1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4 (vect_model_store_cost): Take a vec_load_store_type instead of a
6 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
7 (vect_model_store_cost): Take a vec_load_store_type instead of a
9 (vectorizable_mask_load_store): Update accordingly.
10 (vectorizable_store): Likewise.
11 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
13 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
15 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
16 IFN_MASK_LOAD calls here rather than...
17 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
19 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
20 Alan Hayward <alan.hayward@arm.com>
21 David Sherwood <david.sherwood@arm.com>
23 * expmed.c (extract_bit_field_1): For vector extracts,
24 fall back to extract_bit_field_as_subreg if vec_extract
27 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
28 Alan Hayward <alan.hayward@arm.com>
29 David Sherwood <david.sherwood@arm.com>
31 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
32 they are variable or constant sized.
33 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
34 slots for constant-sized data.
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 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
41 handling COND_EXPRs with boolean comparisons, try to find a better
42 basis for the mask type than the boolean itself.
44 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
46 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
47 is calculated and how it can be overridden.
48 * genmodes.c (max_bitsize_mode_any_mode): New variable.
49 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
51 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
54 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
55 Alan Hayward <alan.hayward@arm.com>
56 David Sherwood <david.sherwood@arm.com>
58 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
59 Remove the mode argument.
60 (aarch64_simd_valid_immediate): Remove the mode and inverse
62 * config/aarch64/iterators.md (bitsize): New iterator.
63 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
64 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
65 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
66 aarch64_simd_valid_immediate.
67 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
68 (aarch64_reg_or_bic_imm): Likewise.
69 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
70 with an insn_type enum and msl with a modifier_type enum.
71 Replace element_width with a scalar_mode. Change the shift
72 to unsigned int. Add constructors for scalar_float_mode and
73 scalar_int_mode elements.
74 (aarch64_vect_float_const_representable_p): Delete.
75 (aarch64_can_const_movi_rtx_p)
76 (aarch64_simd_scalar_immediate_valid_for_move)
77 (aarch64_simd_make_constant): Update call to
78 aarch64_simd_valid_immediate.
79 (aarch64_advsimd_valid_immediate_hs): New function.
80 (aarch64_advsimd_valid_immediate): Likewise.
81 (aarch64_simd_valid_immediate): Remove mode and inverse
82 arguments. Rewrite to use the above. Use const_vec_duplicate_p
83 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
84 and aarch64_float_const_representable_p on the result.
85 (aarch64_output_simd_mov_immediate): Remove mode argument.
86 Update call to aarch64_simd_valid_immediate and use of
88 (aarch64_output_scalar_simd_mov_immediate): Update call
91 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
92 Alan Hayward <alan.hayward@arm.com>
93 David Sherwood <david.sherwood@arm.com>
95 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
96 (mode_nunits): Likewise CONST_MODE_NUNITS.
97 * machmode.def (ADJUST_NUNITS): Document.
98 * genmodes.c (mode_data::need_nunits_adj): New field.
99 (blank_mode): Update accordingly.
100 (adj_nunits): New variable.
101 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
103 (emit_mode_size_inline): Set need_bytesize_adj for all modes
104 listed in adj_nunits.
105 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
106 listed in adj_nunits. Don't emit case statements for such modes.
107 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
108 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
109 nothing if adj_nunits is nonnull.
110 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
111 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
112 (emit_mode_fbit): Update use of print_maybe_const_decl.
113 (emit_move_size): Likewise. Treat the array as non-const
115 (emit_mode_adjustments): Handle adj_nunits.
117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
119 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
120 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
121 (VECTOR_MODES): Use it.
122 (make_vector_modes): Take the prefix as an argument.
124 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
125 Alan Hayward <alan.hayward@arm.com>
126 David Sherwood <david.sherwood@arm.com>
128 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
129 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
130 for MODE_VECTOR_BOOL.
131 * machmode.def (VECTOR_BOOL_MODE): Document.
132 * genmodes.c (VECTOR_BOOL_MODE): New macro.
133 (make_vector_bool_mode): New function.
134 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
136 * lto-streamer-in.c (lto_input_mode_table): Likewise.
137 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
139 * stor-layout.c (int_mode_for_mode): Likewise.
140 * tree.c (build_vector_type_for_mode): Likewise.
141 * varasm.c (output_constant_pool_2): Likewise.
142 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
143 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
144 for MODE_VECTOR_BOOL.
145 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
146 of mode class checks.
147 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
148 instead of a list of mode class checks.
149 (expand_vector_scalar_condition): Likewise.
150 (type_for_widest_vector_mode): Handle BImode as an inner mode.
152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
153 Alan Hayward <alan.hayward@arm.com>
154 David Sherwood <david.sherwood@arm.com>
156 * machmode.h (mode_size): Change from unsigned short to
158 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
159 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
160 or if measurement_type is not polynomial.
161 (fixed_size_mode::includes_p): Check for constant-sized modes.
162 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
163 return a poly_uint16 rather than an unsigned short.
164 (emit_mode_size): Change the type of mode_size from unsigned short
165 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
166 (emit_mode_adjustments): Cope with polynomial vector sizes.
167 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
169 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
171 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
172 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
173 * caller-save.c (setup_save_areas): Likewise.
174 (replace_reg_with_saved_mem): Likewise.
175 * calls.c (emit_library_call_value_1): Likewise.
176 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
177 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
178 (gen_lowpart_for_combine): Likewise.
179 * convert.c (convert_to_integer_1): Likewise.
180 * cse.c (equiv_constant, cse_insn): Likewise.
181 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
182 (cselib_subst_to_values): Likewise.
183 * dce.c (word_dce_process_block): Likewise.
184 * df-problems.c (df_word_lr_mark_ref): Likewise.
185 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
186 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
187 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
188 (rtl_for_decl_location): Likewise.
189 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
190 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
191 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
192 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
193 (expand_expr_real_1): Likewise.
194 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
195 (pad_below): Likewise.
196 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
197 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
198 * ira.c (get_subreg_tracking_sizes): Likewise.
199 * ira-build.c (ira_create_allocno_objects): Likewise.
200 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
201 (ira_sort_regnos_for_alter_reg): Likewise.
202 * ira-costs.c (record_operand_costs): Likewise.
