1 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4 the VEC_PERM_EXPR fold to fail.
6 2018-01-04 Jakub Jelinek <jakub@redhat.com>
9 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
12 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
15 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
18 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
21 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
22 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
24 2018-01-04 Jakub Jelinek <jakub@redhat.com>
27 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
28 is BLKmode and bitpos not zero or mode change is needed.
30 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
33 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
36 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
39 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
40 instead of MULT rtx. Update all corresponding splitters.
42 (*ssub<modesuffix>): Ditto.
44 (*cmp_sadd_di): Update split patterns.
45 (*cmp_sadd_si): Ditto.
46 (*cmp_sadd_sidi): Ditto.
47 (*cmp_ssub_di): Ditto.
48 (*cmp_ssub_si): Ditto.
49 (*cmp_ssub_sidi): Ditto.
50 * config/alpha/predicates.md (const23_operand): New predicate.
51 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
52 Look for ASHIFT, not MULT inner operand.
53 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
55 2018-01-04 Martin Liska <mliska@suse.cz>
58 * gcov.c (output_intermediate_file): Add version to intermediate
60 * doc/gcov.texi: Document new field 'version' in intermediate
61 file format. Fix location of '-k' option of gcov command.
63 2018-01-04 Martin Liska <mliska@suse.cz>
66 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
68 2018-01-04 Jakub Jelinek <jakub@redhat.com>
70 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
72 2018-01-03 Martin Sebor <msebor@redhat.com>
74 PR tree-optimization/83655
75 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
76 checking calls with invalid arguments.
78 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
80 * tree-vect-stmts.c (vect_get_store_rhs): New function.
81 (vectorizable_mask_load_store): Delete.
82 (vectorizable_call): Return false for masked loads and stores.
83 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
84 instead of gimple_assign_rhs1.
85 (vectorizable_load): Handle IFN_MASK_LOAD.
86 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
88 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
90 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
92 (vectorizable_mask_load_store): ...here.
93 (vectorizable_load): ...and here.
95 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
97 * tree-vect-stmts.c (vect_build_all_ones_mask)
98 (vect_build_zero_merge_argument): New functions, split out from...
99 (vectorizable_load): ...here.
101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
103 * tree-vect-stmts.c (vect_check_store_rhs): New function,
105 (vectorizable_mask_load_store): ...here.
106 (vectorizable_store): ...and here.
108 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
110 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
112 (vectorizable_mask_load_store): ...here.
114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
116 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
117 (vect_model_store_cost): Take a vec_load_store_type instead of a
119 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
120 (vect_model_store_cost): Take a vec_load_store_type instead of a
122 (vectorizable_mask_load_store): Update accordingly.
123 (vectorizable_store): Likewise.
124 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
128 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
129 IFN_MASK_LOAD calls here rather than...
130 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
132 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
133 Alan Hayward <alan.hayward@arm.com>
134 David Sherwood <david.sherwood@arm.com>
136 * expmed.c (extract_bit_field_1): For vector extracts,
137 fall back to extract_bit_field_as_subreg if vec_extract
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 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
145 they are variable or constant sized.
146 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
147 slots for constant-sized data.
149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
150 Alan Hayward <alan.hayward@arm.com>
151 David Sherwood <david.sherwood@arm.com>
153 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
154 handling COND_EXPRs with boolean comparisons, try to find a better
155 basis for the mask type than the boolean itself.
157 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
159 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
160 is calculated and how it can be overridden.
161 * genmodes.c (max_bitsize_mode_any_mode): New variable.
162 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
164 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
168 Alan Hayward <alan.hayward@arm.com>
169 David Sherwood <david.sherwood@arm.com>
171 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
172 Remove the mode argument.
173 (aarch64_simd_valid_immediate): Remove the mode and inverse
175 * config/aarch64/iterators.md (bitsize): New iterator.
176 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
177 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
178 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
179 aarch64_simd_valid_immediate.
180 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
181 (aarch64_reg_or_bic_imm): Likewise.
182 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
183 with an insn_type enum and msl with a modifier_type enum.
184 Replace element_width with a scalar_mode. Change the shift
185 to unsigned int. Add constructors for scalar_float_mode and
186 scalar_int_mode elements.
187 (aarch64_vect_float_const_representable_p): Delete.
188 (aarch64_can_const_movi_rtx_p)
189 (aarch64_simd_scalar_immediate_valid_for_move)
190 (aarch64_simd_make_constant): Update call to
191 aarch64_simd_valid_immediate.
192 (aarch64_advsimd_valid_immediate_hs): New function.
193 (aarch64_advsimd_valid_immediate): Likewise.
194 (aarch64_simd_valid_immediate): Remove mode and inverse
195 arguments. Rewrite to use the above. Use const_vec_duplicate_p
196 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
197 and aarch64_float_const_representable_p on the result.
198 (aarch64_output_simd_mov_immediate): Remove mode argument.
199 Update call to aarch64_simd_valid_immediate and use of
201 (aarch64_output_scalar_simd_mov_immediate): Update call
204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
205 Alan Hayward <alan.hayward@arm.com>
206 David Sherwood <david.sherwood@arm.com>
208 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
209 (mode_nunits): Likewise CONST_MODE_NUNITS.
210 * machmode.def (ADJUST_NUNITS): Document.
211 * genmodes.c (mode_data::need_nunits_adj): New field.
212 (blank_mode): Update accordingly.
213 (adj_nunits): New variable.
214 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
216 (emit_mode_size_inline): Set need_bytesize_adj for all modes
217 listed in adj_nunits.
218 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
219 listed in adj_nunits. Don't emit case statements for such modes.
220 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
221 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
222 nothing if adj_nunits is nonnull.
223 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
224 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
225 (emit_mode_fbit): Update use of print_maybe_const_decl.
226 (emit_move_size): Likewise. Treat the array as non-const
228 (emit_mode_adjustments): Handle adj_nunits.
230 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
232 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
233 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
234 (VECTOR_MODES): Use it.
235 (make_vector_modes): Take the prefix as an argument.
237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
238 Alan Hayward <alan.hayward@arm.com>
239 David Sherwood <david.sherwood@arm.com>
241 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
242 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
243 for MODE_VECTOR_BOOL.
244 * machmode.def (VECTOR_BOOL_MODE): Document.
