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 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6 they are variable or constant sized.
7 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
8 slots for constant-sized data.
10 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
11 Alan Hayward <alan.hayward@arm.com>
12 David Sherwood <david.sherwood@arm.com>
14 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
15 handling COND_EXPRs with boolean comparisons, try to find a better
16 basis for the mask type than the boolean itself.
18 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
20 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
21 is calculated and how it can be overridden.
22 * genmodes.c (max_bitsize_mode_any_mode): New variable.
23 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
25 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
28 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
29 Alan Hayward <alan.hayward@arm.com>
30 David Sherwood <david.sherwood@arm.com>
32 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
33 Remove the mode argument.
34 (aarch64_simd_valid_immediate): Remove the mode and inverse
36 * config/aarch64/iterators.md (bitsize): New iterator.
37 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
38 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
39 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
40 aarch64_simd_valid_immediate.
41 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
42 (aarch64_reg_or_bic_imm): Likewise.
43 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
44 with an insn_type enum and msl with a modifier_type enum.
45 Replace element_width with a scalar_mode. Change the shift
46 to unsigned int. Add constructors for scalar_float_mode and
47 scalar_int_mode elements.
48 (aarch64_vect_float_const_representable_p): Delete.
49 (aarch64_can_const_movi_rtx_p)
50 (aarch64_simd_scalar_immediate_valid_for_move)
51 (aarch64_simd_make_constant): Update call to
52 aarch64_simd_valid_immediate.
53 (aarch64_advsimd_valid_immediate_hs): New function.
54 (aarch64_advsimd_valid_immediate): Likewise.
55 (aarch64_simd_valid_immediate): Remove mode and inverse
56 arguments. Rewrite to use the above. Use const_vec_duplicate_p
57 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
58 and aarch64_float_const_representable_p on the result.
59 (aarch64_output_simd_mov_immediate): Remove mode argument.
60 Update call to aarch64_simd_valid_immediate and use of
62 (aarch64_output_scalar_simd_mov_immediate): Update call
65 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
66 Alan Hayward <alan.hayward@arm.com>
67 David Sherwood <david.sherwood@arm.com>
69 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
70 (mode_nunits): Likewise CONST_MODE_NUNITS.
71 * machmode.def (ADJUST_NUNITS): Document.
72 * genmodes.c (mode_data::need_nunits_adj): New field.
73 (blank_mode): Update accordingly.
74 (adj_nunits): New variable.
75 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
77 (emit_mode_size_inline): Set need_bytesize_adj for all modes
79 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
80 listed in adj_nunits. Don't emit case statements for such modes.
81 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
82 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
83 nothing if adj_nunits is nonnull.
84 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
85 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
86 (emit_mode_fbit): Update use of print_maybe_const_decl.
87 (emit_move_size): Likewise. Treat the array as non-const
89 (emit_mode_adjustments): Handle adj_nunits.
91 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
93 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
94 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
95 (VECTOR_MODES): Use it.
96 (make_vector_modes): Take the prefix as an argument.
98 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
99 Alan Hayward <alan.hayward@arm.com>
100 David Sherwood <david.sherwood@arm.com>
102 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
103 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
104 for MODE_VECTOR_BOOL.
105 * machmode.def (VECTOR_BOOL_MODE): Document.
106 * genmodes.c (VECTOR_BOOL_MODE): New macro.
107 (make_vector_bool_mode): New function.
108 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
110 * lto-streamer-in.c (lto_input_mode_table): Likewise.
111 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
113 * stor-layout.c (int_mode_for_mode): Likewise.
114 * tree.c (build_vector_type_for_mode): Likewise.
115 * varasm.c (output_constant_pool_2): Likewise.
116 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
117 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
118 for MODE_VECTOR_BOOL.
119 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
120 of mode class checks.
121 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
122 instead of a list of mode class checks.
123 (expand_vector_scalar_condition): Likewise.
124 (type_for_widest_vector_mode): Handle BImode as an inner mode.
126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
127 Alan Hayward <alan.hayward@arm.com>
128 David Sherwood <david.sherwood@arm.com>
130 * machmode.h (mode_size): Change from unsigned short to
132 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
133 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
134 or if measurement_type is not polynomial.
135 (fixed_size_mode::includes_p): Check for constant-sized modes.
136 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
137 return a poly_uint16 rather than an unsigned short.
138 (emit_mode_size): Change the type of mode_size from unsigned short
139 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
140 (emit_mode_adjustments): Cope with polynomial vector sizes.
141 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
143 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
145 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
146 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
147 * caller-save.c (setup_save_areas): Likewise.
148 (replace_reg_with_saved_mem): Likewise.
149 * calls.c (emit_library_call_value_1): Likewise.
150 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
151 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
152 (gen_lowpart_for_combine): Likewise.
153 * convert.c (convert_to_integer_1): Likewise.
154 * cse.c (equiv_constant, cse_insn): Likewise.
155 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
156 (cselib_subst_to_values): Likewise.
157 * dce.c (word_dce_process_block): Likewise.
158 * df-problems.c (df_word_lr_mark_ref): Likewise.
159 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
160 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
161 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
162 (rtl_for_decl_location): Likewise.
163 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
164 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
165 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
166 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
167 (expand_expr_real_1): Likewise.
168 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
169 (pad_below): Likewise.
170 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
171 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
172 * ira.c (get_subreg_tracking_sizes): Likewise.
173 * ira-build.c (ira_create_allocno_objects): Likewise.
174 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
175 (ira_sort_regnos_for_alter_reg): Likewise.
176 * ira-costs.c (record_operand_costs): Likewise.
177 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
178 (resolve_simple_move): Likewise.
179 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
180 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
181 (lra_constraints): Likewise.
182 (CONST_POOL_OK_P): Reject variable-sized modes.
183 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
184 (add_pseudo_to_slot, lra_spill): Likewise.
185 * omp-low.c (omp_clause_aligned_alignment): Likewise.
186 * optabs-query.c (get_best_extraction_insn): Likewise.
187 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
188 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
189 (expand_mult_highpart, valid_multiword_target_p): Likewise.
