430 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
431 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
432 * rtlanal.c (subreg_get_info): Likewise.
433 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
434 (vect_grouped_load_supported): Likewise.
435 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
436 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
437 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
438 (simplify_const_unary_operation, simplify_binary_operation_1)
439 (simplify_const_binary_operation, simplify_ternary_operation)
440 (test_vector_ops_duplicate, test_vector_ops): Likewise.
441 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
442 instead of CONST_VECTOR_NUNITS.
443 * varasm.c (output_constant_pool_2): Likewise.
444 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
445 explicit-encoded elements in the XVEC for variable-length vectors.
447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
449 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
452 Alan Hayward <alan.hayward@arm.com>
453 David Sherwood <david.sherwood@arm.com>
455 * coretypes.h (fixed_size_mode): Declare.
456 (fixed_size_mode_pod): New typedef.
457 * builtins.h (target_builtins::x_apply_args_mode)
458 (target_builtins::x_apply_result_mode): Change type to
460 * builtins.c (apply_args_size, apply_result_size, result_vector)
461 (expand_builtin_apply_args_1, expand_builtin_apply)
462 (expand_builtin_return): Update accordingly.
464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
466 * cse.c (hash_rtx_cb): Hash only the encoded elements.
467 * cselib.c (cselib_hash_rtx): Likewise.
468 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
469 CONST_VECTOR encoding.
471 2017-01-03 Jakub Jelinek <jakub@redhat.com>
472 Jeff Law <law@redhat.com>
475 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
476 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
477 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
478 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
481 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
482 explicitly probe *sp in a noreturn function if there were any callee
483 register saves or frame pointer is needed.
485 2018-01-03 Jakub Jelinek <jakub@redhat.com>
488 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
489 BLKmode for ternary, binary or unary expressions.
492 * var-tracking.c (delete_vta_debug_insn): New inline function.
493 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
494 insns from get_insns () to NULL instead of each bb separately.
495 Use delete_vta_debug_insn. No longer static.
496 (vt_debug_insns_local, variable_tracking_main_1): Adjust
497 delete_vta_debug_insns callers.
498 * rtl.h (delete_vta_debug_insns): Declare.
499 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
500 instead of variable_tracking_main.
502 2018-01-03 Martin Sebor <msebor@redhat.com>
505 * doc/extend.texi (attribute const): Fix a typo.
506 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
507 issuing -Wsuggest-attribute for void functions.
509 2018-01-03 Martin Sebor <msebor@redhat.com>
511 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
512 offset_int::from instead of wide_int::to_shwi.
513 (maybe_diag_overlap): Remove assertion.
514 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
515 * gimple-ssa-sprintf.c (format_directive): Same.
516 (parse_directive): Same.
517 (sprintf_dom_walker::compute_format_length): Same.
518 (try_substitute_return_value): Same.
520 2017-01-03 Jeff Law <law@redhat.com>
523 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
524 non-constant residual for zero at runtime and avoid probing in
525 that case. Reorganize code for trailing problem to mirror handling
528 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
530 PR tree-optimization/83501
531 * tree-ssa-strlen.c (get_string_cst): New.
532 (handle_char_store): Call get_string_cst.
534 2018-01-03 Martin Liska <mliska@suse.cz>
536 PR tree-optimization/83593
537 * tree-ssa-strlen.c: Include tree-cfg.h.
538 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
539 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
540 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
542 (strlen_dom_walker::before_dom_children): Call
543 gimple_purge_dead_eh_edges. Dump tranformation with details
545 (strlen_dom_walker::before_dom_children): Update call by adding
546 new argument cleanup_eh.
547 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
549 2018-01-03 Martin Liska <mliska@suse.cz>
552 * cif-code.def (VARIADIC_THUNK): New enum value.
553 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
556 2018-01-03 Jan Beulich <jbeulich@suse.com>
558 * sse.md (mov<mode>_internal): Tighten condition for when to use
559 vmovdqu<ssescalarsize> for TI and OI modes.
561 2018-01-03 Jakub Jelinek <jakub@redhat.com>
563 Update copyright years.
565 2018-01-03 Martin Liska <mliska@suse.cz>
568 * ipa-visibility.c (function_and_variable_visibility): Skip
569 functions with noipa attribure.
571 2018-01-03 Jakub Jelinek <jakub@redhat.com>
573 * gcc.c (process_command): Update copyright notice dates.
574 * gcov-dump.c (print_version): Ditto.
575 * gcov.c (print_version): Ditto.
576 * gcov-tool.c (print_version): Ditto.
