1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4 IFN_MASK_LOAD calls here rather than...
5 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
7 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
8 Alan Hayward <alan.hayward@arm.com>
9 David Sherwood <david.sherwood@arm.com>
11 * expmed.c (extract_bit_field_1): For vector extracts,
12 fall back to extract_bit_field_as_subreg if vec_extract
15 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
16 Alan Hayward <alan.hayward@arm.com>
17 David Sherwood <david.sherwood@arm.com>
19 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
20 they are variable or constant sized.
21 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
22 slots for constant-sized data.
24 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
25 Alan Hayward <alan.hayward@arm.com>
26 David Sherwood <david.sherwood@arm.com>
28 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
29 handling COND_EXPRs with boolean comparisons, try to find a better
30 basis for the mask type than the boolean itself.
32 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
34 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
35 is calculated and how it can be overridden.
36 * genmodes.c (max_bitsize_mode_any_mode): New variable.
37 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
39 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
42 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
43 Alan Hayward <alan.hayward@arm.com>
44 David Sherwood <david.sherwood@arm.com>
46 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
47 Remove the mode argument.
48 (aarch64_simd_valid_immediate): Remove the mode and inverse
50 * config/aarch64/iterators.md (bitsize): New iterator.
51 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
52 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
53 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
54 aarch64_simd_valid_immediate.
55 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
56 (aarch64_reg_or_bic_imm): Likewise.
57 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
58 with an insn_type enum and msl with a modifier_type enum.
59 Replace element_width with a scalar_mode. Change the shift
60 to unsigned int. Add constructors for scalar_float_mode and
61 scalar_int_mode elements.
62 (aarch64_vect_float_const_representable_p): Delete.
63 (aarch64_can_const_movi_rtx_p)
64 (aarch64_simd_scalar_immediate_valid_for_move)
65 (aarch64_simd_make_constant): Update call to
66 aarch64_simd_valid_immediate.
67 (aarch64_advsimd_valid_immediate_hs): New function.
68 (aarch64_advsimd_valid_immediate): Likewise.
69 (aarch64_simd_valid_immediate): Remove mode and inverse
70 arguments. Rewrite to use the above. Use const_vec_duplicate_p
71 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
72 and aarch64_float_const_representable_p on the result.
73 (aarch64_output_simd_mov_immediate): Remove mode argument.
74 Update call to aarch64_simd_valid_immediate and use of
76 (aarch64_output_scalar_simd_mov_immediate): Update call
79 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
80 Alan Hayward <alan.hayward@arm.com>
81 David Sherwood <david.sherwood@arm.com>
83 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
84 (mode_nunits): Likewise CONST_MODE_NUNITS.
85 * machmode.def (ADJUST_NUNITS): Document.
86 * genmodes.c (mode_data::need_nunits_adj): New field.
87 (blank_mode): Update accordingly.
88 (adj_nunits): New variable.
89 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
91 (emit_mode_size_inline): Set need_bytesize_adj for all modes
93 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
94 listed in adj_nunits. Don't emit case statements for such modes.
95 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
96 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
97 nothing if adj_nunits is nonnull.
98 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
99 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
100 (emit_mode_fbit): Update use of print_maybe_const_decl.
101 (emit_move_size): Likewise. Treat the array as non-const
103 (emit_mode_adjustments): Handle adj_nunits.
105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
107 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
108 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
109 (VECTOR_MODES): Use it.
110 (make_vector_modes): Take the prefix as an argument.
112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
113 Alan Hayward <alan.hayward@arm.com>
114 David Sherwood <david.sherwood@arm.com>
116 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
117 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
118 for MODE_VECTOR_BOOL.
119 * machmode.def (VECTOR_BOOL_MODE): Document.
120 * genmodes.c (VECTOR_BOOL_MODE): New macro.
121 (make_vector_bool_mode): New function.
122 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
124 * lto-streamer-in.c (lto_input_mode_table): Likewise.
125 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
127 * stor-layout.c (int_mode_for_mode): Likewise.
128 * tree.c (build_vector_type_for_mode): Likewise.
129 * varasm.c (output_constant_pool_2): Likewise.
130 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
131 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
132 for MODE_VECTOR_BOOL.
133 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
134 of mode class checks.
135 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
136 instead of a list of mode class checks.
137 (expand_vector_scalar_condition): Likewise.
138 (type_for_widest_vector_mode): Handle BImode as an inner mode.
140 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
141 Alan Hayward <alan.hayward@arm.com>
142 David Sherwood <david.sherwood@arm.com>
144 * machmode.h (mode_size): Change from unsigned short to
146 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
147 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
148 or if measurement_type is not polynomial.
149 (fixed_size_mode::includes_p): Check for constant-sized modes.
150 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
151 return a poly_uint16 rather than an unsigned short.
152 (emit_mode_size): Change the type of mode_size from unsigned short
153 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
154 (emit_mode_adjustments): Cope with polynomial vector sizes.
155 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
157 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
159 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
160 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
161 * caller-save.c (setup_save_areas): Likewise.
162 (replace_reg_with_saved_mem): Likewise.
163 * calls.c (emit_library_call_value_1): Likewise.
164 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
165 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
166 (gen_lowpart_for_combine): Likewise.
167 * convert.c (convert_to_integer_1): Likewise.
168 * cse.c (equiv_constant, cse_insn): Likewise.
169 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
170 (cselib_subst_to_values): Likewise.
171 * dce.c (word_dce_process_block): Likewise.
172 * df-problems.c (df_word_lr_mark_ref): Likewise.
173 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
174 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
175 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
176 (rtl_for_decl_location): Likewise.
177 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
178 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
179 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
180 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
181 (expand_expr_real_1): Likewise.
182 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
183 (pad_below): Likewise.
184 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
185 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
186 * ira.c (get_subreg_tracking_sizes): Likewise.
187 * ira-build.c (ira_create_allocno_objects): Likewise.
188 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
189 (ira_sort_regnos_for_alter_reg): Likewise.
190 * ira-costs.c (record_operand_costs): Likewise.
191 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
192 (resolve_simple_move): Likewise.
193 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
194 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
195 (lra_constraints): Likewise.
196 (CONST_POOL_OK_P): Reject variable-sized modes.
197 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
198 (add_pseudo_to_slot, lra_spill): Likewise.