190 * recog.c (offsettable_address_addr_space_p): Likewise.
191 * regcprop.c (maybe_mode_change): Likewise.
192 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
193 * regrename.c (build_def_use): Likewise.
194 * regstat.c (dump_reg_info): Likewise.
195 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
196 (find_reloads, find_reloads_subreg_address): Likewise.
197 * reload1.c (eliminate_regs_1): Likewise.
198 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
199 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
200 (simplify_binary_operation_1, simplify_subreg): Likewise.
201 * targhooks.c (default_function_arg_padding): Likewise.
202 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
203 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
204 (verify_gimple_assign_ternary): Likewise.
205 * tree-inline.c (estimate_move_cost): Likewise.
206 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
207 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
208 (get_address_cost_ainc): Likewise.
209 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
210 (vect_supportable_dr_alignment): Likewise.
211 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
212 (vectorizable_reduction): Likewise.
213 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
214 (vectorizable_operation, vectorizable_load): Likewise.
215 * tree.c (build_same_sized_truth_vector_type): Likewise.
216 * valtrack.c (cleanup_auto_inc_dec): Likewise.
217 * var-tracking.c (emit_note_insn_var_location): Likewise.
218 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
219 (ADDR_VEC_ALIGN): Likewise.
221 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
222 Alan Hayward <alan.hayward@arm.com>
223 David Sherwood <david.sherwood@arm.com>
225 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
227 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
228 or if measurement_type is polynomial.
229 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
230 * combine.c (make_extraction): Likewise.
231 * dse.c (find_shift_sequence): Likewise.
232 * dwarf2out.c (mem_loc_descriptor): Likewise.
233 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
234 (extract_bit_field, extract_low_bits): Likewise.
235 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
236 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
237 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
238 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
239 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
240 * reload.c (find_reloads): Likewise.
241 * reload1.c (alter_reg): Likewise.
242 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
243 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
244 * tree-if-conv.c (predicate_mem_writes): Likewise.
245 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
246 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
247 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
248 * valtrack.c (dead_debug_insert_temp): Likewise.
249 * varasm.c (mergeable_constant_section): Likewise.
250 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
253 Alan Hayward <alan.hayward@arm.com>
254 David Sherwood <david.sherwood@arm.com>
256 * expr.c (expand_assignment): Cope with polynomial mode sizes
257 when assigning to a CONCAT.
259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
260 Alan Hayward <alan.hayward@arm.com>
261 David Sherwood <david.sherwood@arm.com>
263 * machmode.h (mode_precision): Change from unsigned short to
265 (mode_to_precision): Return a poly_uint16 rather than an unsigned
267 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
268 or if measurement_type is not polynomial.
269 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
270 in which the mode is already known to be a scalar_int_mode.
271 * genmodes.c (emit_mode_precision): Change the type of mode_precision
272 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
274 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
275 for GET_MODE_PRECISION.
276 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
277 for GET_MODE_PRECISION.
278 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
280 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
281 (expand_field_assignment, make_extraction): Likewise.
282 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
283 (get_last_value): Likewise.
284 * convert.c (convert_to_integer_1): Likewise.
285 * cse.c (cse_insn): Likewise.
286 * expr.c (expand_expr_real_1): Likewise.
287 * lra-constraints.c (simplify_operand_subreg): Likewise.
288 * optabs-query.c (can_atomic_load_p): Likewise.
289 * optabs.c (expand_atomic_load): Likewise.
290 (expand_atomic_store): Likewise.
291 * ree.c (combine_reaching_defs): Likewise.
292 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
293 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
294 * tree.h (type_has_mode_precision_p): Likewise.
295 * ubsan.c (instrument_si_overflow): Likewise.
297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
298 Alan Hayward <alan.hayward@arm.com>
299 David Sherwood <david.sherwood@arm.com>
301 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
302 polynomial numbers of units.
303 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
304 (valid_vector_subparts_p): New function.
305 (build_vector_type): Remove temporary shim and take the number
306 of units as a poly_uint64 rather than an int.
307 (build_opaque_vector_type): Take the number of units as a
308 poly_uint64 rather than an int.
309 * tree.c (build_vector_from_ctor): Handle polynomial
310 TYPE_VECTOR_SUBPARTS.
311 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
312 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
313 (build_vector_from_val): If the number of units is variable,
314 use build_vec_duplicate_cst for constant operands and
315 VEC_DUPLICATE_EXPR otherwise.
316 (make_vector_type): Remove temporary is_constant ().
317 (build_vector_type, build_opaque_vector_type): Take the number of
318 units as a poly_uint64 rather than an int.
319 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
321 * cfgexpand.c (expand_debug_expr): Likewise.
322 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
323 (store_constructor, expand_expr_real_1): Likewise.
324 (const_scalar_mask_from_tree): Likewise.
325 * fold-const-call.c (fold_const_reduction): Likewise.
326 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
327 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
328 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
329 (fold_relational_const): Likewise.
330 (native_interpret_vector): Likewise. Change the size from an
331 int to an unsigned int.
332 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
333 TYPE_VECTOR_SUBPARTS.
334 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
335 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
336 duplicating a non-constant operand into a variable-length vector.
337 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
338 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
339 * ipa-icf.c (sem_variable::equals): Likewise.
340 * match.pd: Likewise.
341 * omp-simd-clone.c (simd_clone_subparts): Likewise.
342 * print-tree.c (print_node): Likewise.
343 * stor-layout.c (layout_type): Likewise.
344 * targhooks.c (default_builtin_vectorization_cost): Likewise.
345 * tree-cfg.c (verify_gimple_comparison): Likewise.
346 (verify_gimple_assign_binary): Likewise.
347 (verify_gimple_assign_ternary): Likewise.
348 (verify_gimple_assign_single): Likewise.
349 * tree-pretty-print.c (dump_generic_node): Likewise.
350 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
351 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
352 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
353 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
354 (vect_shift_permute_load_chain): Likewise.
355 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
356 (expand_vector_condition, optimize_vector_constructor): Likewise.
357 (lower_vec_perm, get_compute_type): Likewise.