577 * gengtype.c (create_file): Ditto.
578 * doc/cpp.texi: Bump @copying's copyright year.
579 * doc/cppinternals.texi: Ditto.
580 * doc/gcc.texi: Ditto.
581 * doc/gccint.texi: Ditto.
582 * doc/gcov.texi: Ditto.
583 * doc/install.texi: Ditto.
584 * doc/invoke.texi: Ditto.
586 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
588 * vector-builder.h (vector_builder::m_full_nelts): Change from
589 unsigned int to poly_uint64.
590 (vector_builder::full_nelts): Update prototype accordingly.
591 (vector_builder::new_vector): Likewise.
592 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
593 (vector_builder::operator ==): Likewise.
594 (vector_builder::finalize): Likewise.
595 * int-vector-builder.h (int_vector_builder::int_vector_builder):
596 Take the number of elements as a poly_uint64 rather than an
598 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
599 from unsigned int to poly_uint64.
600 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
601 (vec_perm_indices::new_vector): Likewise.
602 (vec_perm_indices::length): Likewise.
603 (vec_perm_indices::nelts_per_input): Likewise.
604 (vec_perm_indices::input_nelts): Likewise.
605 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
606 number of elements per input as a poly_uint64 rather than an
607 unsigned int. Use the original encoding for variable-length
608 vectors, rather than clamping each individual element.
609 For the second and subsequent elements in each pattern,
610 clamp the step and base before clamping their sum.
611 (vec_perm_indices::series_p): Handle polynomial element counts.
612 (vec_perm_indices::all_in_range_p): Likewise.
613 (vec_perm_indices_to_tree): Likewise.
614 (vec_perm_indices_to_rtx): Likewise.
615 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
616 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
617 (tree_vector_builder::new_binary_operation): Handle polynomial
618 element counts. Return false if we need to know the number
619 of elements at compile time.
620 * fold-const.c (fold_vec_perm): Punt if the number of elements
621 isn't known at compile time.
623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
625 * vec-perm-indices.h (vec_perm_builder): Change element type
626 from HOST_WIDE_INT to poly_int64.
627 (vec_perm_indices::element_type): Update accordingly.
628 (vec_perm_indices::clamp): Handle polynomial element_types.
629 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
630 (vec_perm_indices::all_in_range_p): Likewise.
631 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
633 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
634 polynomial vec_perm_indices element types.
635 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
636 * fold-const.c (fold_vec_perm): Likewise.
637 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
638 * tree-vect-generic.c (lower_vec_perm): Likewise.
639 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
640 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
641 element type to HOST_WIDE_INT.
643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
644 Alan Hayward <alan.hayward@arm.com>
645 David Sherwood <david.sherwood@arm.com>
647 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
648 rather than an int. Use plus_constant.
649 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
650 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
652 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
653 Alan Hayward <alan.hayward@arm.com>
654 David Sherwood <david.sherwood@arm.com>
656 * calls.c (emit_call_1, expand_call): Change struct_value_size from
657 a HOST_WIDE_INT to a poly_int64.
659 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
660 Alan Hayward <alan.hayward@arm.com>
661 David Sherwood <david.sherwood@arm.com>
663 * calls.c (load_register_parameters): Cope with polynomial
664 mode sizes. Require a constant size for BLKmode parameters
665 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
666 forces a parameter to be padded at the lsb end in order to
667 fill a complete number of words, require the parameter size
668 to be ordered wrt UNITS_PER_WORD.
670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
671 Alan Hayward <alan.hayward@arm.com>
672 David Sherwood <david.sherwood@arm.com>
674 * reload1.c (spill_stack_slot_width): Change element type
675 from unsigned int to poly_uint64_pod.
676 (alter_reg): Treat mode sizes as polynomial.
678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
679 Alan Hayward <alan.hayward@arm.com>
680 David Sherwood <david.sherwood@arm.com>
682 * reload.c (complex_word_subreg_p): New function.
683 (reload_inner_reg_of_subreg, push_reload): Use it.
685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
686 Alan Hayward <alan.hayward@arm.com>
687 David Sherwood <david.sherwood@arm.com>
689 * lra-constraints.c (process_alt_operands): Reject matched
690 operands whose sizes aren't ordered.
691 (match_reload): Refer to this check here.
693 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
694 Alan Hayward <alan.hayward@arm.com>
695 David Sherwood <david.sherwood@arm.com>
697 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
698 that the mode size is in the set {1, 2, 4, 8, 16}.