203 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
204 (resolve_simple_move): Likewise.
205 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
206 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
207 (lra_constraints): Likewise.
208 (CONST_POOL_OK_P): Reject variable-sized modes.
209 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
210 (add_pseudo_to_slot, lra_spill): Likewise.
211 * omp-low.c (omp_clause_aligned_alignment): Likewise.
212 * optabs-query.c (get_best_extraction_insn): Likewise.
213 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
214 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
215 (expand_mult_highpart, valid_multiword_target_p): Likewise.
216 * recog.c (offsettable_address_addr_space_p): Likewise.
217 * regcprop.c (maybe_mode_change): Likewise.
218 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
219 * regrename.c (build_def_use): Likewise.
220 * regstat.c (dump_reg_info): Likewise.
221 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
222 (find_reloads, find_reloads_subreg_address): Likewise.
223 * reload1.c (eliminate_regs_1): Likewise.
224 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
225 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
226 (simplify_binary_operation_1, simplify_subreg): Likewise.
227 * targhooks.c (default_function_arg_padding): Likewise.
228 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
229 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
230 (verify_gimple_assign_ternary): Likewise.
231 * tree-inline.c (estimate_move_cost): Likewise.
232 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
233 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
234 (get_address_cost_ainc): Likewise.
235 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
236 (vect_supportable_dr_alignment): Likewise.
237 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
238 (vectorizable_reduction): Likewise.
239 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
240 (vectorizable_operation, vectorizable_load): Likewise.
241 * tree.c (build_same_sized_truth_vector_type): Likewise.
242 * valtrack.c (cleanup_auto_inc_dec): Likewise.
243 * var-tracking.c (emit_note_insn_var_location): Likewise.
244 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
245 (ADDR_VEC_ALIGN): Likewise.
247 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
248 Alan Hayward <alan.hayward@arm.com>
249 David Sherwood <david.sherwood@arm.com>
251 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
253 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
254 or if measurement_type is polynomial.
255 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
256 * combine.c (make_extraction): Likewise.
257 * dse.c (find_shift_sequence): Likewise.
258 * dwarf2out.c (mem_loc_descriptor): Likewise.
259 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
260 (extract_bit_field, extract_low_bits): Likewise.
261 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
262 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
263 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
264 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
265 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
266 * reload.c (find_reloads): Likewise.
267 * reload1.c (alter_reg): Likewise.
268 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
269 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
270 * tree-if-conv.c (predicate_mem_writes): Likewise.
271 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
272 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
273 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
274 * valtrack.c (dead_debug_insert_temp): Likewise.
275 * varasm.c (mergeable_constant_section): Likewise.
276 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
279 Alan Hayward <alan.hayward@arm.com>
280 David Sherwood <david.sherwood@arm.com>
282 * expr.c (expand_assignment): Cope with polynomial mode sizes
283 when assigning to a CONCAT.
285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
286 Alan Hayward <alan.hayward@arm.com>
287 David Sherwood <david.sherwood@arm.com>
289 * machmode.h (mode_precision): Change from unsigned short to
291 (mode_to_precision): Return a poly_uint16 rather than an unsigned
293 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
294 or if measurement_type is not polynomial.
295 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
296 in which the mode is already known to be a scalar_int_mode.
297 * genmodes.c (emit_mode_precision): Change the type of mode_precision
298 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
300 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
301 for GET_MODE_PRECISION.
302 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
303 for GET_MODE_PRECISION.
304 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
306 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
307 (expand_field_assignment, make_extraction): Likewise.
308 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
309 (get_last_value): Likewise.
310 * convert.c (convert_to_integer_1): Likewise.
311 * cse.c (cse_insn): Likewise.
312 * expr.c (expand_expr_real_1): Likewise.
313 * lra-constraints.c (simplify_operand_subreg): Likewise.
314 * optabs-query.c (can_atomic_load_p): Likewise.
315 * optabs.c (expand_atomic_load): Likewise.
316 (expand_atomic_store): Likewise.
317 * ree.c (combine_reaching_defs): Likewise.
318 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
319 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
320 * tree.h (type_has_mode_precision_p): Likewise.
321 * ubsan.c (instrument_si_overflow): Likewise.
323 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
324 Alan Hayward <alan.hayward@arm.com>
325 David Sherwood <david.sherwood@arm.com>
327 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
328 polynomial numbers of units.
329 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
330 (valid_vector_subparts_p): New function.
331 (build_vector_type): Remove temporary shim and take the number
332 of units as a poly_uint64 rather than an int.
333 (build_opaque_vector_type): Take the number of units as a
334 poly_uint64 rather than an int.
335 * tree.c (build_vector_from_ctor): Handle polynomial
336 TYPE_VECTOR_SUBPARTS.
337 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
338 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
339 (build_vector_from_val): If the number of units is variable,
340 use build_vec_duplicate_cst for constant operands and
341 VEC_DUPLICATE_EXPR otherwise.
342 (make_vector_type): Remove temporary is_constant ().
343 (build_vector_type, build_opaque_vector_type): Take the number of
344 units as a poly_uint64 rather than an int.
345 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
347 * cfgexpand.c (expand_debug_expr): Likewise.
348 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
349 (store_constructor, expand_expr_real_1): Likewise.
350 (const_scalar_mask_from_tree): Likewise.
351 * fold-const-call.c (fold_const_reduction): Likewise.
352 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
353 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
354 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
355 (fold_relational_const): Likewise.
356 (native_interpret_vector): Likewise. Change the size from an
357 int to an unsigned int.
358 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
359 TYPE_VECTOR_SUBPARTS.
360 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
361 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
362 duplicating a non-constant operand into a variable-length vector.
363 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
364 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
365 * ipa-icf.c (sem_variable::equals): Likewise.
366 * match.pd: Likewise.
367 * omp-simd-clone.c (simd_clone_subparts): Likewise.
368 * print-tree.c (print_node): Likewise.
369 * stor-layout.c (layout_type): Likewise.
370 * targhooks.c (default_builtin_vectorization_cost): Likewise.
371 * tree-cfg.c (verify_gimple_comparison): Likewise.
372 (verify_gimple_assign_binary): Likewise.
373 (verify_gimple_assign_ternary): Likewise.
374 (verify_gimple_assign_single): Likewise.