358 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
359 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
360 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
361 (vect_recog_mask_conversion_pattern): Likewise.
362 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
363 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
364 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
365 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
366 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
367 (vectorizable_shift, vectorizable_operation, vectorizable_store)
368 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
369 (supportable_widening_operation): Likewise.
370 (supportable_narrowing_operation): Likewise.
371 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
373 * varasm.c (output_constant): Likewise.
375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
376 Alan Hayward <alan.hayward@arm.com>
377 David Sherwood <david.sherwood@arm.com>
379 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
380 so that both the length == 3 and length != 3 cases set up their
381 own permute vectors. Add comments explaining why we know the
382 number of elements is constant.
383 (vect_permute_load_chain): Likewise.
385 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
386 Alan Hayward <alan.hayward@arm.com>
387 David Sherwood <david.sherwood@arm.com>
389 * machmode.h (mode_nunits): Change from unsigned char to
391 (ONLY_FIXED_SIZE_MODES): New macro.
392 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
393 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
394 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
396 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
397 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
398 or if measurement_type is not polynomial.
399 * genmodes.c (ZERO_COEFFS): New macro.
400 (emit_mode_nunits_inline): Make mode_nunits_inline return a
402 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
403 Use ZERO_COEFFS when emitting initializers.
404 * data-streamer.h (bp_pack_poly_value): New function.
405 (bp_unpack_poly_value): Likewise.
406 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
408 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
410 * tree.c (make_vector_type): Remove temporary shim and make
411 the real function take the number of units as a poly_uint64
413 (build_vector_type_for_mode): Handle polynomial nunits.
414 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
415 * emit-rtl.c (const_vec_series_p_1): Likewise.
416 (gen_rtx_CONST_VECTOR): Likewise.
417 * fold-const.c (test_vec_duplicate_folding): Likewise.
418 * genrecog.c (validate_pattern): Likewise.
419 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
420 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
421 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
422 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
423 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
424 * rtlanal.c (subreg_get_info): Likewise.
425 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
426 (vect_grouped_load_supported): Likewise.
427 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
428 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
429 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
430 (simplify_const_unary_operation, simplify_binary_operation_1)
431 (simplify_const_binary_operation, simplify_ternary_operation)
432 (test_vector_ops_duplicate, test_vector_ops): Likewise.
433 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
434 instead of CONST_VECTOR_NUNITS.
435 * varasm.c (output_constant_pool_2): Likewise.
436 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
437 explicit-encoded elements in the XVEC for variable-length vectors.
439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
441 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
444 Alan Hayward <alan.hayward@arm.com>
445 David Sherwood <david.sherwood@arm.com>
447 * coretypes.h (fixed_size_mode): Declare.
448 (fixed_size_mode_pod): New typedef.
449 * builtins.h (target_builtins::x_apply_args_mode)
450 (target_builtins::x_apply_result_mode): Change type to
452 * builtins.c (apply_args_size, apply_result_size, result_vector)
453 (expand_builtin_apply_args_1, expand_builtin_apply)
454 (expand_builtin_return): Update accordingly.
456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
458 * cse.c (hash_rtx_cb): Hash only the encoded elements.
459 * cselib.c (cselib_hash_rtx): Likewise.
460 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
461 CONST_VECTOR encoding.
463 2017-01-03 Jakub Jelinek <jakub@redhat.com>
464 Jeff Law <law@redhat.com>
467 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
468 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
469 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
470 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
473 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
474 explicitly probe *sp in a noreturn function if there were any callee
475 register saves or frame pointer is needed.
477 2018-01-03 Jakub Jelinek <jakub@redhat.com>
480 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
481 BLKmode for ternary, binary or unary expressions.
484 * var-tracking.c (delete_vta_debug_insn): New inline function.
485 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
486 insns from get_insns () to NULL instead of each bb separately.
487 Use delete_vta_debug_insn. No longer static.
488 (vt_debug_insns_local, variable_tracking_main_1): Adjust
489 delete_vta_debug_insns callers.
490 * rtl.h (delete_vta_debug_insns): Declare.
491 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
492 instead of variable_tracking_main.
494 2018-01-03 Martin Sebor <msebor@redhat.com>
497 * doc/extend.texi (attribute const): Fix a typo.
498 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
499 issuing -Wsuggest-attribute for void functions.
501 2018-01-03 Martin Sebor <msebor@redhat.com>
503 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
504 offset_int::from instead of wide_int::to_shwi.
505 (maybe_diag_overlap): Remove assertion.
506 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
507 * gimple-ssa-sprintf.c (format_directive): Same.
508 (parse_directive): Same.
509 (sprintf_dom_walker::compute_format_length): Same.
510 (try_substitute_return_value): Same.
512 2017-01-03 Jeff Law <law@redhat.com>
515 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
516 non-constant residual for zero at runtime and avoid probing in
517 that case. Reorganize code for trailing problem to mirror handling
520 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
522 PR tree-optimization/83501
523 * tree-ssa-strlen.c (get_string_cst): New.
524 (handle_char_store): Call get_string_cst.
526 2018-01-03 Martin Liska <mliska@suse.cz>
528 PR tree-optimization/83593
529 * tree-ssa-strlen.c: Include tree-cfg.h.
530 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
531 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
532 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
534 (strlen_dom_walker::before_dom_children): Call
535 gimple_purge_dead_eh_edges. Dump tranformation with details
537 (strlen_dom_walker::before_dom_children): Update call by adding
538 new argument cleanup_eh.
539 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
541 2018-01-03 Martin Liska <mliska@suse.cz>
544 * cif-code.def (VARIADIC_THUNK): New enum value.
545 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
548 2018-01-03 Jan Beulich <jbeulich@suse.com>
550 * sse.md (mov<mode>_internal): Tighten condition for when to use
551 vmovdqu<ssescalarsize> for TI and OI modes.
553 2018-01-03 Jakub Jelinek <jakub@redhat.com>
555 Update copyright years.
557 2018-01-03 Martin Liska <mliska@suse.cz>
560 * ipa-visibility.c (function_and_variable_visibility): Skip
561 functions with noipa attribure.