700 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
701 Alan Hayward <alan.hayward@arm.com>
702 David Sherwood <david.sherwood@arm.com>
704 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
705 Use plus_constant instead of gen_rtx_PLUS.
707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
708 Alan Hayward <alan.hayward@arm.com>
709 David Sherwood <david.sherwood@arm.com>
711 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
712 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
713 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
714 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
715 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
716 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
717 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
718 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
719 * config/i386/i386.c (ix86_push_rounding): ...this new function.
720 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
722 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
723 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
724 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
725 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
726 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
727 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
728 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
729 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
730 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
731 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
733 * expr.c (emit_move_resolve_push): Treat the input and result
734 of PUSH_ROUNDING as a poly_int64.
735 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
736 (emit_push_insn): Likewise.
737 * lra-eliminations.c (mark_not_eliminable): Likewise.
738 * recog.c (push_operand): Likewise.
739 * reload1.c (elimination_effects): Likewise.
740 * rtlanal.c (nonzero_bits1): Likewise.
741 * calls.c (store_one_arg): Likewise. Require the padding to be
742 known at compile time.
744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
745 Alan Hayward <alan.hayward@arm.com>
746 David Sherwood <david.sherwood@arm.com>
748 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
749 Use plus_constant instead of gen_rtx_PLUS.
751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
752 Alan Hayward <alan.hayward@arm.com>
753 David Sherwood <david.sherwood@arm.com>
755 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
758 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
759 Alan Hayward <alan.hayward@arm.com>
760 David Sherwood <david.sherwood@arm.com>
762 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
763 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
764 via stack temporaries. Treat the mode size as polynomial too.
766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
767 Alan Hayward <alan.hayward@arm.com>
768 David Sherwood <david.sherwood@arm.com>
770 * expr.c (expand_expr_real_2): When handling conversions involving
771 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
772 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
773 as a poly_uint64 too.
775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
776 Alan Hayward <alan.hayward@arm.com>
777 David Sherwood <david.sherwood@arm.com>
779 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
782 Alan Hayward <alan.hayward@arm.com>
783 David Sherwood <david.sherwood@arm.com>
785 * combine.c (can_change_dest_mode): Handle polynomial
786 REGMODE_NATURAL_SIZE.
787 * expmed.c (store_bit_field_1): Likewise.
788 * expr.c (store_constructor): Likewise.
789 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
790 and polynomial REGMODE_NATURAL_SIZE.
791 (gen_lowpart_common): Likewise.
792 * reginfo.c (record_subregs_of_mode): Likewise.
793 * rtlanal.c (read_modify_subreg_p): Likewise.
795 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
796 Alan Hayward <alan.hayward@arm.com>
797 David Sherwood <david.sherwood@arm.com>
799 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
802 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
803 Alan Hayward <alan.hayward@arm.com>
804 David Sherwood <david.sherwood@arm.com>
806 * match.pd: Cope with polynomial numbers of vector elements.
808 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
809 Alan Hayward <alan.hayward@arm.com>
810 David Sherwood <david.sherwood@arm.com>
812 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
813 in a POINTER_PLUS_EXPR.
815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
816 Alan Hayward <alan.hayward@arm.com>
817 David Sherwood <david.sherwood@arm.com>
819 * omp-simd-clone.c (simd_clone_subparts): New function.
820 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
821 (ipa_simd_modify_function_body): Likewise.
823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
824 Alan Hayward <alan.hayward@arm.com>
825 David Sherwood <david.sherwood@arm.com>
827 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
828 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
829 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
830 (expand_vector_condition, vector_element): Likewise.
831 (subparts_gt): New function.
832 (get_compute_type): Use subparts_gt.
833 (count_type_subparts): Delete.
834 (expand_vector_operations_1): Use subparts_gt instead of
837 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
838 Alan Hayward <alan.hayward@arm.com>
839 David Sherwood <david.sherwood@arm.com>
841 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
842 (vect_compile_time_alias): ...this new function. Do the calculation
843 on poly_ints rather than trees.
844 (vect_prune_runtime_alias_test_list): Update call accordingly.
846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
847 Alan Hayward <alan.hayward@arm.com>
848 David Sherwood <david.sherwood@arm.com>
850 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
852 (vect_schedule_slp_instance): Likewise.
854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
855 Alan Hayward <alan.hayward@arm.com>
856 David Sherwood <david.sherwood@arm.com>
858 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
859 constant and extern definitions for variable-length vectors.
860 (vect_get_constant_vectors): Note that the number of units
861 is known to be constant.