375 * tree-pretty-print.c (dump_generic_node): Likewise.
376 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
377 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
378 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
379 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
380 (vect_shift_permute_load_chain): Likewise.
381 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
382 (expand_vector_condition, optimize_vector_constructor): Likewise.
383 (lower_vec_perm, get_compute_type): Likewise.
384 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
385 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
386 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
387 (vect_recog_mask_conversion_pattern): Likewise.
388 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
389 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
390 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
391 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
392 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
393 (vectorizable_shift, vectorizable_operation, vectorizable_store)
394 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
395 (supportable_widening_operation): Likewise.
396 (supportable_narrowing_operation): Likewise.
397 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
399 * varasm.c (output_constant): Likewise.
401 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
402 Alan Hayward <alan.hayward@arm.com>
403 David Sherwood <david.sherwood@arm.com>
405 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
406 so that both the length == 3 and length != 3 cases set up their
407 own permute vectors. Add comments explaining why we know the
408 number of elements is constant.
409 (vect_permute_load_chain): Likewise.
411 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
412 Alan Hayward <alan.hayward@arm.com>
413 David Sherwood <david.sherwood@arm.com>
415 * machmode.h (mode_nunits): Change from unsigned char to
417 (ONLY_FIXED_SIZE_MODES): New macro.
418 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
419 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
420 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
422 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
423 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
424 or if measurement_type is not polynomial.
425 * genmodes.c (ZERO_COEFFS): New macro.
426 (emit_mode_nunits_inline): Make mode_nunits_inline return a
428 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
429 Use ZERO_COEFFS when emitting initializers.
430 * data-streamer.h (bp_pack_poly_value): New function.
431 (bp_unpack_poly_value): Likewise.
432 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
434 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
436 * tree.c (make_vector_type): Remove temporary shim and make
437 the real function take the number of units as a poly_uint64
439 (build_vector_type_for_mode): Handle polynomial nunits.
440 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
441 * emit-rtl.c (const_vec_series_p_1): Likewise.
442 (gen_rtx_CONST_VECTOR): Likewise.
443 * fold-const.c (test_vec_duplicate_folding): Likewise.
444 * genrecog.c (validate_pattern): Likewise.
445 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
446 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
447 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
448 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
449 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
450 * rtlanal.c (subreg_get_info): Likewise.
451 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
452 (vect_grouped_load_supported): Likewise.
453 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
454 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
455 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
456 (simplify_const_unary_operation, simplify_binary_operation_1)
457 (simplify_const_binary_operation, simplify_ternary_operation)
458 (test_vector_ops_duplicate, test_vector_ops): Likewise.
459 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
460 instead of CONST_VECTOR_NUNITS.
461 * varasm.c (output_constant_pool_2): Likewise.
462 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
463 explicit-encoded elements in the XVEC for variable-length vectors.
465 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
467 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
469 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
470 Alan Hayward <alan.hayward@arm.com>
471 David Sherwood <david.sherwood@arm.com>
473 * coretypes.h (fixed_size_mode): Declare.
474 (fixed_size_mode_pod): New typedef.
475 * builtins.h (target_builtins::x_apply_args_mode)
476 (target_builtins::x_apply_result_mode): Change type to
478 * builtins.c (apply_args_size, apply_result_size, result_vector)
479 (expand_builtin_apply_args_1, expand_builtin_apply)
480 (expand_builtin_return): Update accordingly.
482 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
484 * cse.c (hash_rtx_cb): Hash only the encoded elements.
485 * cselib.c (cselib_hash_rtx): Likewise.
486 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
487 CONST_VECTOR encoding.
489 2017-01-03 Jakub Jelinek <jakub@redhat.com>
490 Jeff Law <law@redhat.com>
493 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
494 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
495 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
496 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
499 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
500 explicitly probe *sp in a noreturn function if there were any callee
501 register saves or frame pointer is needed.
503 2018-01-03 Jakub Jelinek <jakub@redhat.com>
506 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
507 BLKmode for ternary, binary or unary expressions.
510 * var-tracking.c (delete_vta_debug_insn): New inline function.
511 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
512 insns from get_insns () to NULL instead of each bb separately.
513 Use delete_vta_debug_insn. No longer static.
514 (vt_debug_insns_local, variable_tracking_main_1): Adjust
515 delete_vta_debug_insns callers.
516 * rtl.h (delete_vta_debug_insns): Declare.
517 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
518 instead of variable_tracking_main.
520 2018-01-03 Martin Sebor <msebor@redhat.com>
523 * doc/extend.texi (attribute const): Fix a typo.
524 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
525 issuing -Wsuggest-attribute for void functions.
527 2018-01-03 Martin Sebor <msebor@redhat.com>
529 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
530 offset_int::from instead of wide_int::to_shwi.
531 (maybe_diag_overlap): Remove assertion.
532 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
533 * gimple-ssa-sprintf.c (format_directive): Same.
534 (parse_directive): Same.
535 (sprintf_dom_walker::compute_format_length): Same.
536 (try_substitute_return_value): Same.
538 2017-01-03 Jeff Law <law@redhat.com>
541 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
542 non-constant residual for zero at runtime and avoid probing in
543 that case. Reorganize code for trailing problem to mirror handling
546 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
548 PR tree-optimization/83501
549 * tree-ssa-strlen.c (get_string_cst): New.
550 (handle_char_store): Call get_string_cst.
552 2018-01-03 Martin Liska <mliska@suse.cz>
554 PR tree-optimization/83593
555 * tree-ssa-strlen.c: Include tree-cfg.h.
556 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
557 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
558 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
560 (strlen_dom_walker::before_dom_children): Call
561 gimple_purge_dead_eh_edges. Dump tranformation with details
563 (strlen_dom_walker::before_dom_children): Update call by adding
564 new argument cleanup_eh.
565 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
567 2018-01-03 Martin Liska <mliska@suse.cz>
570 * cif-code.def (VARIADIC_THUNK): New enum value.
571 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
574 2018-01-03 Jan Beulich <jbeulich@suse.com>
576 * sse.md (mov<mode>_internal): Tighten condition for when to use
577 vmovdqu<ssescalarsize> for TI and OI modes.