199 * omp-low.c (omp_clause_aligned_alignment): Likewise.
200 * optabs-query.c (get_best_extraction_insn): Likewise.
201 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
202 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
203 (expand_mult_highpart, valid_multiword_target_p): Likewise.
204 * recog.c (offsettable_address_addr_space_p): Likewise.
205 * regcprop.c (maybe_mode_change): Likewise.
206 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
207 * regrename.c (build_def_use): Likewise.
208 * regstat.c (dump_reg_info): Likewise.
209 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
210 (find_reloads, find_reloads_subreg_address): Likewise.
211 * reload1.c (eliminate_regs_1): Likewise.
212 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
213 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
214 (simplify_binary_operation_1, simplify_subreg): Likewise.
215 * targhooks.c (default_function_arg_padding): Likewise.
216 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
217 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
218 (verify_gimple_assign_ternary): Likewise.
219 * tree-inline.c (estimate_move_cost): Likewise.
220 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
221 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
222 (get_address_cost_ainc): Likewise.
223 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
224 (vect_supportable_dr_alignment): Likewise.
225 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
226 (vectorizable_reduction): Likewise.
227 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
228 (vectorizable_operation, vectorizable_load): Likewise.
229 * tree.c (build_same_sized_truth_vector_type): Likewise.
230 * valtrack.c (cleanup_auto_inc_dec): Likewise.
231 * var-tracking.c (emit_note_insn_var_location): Likewise.
232 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
233 (ADDR_VEC_ALIGN): Likewise.
235 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
236 Alan Hayward <alan.hayward@arm.com>
237 David Sherwood <david.sherwood@arm.com>
239 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
241 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
242 or if measurement_type is polynomial.
243 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
244 * combine.c (make_extraction): Likewise.
245 * dse.c (find_shift_sequence): Likewise.
246 * dwarf2out.c (mem_loc_descriptor): Likewise.
247 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
248 (extract_bit_field, extract_low_bits): Likewise.
249 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
250 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
251 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
252 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
253 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
254 * reload.c (find_reloads): Likewise.
255 * reload1.c (alter_reg): Likewise.
256 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
257 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
258 * tree-if-conv.c (predicate_mem_writes): Likewise.
259 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
260 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
261 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
262 * valtrack.c (dead_debug_insert_temp): Likewise.
263 * varasm.c (mergeable_constant_section): Likewise.
264 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
267 Alan Hayward <alan.hayward@arm.com>
268 David Sherwood <david.sherwood@arm.com>
270 * expr.c (expand_assignment): Cope with polynomial mode sizes
271 when assigning to a CONCAT.
273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
274 Alan Hayward <alan.hayward@arm.com>
275 David Sherwood <david.sherwood@arm.com>
277 * machmode.h (mode_precision): Change from unsigned short to
279 (mode_to_precision): Return a poly_uint16 rather than an unsigned
281 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
282 or if measurement_type is not polynomial.
283 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
284 in which the mode is already known to be a scalar_int_mode.
285 * genmodes.c (emit_mode_precision): Change the type of mode_precision
286 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
288 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
289 for GET_MODE_PRECISION.
290 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
291 for GET_MODE_PRECISION.
292 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
294 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
295 (expand_field_assignment, make_extraction): Likewise.
296 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
297 (get_last_value): Likewise.
298 * convert.c (convert_to_integer_1): Likewise.
299 * cse.c (cse_insn): Likewise.
300 * expr.c (expand_expr_real_1): Likewise.
301 * lra-constraints.c (simplify_operand_subreg): Likewise.
302 * optabs-query.c (can_atomic_load_p): Likewise.
303 * optabs.c (expand_atomic_load): Likewise.
304 (expand_atomic_store): Likewise.
305 * ree.c (combine_reaching_defs): Likewise.
306 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
307 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
308 * tree.h (type_has_mode_precision_p): Likewise.
309 * ubsan.c (instrument_si_overflow): Likewise.
311 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
312 Alan Hayward <alan.hayward@arm.com>
313 David Sherwood <david.sherwood@arm.com>
315 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
316 polynomial numbers of units.
317 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
318 (valid_vector_subparts_p): New function.
319 (build_vector_type): Remove temporary shim and take the number
320 of units as a poly_uint64 rather than an int.
321 (build_opaque_vector_type): Take the number of units as a
322 poly_uint64 rather than an int.
323 * tree.c (build_vector_from_ctor): Handle polynomial
324 TYPE_VECTOR_SUBPARTS.
325 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
326 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
327 (build_vector_from_val): If the number of units is variable,
328 use build_vec_duplicate_cst for constant operands and
329 VEC_DUPLICATE_EXPR otherwise.
330 (make_vector_type): Remove temporary is_constant ().
331 (build_vector_type, build_opaque_vector_type): Take the number of
332 units as a poly_uint64 rather than an int.
333 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
335 * cfgexpand.c (expand_debug_expr): Likewise.
336 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
337 (store_constructor, expand_expr_real_1): Likewise.
338 (const_scalar_mask_from_tree): Likewise.
339 * fold-const-call.c (fold_const_reduction): Likewise.
340 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
341 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
342 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
343 (fold_relational_const): Likewise.
344 (native_interpret_vector): Likewise. Change the size from an
345 int to an unsigned int.
346 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
347 TYPE_VECTOR_SUBPARTS.
348 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
349 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
350 duplicating a non-constant operand into a variable-length vector.
351 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
352 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
353 * ipa-icf.c (sem_variable::equals): Likewise.
354 * match.pd: Likewise.
355 * omp-simd-clone.c (simd_clone_subparts): Likewise.
356 * print-tree.c (print_node): Likewise.
357 * stor-layout.c (layout_type): Likewise.
358 * targhooks.c (default_builtin_vectorization_cost): Likewise.
359 * tree-cfg.c (verify_gimple_comparison): Likewise.
360 (verify_gimple_assign_binary): Likewise.
361 (verify_gimple_assign_ternary): Likewise.
362 (verify_gimple_assign_single): Likewise.
363 * tree-pretty-print.c (dump_generic_node): Likewise.
364 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
365 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
366 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
367 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
368 (vect_shift_permute_load_chain): Likewise.
369 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
370 (expand_vector_condition, optimize_vector_constructor): Likewise.