245 * genmodes.c (VECTOR_BOOL_MODE): New macro.
246 (make_vector_bool_mode): New function.
247 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
249 * lto-streamer-in.c (lto_input_mode_table): Likewise.
250 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
252 * stor-layout.c (int_mode_for_mode): Likewise.
253 * tree.c (build_vector_type_for_mode): Likewise.
254 * varasm.c (output_constant_pool_2): Likewise.
255 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
256 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
257 for MODE_VECTOR_BOOL.
258 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
259 of mode class checks.
260 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
261 instead of a list of mode class checks.
262 (expand_vector_scalar_condition): Likewise.
263 (type_for_widest_vector_mode): Handle BImode as an inner mode.
265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
266 Alan Hayward <alan.hayward@arm.com>
267 David Sherwood <david.sherwood@arm.com>
269 * machmode.h (mode_size): Change from unsigned short to
271 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
272 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
273 or if measurement_type is not polynomial.
274 (fixed_size_mode::includes_p): Check for constant-sized modes.
275 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
276 return a poly_uint16 rather than an unsigned short.
277 (emit_mode_size): Change the type of mode_size from unsigned short
278 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
279 (emit_mode_adjustments): Cope with polynomial vector sizes.
280 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
282 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
284 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
285 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
286 * caller-save.c (setup_save_areas): Likewise.
287 (replace_reg_with_saved_mem): Likewise.
288 * calls.c (emit_library_call_value_1): Likewise.
289 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
290 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
291 (gen_lowpart_for_combine): Likewise.
292 * convert.c (convert_to_integer_1): Likewise.
293 * cse.c (equiv_constant, cse_insn): Likewise.
294 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
295 (cselib_subst_to_values): Likewise.
296 * dce.c (word_dce_process_block): Likewise.
297 * df-problems.c (df_word_lr_mark_ref): Likewise.
298 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
299 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
300 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
301 (rtl_for_decl_location): Likewise.
302 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
303 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
304 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
305 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
306 (expand_expr_real_1): Likewise.
307 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
308 (pad_below): Likewise.
309 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
310 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
311 * ira.c (get_subreg_tracking_sizes): Likewise.
312 * ira-build.c (ira_create_allocno_objects): Likewise.
313 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
314 (ira_sort_regnos_for_alter_reg): Likewise.
315 * ira-costs.c (record_operand_costs): Likewise.
316 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
317 (resolve_simple_move): Likewise.
318 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
319 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
320 (lra_constraints): Likewise.
321 (CONST_POOL_OK_P): Reject variable-sized modes.
322 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
323 (add_pseudo_to_slot, lra_spill): Likewise.
324 * omp-low.c (omp_clause_aligned_alignment): Likewise.
325 * optabs-query.c (get_best_extraction_insn): Likewise.
326 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
327 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
328 (expand_mult_highpart, valid_multiword_target_p): Likewise.
329 * recog.c (offsettable_address_addr_space_p): Likewise.
330 * regcprop.c (maybe_mode_change): Likewise.
331 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
332 * regrename.c (build_def_use): Likewise.
333 * regstat.c (dump_reg_info): Likewise.
334 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
335 (find_reloads, find_reloads_subreg_address): Likewise.
336 * reload1.c (eliminate_regs_1): Likewise.
337 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
338 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
339 (simplify_binary_operation_1, simplify_subreg): Likewise.
340 * targhooks.c (default_function_arg_padding): Likewise.
341 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
342 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
343 (verify_gimple_assign_ternary): Likewise.
344 * tree-inline.c (estimate_move_cost): Likewise.
345 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
346 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
347 (get_address_cost_ainc): Likewise.
348 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
349 (vect_supportable_dr_alignment): Likewise.
350 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
351 (vectorizable_reduction): Likewise.
352 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
353 (vectorizable_operation, vectorizable_load): Likewise.
354 * tree.c (build_same_sized_truth_vector_type): Likewise.
355 * valtrack.c (cleanup_auto_inc_dec): Likewise.
356 * var-tracking.c (emit_note_insn_var_location): Likewise.
357 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
358 (ADDR_VEC_ALIGN): Likewise.
360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
361 Alan Hayward <alan.hayward@arm.com>
362 David Sherwood <david.sherwood@arm.com>
364 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
366 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
367 or if measurement_type is polynomial.
368 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
369 * combine.c (make_extraction): Likewise.
370 * dse.c (find_shift_sequence): Likewise.
371 * dwarf2out.c (mem_loc_descriptor): Likewise.
372 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
373 (extract_bit_field, extract_low_bits): Likewise.
374 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
375 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
376 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
377 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
378 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
379 * reload.c (find_reloads): Likewise.
380 * reload1.c (alter_reg): Likewise.
381 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
382 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
383 * tree-if-conv.c (predicate_mem_writes): Likewise.
384 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
385 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
386 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
387 * valtrack.c (dead_debug_insert_temp): Likewise.
388 * varasm.c (mergeable_constant_section): Likewise.
389 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
391 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
392 Alan Hayward <alan.hayward@arm.com>
393 David Sherwood <david.sherwood@arm.com>
395 * expr.c (expand_assignment): Cope with polynomial mode sizes
396 when assigning to a CONCAT.
398 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
399 Alan Hayward <alan.hayward@arm.com>
400 David Sherwood <david.sherwood@arm.com>
402 * machmode.h (mode_precision): Change from unsigned short to
404 (mode_to_precision): Return a poly_uint16 rather than an unsigned
406 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
407 or if measurement_type is not polynomial.
408 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
409 in which the mode is already known to be a scalar_int_mode.
410 * genmodes.c (emit_mode_precision): Change the type of mode_precision
411 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
413 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
414 for GET_MODE_PRECISION.
415 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
416 for GET_MODE_PRECISION.
417 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
419 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
420 (expand_field_assignment, make_extraction): Likewise.
421 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
422 (get_last_value): Likewise.
423 * convert.c (convert_to_integer_1): Likewise.
424 * cse.c (cse_insn): Likewise.
425 * expr.c (expand_expr_real_1): Likewise.
426 * lra-constraints.c (simplify_operand_subreg): Likewise.
427 * optabs-query.c (can_atomic_load_p): Likewise.
428 * optabs.c (expand_atomic_load): Likewise.