579 2018-01-03 Jakub Jelinek <jakub@redhat.com>
581 Update copyright years.
583 2018-01-03 Martin Liska <mliska@suse.cz>
586 * ipa-visibility.c (function_and_variable_visibility): Skip
587 functions with noipa attribure.
589 2018-01-03 Jakub Jelinek <jakub@redhat.com>
591 * gcc.c (process_command): Update copyright notice dates.
592 * gcov-dump.c (print_version): Ditto.
593 * gcov.c (print_version): Ditto.
594 * gcov-tool.c (print_version): Ditto.
595 * gengtype.c (create_file): Ditto.
596 * doc/cpp.texi: Bump @copying's copyright year.
597 * doc/cppinternals.texi: Ditto.
598 * doc/gcc.texi: Ditto.
599 * doc/gccint.texi: Ditto.
600 * doc/gcov.texi: Ditto.
601 * doc/install.texi: Ditto.
602 * doc/invoke.texi: Ditto.
604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
606 * vector-builder.h (vector_builder::m_full_nelts): Change from
607 unsigned int to poly_uint64.
608 (vector_builder::full_nelts): Update prototype accordingly.
609 (vector_builder::new_vector): Likewise.
610 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
611 (vector_builder::operator ==): Likewise.
612 (vector_builder::finalize): Likewise.
613 * int-vector-builder.h (int_vector_builder::int_vector_builder):
614 Take the number of elements as a poly_uint64 rather than an
616 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
617 from unsigned int to poly_uint64.
618 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
619 (vec_perm_indices::new_vector): Likewise.
620 (vec_perm_indices::length): Likewise.
621 (vec_perm_indices::nelts_per_input): Likewise.
622 (vec_perm_indices::input_nelts): Likewise.
623 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
624 number of elements per input as a poly_uint64 rather than an
625 unsigned int. Use the original encoding for variable-length
626 vectors, rather than clamping each individual element.
627 For the second and subsequent elements in each pattern,
628 clamp the step and base before clamping their sum.
629 (vec_perm_indices::series_p): Handle polynomial element counts.
630 (vec_perm_indices::all_in_range_p): Likewise.
631 (vec_perm_indices_to_tree): Likewise.
632 (vec_perm_indices_to_rtx): Likewise.
633 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
634 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
635 (tree_vector_builder::new_binary_operation): Handle polynomial
636 element counts. Return false if we need to know the number
637 of elements at compile time.
638 * fold-const.c (fold_vec_perm): Punt if the number of elements
639 isn't known at compile time.
641 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
643 * vec-perm-indices.h (vec_perm_builder): Change element type
644 from HOST_WIDE_INT to poly_int64.
645 (vec_perm_indices::element_type): Update accordingly.
646 (vec_perm_indices::clamp): Handle polynomial element_types.
647 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
648 (vec_perm_indices::all_in_range_p): Likewise.
649 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
651 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
652 polynomial vec_perm_indices element types.
653 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
654 * fold-const.c (fold_vec_perm): Likewise.
655 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
656 * tree-vect-generic.c (lower_vec_perm): Likewise.
657 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
658 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
659 element type to HOST_WIDE_INT.
661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
662 Alan Hayward <alan.hayward@arm.com>
663 David Sherwood <david.sherwood@arm.com>
665 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
666 rather than an int. Use plus_constant.
667 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
668 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
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 * calls.c (emit_call_1, expand_call): Change struct_value_size from
675 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
682 mode sizes. Require a constant size for BLKmode parameters
683 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
684 forces a parameter to be padded at the lsb end in order to
685 fill a complete number of words, require the parameter size
686 to be ordered wrt UNITS_PER_WORD.
688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
689 Alan Hayward <alan.hayward@arm.com>
690 David Sherwood <david.sherwood@arm.com>
692 * reload1.c (spill_stack_slot_width): Change element type
693 from unsigned int to poly_uint64_pod.
694 (alter_reg): Treat mode sizes as polynomial.
696 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
697 Alan Hayward <alan.hayward@arm.com>
698 David Sherwood <david.sherwood@arm.com>
700 * reload.c (complex_word_subreg_p): New function.
701 (reload_inner_reg_of_subreg, push_reload): Use it.
703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
704 Alan Hayward <alan.hayward@arm.com>
705 David Sherwood <david.sherwood@arm.com>
707 * lra-constraints.c (process_alt_operands): Reject matched
708 operands whose sizes aren't ordered.
709 (match_reload): Refer to this check here.
711 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
712 Alan Hayward <alan.hayward@arm.com>
713 David Sherwood <david.sherwood@arm.com>
715 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
716 that the mode size is in the set {1, 2, 4, 8, 16}.
718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
719 Alan Hayward <alan.hayward@arm.com>
720 David Sherwood <david.sherwood@arm.com>
722 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
723 Use plus_constant instead of gen_rtx_PLUS.
725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
726 Alan Hayward <alan.hayward@arm.com>
727 David Sherwood <david.sherwood@arm.com>
729 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
730 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
731 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
732 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
733 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
734 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
735 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
736 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
737 * config/i386/i386.c (ix86_push_rounding): ...this new function.
738 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
740 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
741 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
742 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
743 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
744 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
745 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
746 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
747 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
748 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
749 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
751 * expr.c (emit_move_resolve_push): Treat the input and result
752 of PUSH_ROUNDING as a poly_int64.
753 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
754 (emit_push_insn): Likewise.
755 * lra-eliminations.c (mark_not_eliminable): Likewise.
756 * recog.c (push_operand): Likewise.
757 * reload1.c (elimination_effects): Likewise.
758 * rtlanal.c (nonzero_bits1): Likewise.
759 * calls.c (store_one_arg): Likewise. Require the padding to be
760 known at compile time.
762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
763 Alan Hayward <alan.hayward@arm.com>
764 David Sherwood <david.sherwood@arm.com>
766 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
767 Use plus_constant instead of gen_rtx_PLUS.
769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
770 Alan Hayward <alan.hayward@arm.com>
771 David Sherwood <david.sherwood@arm.com>
773 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
776 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
777 Alan Hayward <alan.hayward@arm.com>
778 David Sherwood <david.sherwood@arm.com>
780 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
781 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
782 via stack temporaries. Treat the mode size as polynomial too.