371 (lower_vec_perm, get_compute_type): Likewise.
372 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
373 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
374 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
375 (vect_recog_mask_conversion_pattern): Likewise.
376 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
377 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
378 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
379 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
380 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
381 (vectorizable_shift, vectorizable_operation, vectorizable_store)
382 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
383 (supportable_widening_operation): Likewise.
384 (supportable_narrowing_operation): Likewise.
385 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
387 * varasm.c (output_constant): Likewise.
389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
390 Alan Hayward <alan.hayward@arm.com>
391 David Sherwood <david.sherwood@arm.com>
393 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
394 so that both the length == 3 and length != 3 cases set up their
395 own permute vectors. Add comments explaining why we know the
396 number of elements is constant.
397 (vect_permute_load_chain): Likewise.
399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
400 Alan Hayward <alan.hayward@arm.com>
401 David Sherwood <david.sherwood@arm.com>
403 * machmode.h (mode_nunits): Change from unsigned char to
405 (ONLY_FIXED_SIZE_MODES): New macro.
406 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
407 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
408 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
410 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
411 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
412 or if measurement_type is not polynomial.
413 * genmodes.c (ZERO_COEFFS): New macro.
414 (emit_mode_nunits_inline): Make mode_nunits_inline return a
416 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
417 Use ZERO_COEFFS when emitting initializers.
418 * data-streamer.h (bp_pack_poly_value): New function.
419 (bp_unpack_poly_value): Likewise.
420 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
422 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
424 * tree.c (make_vector_type): Remove temporary shim and make
425 the real function take the number of units as a poly_uint64
427 (build_vector_type_for_mode): Handle polynomial nunits.
428 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
429 * emit-rtl.c (const_vec_series_p_1): Likewise.
430 (gen_rtx_CONST_VECTOR): Likewise.
431 * fold-const.c (test_vec_duplicate_folding): Likewise.
432 * genrecog.c (validate_pattern): Likewise.
433 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
434 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
435 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
436 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
437 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
438 * rtlanal.c (subreg_get_info): Likewise.
439 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
440 (vect_grouped_load_supported): Likewise.
441 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
442 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
443 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
444 (simplify_const_unary_operation, simplify_binary_operation_1)
445 (simplify_const_binary_operation, simplify_ternary_operation)
446 (test_vector_ops_duplicate, test_vector_ops): Likewise.
447 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
448 instead of CONST_VECTOR_NUNITS.
449 * varasm.c (output_constant_pool_2): Likewise.
450 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
451 explicit-encoded elements in the XVEC for variable-length vectors.
453 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
455 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
457 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
458 Alan Hayward <alan.hayward@arm.com>
459 David Sherwood <david.sherwood@arm.com>
461 * coretypes.h (fixed_size_mode): Declare.
462 (fixed_size_mode_pod): New typedef.
463 * builtins.h (target_builtins::x_apply_args_mode)
464 (target_builtins::x_apply_result_mode): Change type to
466 * builtins.c (apply_args_size, apply_result_size, result_vector)
467 (expand_builtin_apply_args_1, expand_builtin_apply)
468 (expand_builtin_return): Update accordingly.
470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
472 * cse.c (hash_rtx_cb): Hash only the encoded elements.
473 * cselib.c (cselib_hash_rtx): Likewise.
474 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
475 CONST_VECTOR encoding.
477 2017-01-03 Jakub Jelinek <jakub@redhat.com>
478 Jeff Law <law@redhat.com>
481 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
482 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
483 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
484 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
487 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
488 explicitly probe *sp in a noreturn function if there were any callee
489 register saves or frame pointer is needed.
491 2018-01-03 Jakub Jelinek <jakub@redhat.com>
494 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
495 BLKmode for ternary, binary or unary expressions.
498 * var-tracking.c (delete_vta_debug_insn): New inline function.
499 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
500 insns from get_insns () to NULL instead of each bb separately.
501 Use delete_vta_debug_insn. No longer static.
502 (vt_debug_insns_local, variable_tracking_main_1): Adjust
503 delete_vta_debug_insns callers.
504 * rtl.h (delete_vta_debug_insns): Declare.
505 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
506 instead of variable_tracking_main.
508 2018-01-03 Martin Sebor <msebor@redhat.com>
511 * doc/extend.texi (attribute const): Fix a typo.
512 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
513 issuing -Wsuggest-attribute for void functions.
515 2018-01-03 Martin Sebor <msebor@redhat.com>
517 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
518 offset_int::from instead of wide_int::to_shwi.
519 (maybe_diag_overlap): Remove assertion.
520 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
521 * gimple-ssa-sprintf.c (format_directive): Same.
522 (parse_directive): Same.
523 (sprintf_dom_walker::compute_format_length): Same.
524 (try_substitute_return_value): Same.
526 2017-01-03 Jeff Law <law@redhat.com>
529 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
530 non-constant residual for zero at runtime and avoid probing in
531 that case. Reorganize code for trailing problem to mirror handling
534 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
536 PR tree-optimization/83501
537 * tree-ssa-strlen.c (get_string_cst): New.
538 (handle_char_store): Call get_string_cst.
540 2018-01-03 Martin Liska <mliska@suse.cz>
542 PR tree-optimization/83593
543 * tree-ssa-strlen.c: Include tree-cfg.h.
544 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
545 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
546 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
548 (strlen_dom_walker::before_dom_children): Call
549 gimple_purge_dead_eh_edges. Dump tranformation with details
551 (strlen_dom_walker::before_dom_children): Update call by adding
552 new argument cleanup_eh.
553 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
555 2018-01-03 Martin Liska <mliska@suse.cz>
558 * cif-code.def (VARIADIC_THUNK): New enum value.
559 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
562 2018-01-03 Jan Beulich <jbeulich@suse.com>
564 * sse.md (mov<mode>_internal): Tighten condition for when to use
565 vmovdqu<ssescalarsize> for TI and OI modes.
567 2018-01-03 Jakub Jelinek <jakub@redhat.com>
569 Update copyright years.
571 2018-01-03 Martin Liska <mliska@suse.cz>
574 * ipa-visibility.c (function_and_variable_visibility): Skip
575 functions with noipa attribure.
577 2018-01-03 Jakub Jelinek <jakub@redhat.com>
579 * gcc.c (process_command): Update copyright notice dates.