429 (expand_atomic_store): Likewise.
430 * ree.c (combine_reaching_defs): Likewise.
431 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
432 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
433 * tree.h (type_has_mode_precision_p): Likewise.
434 * ubsan.c (instrument_si_overflow): Likewise.
436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
437 Alan Hayward <alan.hayward@arm.com>
438 David Sherwood <david.sherwood@arm.com>
440 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
441 polynomial numbers of units.
442 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
443 (valid_vector_subparts_p): New function.
444 (build_vector_type): Remove temporary shim and take the number
445 of units as a poly_uint64 rather than an int.
446 (build_opaque_vector_type): Take the number of units as a
447 poly_uint64 rather than an int.
448 * tree.c (build_vector_from_ctor): Handle polynomial
449 TYPE_VECTOR_SUBPARTS.
450 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
451 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
452 (build_vector_from_val): If the number of units is variable,
453 use build_vec_duplicate_cst for constant operands and
454 VEC_DUPLICATE_EXPR otherwise.
455 (make_vector_type): Remove temporary is_constant ().
456 (build_vector_type, build_opaque_vector_type): Take the number of
457 units as a poly_uint64 rather than an int.
458 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
460 * cfgexpand.c (expand_debug_expr): Likewise.
461 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
462 (store_constructor, expand_expr_real_1): Likewise.
463 (const_scalar_mask_from_tree): Likewise.
464 * fold-const-call.c (fold_const_reduction): Likewise.
465 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
466 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
467 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
468 (fold_relational_const): Likewise.
469 (native_interpret_vector): Likewise. Change the size from an
470 int to an unsigned int.
471 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
472 TYPE_VECTOR_SUBPARTS.
473 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
474 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
475 duplicating a non-constant operand into a variable-length vector.
476 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
477 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
478 * ipa-icf.c (sem_variable::equals): Likewise.
479 * match.pd: Likewise.
480 * omp-simd-clone.c (simd_clone_subparts): Likewise.
481 * print-tree.c (print_node): Likewise.
482 * stor-layout.c (layout_type): Likewise.
483 * targhooks.c (default_builtin_vectorization_cost): Likewise.
484 * tree-cfg.c (verify_gimple_comparison): Likewise.
485 (verify_gimple_assign_binary): Likewise.
486 (verify_gimple_assign_ternary): Likewise.
487 (verify_gimple_assign_single): Likewise.
488 * tree-pretty-print.c (dump_generic_node): Likewise.
489 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
490 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
491 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
492 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
493 (vect_shift_permute_load_chain): Likewise.
494 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
495 (expand_vector_condition, optimize_vector_constructor): Likewise.
496 (lower_vec_perm, get_compute_type): Likewise.
497 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
498 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
499 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
500 (vect_recog_mask_conversion_pattern): Likewise.
501 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
502 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
503 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
504 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
505 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
506 (vectorizable_shift, vectorizable_operation, vectorizable_store)
507 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
508 (supportable_widening_operation): Likewise.
509 (supportable_narrowing_operation): Likewise.
510 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
512 * varasm.c (output_constant): Likewise.
514 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
515 Alan Hayward <alan.hayward@arm.com>
516 David Sherwood <david.sherwood@arm.com>
518 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
519 so that both the length == 3 and length != 3 cases set up their
520 own permute vectors. Add comments explaining why we know the
521 number of elements is constant.
522 (vect_permute_load_chain): Likewise.
524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
525 Alan Hayward <alan.hayward@arm.com>
526 David Sherwood <david.sherwood@arm.com>
528 * machmode.h (mode_nunits): Change from unsigned char to
530 (ONLY_FIXED_SIZE_MODES): New macro.
531 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
532 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
533 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
535 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
536 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
537 or if measurement_type is not polynomial.
538 * genmodes.c (ZERO_COEFFS): New macro.
539 (emit_mode_nunits_inline): Make mode_nunits_inline return a
541 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
542 Use ZERO_COEFFS when emitting initializers.
543 * data-streamer.h (bp_pack_poly_value): New function.
544 (bp_unpack_poly_value): Likewise.
545 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
547 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
549 * tree.c (make_vector_type): Remove temporary shim and make
550 the real function take the number of units as a poly_uint64
552 (build_vector_type_for_mode): Handle polynomial nunits.
553 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
554 * emit-rtl.c (const_vec_series_p_1): Likewise.
555 (gen_rtx_CONST_VECTOR): Likewise.
556 * fold-const.c (test_vec_duplicate_folding): Likewise.
557 * genrecog.c (validate_pattern): Likewise.
558 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
559 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
560 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
561 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
562 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
563 * rtlanal.c (subreg_get_info): Likewise.
564 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
565 (vect_grouped_load_supported): Likewise.
566 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
567 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
568 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
569 (simplify_const_unary_operation, simplify_binary_operation_1)
570 (simplify_const_binary_operation, simplify_ternary_operation)
571 (test_vector_ops_duplicate, test_vector_ops): Likewise.
572 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
573 instead of CONST_VECTOR_NUNITS.
574 * varasm.c (output_constant_pool_2): Likewise.
575 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
576 explicit-encoded elements in the XVEC for variable-length vectors.
578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
580 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
582 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
583 Alan Hayward <alan.hayward@arm.com>
584 David Sherwood <david.sherwood@arm.com>
586 * coretypes.h (fixed_size_mode): Declare.
587 (fixed_size_mode_pod): New typedef.
588 * builtins.h (target_builtins::x_apply_args_mode)
589 (target_builtins::x_apply_result_mode): Change type to
591 * builtins.c (apply_args_size, apply_result_size, result_vector)
592 (expand_builtin_apply_args_1, expand_builtin_apply)
593 (expand_builtin_return): Update accordingly.
595 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
597 * cse.c (hash_rtx_cb): Hash only the encoded elements.
598 * cselib.c (cselib_hash_rtx): Likewise.
599 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
600 CONST_VECTOR encoding.
602 2017-01-03 Jakub Jelinek <jakub@redhat.com>
603 Jeff Law <law@redhat.com>
606 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
607 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
608 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
609 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
612 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
613 explicitly probe *sp in a noreturn function if there were any callee
614 register saves or frame pointer is needed.