563 2018-01-03 Jakub Jelinek <jakub@redhat.com>
565 * gcc.c (process_command): Update copyright notice dates.
566 * gcov-dump.c (print_version): Ditto.
567 * gcov.c (print_version): Ditto.
568 * gcov-tool.c (print_version): Ditto.
569 * gengtype.c (create_file): Ditto.
570 * doc/cpp.texi: Bump @copying's copyright year.
571 * doc/cppinternals.texi: Ditto.
572 * doc/gcc.texi: Ditto.
573 * doc/gccint.texi: Ditto.
574 * doc/gcov.texi: Ditto.
575 * doc/install.texi: Ditto.
576 * doc/invoke.texi: Ditto.
578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
580 * vector-builder.h (vector_builder::m_full_nelts): Change from
581 unsigned int to poly_uint64.
582 (vector_builder::full_nelts): Update prototype accordingly.
583 (vector_builder::new_vector): Likewise.
584 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
585 (vector_builder::operator ==): Likewise.
586 (vector_builder::finalize): Likewise.
587 * int-vector-builder.h (int_vector_builder::int_vector_builder):
588 Take the number of elements as a poly_uint64 rather than an
590 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
591 from unsigned int to poly_uint64.
592 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
593 (vec_perm_indices::new_vector): Likewise.
594 (vec_perm_indices::length): Likewise.
595 (vec_perm_indices::nelts_per_input): Likewise.
596 (vec_perm_indices::input_nelts): Likewise.
597 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
598 number of elements per input as a poly_uint64 rather than an
599 unsigned int. Use the original encoding for variable-length
600 vectors, rather than clamping each individual element.
601 For the second and subsequent elements in each pattern,
602 clamp the step and base before clamping their sum.
603 (vec_perm_indices::series_p): Handle polynomial element counts.
604 (vec_perm_indices::all_in_range_p): Likewise.
605 (vec_perm_indices_to_tree): Likewise.
606 (vec_perm_indices_to_rtx): Likewise.
607 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
608 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
609 (tree_vector_builder::new_binary_operation): Handle polynomial
610 element counts. Return false if we need to know the number
611 of elements at compile time.
612 * fold-const.c (fold_vec_perm): Punt if the number of elements
613 isn't known at compile time.
615 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
617 * vec-perm-indices.h (vec_perm_builder): Change element type
618 from HOST_WIDE_INT to poly_int64.
619 (vec_perm_indices::element_type): Update accordingly.
620 (vec_perm_indices::clamp): Handle polynomial element_types.
621 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
622 (vec_perm_indices::all_in_range_p): Likewise.
623 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
625 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
626 polynomial vec_perm_indices element types.
627 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
628 * fold-const.c (fold_vec_perm): Likewise.
629 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
630 * tree-vect-generic.c (lower_vec_perm): Likewise.
631 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
632 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
633 element type to HOST_WIDE_INT.
635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
636 Alan Hayward <alan.hayward@arm.com>
637 David Sherwood <david.sherwood@arm.com>
639 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
640 rather than an int. Use plus_constant.
641 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
642 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
645 Alan Hayward <alan.hayward@arm.com>
646 David Sherwood <david.sherwood@arm.com>
648 * calls.c (emit_call_1, expand_call): Change struct_value_size from
649 a HOST_WIDE_INT to a poly_int64.
651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
652 Alan Hayward <alan.hayward@arm.com>
653 David Sherwood <david.sherwood@arm.com>
655 * calls.c (load_register_parameters): Cope with polynomial
656 mode sizes. Require a constant size for BLKmode parameters
657 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
658 forces a parameter to be padded at the lsb end in order to
659 fill a complete number of words, require the parameter size
660 to be ordered wrt UNITS_PER_WORD.
662 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
663 Alan Hayward <alan.hayward@arm.com>
664 David Sherwood <david.sherwood@arm.com>
666 * reload1.c (spill_stack_slot_width): Change element type
667 from unsigned int to poly_uint64_pod.
668 (alter_reg): Treat mode sizes as polynomial.
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 * reload.c (complex_word_subreg_p): New function.
675 (reload_inner_reg_of_subreg, push_reload): Use it.
677 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
678 Alan Hayward <alan.hayward@arm.com>
679 David Sherwood <david.sherwood@arm.com>
681 * lra-constraints.c (process_alt_operands): Reject matched
682 operands whose sizes aren't ordered.
683 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
690 that the mode size is in the set {1, 2, 4, 8, 16}.
692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
693 Alan Hayward <alan.hayward@arm.com>
694 David Sherwood <david.sherwood@arm.com>
696 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
697 Use plus_constant instead of gen_rtx_PLUS.
699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
700 Alan Hayward <alan.hayward@arm.com>
701 David Sherwood <david.sherwood@arm.com>
703 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
704 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
705 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
706 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
707 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
708 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
709 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
710 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
711 * config/i386/i386.c (ix86_push_rounding): ...this new function.
712 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
714 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
715 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
716 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
717 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
718 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
719 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
720 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
721 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
722 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
723 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
725 * expr.c (emit_move_resolve_push): Treat the input and result
726 of PUSH_ROUNDING as a poly_int64.
727 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
728 (emit_push_insn): Likewise.
729 * lra-eliminations.c (mark_not_eliminable): Likewise.
730 * recog.c (push_operand): Likewise.
731 * reload1.c (elimination_effects): Likewise.
732 * rtlanal.c (nonzero_bits1): Likewise.
733 * calls.c (store_one_arg): Likewise. Require the padding to be
734 known at compile time.
736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
737 Alan Hayward <alan.hayward@arm.com>
738 David Sherwood <david.sherwood@arm.com>
740 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
741 Use plus_constant instead of gen_rtx_PLUS.
743 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
744 Alan Hayward <alan.hayward@arm.com>
745 David Sherwood <david.sherwood@arm.com>
747 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
750 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
751 Alan Hayward <alan.hayward@arm.com>
752 David Sherwood <david.sherwood@arm.com>
754 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
755 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
756 via stack temporaries. Treat the mode size as polynomial too.
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_2): When handling conversions involving
763 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
764 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
765 as a poly_uint64 too.