863 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
864 Alan Hayward <alan.hayward@arm.com>
865 David Sherwood <david.sherwood@arm.com>
867 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
868 of units as polynomial. Choose between WIDE and NARROW based
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 * tree-vect-stmts.c (simd_clone_subparts): New function.
876 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 * tree-vect-stmts.c (vectorizable_call): Treat the number of
883 vectors as polynomial. Use build_index_vector for
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-stmts.c (get_load_store_type): Treat the number of
891 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
892 for variable-length vectors.
893 (vectorizable_mask_load_store): Treat the number of units as
894 polynomial, asserting that it is constant if the condition has
895 already been enforced.
896 (vectorizable_store, vectorizable_load): Likewise.
898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
899 Alan Hayward <alan.hayward@arm.com>
900 David Sherwood <david.sherwood@arm.com>
902 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
903 of units as polynomial. Punt if we can't tell at compile time
904 which vector contains the final result.
906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
907 Alan Hayward <alan.hayward@arm.com>
908 David Sherwood <david.sherwood@arm.com>
910 * tree-vect-loop.c (vectorizable_induction): Treat the number
911 of units as polynomial. Punt on SLP inductions. Use an integer
912 VEC_SERIES_EXPR for variable-length integer reductions. Use a
913 cast of such a series for variable-length floating-point
916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
917 Alan Hayward <alan.hayward@arm.com>
918 David Sherwood <david.sherwood@arm.com>
920 * tree.h (build_index_vector): Declare.
921 * tree.c (build_index_vector): New function.
922 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
923 of units as polynomial, forcibly converting it to a constant if
924 vectorizable_reduction has already enforced the condition.
925 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
926 to create a {1,2,3,...} vector.
927 (vectorizable_reduction): Treat the number of units as polynomial.
928 Choose vectype_in based on the largest scalar element size rather
929 than the smallest number of units. Enforce the restrictions
932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
933 Alan Hayward <alan.hayward@arm.com>
934 David Sherwood <david.sherwood@arm.com>
936 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
937 number of units as polynomial.
939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
940 Alan Hayward <alan.hayward@arm.com>
941 David Sherwood <david.sherwood@arm.com>
943 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
944 * target.def (autovectorize_vector_sizes): Return the vector sizes
945 by pointer, using vector_sizes rather than a bitmask.
946 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
947 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
948 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
950 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
951 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
952 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
953 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
954 * omp-general.c (omp_max_vf): Likewise.
955 * omp-low.c (omp_clause_aligned_alignment): Likewise.
956 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
957 * tree-vect-loop.c (vect_analyze_loop): Likewise.
958 * tree-vect-slp.c (vect_slp_bb): Likewise.
959 * doc/tm.texi: Regenerate.
960 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
962 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
963 the vector size as a poly_uint64 rather than an unsigned int.
964 (current_vector_size): Change from an unsigned int to a poly_uint64.
965 (get_vectype_for_scalar_type): Update accordingly.
966 * tree.h (build_truth_vector_type): Take the size and number of
967 units as a poly_uint64 rather than an unsigned int.
968 (build_vector_type): Add a temporary overload that takes
969 the number of units as a poly_uint64 rather than an unsigned int.
970 * tree.c (make_vector_type): Likewise.
971 (build_truth_vector_type): Take the number of units as a poly_uint64
972 rather than an unsigned int.
974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
975 Alan Hayward <alan.hayward@arm.com>
976 David Sherwood <david.sherwood@arm.com>
978 * target.def (get_mask_mode): Take the number of units and length
979 as poly_uint64s rather than unsigned ints.
980 * targhooks.h (default_get_mask_mode): Update accordingly.
981 * targhooks.c (default_get_mask_mode): Likewise.
982 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
983 * doc/tm.texi: Regenerate.
985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
986 Alan Hayward <alan.hayward@arm.com>
987 David Sherwood <david.sherwood@arm.com>
989 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
990 * omp-general.c (omp_max_vf): Likewise.
991 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
992 (expand_omp_simd): Handle polynomial safelen.
993 * omp-low.c (omplow_simd_context): Add a default constructor.
994 (omplow_simd_context::max_vf): Change from int to poly_uint64.
995 (lower_rec_simd_input_clauses): Update accordingly.
996 (lower_rec_input_clauses): Likewise.
998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
999 Alan Hayward <alan.hayward@arm.com>
1000 David Sherwood <david.sherwood@arm.com>
1002 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1003 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1004 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1005 (vect_analyze_slp_cost): Likewise.
1006 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1007 (vect_model_load_cost): Likewise.