580 * gcov-dump.c (print_version): Ditto.
581 * gcov.c (print_version): Ditto.
582 * gcov-tool.c (print_version): Ditto.
583 * gengtype.c (create_file): Ditto.
584 * doc/cpp.texi: Bump @copying's copyright year.
585 * doc/cppinternals.texi: Ditto.
586 * doc/gcc.texi: Ditto.
587 * doc/gccint.texi: Ditto.
588 * doc/gcov.texi: Ditto.
589 * doc/install.texi: Ditto.
590 * doc/invoke.texi: Ditto.
592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
594 * vector-builder.h (vector_builder::m_full_nelts): Change from
595 unsigned int to poly_uint64.
596 (vector_builder::full_nelts): Update prototype accordingly.
597 (vector_builder::new_vector): Likewise.
598 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
599 (vector_builder::operator ==): Likewise.
600 (vector_builder::finalize): Likewise.
601 * int-vector-builder.h (int_vector_builder::int_vector_builder):
602 Take the number of elements as a poly_uint64 rather than an
604 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
605 from unsigned int to poly_uint64.
606 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
607 (vec_perm_indices::new_vector): Likewise.
608 (vec_perm_indices::length): Likewise.
609 (vec_perm_indices::nelts_per_input): Likewise.
610 (vec_perm_indices::input_nelts): Likewise.
611 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
612 number of elements per input as a poly_uint64 rather than an
613 unsigned int. Use the original encoding for variable-length
614 vectors, rather than clamping each individual element.
615 For the second and subsequent elements in each pattern,
616 clamp the step and base before clamping their sum.
617 (vec_perm_indices::series_p): Handle polynomial element counts.
618 (vec_perm_indices::all_in_range_p): Likewise.
619 (vec_perm_indices_to_tree): Likewise.
620 (vec_perm_indices_to_rtx): Likewise.
621 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
622 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
623 (tree_vector_builder::new_binary_operation): Handle polynomial
624 element counts. Return false if we need to know the number
625 of elements at compile time.
626 * fold-const.c (fold_vec_perm): Punt if the number of elements
627 isn't known at compile time.
629 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
631 * vec-perm-indices.h (vec_perm_builder): Change element type
632 from HOST_WIDE_INT to poly_int64.
633 (vec_perm_indices::element_type): Update accordingly.
634 (vec_perm_indices::clamp): Handle polynomial element_types.
635 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
636 (vec_perm_indices::all_in_range_p): Likewise.
637 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
639 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
640 polynomial vec_perm_indices element types.
641 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
642 * fold-const.c (fold_vec_perm): Likewise.
643 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
644 * tree-vect-generic.c (lower_vec_perm): Likewise.
645 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
646 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
647 element type to HOST_WIDE_INT.
649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
650 Alan Hayward <alan.hayward@arm.com>
651 David Sherwood <david.sherwood@arm.com>
653 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
654 rather than an int. Use plus_constant.
655 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
656 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
658 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
659 Alan Hayward <alan.hayward@arm.com>
660 David Sherwood <david.sherwood@arm.com>
662 * calls.c (emit_call_1, expand_call): Change struct_value_size from
663 a HOST_WIDE_INT to a poly_int64.
665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
666 Alan Hayward <alan.hayward@arm.com>
667 David Sherwood <david.sherwood@arm.com>
669 * calls.c (load_register_parameters): Cope with polynomial
670 mode sizes. Require a constant size for BLKmode parameters
671 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
672 forces a parameter to be padded at the lsb end in order to
673 fill a complete number of words, require the parameter size
674 to be ordered wrt UNITS_PER_WORD.
676 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
677 Alan Hayward <alan.hayward@arm.com>
678 David Sherwood <david.sherwood@arm.com>
680 * reload1.c (spill_stack_slot_width): Change element type
681 from unsigned int to poly_uint64_pod.
682 (alter_reg): Treat mode sizes as polynomial.
684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
685 Alan Hayward <alan.hayward@arm.com>
686 David Sherwood <david.sherwood@arm.com>
688 * reload.c (complex_word_subreg_p): New function.
689 (reload_inner_reg_of_subreg, push_reload): Use it.
691 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
692 Alan Hayward <alan.hayward@arm.com>
693 David Sherwood <david.sherwood@arm.com>
695 * lra-constraints.c (process_alt_operands): Reject matched
696 operands whose sizes aren't ordered.
697 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
704 that the mode size is in the set {1, 2, 4, 8, 16}.
706 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
707 Alan Hayward <alan.hayward@arm.com>
708 David Sherwood <david.sherwood@arm.com>
710 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
711 Use plus_constant instead of gen_rtx_PLUS.
713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
714 Alan Hayward <alan.hayward@arm.com>
715 David Sherwood <david.sherwood@arm.com>
717 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
718 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
719 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
720 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
721 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
722 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
723 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
724 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
725 * config/i386/i386.c (ix86_push_rounding): ...this new function.
726 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
728 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
729 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
730 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
731 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
732 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
733 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
734 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
735 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
736 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
737 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
739 * expr.c (emit_move_resolve_push): Treat the input and result
740 of PUSH_ROUNDING as a poly_int64.
741 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
742 (emit_push_insn): Likewise.
743 * lra-eliminations.c (mark_not_eliminable): Likewise.
744 * recog.c (push_operand): Likewise.
745 * reload1.c (elimination_effects): Likewise.
746 * rtlanal.c (nonzero_bits1): Likewise.
747 * calls.c (store_one_arg): Likewise. Require the padding to be
748 known at compile time.
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 (emit_single_push_insn_1): Treat mode sizes as polynomial.
755 Use plus_constant instead of gen_rtx_PLUS.
757 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
758 Alan Hayward <alan.hayward@arm.com>
759 David Sherwood <david.sherwood@arm.com>
761 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
765 Alan Hayward <alan.hayward@arm.com>
766 David Sherwood <david.sherwood@arm.com>
768 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
769 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
770 via stack temporaries. Treat the mode size as polynomial too.
772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
773 Alan Hayward <alan.hayward@arm.com>
774 David Sherwood <david.sherwood@arm.com>
776 * expr.c (expand_expr_real_2): When handling conversions involving
777 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
778 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
779 as a poly_uint64 too.