616 2018-01-03 Jakub Jelinek <jakub@redhat.com>
619 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
620 BLKmode for ternary, binary or unary expressions.
623 * var-tracking.c (delete_vta_debug_insn): New inline function.
624 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
625 insns from get_insns () to NULL instead of each bb separately.
626 Use delete_vta_debug_insn. No longer static.
627 (vt_debug_insns_local, variable_tracking_main_1): Adjust
628 delete_vta_debug_insns callers.
629 * rtl.h (delete_vta_debug_insns): Declare.
630 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
631 instead of variable_tracking_main.
633 2018-01-03 Martin Sebor <msebor@redhat.com>
635 PR tree-optimization/83603
636 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
637 arguments past the endof the argument list in functions declared
639 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
640 Avoid checking when arguments are null.
642 2018-01-03 Martin Sebor <msebor@redhat.com>
645 * doc/extend.texi (attribute const): Fix a typo.
646 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
647 issuing -Wsuggest-attribute for void functions.
649 2018-01-03 Martin Sebor <msebor@redhat.com>
651 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
652 offset_int::from instead of wide_int::to_shwi.
653 (maybe_diag_overlap): Remove assertion.
654 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
655 * gimple-ssa-sprintf.c (format_directive): Same.
656 (parse_directive): Same.
657 (sprintf_dom_walker::compute_format_length): Same.
658 (try_substitute_return_value): Same.
660 2017-01-03 Jeff Law <law@redhat.com>
663 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
664 non-constant residual for zero at runtime and avoid probing in
665 that case. Reorganize code for trailing problem to mirror handling
668 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
670 PR tree-optimization/83501
671 * tree-ssa-strlen.c (get_string_cst): New.
672 (handle_char_store): Call get_string_cst.
674 2018-01-03 Martin Liska <mliska@suse.cz>
676 PR tree-optimization/83593
677 * tree-ssa-strlen.c: Include tree-cfg.h.
678 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
679 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
680 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
682 (strlen_dom_walker::before_dom_children): Call
683 gimple_purge_dead_eh_edges. Dump tranformation with details
685 (strlen_dom_walker::before_dom_children): Update call by adding
686 new argument cleanup_eh.
687 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
689 2018-01-03 Martin Liska <mliska@suse.cz>
692 * cif-code.def (VARIADIC_THUNK): New enum value.
693 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
696 2018-01-03 Jan Beulich <jbeulich@suse.com>
698 * sse.md (mov<mode>_internal): Tighten condition for when to use
699 vmovdqu<ssescalarsize> for TI and OI modes.
701 2018-01-03 Jakub Jelinek <jakub@redhat.com>
703 Update copyright years.
705 2018-01-03 Martin Liska <mliska@suse.cz>
708 * ipa-visibility.c (function_and_variable_visibility): Skip
709 functions with noipa attribure.
711 2018-01-03 Jakub Jelinek <jakub@redhat.com>
713 * gcc.c (process_command): Update copyright notice dates.
714 * gcov-dump.c (print_version): Ditto.
715 * gcov.c (print_version): Ditto.
716 * gcov-tool.c (print_version): Ditto.
717 * gengtype.c (create_file): Ditto.
718 * doc/cpp.texi: Bump @copying's copyright year.
719 * doc/cppinternals.texi: Ditto.
720 * doc/gcc.texi: Ditto.
721 * doc/gccint.texi: Ditto.
722 * doc/gcov.texi: Ditto.
723 * doc/install.texi: Ditto.
724 * doc/invoke.texi: Ditto.
726 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
728 * vector-builder.h (vector_builder::m_full_nelts): Change from
729 unsigned int to poly_uint64.
730 (vector_builder::full_nelts): Update prototype accordingly.
731 (vector_builder::new_vector): Likewise.
732 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
733 (vector_builder::operator ==): Likewise.
734 (vector_builder::finalize): Likewise.
735 * int-vector-builder.h (int_vector_builder::int_vector_builder):
736 Take the number of elements as a poly_uint64 rather than an
738 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
739 from unsigned int to poly_uint64.
740 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
741 (vec_perm_indices::new_vector): Likewise.
742 (vec_perm_indices::length): Likewise.
743 (vec_perm_indices::nelts_per_input): Likewise.
744 (vec_perm_indices::input_nelts): Likewise.
745 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
746 number of elements per input as a poly_uint64 rather than an
747 unsigned int. Use the original encoding for variable-length
748 vectors, rather than clamping each individual element.
749 For the second and subsequent elements in each pattern,
750 clamp the step and base before clamping their sum.
751 (vec_perm_indices::series_p): Handle polynomial element counts.
752 (vec_perm_indices::all_in_range_p): Likewise.
753 (vec_perm_indices_to_tree): Likewise.
754 (vec_perm_indices_to_rtx): Likewise.
755 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
756 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
757 (tree_vector_builder::new_binary_operation): Handle polynomial
758 element counts. Return false if we need to know the number
759 of elements at compile time.
760 * fold-const.c (fold_vec_perm): Punt if the number of elements
761 isn't known at compile time.
763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
765 * vec-perm-indices.h (vec_perm_builder): Change element type
766 from HOST_WIDE_INT to poly_int64.
767 (vec_perm_indices::element_type): Update accordingly.
768 (vec_perm_indices::clamp): Handle polynomial element_types.
769 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
770 (vec_perm_indices::all_in_range_p): Likewise.
771 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
773 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
774 polynomial vec_perm_indices element types.
775 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
776 * fold-const.c (fold_vec_perm): Likewise.
777 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
778 * tree-vect-generic.c (lower_vec_perm): Likewise.
779 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
780 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
781 element type to HOST_WIDE_INT.
783 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
784 Alan Hayward <alan.hayward@arm.com>
785 David Sherwood <david.sherwood@arm.com>
787 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
788 rather than an int. Use plus_constant.
789 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
790 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
793 Alan Hayward <alan.hayward@arm.com>
794 David Sherwood <david.sherwood@arm.com>
796 * calls.c (emit_call_1, expand_call): Change struct_value_size from
797 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
804 mode sizes. Require a constant size for BLKmode parameters
805 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
806 forces a parameter to be padded at the lsb end in order to
807 fill a complete number of words, require the parameter size
808 to be ordered wrt UNITS_PER_WORD.