1009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1010 Alan Hayward <alan.hayward@arm.com>
1011 David Sherwood <david.sherwood@arm.com>
1013 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1014 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1015 from an unsigned int * to a poly_uint64_pod *.
1016 (calculate_unrolling_factor): New function.
1017 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1019 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1020 Alan Hayward <alan.hayward@arm.com>
1021 David Sherwood <david.sherwood@arm.com>
1023 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1024 from an unsigned int to a poly_uint64.
1025 (_loop_vec_info::slp_unrolling_factor): Likewise.
1026 (_loop_vec_info::vectorization_factor): Change from an int
1028 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1029 (vect_get_num_vectors): New function.
1030 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1031 (vect_get_num_copies): Use vect_get_num_vectors.
1032 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1033 to an unsigned int *.
1034 (vect_analyze_data_refs): Change min_vf from an int * to a
1036 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1037 than an unsigned HOST_WIDE_INT.
1038 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1039 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1040 to an unsigned int *.
1041 (vect_analyze_data_ref_dependences): Likewise.
1042 (vect_compute_data_ref_alignment): Handle polynomial vf.
1043 (vect_enhance_data_refs_alignment): Likewise.
1044 (vect_prune_runtime_alias_test_list): Likewise.
1045 (vect_shift_permute_load_chain): Likewise.
1046 (vect_supportable_dr_alignment): Likewise.
1047 (dependence_distance_ge_vf): Take the vectorization factor as a
1048 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1049 (vect_analyze_data_refs): Change min_vf from an int * to a
1051 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1052 vfm1 as a poly_uint64 rather than an int. Make the same change
1053 for the returned bound_scalar.
1054 (vect_gen_vector_loop_niters): Handle polynomial vf.
1055 (vect_do_peeling): Likewise. Update call to
1056 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1057 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1059 * tree-vect-loop.c (vect_determine_vectorization_factor)
1060 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1061 (vect_get_known_peeling_cost): Likewise.
1062 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1063 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1064 (vect_transform_loop): Likewise. Use the lowest possible VF when
1065 updating the upper bounds of the loop.
1066 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1068 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1069 polynomial unroll factors.
1070 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1071 (vect_make_slp_decision): Likewise.
1072 (vect_supported_load_permutation_p): Likewise, and polynomial
1074 (vect_analyze_slp_cost): Handle polynomial vf.
1075 (vect_slp_analyze_node_operations): Likewise.
1076 (vect_slp_analyze_bb_1): Likewise.
1077 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1078 than an unsigned HOST_WIDE_INT.
1079 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1080 (vectorizable_load): Handle polynomial vf.
1081 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1083 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1086 Alan Hayward <alan.hayward@arm.com>
1087 David Sherwood <david.sherwood@arm.com>
1089 * match.pd: Handle bit operations involving three constants
1090 and try to fold one pair.
1092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1094 * tree-vect-loop-manip.c: Include gimple-fold.h.
1095 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1096 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1097 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1098 Add a path that uses a step of VF instead of 1, but disable it
1100 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1101 and niters_no_overflow parameters. Update calls to
1102 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1103 Create a new SSA name if the latter choses to use a ste other
1104 than zero, and return it via niters_vector_mult_vf_var.
1105 * tree-vect-loop.c (vect_transform_loop): Update calls to
1106 vect_do_peeling, vect_gen_vector_loop_niters and
1107 slpeel_make_loop_iterate_ntimes.
1108 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1109 (vect_gen_vector_loop_niters): Update declarations after above changes.
1111 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1113 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1114 128-bit round to integer instructions.
1115 (ceil<mode>2): Likewise.
1116 (btrunc<mode>2): Likewise.
1117 (round<mode>2): Likewise.
1119 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1121 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1122 unaligned VSX load/store on P8/P9.
1123 (expand_block_clear): Allow the use of unaligned VSX
1124 load/store on P8/P9.
1126 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1128 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1130 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1131 swap associated with both a load and a store.
1133 2018-01-02 Andrew Waterman <andrew@sifive.com>
1135 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1136 * config/riscv/riscv.md (clear_cache): Use it.
1138 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1140 * web.c: Remove out-of-date comment.
1142 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1144 * expr.c (fixup_args_size_notes): Check that any existing
1145 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1146 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1147 (emit_single_push_insn): ...here.
1149 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1151 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1152 (const_vector_encoded_nelts): New function.
1153 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1154 (const_vector_int_elt, const_vector_elt): Declare.
1155 * emit-rtl.c (const_vector_int_elt_1): New function.