767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
768 Alan Hayward <alan.hayward@arm.com>
769 David Sherwood <david.sherwood@arm.com>
771 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
774 Alan Hayward <alan.hayward@arm.com>
775 David Sherwood <david.sherwood@arm.com>
777 * combine.c (can_change_dest_mode): Handle polynomial
778 REGMODE_NATURAL_SIZE.
779 * expmed.c (store_bit_field_1): Likewise.
780 * expr.c (store_constructor): Likewise.
781 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
782 and polynomial REGMODE_NATURAL_SIZE.
783 (gen_lowpart_common): Likewise.
784 * reginfo.c (record_subregs_of_mode): Likewise.
785 * rtlanal.c (read_modify_subreg_p): Likewise.
787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
788 Alan Hayward <alan.hayward@arm.com>
789 David Sherwood <david.sherwood@arm.com>
791 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
795 Alan Hayward <alan.hayward@arm.com>
796 David Sherwood <david.sherwood@arm.com>
798 * match.pd: Cope with polynomial numbers of vector elements.
800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
801 Alan Hayward <alan.hayward@arm.com>
802 David Sherwood <david.sherwood@arm.com>
804 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
805 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
812 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
813 (ipa_simd_modify_function_body): Likewise.
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 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
820 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
821 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
822 (expand_vector_condition, vector_element): Likewise.
823 (subparts_gt): New function.
824 (get_compute_type): Use subparts_gt.
825 (count_type_subparts): Delete.
826 (expand_vector_operations_1): Use subparts_gt instead of
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 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
834 (vect_compile_time_alias): ...this new function. Do the calculation
835 on poly_ints rather than trees.
836 (vect_prune_runtime_alias_test_list): Update call accordingly.
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 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
844 (vect_schedule_slp_instance): Likewise.
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_get_and_check_slp_defs): Reject
851 constant and extern definitions for variable-length vectors.
852 (vect_get_constant_vectors): Note that the number of units
853 is known to be constant.
855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
856 Alan Hayward <alan.hayward@arm.com>
857 David Sherwood <david.sherwood@arm.com>
859 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
860 of units as polynomial. Choose between WIDE and NARROW based
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 (simd_clone_subparts): New function.
868 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
871 Alan Hayward <alan.hayward@arm.com>
872 David Sherwood <david.sherwood@arm.com>
874 * tree-vect-stmts.c (vectorizable_call): Treat the number of
875 vectors as polynomial. Use build_index_vector for
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 (get_load_store_type): Treat the number of
883 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
884 for variable-length vectors.
885 (vectorizable_mask_load_store): Treat the number of units as
886 polynomial, asserting that it is constant if the condition has
887 already been enforced.
888 (vectorizable_store, vectorizable_load): Likewise.
890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
891 Alan Hayward <alan.hayward@arm.com>
892 David Sherwood <david.sherwood@arm.com>
894 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
895 of units as polynomial. Punt if we can't tell at compile time
896 which vector contains the final result.
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_induction): Treat the number
903 of units as polynomial. Punt on SLP inductions. Use an integer
904 VEC_SERIES_EXPR for variable-length integer reductions. Use a
905 cast of such a series for variable-length floating-point
908 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
909 Alan Hayward <alan.hayward@arm.com>
910 David Sherwood <david.sherwood@arm.com>
912 * tree.h (build_index_vector): Declare.
913 * tree.c (build_index_vector): New function.
914 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
915 of units as polynomial, forcibly converting it to a constant if
916 vectorizable_reduction has already enforced the condition.
917 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
918 to create a {1,2,3,...} vector.
919 (vectorizable_reduction): Treat the number of units as polynomial.
920 Choose vectype_in based on the largest scalar element size rather
921 than the smallest number of units. Enforce the restrictions
924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
925 Alan Hayward <alan.hayward@arm.com>
926 David Sherwood <david.sherwood@arm.com>
928 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
929 number of units as polynomial.
931 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
932 Alan Hayward <alan.hayward@arm.com>
933 David Sherwood <david.sherwood@arm.com>
935 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
936 * target.def (autovectorize_vector_sizes): Return the vector sizes
937 by pointer, using vector_sizes rather than a bitmask.
938 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
939 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
940 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
942 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
943 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
944 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
945 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
946 * omp-general.c (omp_max_vf): Likewise.
947 * omp-low.c (omp_clause_aligned_alignment): Likewise.
948 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
949 * tree-vect-loop.c (vect_analyze_loop): Likewise.
950 * tree-vect-slp.c (vect_slp_bb): Likewise.
951 * doc/tm.texi: Regenerate.
952 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
954 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
955 the vector size as a poly_uint64 rather than an unsigned int.
956 (current_vector_size): Change from an unsigned int to a poly_uint64.
957 (get_vectype_for_scalar_type): Update accordingly.
958 * tree.h (build_truth_vector_type): Take the size and number of
959 units as a poly_uint64 rather than an unsigned int.
960 (build_vector_type): Add a temporary overload that takes
961 the number of units as a poly_uint64 rather than an unsigned int.
962 * tree.c (make_vector_type): Likewise.
963 (build_truth_vector_type): Take the number of units as a poly_uint64
964 rather than an unsigned int.
966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
967 Alan Hayward <alan.hayward@arm.com>
968 David Sherwood <david.sherwood@arm.com>
970 * target.def (get_mask_mode): Take the number of units and length
971 as poly_uint64s rather than unsigned ints.
972 * targhooks.h (default_get_mask_mode): Update accordingly.
973 * targhooks.c (default_get_mask_mode): Likewise.
974 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
975 * doc/tm.texi: Regenerate.
977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
978 Alan Hayward <alan.hayward@arm.com>
979 David Sherwood <david.sherwood@arm.com>
981 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
982 * omp-general.c (omp_max_vf): Likewise.
983 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
984 (expand_omp_simd): Handle polynomial safelen.
985 * omp-low.c (omplow_simd_context): Add a default constructor.
986 (omplow_simd_context::max_vf): Change from int to poly_uint64.
987 (lower_rec_simd_input_clauses): Update accordingly.