781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
782 Alan Hayward <alan.hayward@arm.com>
783 David Sherwood <david.sherwood@arm.com>
785 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
792 REGMODE_NATURAL_SIZE.
793 * expmed.c (store_bit_field_1): Likewise.
794 * expr.c (store_constructor): Likewise.
795 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
796 and polynomial REGMODE_NATURAL_SIZE.
797 (gen_lowpart_common): Likewise.
798 * reginfo.c (record_subregs_of_mode): Likewise.
799 * rtlanal.c (read_modify_subreg_p): Likewise.
801 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
802 Alan Hayward <alan.hayward@arm.com>
803 David Sherwood <david.sherwood@arm.com>
805 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
808 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
809 Alan Hayward <alan.hayward@arm.com>
810 David Sherwood <david.sherwood@arm.com>
812 * match.pd: Cope with polynomial numbers of vector elements.
814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
815 Alan Hayward <alan.hayward@arm.com>
816 David Sherwood <david.sherwood@arm.com>
818 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
819 in a POINTER_PLUS_EXPR.
821 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
822 Alan Hayward <alan.hayward@arm.com>
823 David Sherwood <david.sherwood@arm.com>
825 * omp-simd-clone.c (simd_clone_subparts): New function.
826 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
827 (ipa_simd_modify_function_body): Likewise.
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-generic.c (nunits_for_known_piecewise_op): New function.
834 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
835 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
836 (expand_vector_condition, vector_element): Likewise.
837 (subparts_gt): New function.
838 (get_compute_type): Use subparts_gt.
839 (count_type_subparts): Delete.
840 (expand_vector_operations_1): Use subparts_gt instead of
843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
844 Alan Hayward <alan.hayward@arm.com>
845 David Sherwood <david.sherwood@arm.com>
847 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
848 (vect_compile_time_alias): ...this new function. Do the calculation
849 on poly_ints rather than trees.
850 (vect_prune_runtime_alias_test_list): Update call accordingly.
852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
853 Alan Hayward <alan.hayward@arm.com>
854 David Sherwood <david.sherwood@arm.com>
856 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
858 (vect_schedule_slp_instance): Likewise.
860 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
861 Alan Hayward <alan.hayward@arm.com>
862 David Sherwood <david.sherwood@arm.com>
864 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
865 constant and extern definitions for variable-length vectors.
866 (vect_get_constant_vectors): Note that the number of units
867 is known to be constant.
869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
870 Alan Hayward <alan.hayward@arm.com>
871 David Sherwood <david.sherwood@arm.com>
873 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
874 of units as polynomial. Choose between WIDE and NARROW based
877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
878 Alan Hayward <alan.hayward@arm.com>
879 David Sherwood <david.sherwood@arm.com>
881 * tree-vect-stmts.c (simd_clone_subparts): New function.
882 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
885 Alan Hayward <alan.hayward@arm.com>
886 David Sherwood <david.sherwood@arm.com>
888 * tree-vect-stmts.c (vectorizable_call): Treat the number of
889 vectors as polynomial. Use build_index_vector for
892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
893 Alan Hayward <alan.hayward@arm.com>
894 David Sherwood <david.sherwood@arm.com>
896 * tree-vect-stmts.c (get_load_store_type): Treat the number of
897 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
898 for variable-length vectors.
899 (vectorizable_mask_load_store): Treat the number of units as
900 polynomial, asserting that it is constant if the condition has
901 already been enforced.
902 (vectorizable_store, vectorizable_load): Likewise.
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-loop.c (vectorizable_live_operation): Treat the number
909 of units as polynomial. Punt if we can't tell at compile time
910 which vector contains the final result.
912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
913 Alan Hayward <alan.hayward@arm.com>
914 David Sherwood <david.sherwood@arm.com>
916 * tree-vect-loop.c (vectorizable_induction): Treat the number
917 of units as polynomial. Punt on SLP inductions. Use an integer
918 VEC_SERIES_EXPR for variable-length integer reductions. Use a
919 cast of such a series for variable-length floating-point
922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
923 Alan Hayward <alan.hayward@arm.com>
924 David Sherwood <david.sherwood@arm.com>
926 * tree.h (build_index_vector): Declare.
927 * tree.c (build_index_vector): New function.
928 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
929 of units as polynomial, forcibly converting it to a constant if
930 vectorizable_reduction has already enforced the condition.
931 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
932 to create a {1,2,3,...} vector.
933 (vectorizable_reduction): Treat the number of units as polynomial.
934 Choose vectype_in based on the largest scalar element size rather
935 than the smallest number of units. Enforce the restrictions
938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
939 Alan Hayward <alan.hayward@arm.com>
940 David Sherwood <david.sherwood@arm.com>
942 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
943 number of units as polynomial.
945 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
946 Alan Hayward <alan.hayward@arm.com>
947 David Sherwood <david.sherwood@arm.com>
949 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
950 * target.def (autovectorize_vector_sizes): Return the vector sizes
951 by pointer, using vector_sizes rather than a bitmask.
952 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
953 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
954 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
956 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
957 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
958 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
959 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
960 * omp-general.c (omp_max_vf): Likewise.
961 * omp-low.c (omp_clause_aligned_alignment): Likewise.
962 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
963 * tree-vect-loop.c (vect_analyze_loop): Likewise.
964 * tree-vect-slp.c (vect_slp_bb): Likewise.
965 * doc/tm.texi: Regenerate.
966 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
968 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
969 the vector size as a poly_uint64 rather than an unsigned int.
970 (current_vector_size): Change from an unsigned int to a poly_uint64.
971 (get_vectype_for_scalar_type): Update accordingly.
972 * tree.h (build_truth_vector_type): Take the size and number of
973 units as a poly_uint64 rather than an unsigned int.
974 (build_vector_type): Add a temporary overload that takes
975 the number of units as a poly_uint64 rather than an unsigned int.
976 * tree.c (make_vector_type): Likewise.
977 (build_truth_vector_type): Take the number of units as a poly_uint64
978 rather than an unsigned int.
980 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
981 Alan Hayward <alan.hayward@arm.com>
982 David Sherwood <david.sherwood@arm.com>
984 * target.def (get_mask_mode): Take the number of units and length
985 as poly_uint64s rather than unsigned ints.
986 * targhooks.h (default_get_mask_mode): Update accordingly.