1156 (const_vector_elt): Likewise.
1157 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1158 of CONST_VECTOR_ELT.
1160 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1162 * expr.c: Include rtx-vector-builder.h.
1163 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1164 directly on the tree encoding.
1165 (const_vector_from_tree): Likewise.
1166 * optabs.c: Include rtx-vector-builder.h.
1167 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1168 sequence of "u" values.
1169 * vec-perm-indices.c: Include rtx-vector-builder.h.
1170 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1171 directly on the vec_perm_indices encoding.
1173 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1175 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1176 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1177 * rtx-vector-builder.h: New file.
1178 * rtx-vector-builder.c: Likewise.
1179 * rtl.h (rtx_def::u2): Add a const_vector field.
1180 (CONST_VECTOR_NPATTERNS): New macro.
1181 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1182 (CONST_VECTOR_DUPLICATE_P): Likewise.
1183 (CONST_VECTOR_STEPPED_P): Likewise.
1184 (CONST_VECTOR_ENCODED_ELT): Likewise.
1185 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1186 (unwrap_const_vec_duplicate): Likewise.
1187 (const_vec_series_p): Check for a non-duplicated vector encoding.
1188 Say that the function only returns true for integer vectors.
1189 * emit-rtl.c: Include rtx-vector-builder.h.
1190 (gen_const_vec_duplicate_1): Delete.
1191 (gen_const_vector): Call gen_const_vec_duplicate instead of
1192 gen_const_vec_duplicate_1.
1193 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1194 (gen_const_vec_duplicate): Use rtx_vector_builder.
1195 (gen_const_vec_series): Likewise.
1196 (gen_rtx_CONST_VECTOR): Likewise.
1197 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1198 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1199 Build a new vector rather than modifying a CONST_VECTOR in-place.
1200 (handle_special_swappables): Update call accordingly.
1201 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1202 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1203 Build a new vector rather than modifying a CONST_VECTOR in-place.
1204 (handle_special_swappables): Update call accordingly.
1206 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1208 * simplify-rtx.c (simplify_const_binary_operation): Use
1209 CONST_VECTOR_ELT instead of XVECEXP.
1211 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1213 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1214 the selector elements to be different from the data elements
1215 if the selector is a VECTOR_CST.
1216 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1217 ssizetype for the selector.
1219 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1221 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1222 before testing each element individually.
1223 * tree-vect-generic.c (lower_vec_perm): Likewise.
1225 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1227 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1228 * selftest-run-tests.c (selftest::run_tests): Call it.
1229 * vector-builder.h (vector_builder::operator ==): New function.
1230 (vector_builder::operator !=): Likewise.
1231 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1232 (vec_perm_indices::all_from_input_p): New function.
1233 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1234 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1235 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1236 instead of reading the VECTOR_CST directly. Detect whether both
1237 vector inputs are the same before constructing the vec_perm_indices,
1238 and update the number of inputs argument accordingly. Use the
1239 utility functions added above. Only construct sel2 if we need to.
1241 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1243 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1244 the broadcast of the low byte.
1245 (expand_mult_highpart): Use an explicit encoding for the permutes.
1246 * optabs-query.c (can_mult_highpart_p): Likewise.
1247 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1248 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1249 (vectorizable_bswap): Likewise.
1250 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1251 explicit encoding for the power-of-2 permutes.
1252 (vect_permute_store_chain): Likewise.
1253 (vect_grouped_load_supported): Likewise.
1254 (vect_permute_load_chain): Likewise.
1256 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1258 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1259 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1260 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1261 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1262 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1263 (vect_gen_perm_mask_any): Likewise.
1265 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1267 * int-vector-builder.h: New file.
1268 * vec-perm-indices.h: Include int-vector-builder.h.
1269 (vec_perm_indices): Redefine as an int_vector_builder.
1270 (auto_vec_perm_indices): Delete.
1271 (vec_perm_builder): Redefine as a stand-alone class.
1272 (vec_perm_indices::vec_perm_indices): New function.
1273 (vec_perm_indices::clamp): Likewise.
1274 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1275 (vec_perm_indices::new_vector): New function.
1276 (vec_perm_indices::new_expanded_vector): Update for new
1277 vec_perm_indices class.
1278 (vec_perm_indices::rotate_inputs): New function.
1279 (vec_perm_indices::all_in_range_p): Operate directly on the
1280 encoded form, without computing elided elements.
1281 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1282 encoding. Update for new vec_perm_indices class.
1283 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1284 the given vec_perm_builder.