784 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
785 Alan Hayward <alan.hayward@arm.com>
786 David Sherwood <david.sherwood@arm.com>
788 * expr.c (expand_expr_real_2): When handling conversions involving
789 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
790 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
791 as a poly_uint64 too.
793 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
794 Alan Hayward <alan.hayward@arm.com>
795 David Sherwood <david.sherwood@arm.com>
797 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
800 Alan Hayward <alan.hayward@arm.com>
801 David Sherwood <david.sherwood@arm.com>
803 * combine.c (can_change_dest_mode): Handle polynomial
804 REGMODE_NATURAL_SIZE.
805 * expmed.c (store_bit_field_1): Likewise.
806 * expr.c (store_constructor): Likewise.
807 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
808 and polynomial REGMODE_NATURAL_SIZE.
809 (gen_lowpart_common): Likewise.
810 * reginfo.c (record_subregs_of_mode): Likewise.
811 * rtlanal.c (read_modify_subreg_p): Likewise.
813 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
814 Alan Hayward <alan.hayward@arm.com>
815 David Sherwood <david.sherwood@arm.com>
817 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
820 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
821 Alan Hayward <alan.hayward@arm.com>
822 David Sherwood <david.sherwood@arm.com>
824 * match.pd: Cope with polynomial numbers of vector elements.
826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
827 Alan Hayward <alan.hayward@arm.com>
828 David Sherwood <david.sherwood@arm.com>
830 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
831 in a POINTER_PLUS_EXPR.
833 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
834 Alan Hayward <alan.hayward@arm.com>
835 David Sherwood <david.sherwood@arm.com>
837 * omp-simd-clone.c (simd_clone_subparts): New function.
838 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
839 (ipa_simd_modify_function_body): Likewise.
841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
842 Alan Hayward <alan.hayward@arm.com>
843 David Sherwood <david.sherwood@arm.com>
845 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
846 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
847 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
848 (expand_vector_condition, vector_element): Likewise.
849 (subparts_gt): New function.
850 (get_compute_type): Use subparts_gt.
851 (count_type_subparts): Delete.
852 (expand_vector_operations_1): Use subparts_gt instead of
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-data-refs.c (vect_no_alias_p): Replace with...
860 (vect_compile_time_alias): ...this new function. Do the calculation
861 on poly_ints rather than trees.
862 (vect_prune_runtime_alias_test_list): Update call accordingly.
864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
865 Alan Hayward <alan.hayward@arm.com>
866 David Sherwood <david.sherwood@arm.com>
868 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
870 (vect_schedule_slp_instance): Likewise.
872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
873 Alan Hayward <alan.hayward@arm.com>
874 David Sherwood <david.sherwood@arm.com>
876 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
877 constant and extern definitions for variable-length vectors.
878 (vect_get_constant_vectors): Note that the number of units
879 is known to be constant.
881 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
882 Alan Hayward <alan.hayward@arm.com>
883 David Sherwood <david.sherwood@arm.com>
885 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
886 of units as polynomial. Choose between WIDE and NARROW based
889 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
890 Alan Hayward <alan.hayward@arm.com>
891 David Sherwood <david.sherwood@arm.com>
893 * tree-vect-stmts.c (simd_clone_subparts): New function.
894 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
896 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
897 Alan Hayward <alan.hayward@arm.com>
898 David Sherwood <david.sherwood@arm.com>
900 * tree-vect-stmts.c (vectorizable_call): Treat the number of
901 vectors as polynomial. Use build_index_vector for
904 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
905 Alan Hayward <alan.hayward@arm.com>
906 David Sherwood <david.sherwood@arm.com>
908 * tree-vect-stmts.c (get_load_store_type): Treat the number of
909 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
910 for variable-length vectors.
911 (vectorizable_mask_load_store): Treat the number of units as
912 polynomial, asserting that it is constant if the condition has
913 already been enforced.
914 (vectorizable_store, vectorizable_load): Likewise.
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-vect-loop.c (vectorizable_live_operation): Treat the number
921 of units as polynomial. Punt if we can't tell at compile time
922 which vector contains the final result.
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-loop.c (vectorizable_induction): Treat the number
929 of units as polynomial. Punt on SLP inductions. Use an integer
930 VEC_SERIES_EXPR for variable-length integer reductions. Use a
931 cast of such a series for variable-length floating-point
934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
935 Alan Hayward <alan.hayward@arm.com>
936 David Sherwood <david.sherwood@arm.com>
938 * tree.h (build_index_vector): Declare.
939 * tree.c (build_index_vector): New function.
940 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
941 of units as polynomial, forcibly converting it to a constant if
942 vectorizable_reduction has already enforced the condition.
943 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
944 to create a {1,2,3,...} vector.
945 (vectorizable_reduction): Treat the number of units as polynomial.
946 Choose vectype_in based on the largest scalar element size rather
947 than the smallest number of units. Enforce the restrictions
950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
951 Alan Hayward <alan.hayward@arm.com>
952 David Sherwood <david.sherwood@arm.com>
954 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
955 number of units as polynomial.
957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
958 Alan Hayward <alan.hayward@arm.com>
959 David Sherwood <david.sherwood@arm.com>
961 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
962 * target.def (autovectorize_vector_sizes): Return the vector sizes
963 by pointer, using vector_sizes rather than a bitmask.
964 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
965 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
966 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
968 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
969 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
970 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
971 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
972 * omp-general.c (omp_max_vf): Likewise.
973 * omp-low.c (omp_clause_aligned_alignment): Likewise.
974 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
975 * tree-vect-loop.c (vect_analyze_loop): Likewise.
976 * tree-vect-slp.c (vect_slp_bb): Likewise.
977 * doc/tm.texi: Regenerate.
978 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
980 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
981 the vector size as a poly_uint64 rather than an unsigned int.
982 (current_vector_size): Change from an unsigned int to a poly_uint64.
983 (get_vectype_for_scalar_type): Update accordingly.
984 * tree.h (build_truth_vector_type): Take the size and number of
985 units as a poly_uint64 rather than an unsigned int.
986 (build_vector_type): Add a temporary overload that takes
987 the number of units as a poly_uint64 rather than an unsigned int.
988 * tree.c (make_vector_type): Likewise.