987 * targhooks.c (default_get_mask_mode): Likewise.
988 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
989 * doc/tm.texi: Regenerate.
991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
992 Alan Hayward <alan.hayward@arm.com>
993 David Sherwood <david.sherwood@arm.com>
995 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
996 * omp-general.c (omp_max_vf): Likewise.
997 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
998 (expand_omp_simd): Handle polynomial safelen.
999 * omp-low.c (omplow_simd_context): Add a default constructor.
1000 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1001 (lower_rec_simd_input_clauses): Update accordingly.
1002 (lower_rec_input_clauses): Likewise.
1004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1005 Alan Hayward <alan.hayward@arm.com>
1006 David Sherwood <david.sherwood@arm.com>
1008 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1009 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1010 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1011 (vect_analyze_slp_cost): Likewise.
1012 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1013 (vect_model_load_cost): Likewise.
1015 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1016 Alan Hayward <alan.hayward@arm.com>
1017 David Sherwood <david.sherwood@arm.com>
1019 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1020 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1021 from an unsigned int * to a poly_uint64_pod *.
1022 (calculate_unrolling_factor): New function.
1023 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1025 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1026 Alan Hayward <alan.hayward@arm.com>
1027 David Sherwood <david.sherwood@arm.com>
1029 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1030 from an unsigned int to a poly_uint64.
1031 (_loop_vec_info::slp_unrolling_factor): Likewise.
1032 (_loop_vec_info::vectorization_factor): Change from an int
1034 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1035 (vect_get_num_vectors): New function.
1036 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1037 (vect_get_num_copies): Use vect_get_num_vectors.
1038 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1039 to an unsigned int *.
1040 (vect_analyze_data_refs): Change min_vf from an int * to a
1042 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1043 than an unsigned HOST_WIDE_INT.
1044 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1045 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1046 to an unsigned int *.
1047 (vect_analyze_data_ref_dependences): Likewise.
1048 (vect_compute_data_ref_alignment): Handle polynomial vf.
1049 (vect_enhance_data_refs_alignment): Likewise.
1050 (vect_prune_runtime_alias_test_list): Likewise.
1051 (vect_shift_permute_load_chain): Likewise.
1052 (vect_supportable_dr_alignment): Likewise.
1053 (dependence_distance_ge_vf): Take the vectorization factor as a
1054 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1055 (vect_analyze_data_refs): Change min_vf from an int * to a
1057 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1058 vfm1 as a poly_uint64 rather than an int. Make the same change
1059 for the returned bound_scalar.
1060 (vect_gen_vector_loop_niters): Handle polynomial vf.
1061 (vect_do_peeling): Likewise. Update call to
1062 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1063 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1065 * tree-vect-loop.c (vect_determine_vectorization_factor)
1066 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1067 (vect_get_known_peeling_cost): Likewise.
1068 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1069 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1070 (vect_transform_loop): Likewise. Use the lowest possible VF when
1071 updating the upper bounds of the loop.
1072 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1074 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1075 polynomial unroll factors.
1076 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1077 (vect_make_slp_decision): Likewise.
1078 (vect_supported_load_permutation_p): Likewise, and polynomial
1080 (vect_analyze_slp_cost): Handle polynomial vf.
1081 (vect_slp_analyze_node_operations): Likewise.
1082 (vect_slp_analyze_bb_1): Likewise.
1083 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1084 than an unsigned HOST_WIDE_INT.
1085 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1086 (vectorizable_load): Handle polynomial vf.
1087 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1089 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1092 Alan Hayward <alan.hayward@arm.com>
1093 David Sherwood <david.sherwood@arm.com>
1095 * match.pd: Handle bit operations involving three constants
1096 and try to fold one pair.
1098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1100 * tree-vect-loop-manip.c: Include gimple-fold.h.
1101 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1102 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1103 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1104 Add a path that uses a step of VF instead of 1, but disable it
1106 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1107 and niters_no_overflow parameters. Update calls to
1108 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1109 Create a new SSA name if the latter choses to use a ste other
1110 than zero, and return it via niters_vector_mult_vf_var.
1111 * tree-vect-loop.c (vect_transform_loop): Update calls to
1112 vect_do_peeling, vect_gen_vector_loop_niters and
1113 slpeel_make_loop_iterate_ntimes.
1114 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1115 (vect_gen_vector_loop_niters): Update declarations after above changes.
1117 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1119 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1120 128-bit round to integer instructions.
1121 (ceil<mode>2): Likewise.
1122 (btrunc<mode>2): Likewise.
1123 (round<mode>2): Likewise.
1125 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1127 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1128 unaligned VSX load/store on P8/P9.
1129 (expand_block_clear): Allow the use of unaligned VSX
1130 load/store on P8/P9.
1132 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1134 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1136 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1137 swap associated with both a load and a store.
1139 2018-01-02 Andrew Waterman <andrew@sifive.com>
1141 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1142 * config/riscv/riscv.md (clear_cache): Use it.
1144 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1146 * web.c: Remove out-of-date comment.
1148 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1150 * expr.c (fixup_args_size_notes): Check that any existing
1151 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1152 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1153 (emit_single_push_insn): ...here.
1155 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1157 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1158 (const_vector_encoded_nelts): New function.
1159 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1160 (const_vector_int_elt, const_vector_elt): Declare.
1161 * emit-rtl.c (const_vector_int_elt_1): New function.
1162 (const_vector_elt): Likewise.
1163 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1164 of CONST_VECTOR_ELT.
1166 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1168 * expr.c: Include rtx-vector-builder.h.
1169 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1170 directly on the tree encoding.
1171 (const_vector_from_tree): Likewise.
1172 * optabs.c: Include rtx-vector-builder.h.
1173 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1174 sequence of "u" values.
1175 * vec-perm-indices.c: Include rtx-vector-builder.h.
1176 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1177 directly on the vec_perm_indices encoding.
1179 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1181 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1182 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1183 * rtx-vector-builder.h: New file.
1184 * rtx-vector-builder.c: Likewise.
1185 * rtl.h (rtx_def::u2): Add a const_vector field.
1186 (CONST_VECTOR_NPATTERNS): New macro.
1187 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1188 (CONST_VECTOR_DUPLICATE_P): Likewise.