1285 (expand_vec_perm_var): Update vec_perm_builder constructor.
1286 (expand_mult_highpart): Use vec_perm_builder instead of
1287 auto_vec_perm_indices.
1288 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1289 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1290 or double series encoding as appropriate.
1291 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1292 vec_perm_indices instead of auto_vec_perm_indices.
1293 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1294 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1295 (vect_permute_store_chain): Likewise.
1296 (vect_grouped_load_supported): Likewise.
1297 (vect_permute_load_chain): Likewise.
1298 (vect_shift_permute_load_chain): Likewise.
1299 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1300 (vect_transform_slp_perm_load): Likewise.
1301 (vect_schedule_slp_instance): Likewise.
1302 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1303 (vectorizable_mask_load_store): Likewise.
1304 (vectorizable_bswap): Likewise.
1305 (vectorizable_store): Likewise.
1306 (vectorizable_load): Likewise.
1307 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1308 vec_perm_indices instead of auto_vec_perm_indices. Use
1309 tree_to_vec_perm_builder to read the vector from a tree.
1310 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1311 vec_perm_builder instead of a vec_perm_indices.
1312 (have_whole_vector_shift): Use vec_perm_builder and
1313 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1314 truncation to calc_vec_perm_mask_for_shift.
1315 (vect_create_epilog_for_reduction): Likewise.
1316 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1317 from auto_vec_perm_indices to vec_perm_indices.
1318 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1319 instead of changing individual elements.
1320 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1321 the vector in d.perm.
1322 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1323 from auto_vec_perm_indices to vec_perm_indices.
1324 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1325 instead of changing individual elements.
1326 (arm_vectorize_vec_perm_const): Use new_vector to install
1327 the vector in d.perm.
1328 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1329 Update vec_perm_builder constructor.
1330 (rs6000_expand_interleave): Likewise.
1331 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1332 (rs6000_expand_interleave): Likewise.
1334 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1336 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1337 to qimode could truncate the indices.
1338 * optabs.c (expand_vec_perm_var): Likewise.
1340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1342 * Makefile.in (OBJS): Add vec-perm-indices.o.
1343 * vec-perm-indices.h: New file.
1344 * vec-perm-indices.c: Likewise.
1345 * target.h (vec_perm_indices): Replace with a forward class
1347 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1348 * optabs.h: Include vec-perm-indices.h.
1349 (expand_vec_perm): Delete.
1350 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1351 (expand_vec_perm_const): Declare.
1352 * target.def (vec_perm_const_ok): Replace with...
1353 (vec_perm_const): ...this new hook.
1354 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1355 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1356 * doc/tm.texi: Regenerate.
1357 * optabs.def (vec_perm_const): Delete.
1358 * doc/md.texi (vec_perm_const): Likewise.
1359 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1360 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1361 expand_vec_perm for constant permutation vectors. Assert that
1362 the mode of variable permutation vectors is the integer equivalent
1363 of the mode that is being permuted.
1364 * optabs-query.h (selector_fits_mode_p): Declare.
1365 * optabs-query.c: Include vec-perm-indices.h.
1366 (selector_fits_mode_p): New function.
1367 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1368 is defined, instead of checking whether the vec_perm_const_optab
1369 exists. Use targetm.vectorize.vec_perm_const instead of
1370 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1371 fit in the vector mode before using a variable permute.
1372 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1373 vec_perm_indices instead of an rtx.
1374 (expand_vec_perm): Replace with...
1375 (expand_vec_perm_const): ...this new function. Take the selector
1376 as a vec_perm_indices rather than an rtx. Also take the mode of
1377 the selector. Update call to shift_amt_for_vec_perm_mask.
1378 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1379 Use vec_perm_indices::new_expanded_vector to expand the original
1380 selector into bytes. Check whether the indices fit in the vector
1381 mode before using a variable permute.
1382 (expand_vec_perm_var): Make global.
1383 (expand_mult_highpart): Use expand_vec_perm_const.
1384 * fold-const.c: Includes vec-perm-indices.h.
1385 * tree-ssa-forwprop.c: Likewise.
1386 * tree-vect-data-refs.c: Likewise.
1387 * tree-vect-generic.c: Likewise.
1388 * tree-vect-loop.c: Likewise.
1389 * tree-vect-slp.c: Likewise.
1390 * tree-vect-stmts.c: Likewise.
1391 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1393 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1394 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1395 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1396 (aarch64_vectorize_vec_perm_const): ...this new function.