989 (build_truth_vector_type): Take the number of units as a poly_uint64
990 rather than an unsigned int.
992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
993 Alan Hayward <alan.hayward@arm.com>
994 David Sherwood <david.sherwood@arm.com>
996 * target.def (get_mask_mode): Take the number of units and length
997 as poly_uint64s rather than unsigned ints.
998 * targhooks.h (default_get_mask_mode): Update accordingly.
999 * targhooks.c (default_get_mask_mode): Likewise.
1000 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1001 * doc/tm.texi: Regenerate.
1003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1004 Alan Hayward <alan.hayward@arm.com>
1005 David Sherwood <david.sherwood@arm.com>
1007 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1008 * omp-general.c (omp_max_vf): Likewise.
1009 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1010 (expand_omp_simd): Handle polynomial safelen.
1011 * omp-low.c (omplow_simd_context): Add a default constructor.
1012 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1013 (lower_rec_simd_input_clauses): Update accordingly.
1014 (lower_rec_input_clauses): Likewise.
1016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1017 Alan Hayward <alan.hayward@arm.com>
1018 David Sherwood <david.sherwood@arm.com>
1020 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1021 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1022 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1023 (vect_analyze_slp_cost): Likewise.
1024 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1025 (vect_model_load_cost): Likewise.
1027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1028 Alan Hayward <alan.hayward@arm.com>
1029 David Sherwood <david.sherwood@arm.com>
1031 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1032 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1033 from an unsigned int * to a poly_uint64_pod *.
1034 (calculate_unrolling_factor): New function.
1035 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1038 Alan Hayward <alan.hayward@arm.com>
1039 David Sherwood <david.sherwood@arm.com>
1041 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1042 from an unsigned int to a poly_uint64.
1043 (_loop_vec_info::slp_unrolling_factor): Likewise.
1044 (_loop_vec_info::vectorization_factor): Change from an int
1046 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1047 (vect_get_num_vectors): New function.
1048 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1049 (vect_get_num_copies): Use vect_get_num_vectors.
1050 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1051 to an unsigned int *.
1052 (vect_analyze_data_refs): Change min_vf from an int * to a
1054 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1055 than an unsigned HOST_WIDE_INT.
1056 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1057 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1058 to an unsigned int *.
1059 (vect_analyze_data_ref_dependences): Likewise.
1060 (vect_compute_data_ref_alignment): Handle polynomial vf.
1061 (vect_enhance_data_refs_alignment): Likewise.
1062 (vect_prune_runtime_alias_test_list): Likewise.
1063 (vect_shift_permute_load_chain): Likewise.
1064 (vect_supportable_dr_alignment): Likewise.
1065 (dependence_distance_ge_vf): Take the vectorization factor as a
1066 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1067 (vect_analyze_data_refs): Change min_vf from an int * to a
1069 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1070 vfm1 as a poly_uint64 rather than an int. Make the same change
1071 for the returned bound_scalar.
1072 (vect_gen_vector_loop_niters): Handle polynomial vf.
1073 (vect_do_peeling): Likewise. Update call to
1074 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1075 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1077 * tree-vect-loop.c (vect_determine_vectorization_factor)
1078 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1079 (vect_get_known_peeling_cost): Likewise.
1080 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1081 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1082 (vect_transform_loop): Likewise. Use the lowest possible VF when
1083 updating the upper bounds of the loop.
1084 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1086 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1087 polynomial unroll factors.
1088 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1089 (vect_make_slp_decision): Likewise.
1090 (vect_supported_load_permutation_p): Likewise, and polynomial
1092 (vect_analyze_slp_cost): Handle polynomial vf.
1093 (vect_slp_analyze_node_operations): Likewise.
1094 (vect_slp_analyze_bb_1): Likewise.
1095 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1096 than an unsigned HOST_WIDE_INT.
1097 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1098 (vectorizable_load): Handle polynomial vf.
1099 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1101 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1104 Alan Hayward <alan.hayward@arm.com>
1105 David Sherwood <david.sherwood@arm.com>
1107 * match.pd: Handle bit operations involving three constants
1108 and try to fold one pair.
1110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1112 * tree-vect-loop-manip.c: Include gimple-fold.h.
1113 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1114 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1115 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1116 Add a path that uses a step of VF instead of 1, but disable it
1118 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1119 and niters_no_overflow parameters. Update calls to
1120 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1121 Create a new SSA name if the latter choses to use a ste other
1122 than zero, and return it via niters_vector_mult_vf_var.
1123 * tree-vect-loop.c (vect_transform_loop): Update calls to
1124 vect_do_peeling, vect_gen_vector_loop_niters and
1125 slpeel_make_loop_iterate_ntimes.
1126 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1127 (vect_gen_vector_loop_niters): Update declarations after above changes.
1129 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1131 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1132 128-bit round to integer instructions.
1133 (ceil<mode>2): Likewise.
1134 (btrunc<mode>2): Likewise.
1135 (round<mode>2): Likewise.
1137 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1139 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1140 unaligned VSX load/store on P8/P9.
1141 (expand_block_clear): Allow the use of unaligned VSX
1142 load/store on P8/P9.
1144 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1146 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1148 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1149 swap associated with both a load and a store.
1151 2018-01-02 Andrew Waterman <andrew@sifive.com>
1153 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1154 * config/riscv/riscv.md (clear_cache): Use it.
1156 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1158 * web.c: Remove out-of-date comment.
1160 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1162 * expr.c (fixup_args_size_notes): Check that any existing
1163 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1164 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1165 (emit_single_push_insn): ...here.
1167 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1169 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1170 (const_vector_encoded_nelts): New function.
1171 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1172 (const_vector_int_elt, const_vector_elt): Declare.
1173 * emit-rtl.c (const_vector_int_elt_1): New function.
1174 (const_vector_elt): Likewise.
1175 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1176 of CONST_VECTOR_ELT.
1178 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1180 * expr.c: Include rtx-vector-builder.h.
1181 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1182 directly on the tree encoding.
1183 (const_vector_from_tree): Likewise.
1184 * optabs.c: Include rtx-vector-builder.h.
1185 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1186 sequence of "u" values.
1187 * vec-perm-indices.c: Include rtx-vector-builder.h.
1188 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1189 directly on the vec_perm_indices encoding.