1189 (CONST_VECTOR_STEPPED_P): Likewise.
1190 (CONST_VECTOR_ENCODED_ELT): Likewise.
1191 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1192 (unwrap_const_vec_duplicate): Likewise.
1193 (const_vec_series_p): Check for a non-duplicated vector encoding.
1194 Say that the function only returns true for integer vectors.
1195 * emit-rtl.c: Include rtx-vector-builder.h.
1196 (gen_const_vec_duplicate_1): Delete.
1197 (gen_const_vector): Call gen_const_vec_duplicate instead of
1198 gen_const_vec_duplicate_1.
1199 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1200 (gen_const_vec_duplicate): Use rtx_vector_builder.
1201 (gen_const_vec_series): Likewise.
1202 (gen_rtx_CONST_VECTOR): Likewise.
1203 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1204 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1205 Build a new vector rather than modifying a CONST_VECTOR in-place.
1206 (handle_special_swappables): Update call accordingly.
1207 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1208 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1209 Build a new vector rather than modifying a CONST_VECTOR in-place.
1210 (handle_special_swappables): Update call accordingly.
1212 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1214 * simplify-rtx.c (simplify_const_binary_operation): Use
1215 CONST_VECTOR_ELT instead of XVECEXP.
1217 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1219 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1220 the selector elements to be different from the data elements
1221 if the selector is a VECTOR_CST.
1222 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1223 ssizetype for the selector.
1225 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1227 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1228 before testing each element individually.
1229 * tree-vect-generic.c (lower_vec_perm): Likewise.
1231 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1233 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1234 * selftest-run-tests.c (selftest::run_tests): Call it.
1235 * vector-builder.h (vector_builder::operator ==): New function.
1236 (vector_builder::operator !=): Likewise.
1237 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1238 (vec_perm_indices::all_from_input_p): New function.
1239 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1240 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1241 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1242 instead of reading the VECTOR_CST directly. Detect whether both
1243 vector inputs are the same before constructing the vec_perm_indices,
1244 and update the number of inputs argument accordingly. Use the
1245 utility functions added above. Only construct sel2 if we need to.
1247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1249 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1250 the broadcast of the low byte.
1251 (expand_mult_highpart): Use an explicit encoding for the permutes.
1252 * optabs-query.c (can_mult_highpart_p): Likewise.
1253 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1254 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1255 (vectorizable_bswap): Likewise.
1256 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1257 explicit encoding for the power-of-2 permutes.
1258 (vect_permute_store_chain): Likewise.
1259 (vect_grouped_load_supported): Likewise.
1260 (vect_permute_load_chain): Likewise.
1262 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1264 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1265 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1266 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1267 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1268 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1269 (vect_gen_perm_mask_any): Likewise.
1271 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1273 * int-vector-builder.h: New file.
1274 * vec-perm-indices.h: Include int-vector-builder.h.
1275 (vec_perm_indices): Redefine as an int_vector_builder.
1276 (auto_vec_perm_indices): Delete.
1277 (vec_perm_builder): Redefine as a stand-alone class.
1278 (vec_perm_indices::vec_perm_indices): New function.
1279 (vec_perm_indices::clamp): Likewise.
1280 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1281 (vec_perm_indices::new_vector): New function.
1282 (vec_perm_indices::new_expanded_vector): Update for new
1283 vec_perm_indices class.
1284 (vec_perm_indices::rotate_inputs): New function.
1285 (vec_perm_indices::all_in_range_p): Operate directly on the
1286 encoded form, without computing elided elements.
1287 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1288 encoding. Update for new vec_perm_indices class.
1289 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1290 the given vec_perm_builder.
1291 (expand_vec_perm_var): Update vec_perm_builder constructor.
1292 (expand_mult_highpart): Use vec_perm_builder instead of
1293 auto_vec_perm_indices.
1294 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1295 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1296 or double series encoding as appropriate.
1297 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1298 vec_perm_indices instead of auto_vec_perm_indices.
1299 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1300 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1301 (vect_permute_store_chain): Likewise.
1302 (vect_grouped_load_supported): Likewise.
1303 (vect_permute_load_chain): Likewise.
1304 (vect_shift_permute_load_chain): Likewise.
1305 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1306 (vect_transform_slp_perm_load): Likewise.
1307 (vect_schedule_slp_instance): Likewise.
1308 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1309 (vectorizable_mask_load_store): Likewise.
1310 (vectorizable_bswap): Likewise.
1311 (vectorizable_store): Likewise.
1312 (vectorizable_load): Likewise.
1313 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1314 vec_perm_indices instead of auto_vec_perm_indices. Use
1315 tree_to_vec_perm_builder to read the vector from a tree.
1316 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1317 vec_perm_builder instead of a vec_perm_indices.
1318 (have_whole_vector_shift): Use vec_perm_builder and
1319 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1320 truncation to calc_vec_perm_mask_for_shift.
1321 (vect_create_epilog_for_reduction): Likewise.
1322 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1323 from auto_vec_perm_indices to vec_perm_indices.
1324 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1325 instead of changing individual elements.
1326 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1327 the vector in d.perm.
1328 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1329 from auto_vec_perm_indices to vec_perm_indices.
1330 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1331 instead of changing individual elements.
1332 (arm_vectorize_vec_perm_const): Use new_vector to install
1333 the vector in d.perm.
1334 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1335 Update vec_perm_builder constructor.
1336 (rs6000_expand_interleave): Likewise.
1337 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1338 (rs6000_expand_interleave): Likewise.
1340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1342 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1343 to qimode could truncate the indices.
1344 * optabs.c (expand_vec_perm_var): Likewise.
1346 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1348 * Makefile.in (OBJS): Add vec-perm-indices.o.
1349 * vec-perm-indices.h: New file.
1350 * vec-perm-indices.c: Likewise.
1351 * target.h (vec_perm_indices): Replace with a forward class
1353 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1354 * optabs.h: Include vec-perm-indices.h.
1355 (expand_vec_perm): Delete.
1356 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1357 (expand_vec_perm_const): Declare.
1358 * target.def (vec_perm_const_ok): Replace with...
1359 (vec_perm_const): ...this new hook.
1360 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1361 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1362 * doc/tm.texi: Regenerate.
1363 * optabs.def (vec_perm_const): Delete.