810 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
811 Alan Hayward <alan.hayward@arm.com>
812 David Sherwood <david.sherwood@arm.com>
814 * reload1.c (spill_stack_slot_width): Change element type
815 from unsigned int to poly_uint64_pod.
816 (alter_reg): Treat mode sizes as polynomial.
818 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
819 Alan Hayward <alan.hayward@arm.com>
820 David Sherwood <david.sherwood@arm.com>
822 * reload.c (complex_word_subreg_p): New function.
823 (reload_inner_reg_of_subreg, push_reload): Use it.
825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
826 Alan Hayward <alan.hayward@arm.com>
827 David Sherwood <david.sherwood@arm.com>
829 * lra-constraints.c (process_alt_operands): Reject matched
830 operands whose sizes aren't ordered.
831 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
838 that the mode size is in the set {1, 2, 4, 8, 16}.
840 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
841 Alan Hayward <alan.hayward@arm.com>
842 David Sherwood <david.sherwood@arm.com>
844 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
845 Use plus_constant instead of gen_rtx_PLUS.
847 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
848 Alan Hayward <alan.hayward@arm.com>
849 David Sherwood <david.sherwood@arm.com>
851 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
852 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
853 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
854 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
855 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
856 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
857 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
858 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
859 * config/i386/i386.c (ix86_push_rounding): ...this new function.
860 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
862 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
863 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
864 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
865 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
866 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
867 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
868 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
869 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
870 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
871 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
873 * expr.c (emit_move_resolve_push): Treat the input and result
874 of PUSH_ROUNDING as a poly_int64.
875 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
876 (emit_push_insn): Likewise.
877 * lra-eliminations.c (mark_not_eliminable): Likewise.
878 * recog.c (push_operand): Likewise.
879 * reload1.c (elimination_effects): Likewise.
880 * rtlanal.c (nonzero_bits1): Likewise.
881 * calls.c (store_one_arg): Likewise. Require the padding to be
882 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
889 Use plus_constant instead of gen_rtx_PLUS.
891 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
892 Alan Hayward <alan.hayward@arm.com>
893 David Sherwood <david.sherwood@arm.com>
895 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
899 Alan Hayward <alan.hayward@arm.com>
900 David Sherwood <david.sherwood@arm.com>
902 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
903 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
904 via stack temporaries. Treat the mode size as polynomial too.
906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
907 Alan Hayward <alan.hayward@arm.com>
908 David Sherwood <david.sherwood@arm.com>
910 * expr.c (expand_expr_real_2): When handling conversions involving
911 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
912 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
913 as a poly_uint64 too.
915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
916 Alan Hayward <alan.hayward@arm.com>
917 David Sherwood <david.sherwood@arm.com>
919 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
921 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
922 Alan Hayward <alan.hayward@arm.com>
923 David Sherwood <david.sherwood@arm.com>
925 * combine.c (can_change_dest_mode): Handle polynomial
926 REGMODE_NATURAL_SIZE.
927 * expmed.c (store_bit_field_1): Likewise.
928 * expr.c (store_constructor): Likewise.
929 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
930 and polynomial REGMODE_NATURAL_SIZE.
931 (gen_lowpart_common): Likewise.
932 * reginfo.c (record_subregs_of_mode): Likewise.
933 * rtlanal.c (read_modify_subreg_p): Likewise.
935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
936 Alan Hayward <alan.hayward@arm.com>
937 David Sherwood <david.sherwood@arm.com>
939 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
943 Alan Hayward <alan.hayward@arm.com>
944 David Sherwood <david.sherwood@arm.com>
946 * match.pd: Cope with polynomial numbers of vector elements.
948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
949 Alan Hayward <alan.hayward@arm.com>
950 David Sherwood <david.sherwood@arm.com>
952 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
953 in a POINTER_PLUS_EXPR.
955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
956 Alan Hayward <alan.hayward@arm.com>
957 David Sherwood <david.sherwood@arm.com>
959 * omp-simd-clone.c (simd_clone_subparts): New function.
960 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
961 (ipa_simd_modify_function_body): Likewise.
963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
964 Alan Hayward <alan.hayward@arm.com>
965 David Sherwood <david.sherwood@arm.com>
967 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
968 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
969 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
970 (expand_vector_condition, vector_element): Likewise.
971 (subparts_gt): New function.
972 (get_compute_type): Use subparts_gt.
973 (count_type_subparts): Delete.
974 (expand_vector_operations_1): Use subparts_gt instead of
977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
978 Alan Hayward <alan.hayward@arm.com>
979 David Sherwood <david.sherwood@arm.com>
981 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
982 (vect_compile_time_alias): ...this new function. Do the calculation
983 on poly_ints rather than trees.
984 (vect_prune_runtime_alias_test_list): Update call accordingly.
986 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
987 Alan Hayward <alan.hayward@arm.com>
988 David Sherwood <david.sherwood@arm.com>
990 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
992 (vect_schedule_slp_instance): Likewise.
994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
995 Alan Hayward <alan.hayward@arm.com>
996 David Sherwood <david.sherwood@arm.com>
998 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
999 constant and extern definitions for variable-length vectors.
1000 (vect_get_constant_vectors): Note that the number of units
1001 is known to be constant.
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 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1008 of units as polynomial. Choose between WIDE and NARROW based
1011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1012 Alan Hayward <alan.hayward@arm.com>
1013 David Sherwood <david.sherwood@arm.com>
1015 * tree-vect-stmts.c (simd_clone_subparts): New function.
1016 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1019 Alan Hayward <alan.hayward@arm.com>
1020 David Sherwood <david.sherwood@arm.com>
1022 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1023 vectors as polynomial. Use build_index_vector for
1026 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1027 Alan Hayward <alan.hayward@arm.com>
1028 David Sherwood <david.sherwood@arm.com>
1030 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1031 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1032 for variable-length vectors.
1033 (vectorizable_mask_load_store): Treat the number of units as
1034 polynomial, asserting that it is constant if the condition has
1035 already been enforced.
1036 (vectorizable_store, vectorizable_load): Likewise.
1038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1039 Alan Hayward <alan.hayward@arm.com>
1040 David Sherwood <david.sherwood@arm.com>
1042 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1043 of units as polynomial. Punt if we can't tell at compile time
1044 which vector contains the final result.