988 (lower_rec_input_clauses): Likewise.
990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
991 Alan Hayward <alan.hayward@arm.com>
992 David Sherwood <david.sherwood@arm.com>
994 * tree-vectorizer.h (vect_nunits_for_cost): New function.
995 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
996 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
997 (vect_analyze_slp_cost): Likewise.
998 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
999 (vect_model_load_cost): Likewise.
1001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1002 Alan Hayward <alan.hayward@arm.com>
1003 David Sherwood <david.sherwood@arm.com>
1005 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1006 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1007 from an unsigned int * to a poly_uint64_pod *.
1008 (calculate_unrolling_factor): New function.
1009 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1012 Alan Hayward <alan.hayward@arm.com>
1013 David Sherwood <david.sherwood@arm.com>
1015 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1016 from an unsigned int to a poly_uint64.
1017 (_loop_vec_info::slp_unrolling_factor): Likewise.
1018 (_loop_vec_info::vectorization_factor): Change from an int
1020 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1021 (vect_get_num_vectors): New function.
1022 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1023 (vect_get_num_copies): Use vect_get_num_vectors.
1024 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1025 to an unsigned int *.
1026 (vect_analyze_data_refs): Change min_vf from an int * to a
1028 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1029 than an unsigned HOST_WIDE_INT.
1030 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1031 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1032 to an unsigned int *.
1033 (vect_analyze_data_ref_dependences): Likewise.
1034 (vect_compute_data_ref_alignment): Handle polynomial vf.
1035 (vect_enhance_data_refs_alignment): Likewise.
1036 (vect_prune_runtime_alias_test_list): Likewise.
1037 (vect_shift_permute_load_chain): Likewise.
1038 (vect_supportable_dr_alignment): Likewise.
1039 (dependence_distance_ge_vf): Take the vectorization factor as a
1040 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1041 (vect_analyze_data_refs): Change min_vf from an int * to a
1043 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1044 vfm1 as a poly_uint64 rather than an int. Make the same change
1045 for the returned bound_scalar.
1046 (vect_gen_vector_loop_niters): Handle polynomial vf.
1047 (vect_do_peeling): Likewise. Update call to
1048 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1049 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1051 * tree-vect-loop.c (vect_determine_vectorization_factor)
1052 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1053 (vect_get_known_peeling_cost): Likewise.
1054 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1055 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1056 (vect_transform_loop): Likewise. Use the lowest possible VF when
1057 updating the upper bounds of the loop.
1058 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1060 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1061 polynomial unroll factors.
1062 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1063 (vect_make_slp_decision): Likewise.
1064 (vect_supported_load_permutation_p): Likewise, and polynomial
1066 (vect_analyze_slp_cost): Handle polynomial vf.
1067 (vect_slp_analyze_node_operations): Likewise.
1068 (vect_slp_analyze_bb_1): Likewise.
1069 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1070 than an unsigned HOST_WIDE_INT.
1071 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1072 (vectorizable_load): Handle polynomial vf.
1073 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1075 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1078 Alan Hayward <alan.hayward@arm.com>
1079 David Sherwood <david.sherwood@arm.com>
1081 * match.pd: Handle bit operations involving three constants
1082 and try to fold one pair.
1084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1086 * tree-vect-loop-manip.c: Include gimple-fold.h.
1087 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1088 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1089 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1090 Add a path that uses a step of VF instead of 1, but disable it
1092 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1093 and niters_no_overflow parameters. Update calls to
1094 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1095 Create a new SSA name if the latter choses to use a ste other
1096 than zero, and return it via niters_vector_mult_vf_var.
1097 * tree-vect-loop.c (vect_transform_loop): Update calls to
1098 vect_do_peeling, vect_gen_vector_loop_niters and
1099 slpeel_make_loop_iterate_ntimes.
1100 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1101 (vect_gen_vector_loop_niters): Update declarations after above changes.
1103 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1105 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1106 128-bit round to integer instructions.
1107 (ceil<mode>2): Likewise.
1108 (btrunc<mode>2): Likewise.
1109 (round<mode>2): Likewise.
1111 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1113 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1114 unaligned VSX load/store on P8/P9.
1115 (expand_block_clear): Allow the use of unaligned VSX
1116 load/store on P8/P9.
1118 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1120 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1122 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1123 swap associated with both a load and a store.
1125 2018-01-02 Andrew Waterman <andrew@sifive.com>
1127 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1128 * config/riscv/riscv.md (clear_cache): Use it.
1130 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1132 * web.c: Remove out-of-date comment.
1134 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1136 * expr.c (fixup_args_size_notes): Check that any existing
1137 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1138 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1139 (emit_single_push_insn): ...here.
1141 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1143 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1144 (const_vector_encoded_nelts): New function.
1145 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1146 (const_vector_int_elt, const_vector_elt): Declare.
1147 * emit-rtl.c (const_vector_int_elt_1): New function.
1148 (const_vector_elt): Likewise.
1149 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1150 of CONST_VECTOR_ELT.
1152 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1154 * expr.c: Include rtx-vector-builder.h.
1155 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1156 directly on the tree encoding.
1157 (const_vector_from_tree): Likewise.
1158 * optabs.c: Include rtx-vector-builder.h.
1159 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1160 sequence of "u" values.
1161 * vec-perm-indices.c: Include rtx-vector-builder.h.
1162 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1163 directly on the vec_perm_indices encoding.
1165 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1167 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1168 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1169 * rtx-vector-builder.h: New file.
1170 * rtx-vector-builder.c: Likewise.
1171 * rtl.h (rtx_def::u2): Add a const_vector field.
1172 (CONST_VECTOR_NPATTERNS): New macro.
1173 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1174 (CONST_VECTOR_DUPLICATE_P): Likewise.
1175 (CONST_VECTOR_STEPPED_P): Likewise.
1176 (CONST_VECTOR_ENCODED_ELT): Likewise.
1177 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1178 (unwrap_const_vec_duplicate): Likewise.
1179 (const_vec_series_p): Check for a non-duplicated vector encoding.
1180 Say that the function only returns true for integer vectors.
1181 * emit-rtl.c: Include rtx-vector-builder.h.