1191 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1193 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1194 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1195 * rtx-vector-builder.h: New file.
1196 * rtx-vector-builder.c: Likewise.
1197 * rtl.h (rtx_def::u2): Add a const_vector field.
1198 (CONST_VECTOR_NPATTERNS): New macro.
1199 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1200 (CONST_VECTOR_DUPLICATE_P): Likewise.
1201 (CONST_VECTOR_STEPPED_P): Likewise.
1202 (CONST_VECTOR_ENCODED_ELT): Likewise.
1203 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1204 (unwrap_const_vec_duplicate): Likewise.
1205 (const_vec_series_p): Check for a non-duplicated vector encoding.
1206 Say that the function only returns true for integer vectors.
1207 * emit-rtl.c: Include rtx-vector-builder.h.
1208 (gen_const_vec_duplicate_1): Delete.
1209 (gen_const_vector): Call gen_const_vec_duplicate instead of
1210 gen_const_vec_duplicate_1.
1211 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1212 (gen_const_vec_duplicate): Use rtx_vector_builder.
1213 (gen_const_vec_series): Likewise.
1214 (gen_rtx_CONST_VECTOR): Likewise.
1215 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1216 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1217 Build a new vector rather than modifying a CONST_VECTOR in-place.
1218 (handle_special_swappables): Update call accordingly.
1219 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1220 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1221 Build a new vector rather than modifying a CONST_VECTOR in-place.
1222 (handle_special_swappables): Update call accordingly.
1224 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1226 * simplify-rtx.c (simplify_const_binary_operation): Use
1227 CONST_VECTOR_ELT instead of XVECEXP.
1229 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1231 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1232 the selector elements to be different from the data elements
1233 if the selector is a VECTOR_CST.
1234 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1235 ssizetype for the selector.
1237 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1239 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1240 before testing each element individually.
1241 * tree-vect-generic.c (lower_vec_perm): Likewise.
1243 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1245 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1246 * selftest-run-tests.c (selftest::run_tests): Call it.
1247 * vector-builder.h (vector_builder::operator ==): New function.
1248 (vector_builder::operator !=): Likewise.
1249 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1250 (vec_perm_indices::all_from_input_p): New function.
1251 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1252 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1253 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1254 instead of reading the VECTOR_CST directly. Detect whether both
1255 vector inputs are the same before constructing the vec_perm_indices,
1256 and update the number of inputs argument accordingly. Use the
1257 utility functions added above. Only construct sel2 if we need to.
1259 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1261 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1262 the broadcast of the low byte.
1263 (expand_mult_highpart): Use an explicit encoding for the permutes.
1264 * optabs-query.c (can_mult_highpart_p): Likewise.
1265 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1266 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1267 (vectorizable_bswap): Likewise.
1268 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1269 explicit encoding for the power-of-2 permutes.
1270 (vect_permute_store_chain): Likewise.
1271 (vect_grouped_load_supported): Likewise.
1272 (vect_permute_load_chain): Likewise.
1274 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1276 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1277 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1278 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1279 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1280 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1281 (vect_gen_perm_mask_any): Likewise.
1283 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1285 * int-vector-builder.h: New file.
1286 * vec-perm-indices.h: Include int-vector-builder.h.
1287 (vec_perm_indices): Redefine as an int_vector_builder.
1288 (auto_vec_perm_indices): Delete.
1289 (vec_perm_builder): Redefine as a stand-alone class.
1290 (vec_perm_indices::vec_perm_indices): New function.
1291 (vec_perm_indices::clamp): Likewise.
1292 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1293 (vec_perm_indices::new_vector): New function.
1294 (vec_perm_indices::new_expanded_vector): Update for new
1295 vec_perm_indices class.
1296 (vec_perm_indices::rotate_inputs): New function.
1297 (vec_perm_indices::all_in_range_p): Operate directly on the
1298 encoded form, without computing elided elements.
1299 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1300 encoding. Update for new vec_perm_indices class.
1301 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1302 the given vec_perm_builder.
1303 (expand_vec_perm_var): Update vec_perm_builder constructor.
1304 (expand_mult_highpart): Use vec_perm_builder instead of
1305 auto_vec_perm_indices.
1306 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1307 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1308 or double series encoding as appropriate.
1309 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1310 vec_perm_indices instead of auto_vec_perm_indices.
1311 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1312 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1313 (vect_permute_store_chain): Likewise.
1314 (vect_grouped_load_supported): Likewise.
1315 (vect_permute_load_chain): Likewise.
1316 (vect_shift_permute_load_chain): Likewise.
1317 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1318 (vect_transform_slp_perm_load): Likewise.
1319 (vect_schedule_slp_instance): Likewise.
1320 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1321 (vectorizable_mask_load_store): Likewise.
1322 (vectorizable_bswap): Likewise.
1323 (vectorizable_store): Likewise.
1324 (vectorizable_load): Likewise.
1325 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1326 vec_perm_indices instead of auto_vec_perm_indices. Use
1327 tree_to_vec_perm_builder to read the vector from a tree.
1328 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1329 vec_perm_builder instead of a vec_perm_indices.
1330 (have_whole_vector_shift): Use vec_perm_builder and
1331 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1332 truncation to calc_vec_perm_mask_for_shift.
1333 (vect_create_epilog_for_reduction): Likewise.
1334 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1335 from auto_vec_perm_indices to vec_perm_indices.
1336 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1337 instead of changing individual elements.
1338 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1339 the vector in d.perm.
1340 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1341 from auto_vec_perm_indices to vec_perm_indices.
1342 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1343 instead of changing individual elements.
1344 (arm_vectorize_vec_perm_const): Use new_vector to install
1345 the vector in d.perm.
1346 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1347 Update vec_perm_builder constructor.
1348 (rs6000_expand_interleave): Likewise.
1349 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1350 (rs6000_expand_interleave): Likewise.
1352 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1354 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1355 to qimode could truncate the indices.
1356 * optabs.c (expand_vec_perm_var): Likewise.
1358 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1360 * Makefile.in (OBJS): Add vec-perm-indices.o.
1361 * vec-perm-indices.h: New file.
1362 * vec-perm-indices.c: Likewise.
1363 * target.h (vec_perm_indices): Replace with a forward class
1365 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1366 * optabs.h: Include vec-perm-indices.h.