1364 * doc/md.texi (vec_perm_const): Likewise.
1365 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1366 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1367 expand_vec_perm for constant permutation vectors. Assert that
1368 the mode of variable permutation vectors is the integer equivalent
1369 of the mode that is being permuted.
1370 * optabs-query.h (selector_fits_mode_p): Declare.
1371 * optabs-query.c: Include vec-perm-indices.h.
1372 (selector_fits_mode_p): New function.
1373 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1374 is defined, instead of checking whether the vec_perm_const_optab
1375 exists. Use targetm.vectorize.vec_perm_const instead of
1376 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1377 fit in the vector mode before using a variable permute.
1378 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1379 vec_perm_indices instead of an rtx.
1380 (expand_vec_perm): Replace with...
1381 (expand_vec_perm_const): ...this new function. Take the selector
1382 as a vec_perm_indices rather than an rtx. Also take the mode of
1383 the selector. Update call to shift_amt_for_vec_perm_mask.
1384 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1385 Use vec_perm_indices::new_expanded_vector to expand the original
1386 selector into bytes. Check whether the indices fit in the vector
1387 mode before using a variable permute.
1388 (expand_vec_perm_var): Make global.
1389 (expand_mult_highpart): Use expand_vec_perm_const.
1390 * fold-const.c: Includes vec-perm-indices.h.
1391 * tree-ssa-forwprop.c: Likewise.
1392 * tree-vect-data-refs.c: Likewise.
1393 * tree-vect-generic.c: Likewise.
1394 * tree-vect-loop.c: Likewise.
1395 * tree-vect-slp.c: Likewise.
1396 * tree-vect-stmts.c: Likewise.
1397 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1399 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1400 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1401 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1402 (aarch64_vectorize_vec_perm_const): ...this new function.
1403 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1404 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1405 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1406 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1407 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1408 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1409 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1411 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1412 check for NEON modes.
1413 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1414 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1415 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1416 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1418 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1419 the old VEC_PERM_CONST conditions.
1420 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1421 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1422 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1423 (ia64_vectorize_vec_perm_const_ok): Merge into...
1424 (ia64_vectorize_vec_perm_const): ...this new function.
1425 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1426 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1427 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1428 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1429 * config/mips/mips.c (mips_expand_vec_perm_const)
1430 (mips_vectorize_vec_perm_const_ok): Merge into...
1431 (mips_vectorize_vec_perm_const): ...this new function.
1432 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1433 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1434 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1435 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1436 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1437 (rs6000_expand_vec_perm_const): Delete.
1438 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1440 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1441 (altivec_expand_vec_perm_const_le): Take each operand individually.
1442 Operate on constant selectors rather than rtxes.
1443 (altivec_expand_vec_perm_const): Likewise. Update call to
1444 altivec_expand_vec_perm_const_le.
1445 (rs6000_expand_vec_perm_const): Delete.
1446 (rs6000_vectorize_vec_perm_const_ok): Delete.
1447 (rs6000_vectorize_vec_perm_const): New function.
1448 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1449 an element count and rtx array.
1450 (rs6000_expand_extract_even): Update call accordingly.
1451 (rs6000_expand_interleave): Likewise.
1452 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1453 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1454 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1455 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1456 (rs6000_expand_vec_perm_const): Delete.
1457 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1458 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1459 (altivec_expand_vec_perm_const_le): Take each operand individually.
1460 Operate on constant selectors rather than rtxes.
1461 (altivec_expand_vec_perm_const): Likewise. Update call to
1462 altivec_expand_vec_perm_const_le.
1463 (rs6000_expand_vec_perm_const): Delete.
1464 (rs6000_vectorize_vec_perm_const_ok): Delete.
1465 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1466 reference to the SPE evmerge intructions.
1467 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1468 an element count and rtx array.
1469 (rs6000_expand_extract_even): Update call accordingly.
1470 (rs6000_expand_interleave): Likewise.
1471 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1472 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1474 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1476 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1478 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1479 vector mode and that that mode matches the mode of the data
1481 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1482 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1483 directly using expand_vec_perm_1 when forcing selectors into
1485 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1487 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1489 * optabs-query.h (can_vec_perm_p): Delete.
1490 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1491 * optabs-query.c (can_vec_perm_p): Split into...
1492 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1493 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1494 particular selector is valid.
1495 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1496 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1497 (vect_grouped_load_supported): Likewise.
1498 (vect_shift_permute_load_chain): Likewise.
1499 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1500 (vect_transform_slp_perm_load): Likewise.
1501 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1502 (vectorizable_bswap): Likewise.
1503 (vect_gen_perm_mask_checked): Likewise.
1504 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1505 implementations of variable permutation vectors into account
1506 when deciding which selector to use.
1507 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1508 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1509 with a false third argument.
1510 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1511 to test whether the constant selector is valid and can_vec_perm_var_p
1512 to test whether a variable selector is valid.
1514 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1516 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1517 * optabs-query.c (can_vec_perm_p): Likewise.
1518 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1519 instead of vec_perm_indices.
1520 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1521 (vect_gen_perm_mask_checked): Likewise,
1522 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1523 (vect_gen_perm_mask_checked): Likewise,
1525 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1527 * optabs-query.h (qimode_for_vec_perm): Declare.
1528 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1529 (qimode_for_vec_perm): ...this new function.
1530 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1532 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1534 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1535 does not have a conditional at the top.
1537 2018-01-02 Richard Biener <rguenther@suse.de>
1539 * ipa-inline.c (big_speedup_p): Fix expression.
1541 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1544 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1547 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1551 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1552 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1553 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1554 cond_taken_branch_cost 3->4.
1556 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1558 PR tree-optimization/83581
1559 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1560 TODO_cleanup_cfg if any changes have been made.
1563 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1564 convert_modes if target mode has the right side, but different mode
1568 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1569 last argument when extracting from CONCAT. If either from_real or
1570 from_imag is NULL, use expansion through memory. If result is not
1571 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1572 the parts directly to inner mode, if even that fails, use expansion
1576 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1577 check for bswap in mode rather than HImode and use that in expand_unop
1580 Copyright (C) 2018 Free Software Foundation, Inc.
1582 Copying and distribution of this file, with or without modification,
1583 are permitted in any medium without royalty provided the copyright
1584 notice and this notice are preserved.