1046 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1047 Alan Hayward <alan.hayward@arm.com>
1048 David Sherwood <david.sherwood@arm.com>
1050 * tree-vect-loop.c (vectorizable_induction): Treat the number
1051 of units as polynomial. Punt on SLP inductions. Use an integer
1052 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1053 cast of such a series for variable-length floating-point
1056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1057 Alan Hayward <alan.hayward@arm.com>
1058 David Sherwood <david.sherwood@arm.com>
1060 * tree.h (build_index_vector): Declare.
1061 * tree.c (build_index_vector): New function.
1062 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1063 of units as polynomial, forcibly converting it to a constant if
1064 vectorizable_reduction has already enforced the condition.
1065 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1066 to create a {1,2,3,...} vector.
1067 (vectorizable_reduction): Treat the number of units as polynomial.
1068 Choose vectype_in based on the largest scalar element size rather
1069 than the smallest number of units. Enforce the restrictions
1072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1073 Alan Hayward <alan.hayward@arm.com>
1074 David Sherwood <david.sherwood@arm.com>
1076 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1077 number of units as polynomial.
1079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1080 Alan Hayward <alan.hayward@arm.com>
1081 David Sherwood <david.sherwood@arm.com>
1083 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1084 * target.def (autovectorize_vector_sizes): Return the vector sizes
1085 by pointer, using vector_sizes rather than a bitmask.
1086 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1087 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1088 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1090 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1091 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1092 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1093 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1094 * omp-general.c (omp_max_vf): Likewise.
1095 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1096 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1097 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1098 * tree-vect-slp.c (vect_slp_bb): Likewise.
1099 * doc/tm.texi: Regenerate.
1100 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1102 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1103 the vector size as a poly_uint64 rather than an unsigned int.
1104 (current_vector_size): Change from an unsigned int to a poly_uint64.
1105 (get_vectype_for_scalar_type): Update accordingly.
1106 * tree.h (build_truth_vector_type): Take the size and number of
1107 units as a poly_uint64 rather than an unsigned int.
1108 (build_vector_type): Add a temporary overload that takes
1109 the number of units as a poly_uint64 rather than an unsigned int.
1110 * tree.c (make_vector_type): Likewise.
1111 (build_truth_vector_type): Take the number of units as a poly_uint64
1112 rather than an unsigned int.
1114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1115 Alan Hayward <alan.hayward@arm.com>
1116 David Sherwood <david.sherwood@arm.com>
1118 * target.def (get_mask_mode): Take the number of units and length
1119 as poly_uint64s rather than unsigned ints.
1120 * targhooks.h (default_get_mask_mode): Update accordingly.
1121 * targhooks.c (default_get_mask_mode): Likewise.
1122 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1123 * doc/tm.texi: Regenerate.
1125 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1126 Alan Hayward <alan.hayward@arm.com>
1127 David Sherwood <david.sherwood@arm.com>
1129 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1130 * omp-general.c (omp_max_vf): Likewise.
1131 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1132 (expand_omp_simd): Handle polynomial safelen.
1133 * omp-low.c (omplow_simd_context): Add a default constructor.
1134 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1135 (lower_rec_simd_input_clauses): Update accordingly.
1136 (lower_rec_input_clauses): Likewise.
1138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1139 Alan Hayward <alan.hayward@arm.com>
1140 David Sherwood <david.sherwood@arm.com>
1142 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1143 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1144 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1145 (vect_analyze_slp_cost): Likewise.
1146 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1147 (vect_model_load_cost): Likewise.
1149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1150 Alan Hayward <alan.hayward@arm.com>
1151 David Sherwood <david.sherwood@arm.com>
1153 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1154 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1155 from an unsigned int * to a poly_uint64_pod *.
1156 (calculate_unrolling_factor): New function.
1157 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1160 Alan Hayward <alan.hayward@arm.com>
1161 David Sherwood <david.sherwood@arm.com>
1163 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1164 from an unsigned int to a poly_uint64.
1165 (_loop_vec_info::slp_unrolling_factor): Likewise.
1166 (_loop_vec_info::vectorization_factor): Change from an int
1168 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1169 (vect_get_num_vectors): New function.
1170 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1171 (vect_get_num_copies): Use vect_get_num_vectors.
1172 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1173 to an unsigned int *.
1174 (vect_analyze_data_refs): Change min_vf from an int * to a
1176 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1177 than an unsigned HOST_WIDE_INT.
1178 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1179 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1180 to an unsigned int *.
1181 (vect_analyze_data_ref_dependences): Likewise.
1182 (vect_compute_data_ref_alignment): Handle polynomial vf.
1183 (vect_enhance_data_refs_alignment): Likewise.
1184 (vect_prune_runtime_alias_test_list): Likewise.
1185 (vect_shift_permute_load_chain): Likewise.
1186 (vect_supportable_dr_alignment): Likewise.
1187 (dependence_distance_ge_vf): Take the vectorization factor as a
1188 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1189 (vect_analyze_data_refs): Change min_vf from an int * to a
1191 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1192 vfm1 as a poly_uint64 rather than an int. Make the same change
1193 for the returned bound_scalar.
1194 (vect_gen_vector_loop_niters): Handle polynomial vf.
1195 (vect_do_peeling): Likewise. Update call to
1196 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1197 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1199 * tree-vect-loop.c (vect_determine_vectorization_factor)
1200 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1201 (vect_get_known_peeling_cost): Likewise.
1202 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1203 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1204 (vect_transform_loop): Likewise. Use the lowest possible VF when
1205 updating the upper bounds of the loop.
1206 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1208 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1209 polynomial unroll factors.
1210 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1211 (vect_make_slp_decision): Likewise.
1212 (vect_supported_load_permutation_p): Likewise, and polynomial
1214 (vect_analyze_slp_cost): Handle polynomial vf.
1215 (vect_slp_analyze_node_operations): Likewise.
1216 (vect_slp_analyze_bb_1): Likewise.
1217 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1218 than an unsigned HOST_WIDE_INT.
1219 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1220 (vectorizable_load): Handle polynomial vf.
1221 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1223 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1226 Alan Hayward <alan.hayward@arm.com>
1227 David Sherwood <david.sherwood@arm.com>
1229 * match.pd: Handle bit operations involving three constants
1230 and try to fold one pair.