1182 (gen_const_vec_duplicate_1): Delete.
1183 (gen_const_vector): Call gen_const_vec_duplicate instead of
1184 gen_const_vec_duplicate_1.
1185 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1186 (gen_const_vec_duplicate): Use rtx_vector_builder.
1187 (gen_const_vec_series): Likewise.
1188 (gen_rtx_CONST_VECTOR): Likewise.
1189 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1190 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1191 Build a new vector rather than modifying a CONST_VECTOR in-place.
1192 (handle_special_swappables): Update call accordingly.
1193 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1194 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1195 Build a new vector rather than modifying a CONST_VECTOR in-place.
1196 (handle_special_swappables): Update call accordingly.
1198 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1200 * simplify-rtx.c (simplify_const_binary_operation): Use
1201 CONST_VECTOR_ELT instead of XVECEXP.
1203 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1205 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1206 the selector elements to be different from the data elements
1207 if the selector is a VECTOR_CST.
1208 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1209 ssizetype for the selector.
1211 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1213 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1214 before testing each element individually.
1215 * tree-vect-generic.c (lower_vec_perm): Likewise.
1217 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1219 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1220 * selftest-run-tests.c (selftest::run_tests): Call it.
1221 * vector-builder.h (vector_builder::operator ==): New function.
1222 (vector_builder::operator !=): Likewise.
1223 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1224 (vec_perm_indices::all_from_input_p): New function.
1225 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1226 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1227 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1228 instead of reading the VECTOR_CST directly. Detect whether both
1229 vector inputs are the same before constructing the vec_perm_indices,
1230 and update the number of inputs argument accordingly. Use the
1231 utility functions added above. Only construct sel2 if we need to.
1233 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1235 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1236 the broadcast of the low byte.
1237 (expand_mult_highpart): Use an explicit encoding for the permutes.
1238 * optabs-query.c (can_mult_highpart_p): Likewise.
1239 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1240 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1241 (vectorizable_bswap): Likewise.
1242 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1243 explicit encoding for the power-of-2 permutes.
1244 (vect_permute_store_chain): Likewise.
1245 (vect_grouped_load_supported): Likewise.
1246 (vect_permute_load_chain): Likewise.
1248 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1250 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1251 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1252 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1253 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1254 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1255 (vect_gen_perm_mask_any): Likewise.
1257 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1259 * int-vector-builder.h: New file.
1260 * vec-perm-indices.h: Include int-vector-builder.h.
1261 (vec_perm_indices): Redefine as an int_vector_builder.
1262 (auto_vec_perm_indices): Delete.
1263 (vec_perm_builder): Redefine as a stand-alone class.
1264 (vec_perm_indices::vec_perm_indices): New function.
1265 (vec_perm_indices::clamp): Likewise.
1266 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1267 (vec_perm_indices::new_vector): New function.
1268 (vec_perm_indices::new_expanded_vector): Update for new
1269 vec_perm_indices class.
1270 (vec_perm_indices::rotate_inputs): New function.
1271 (vec_perm_indices::all_in_range_p): Operate directly on the
1272 encoded form, without computing elided elements.
1273 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1274 encoding. Update for new vec_perm_indices class.
1275 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1276 the given vec_perm_builder.
1277 (expand_vec_perm_var): Update vec_perm_builder constructor.
1278 (expand_mult_highpart): Use vec_perm_builder instead of
1279 auto_vec_perm_indices.
1280 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1281 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1282 or double series encoding as appropriate.
1283 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1284 vec_perm_indices instead of auto_vec_perm_indices.
1285 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1286 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1287 (vect_permute_store_chain): Likewise.
1288 (vect_grouped_load_supported): Likewise.
1289 (vect_permute_load_chain): Likewise.
1290 (vect_shift_permute_load_chain): Likewise.
1291 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1292 (vect_transform_slp_perm_load): Likewise.
1293 (vect_schedule_slp_instance): Likewise.
1294 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1295 (vectorizable_mask_load_store): Likewise.
1296 (vectorizable_bswap): Likewise.
1297 (vectorizable_store): Likewise.
1298 (vectorizable_load): Likewise.
1299 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1300 vec_perm_indices instead of auto_vec_perm_indices. Use
1301 tree_to_vec_perm_builder to read the vector from a tree.
1302 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1303 vec_perm_builder instead of a vec_perm_indices.
1304 (have_whole_vector_shift): Use vec_perm_builder and
1305 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1306 truncation to calc_vec_perm_mask_for_shift.
1307 (vect_create_epilog_for_reduction): Likewise.
1308 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1309 from auto_vec_perm_indices to vec_perm_indices.
1310 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1311 instead of changing individual elements.
1312 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1313 the vector in d.perm.
1314 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1315 from auto_vec_perm_indices to vec_perm_indices.
1316 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1317 instead of changing individual elements.
1318 (arm_vectorize_vec_perm_const): Use new_vector to install
1319 the vector in d.perm.
1320 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1321 Update vec_perm_builder constructor.
1322 (rs6000_expand_interleave): Likewise.
1323 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1324 (rs6000_expand_interleave): Likewise.
1326 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1328 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1329 to qimode could truncate the indices.
1330 * optabs.c (expand_vec_perm_var): Likewise.
1332 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1334 * Makefile.in (OBJS): Add vec-perm-indices.o.
1335 * vec-perm-indices.h: New file.
1336 * vec-perm-indices.c: Likewise.
1337 * target.h (vec_perm_indices): Replace with a forward class
1339 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1340 * optabs.h: Include vec-perm-indices.h.
1341 (expand_vec_perm): Delete.
1342 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1343 (expand_vec_perm_const): Declare.
1344 * target.def (vec_perm_const_ok): Replace with...
1345 (vec_perm_const): ...this new hook.
1346 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1347 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1348 * doc/tm.texi: Regenerate.
1349 * optabs.def (vec_perm_const): Delete.
1350 * doc/md.texi (vec_perm_const): Likewise.
1351 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1352 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1353 expand_vec_perm for constant permutation vectors. Assert that
1354 the mode of variable permutation vectors is the integer equivalent
1355 of the mode that is being permuted.