1397 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1398 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1399 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1400 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1401 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1402 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1403 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1405 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1406 check for NEON modes.
1407 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1408 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1409 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1410 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1412 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1413 the old VEC_PERM_CONST conditions.
1414 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1415 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1416 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1417 (ia64_vectorize_vec_perm_const_ok): Merge into...
1418 (ia64_vectorize_vec_perm_const): ...this new function.
1419 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1420 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1421 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1422 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1423 * config/mips/mips.c (mips_expand_vec_perm_const)
1424 (mips_vectorize_vec_perm_const_ok): Merge into...
1425 (mips_vectorize_vec_perm_const): ...this new function.
1426 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1427 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1428 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1429 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1430 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1431 (rs6000_expand_vec_perm_const): Delete.
1432 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1434 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1435 (altivec_expand_vec_perm_const_le): Take each operand individually.
1436 Operate on constant selectors rather than rtxes.
1437 (altivec_expand_vec_perm_const): Likewise. Update call to
1438 altivec_expand_vec_perm_const_le.
1439 (rs6000_expand_vec_perm_const): Delete.
1440 (rs6000_vectorize_vec_perm_const_ok): Delete.
1441 (rs6000_vectorize_vec_perm_const): New function.
1442 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1443 an element count and rtx array.
1444 (rs6000_expand_extract_even): Update call accordingly.
1445 (rs6000_expand_interleave): Likewise.
1446 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1447 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1448 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1449 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1450 (rs6000_expand_vec_perm_const): Delete.
1451 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1452 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1453 (altivec_expand_vec_perm_const_le): Take each operand individually.
1454 Operate on constant selectors rather than rtxes.
1455 (altivec_expand_vec_perm_const): Likewise. Update call to
1456 altivec_expand_vec_perm_const_le.
1457 (rs6000_expand_vec_perm_const): Delete.
1458 (rs6000_vectorize_vec_perm_const_ok): Delete.
1459 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1460 reference to the SPE evmerge intructions.
1461 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1462 an element count and rtx array.
1463 (rs6000_expand_extract_even): Update call accordingly.
1464 (rs6000_expand_interleave): Likewise.
1465 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1466 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1468 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1470 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1472 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1473 vector mode and that that mode matches the mode of the data
1475 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1476 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1477 directly using expand_vec_perm_1 when forcing selectors into
1479 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1481 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1483 * optabs-query.h (can_vec_perm_p): Delete.
1484 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1485 * optabs-query.c (can_vec_perm_p): Split into...
1486 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1487 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1488 particular selector is valid.
1489 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1490 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1491 (vect_grouped_load_supported): Likewise.
1492 (vect_shift_permute_load_chain): Likewise.
1493 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1494 (vect_transform_slp_perm_load): Likewise.
1495 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1496 (vectorizable_bswap): Likewise.
1497 (vect_gen_perm_mask_checked): Likewise.
1498 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1499 implementations of variable permutation vectors into account
1500 when deciding which selector to use.
1501 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1502 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1503 with a false third argument.
1504 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1505 to test whether the constant selector is valid and can_vec_perm_var_p
1506 to test whether a variable selector is valid.
1508 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1510 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1511 * optabs-query.c (can_vec_perm_p): Likewise.
1512 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1513 instead of vec_perm_indices.
1514 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1515 (vect_gen_perm_mask_checked): Likewise,
1516 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1517 (vect_gen_perm_mask_checked): Likewise,
1519 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1521 * optabs-query.h (qimode_for_vec_perm): Declare.
1522 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1523 (qimode_for_vec_perm): ...this new function.
1524 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1526 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1528 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1529 does not have a conditional at the top.
1531 2018-01-02 Richard Biener <rguenther@suse.de>
1533 * ipa-inline.c (big_speedup_p): Fix expression.
1535 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1538 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1541 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1545 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1546 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1547 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1548 cond_taken_branch_cost 3->4.
1550 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1552 PR tree-optimization/83581
1553 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1554 TODO_cleanup_cfg if any changes have been made.
1557 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1558 convert_modes if target mode has the right side, but different mode
1562 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1563 last argument when extracting from CONCAT. If either from_real or
1564 from_imag is NULL, use expansion through memory. If result is not
1565 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1566 the parts directly to inner mode, if even that fails, use expansion
1570 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1571 check for bswap in mode rather than HImode and use that in expand_unop
1574 Copyright (C) 2018 Free Software Foundation, Inc.
1576 Copying and distribution of this file, with or without modification,
1577 are permitted in any medium without royalty provided the copyright
1578 notice and this notice are preserved.