1232 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1234 * tree-vect-loop-manip.c: Include gimple-fold.h.
1235 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1236 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1237 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1238 Add a path that uses a step of VF instead of 1, but disable it
1240 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1241 and niters_no_overflow parameters. Update calls to
1242 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1243 Create a new SSA name if the latter choses to use a ste other
1244 than zero, and return it via niters_vector_mult_vf_var.
1245 * tree-vect-loop.c (vect_transform_loop): Update calls to
1246 vect_do_peeling, vect_gen_vector_loop_niters and
1247 slpeel_make_loop_iterate_ntimes.
1248 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1249 (vect_gen_vector_loop_niters): Update declarations after above changes.
1251 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1253 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1254 128-bit round to integer instructions.
1255 (ceil<mode>2): Likewise.
1256 (btrunc<mode>2): Likewise.
1257 (round<mode>2): Likewise.
1259 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1261 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1262 unaligned VSX load/store on P8/P9.
1263 (expand_block_clear): Allow the use of unaligned VSX
1264 load/store on P8/P9.
1266 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1268 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1270 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1271 swap associated with both a load and a store.
1273 2018-01-02 Andrew Waterman <andrew@sifive.com>
1275 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1276 * config/riscv/riscv.md (clear_cache): Use it.
1278 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1280 * web.c: Remove out-of-date comment.
1282 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1284 * expr.c (fixup_args_size_notes): Check that any existing
1285 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1286 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1287 (emit_single_push_insn): ...here.
1289 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1291 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1292 (const_vector_encoded_nelts): New function.
1293 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1294 (const_vector_int_elt, const_vector_elt): Declare.
1295 * emit-rtl.c (const_vector_int_elt_1): New function.
1296 (const_vector_elt): Likewise.
1297 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1298 of CONST_VECTOR_ELT.
1300 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1302 * expr.c: Include rtx-vector-builder.h.
1303 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1304 directly on the tree encoding.
1305 (const_vector_from_tree): Likewise.
1306 * optabs.c: Include rtx-vector-builder.h.
1307 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1308 sequence of "u" values.
1309 * vec-perm-indices.c: Include rtx-vector-builder.h.
1310 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1311 directly on the vec_perm_indices encoding.
1313 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1315 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1316 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1317 * rtx-vector-builder.h: New file.
1318 * rtx-vector-builder.c: Likewise.
1319 * rtl.h (rtx_def::u2): Add a const_vector field.
1320 (CONST_VECTOR_NPATTERNS): New macro.
1321 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1322 (CONST_VECTOR_DUPLICATE_P): Likewise.
1323 (CONST_VECTOR_STEPPED_P): Likewise.
1324 (CONST_VECTOR_ENCODED_ELT): Likewise.
1325 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1326 (unwrap_const_vec_duplicate): Likewise.
1327 (const_vec_series_p): Check for a non-duplicated vector encoding.
1328 Say that the function only returns true for integer vectors.
1329 * emit-rtl.c: Include rtx-vector-builder.h.
1330 (gen_const_vec_duplicate_1): Delete.
1331 (gen_const_vector): Call gen_const_vec_duplicate instead of
1332 gen_const_vec_duplicate_1.
1333 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1334 (gen_const_vec_duplicate): Use rtx_vector_builder.
1335 (gen_const_vec_series): Likewise.
1336 (gen_rtx_CONST_VECTOR): Likewise.
1337 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1338 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1339 Build a new vector rather than modifying a CONST_VECTOR in-place.
1340 (handle_special_swappables): Update call accordingly.
1341 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1342 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1343 Build a new vector rather than modifying a CONST_VECTOR in-place.
1344 (handle_special_swappables): Update call accordingly.
1346 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1348 * simplify-rtx.c (simplify_const_binary_operation): Use
1349 CONST_VECTOR_ELT instead of XVECEXP.
1351 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1353 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1354 the selector elements to be different from the data elements
1355 if the selector is a VECTOR_CST.
1356 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1357 ssizetype for the selector.
1359 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1361 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1362 before testing each element individually.
1363 * tree-vect-generic.c (lower_vec_perm): Likewise.
1365 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1367 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1368 * selftest-run-tests.c (selftest::run_tests): Call it.
1369 * vector-builder.h (vector_builder::operator ==): New function.
1370 (vector_builder::operator !=): Likewise.
1371 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1372 (vec_perm_indices::all_from_input_p): New function.
1373 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1374 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1375 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1376 instead of reading the VECTOR_CST directly. Detect whether both
1377 vector inputs are the same before constructing the vec_perm_indices,
1378 and update the number of inputs argument accordingly. Use the
1379 utility functions added above. Only construct sel2 if we need to.
1381 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1383 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1384 the broadcast of the low byte.
1385 (expand_mult_highpart): Use an explicit encoding for the permutes.
1386 * optabs-query.c (can_mult_highpart_p): Likewise.
1387 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1388 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1389 (vectorizable_bswap): Likewise.
1390 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1391 explicit encoding for the power-of-2 permutes.
1392 (vect_permute_store_chain): Likewise.
1393 (vect_grouped_load_supported): Likewise.
1394 (vect_permute_load_chain): Likewise.
1396 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1398 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1399 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1400 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1401 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1402 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1403 (vect_gen_perm_mask_any): Likewise.
1405 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1407 * int-vector-builder.h: New file.
1408 * vec-perm-indices.h: Include int-vector-builder.h.
1409 (vec_perm_indices): Redefine as an int_vector_builder.
1410 (auto_vec_perm_indices): Delete.
1411 (vec_perm_builder): Redefine as a stand-alone class.
1412 (vec_perm_indices::vec_perm_indices): New function.
1413 (vec_perm_indices::clamp): Likewise.
1414 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1415 (vec_perm_indices::new_vector): New function.
1416 (vec_perm_indices::new_expanded_vector): Update for new
1417 vec_perm_indices class.
1418 (vec_perm_indices::rotate_inputs): New function.
1419 (vec_perm_indices::all_in_range_p): Operate directly on the
1420 encoded form, without computing elided elements.
1421 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1422 encoding. Update for new vec_perm_indices class.
1423 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1424 the given vec_perm_builder.
1425 (expand_vec_perm_var): Update vec_perm_builder constructor.