1367 (expand_vec_perm): Delete.
1368 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1369 (expand_vec_perm_const): Declare.
1370 * target.def (vec_perm_const_ok): Replace with...
1371 (vec_perm_const): ...this new hook.
1372 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1373 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1374 * doc/tm.texi: Regenerate.
1375 * optabs.def (vec_perm_const): Delete.
1376 * doc/md.texi (vec_perm_const): Likewise.
1377 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1378 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1379 expand_vec_perm for constant permutation vectors. Assert that
1380 the mode of variable permutation vectors is the integer equivalent
1381 of the mode that is being permuted.
1382 * optabs-query.h (selector_fits_mode_p): Declare.
1383 * optabs-query.c: Include vec-perm-indices.h.
1384 (selector_fits_mode_p): New function.
1385 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1386 is defined, instead of checking whether the vec_perm_const_optab
1387 exists. Use targetm.vectorize.vec_perm_const instead of
1388 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1389 fit in the vector mode before using a variable permute.
1390 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1391 vec_perm_indices instead of an rtx.
1392 (expand_vec_perm): Replace with...
1393 (expand_vec_perm_const): ...this new function. Take the selector
1394 as a vec_perm_indices rather than an rtx. Also take the mode of
1395 the selector. Update call to shift_amt_for_vec_perm_mask.
1396 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1397 Use vec_perm_indices::new_expanded_vector to expand the original
1398 selector into bytes. Check whether the indices fit in the vector
1399 mode before using a variable permute.
1400 (expand_vec_perm_var): Make global.
1401 (expand_mult_highpart): Use expand_vec_perm_const.
1402 * fold-const.c: Includes vec-perm-indices.h.
1403 * tree-ssa-forwprop.c: Likewise.
1404 * tree-vect-data-refs.c: Likewise.
1405 * tree-vect-generic.c: Likewise.
1406 * tree-vect-loop.c: Likewise.
1407 * tree-vect-slp.c: Likewise.
1408 * tree-vect-stmts.c: Likewise.
1409 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1411 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1412 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1413 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1414 (aarch64_vectorize_vec_perm_const): ...this new function.
1415 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1416 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1417 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1418 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1419 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1420 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1421 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1423 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1424 check for NEON modes.
1425 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1426 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1427 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1428 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1430 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1431 the old VEC_PERM_CONST conditions.
1432 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1433 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1434 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1435 (ia64_vectorize_vec_perm_const_ok): Merge into...
1436 (ia64_vectorize_vec_perm_const): ...this new function.
1437 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1438 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1439 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1440 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1441 * config/mips/mips.c (mips_expand_vec_perm_const)
1442 (mips_vectorize_vec_perm_const_ok): Merge into...
1443 (mips_vectorize_vec_perm_const): ...this new function.
1444 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1445 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1446 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1447 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1448 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1449 (rs6000_expand_vec_perm_const): Delete.
1450 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
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.
1460 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1461 an element count and rtx array.
1462 (rs6000_expand_extract_even): Update call accordingly.
1463 (rs6000_expand_interleave): Likewise.
1464 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1465 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1466 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1467 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1468 (rs6000_expand_vec_perm_const): Delete.
1469 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1470 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1471 (altivec_expand_vec_perm_const_le): Take each operand individually.
1472 Operate on constant selectors rather than rtxes.
1473 (altivec_expand_vec_perm_const): Likewise. Update call to
1474 altivec_expand_vec_perm_const_le.
1475 (rs6000_expand_vec_perm_const): Delete.
1476 (rs6000_vectorize_vec_perm_const_ok): Delete.
1477 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1478 reference to the SPE evmerge intructions.
1479 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1480 an element count and rtx array.
1481 (rs6000_expand_extract_even): Update call accordingly.
1482 (rs6000_expand_interleave): Likewise.
1483 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1484 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1486 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1488 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1490 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1491 vector mode and that that mode matches the mode of the data
1493 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1494 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1495 directly using expand_vec_perm_1 when forcing selectors into
1497 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1499 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1501 * optabs-query.h (can_vec_perm_p): Delete.
1502 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1503 * optabs-query.c (can_vec_perm_p): Split into...
1504 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1505 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1506 particular selector is valid.
1507 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1508 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1509 (vect_grouped_load_supported): Likewise.
1510 (vect_shift_permute_load_chain): Likewise.
1511 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1512 (vect_transform_slp_perm_load): Likewise.
1513 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1514 (vectorizable_bswap): Likewise.
1515 (vect_gen_perm_mask_checked): Likewise.
1516 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1517 implementations of variable permutation vectors into account
1518 when deciding which selector to use.
1519 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1520 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1521 with a false third argument.
1522 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1523 to test whether the constant selector is valid and can_vec_perm_var_p
1524 to test whether a variable selector is valid.
1526 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1528 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1529 * optabs-query.c (can_vec_perm_p): Likewise.
1530 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1531 instead of vec_perm_indices.
1532 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1533 (vect_gen_perm_mask_checked): Likewise,
1534 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1535 (vect_gen_perm_mask_checked): Likewise,
1537 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1539 * optabs-query.h (qimode_for_vec_perm): Declare.
1540 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1541 (qimode_for_vec_perm): ...this new function.
1542 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1544 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1546 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1547 does not have a conditional at the top.
1549 2018-01-02 Richard Biener <rguenther@suse.de>
1551 * ipa-inline.c (big_speedup_p): Fix expression.
1553 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1556 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1559 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1563 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1564 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1565 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1566 cond_taken_branch_cost 3->4.
1568 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1570 PR tree-optimization/83581
1571 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1572 TODO_cleanup_cfg if any changes have been made.
1575 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1576 convert_modes if target mode has the right side, but different mode
1580 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1581 last argument when extracting from CONCAT. If either from_real or
1582 from_imag is NULL, use expansion through memory. If result is not
1583 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1584 the parts directly to inner mode, if even that fails, use expansion
1588 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1589 check for bswap in mode rather than HImode and use that in expand_unop
1592 Copyright (C) 2018 Free Software Foundation, Inc.
1594 Copying and distribution of this file, with or without modification,
1595 are permitted in any medium without royalty provided the copyright
1596 notice and this notice are preserved.