1356 * optabs-query.h (selector_fits_mode_p): Declare.
1357 * optabs-query.c: Include vec-perm-indices.h.
1358 (selector_fits_mode_p): New function.
1359 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1360 is defined, instead of checking whether the vec_perm_const_optab
1361 exists. Use targetm.vectorize.vec_perm_const instead of
1362 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1363 fit in the vector mode before using a variable permute.
1364 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1365 vec_perm_indices instead of an rtx.
1366 (expand_vec_perm): Replace with...
1367 (expand_vec_perm_const): ...this new function. Take the selector
1368 as a vec_perm_indices rather than an rtx. Also take the mode of
1369 the selector. Update call to shift_amt_for_vec_perm_mask.
1370 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1371 Use vec_perm_indices::new_expanded_vector to expand the original
1372 selector into bytes. Check whether the indices fit in the vector
1373 mode before using a variable permute.
1374 (expand_vec_perm_var): Make global.
1375 (expand_mult_highpart): Use expand_vec_perm_const.
1376 * fold-const.c: Includes vec-perm-indices.h.
1377 * tree-ssa-forwprop.c: Likewise.
1378 * tree-vect-data-refs.c: Likewise.
1379 * tree-vect-generic.c: Likewise.
1380 * tree-vect-loop.c: Likewise.
1381 * tree-vect-slp.c: Likewise.
1382 * tree-vect-stmts.c: Likewise.
1383 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1385 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1386 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1387 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1388 (aarch64_vectorize_vec_perm_const): ...this new function.
1389 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1390 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1391 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1392 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1393 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1394 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1395 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1397 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1398 check for NEON modes.
1399 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1400 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1401 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1402 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1404 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1405 the old VEC_PERM_CONST conditions.
1406 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1407 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1408 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1409 (ia64_vectorize_vec_perm_const_ok): Merge into...
1410 (ia64_vectorize_vec_perm_const): ...this new function.
1411 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1412 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1413 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1414 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1415 * config/mips/mips.c (mips_expand_vec_perm_const)
1416 (mips_vectorize_vec_perm_const_ok): Merge into...
1417 (mips_vectorize_vec_perm_const): ...this new function.
1418 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1419 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1420 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1421 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1422 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1423 (rs6000_expand_vec_perm_const): Delete.
1424 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1426 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1427 (altivec_expand_vec_perm_const_le): Take each operand individually.
1428 Operate on constant selectors rather than rtxes.
1429 (altivec_expand_vec_perm_const): Likewise. Update call to
1430 altivec_expand_vec_perm_const_le.
1431 (rs6000_expand_vec_perm_const): Delete.
1432 (rs6000_vectorize_vec_perm_const_ok): Delete.
1433 (rs6000_vectorize_vec_perm_const): New function.
1434 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1435 an element count and rtx array.
1436 (rs6000_expand_extract_even): Update call accordingly.
1437 (rs6000_expand_interleave): Likewise.
1438 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1439 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1440 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1441 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1442 (rs6000_expand_vec_perm_const): Delete.
1443 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1444 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1445 (altivec_expand_vec_perm_const_le): Take each operand individually.
1446 Operate on constant selectors rather than rtxes.
1447 (altivec_expand_vec_perm_const): Likewise. Update call to
1448 altivec_expand_vec_perm_const_le.
1449 (rs6000_expand_vec_perm_const): Delete.
1450 (rs6000_vectorize_vec_perm_const_ok): Delete.
1451 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1452 reference to the SPE evmerge intructions.
1453 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1454 an element count and rtx array.
1455 (rs6000_expand_extract_even): Update call accordingly.
1456 (rs6000_expand_interleave): Likewise.
1457 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1458 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1460 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1462 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1464 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1465 vector mode and that that mode matches the mode of the data
1467 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1468 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1469 directly using expand_vec_perm_1 when forcing selectors into
1471 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1473 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1475 * optabs-query.h (can_vec_perm_p): Delete.
1476 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1477 * optabs-query.c (can_vec_perm_p): Split into...
1478 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1479 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1480 particular selector is valid.
1481 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1482 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1483 (vect_grouped_load_supported): Likewise.
1484 (vect_shift_permute_load_chain): Likewise.
1485 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1486 (vect_transform_slp_perm_load): Likewise.
1487 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1488 (vectorizable_bswap): Likewise.
1489 (vect_gen_perm_mask_checked): Likewise.
1490 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1491 implementations of variable permutation vectors into account
1492 when deciding which selector to use.
1493 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1494 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1495 with a false third argument.
1496 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1497 to test whether the constant selector is valid and can_vec_perm_var_p
1498 to test whether a variable selector is valid.
1500 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1502 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1503 * optabs-query.c (can_vec_perm_p): Likewise.
1504 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1505 instead of vec_perm_indices.
1506 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1507 (vect_gen_perm_mask_checked): Likewise,
1508 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1509 (vect_gen_perm_mask_checked): Likewise,
1511 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1513 * optabs-query.h (qimode_for_vec_perm): Declare.
1514 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1515 (qimode_for_vec_perm): ...this new function.
1516 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1518 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1520 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1521 does not have a conditional at the top.
1523 2018-01-02 Richard Biener <rguenther@suse.de>
1525 * ipa-inline.c (big_speedup_p): Fix expression.
1527 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1530 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1533 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1537 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1538 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1539 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1540 cond_taken_branch_cost 3->4.
1542 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1544 PR tree-optimization/83581
1545 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1546 TODO_cleanup_cfg if any changes have been made.
1549 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1550 convert_modes if target mode has the right side, but different mode
1554 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1555 last argument when extracting from CONCAT. If either from_real or
1556 from_imag is NULL, use expansion through memory. If result is not
1557 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1558 the parts directly to inner mode, if even that fails, use expansion
1562 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1563 check for bswap in mode rather than HImode and use that in expand_unop
1566 Copyright (C) 2018 Free Software Foundation, Inc.
1568 Copying and distribution of this file, with or without modification,
1569 are permitted in any medium without royalty provided the copyright
1570 notice and this notice are preserved.