1426 (expand_mult_highpart): Use vec_perm_builder instead of
1427 auto_vec_perm_indices.
1428 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1429 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1430 or double series encoding as appropriate.
1431 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1432 vec_perm_indices instead of auto_vec_perm_indices.
1433 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1434 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1435 (vect_permute_store_chain): Likewise.
1436 (vect_grouped_load_supported): Likewise.
1437 (vect_permute_load_chain): Likewise.
1438 (vect_shift_permute_load_chain): Likewise.
1439 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1440 (vect_transform_slp_perm_load): Likewise.
1441 (vect_schedule_slp_instance): Likewise.
1442 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1443 (vectorizable_mask_load_store): Likewise.
1444 (vectorizable_bswap): Likewise.
1445 (vectorizable_store): Likewise.
1446 (vectorizable_load): Likewise.
1447 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1448 vec_perm_indices instead of auto_vec_perm_indices. Use
1449 tree_to_vec_perm_builder to read the vector from a tree.
1450 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1451 vec_perm_builder instead of a vec_perm_indices.
1452 (have_whole_vector_shift): Use vec_perm_builder and
1453 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1454 truncation to calc_vec_perm_mask_for_shift.
1455 (vect_create_epilog_for_reduction): Likewise.
1456 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1457 from auto_vec_perm_indices to vec_perm_indices.
1458 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1459 instead of changing individual elements.
1460 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1461 the vector in d.perm.
1462 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1463 from auto_vec_perm_indices to vec_perm_indices.
1464 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1465 instead of changing individual elements.
1466 (arm_vectorize_vec_perm_const): Use new_vector to install
1467 the vector in d.perm.
1468 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1469 Update vec_perm_builder constructor.
1470 (rs6000_expand_interleave): Likewise.
1471 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1472 (rs6000_expand_interleave): Likewise.
1474 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1476 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1477 to qimode could truncate the indices.
1478 * optabs.c (expand_vec_perm_var): Likewise.
1480 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1482 * Makefile.in (OBJS): Add vec-perm-indices.o.
1483 * vec-perm-indices.h: New file.
1484 * vec-perm-indices.c: Likewise.
1485 * target.h (vec_perm_indices): Replace with a forward class
1487 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1488 * optabs.h: Include vec-perm-indices.h.
1489 (expand_vec_perm): Delete.
1490 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1491 (expand_vec_perm_const): Declare.
1492 * target.def (vec_perm_const_ok): Replace with...
1493 (vec_perm_const): ...this new hook.
1494 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1495 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1496 * doc/tm.texi: Regenerate.
1497 * optabs.def (vec_perm_const): Delete.
1498 * doc/md.texi (vec_perm_const): Likewise.
1499 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1500 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1501 expand_vec_perm for constant permutation vectors. Assert that
1502 the mode of variable permutation vectors is the integer equivalent
1503 of the mode that is being permuted.
1504 * optabs-query.h (selector_fits_mode_p): Declare.
1505 * optabs-query.c: Include vec-perm-indices.h.
1506 (selector_fits_mode_p): New function.
1507 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1508 is defined, instead of checking whether the vec_perm_const_optab
1509 exists. Use targetm.vectorize.vec_perm_const instead of
1510 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1511 fit in the vector mode before using a variable permute.
1512 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1513 vec_perm_indices instead of an rtx.
1514 (expand_vec_perm): Replace with...
1515 (expand_vec_perm_const): ...this new function. Take the selector
1516 as a vec_perm_indices rather than an rtx. Also take the mode of
1517 the selector. Update call to shift_amt_for_vec_perm_mask.
1518 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1519 Use vec_perm_indices::new_expanded_vector to expand the original
1520 selector into bytes. Check whether the indices fit in the vector
1521 mode before using a variable permute.
1522 (expand_vec_perm_var): Make global.
1523 (expand_mult_highpart): Use expand_vec_perm_const.
1524 * fold-const.c: Includes vec-perm-indices.h.
1525 * tree-ssa-forwprop.c: Likewise.
1526 * tree-vect-data-refs.c: Likewise.
1527 * tree-vect-generic.c: Likewise.
1528 * tree-vect-loop.c: Likewise.
1529 * tree-vect-slp.c: Likewise.
1530 * tree-vect-stmts.c: Likewise.
1531 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1533 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1534 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1535 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1536 (aarch64_vectorize_vec_perm_const): ...this new function.
1537 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1538 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1539 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1540 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1541 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1542 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1543 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1545 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1546 check for NEON modes.
1547 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1548 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1549 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1550 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1552 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1553 the old VEC_PERM_CONST conditions.
1554 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1555 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1556 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1557 (ia64_vectorize_vec_perm_const_ok): Merge into...
1558 (ia64_vectorize_vec_perm_const): ...this new function.
1559 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1560 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1561 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1562 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1563 * config/mips/mips.c (mips_expand_vec_perm_const)
1564 (mips_vectorize_vec_perm_const_ok): Merge into...
1565 (mips_vectorize_vec_perm_const): ...this new function.
1566 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1567 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1568 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1569 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1570 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1571 (rs6000_expand_vec_perm_const): Delete.
1572 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1574 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1575 (altivec_expand_vec_perm_const_le): Take each operand individually.
1576 Operate on constant selectors rather than rtxes.
1577 (altivec_expand_vec_perm_const): Likewise. Update call to
1578 altivec_expand_vec_perm_const_le.
1579 (rs6000_expand_vec_perm_const): Delete.
1580 (rs6000_vectorize_vec_perm_const_ok): Delete.
1581 (rs6000_vectorize_vec_perm_const): New function.
1582 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1583 an element count and rtx array.
1584 (rs6000_expand_extract_even): Update call accordingly.
1585 (rs6000_expand_interleave): Likewise.
1586 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1587 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1588 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1589 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1590 (rs6000_expand_vec_perm_const): Delete.
1591 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1592 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1593 (altivec_expand_vec_perm_const_le): Take each operand individually.
1594 Operate on constant selectors rather than rtxes.
1595 (altivec_expand_vec_perm_const): Likewise. Update call to
1596 altivec_expand_vec_perm_const_le.
1597 (rs6000_expand_vec_perm_const): Delete.
1598 (rs6000_vectorize_vec_perm_const_ok): Delete.
1599 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1600 reference to the SPE evmerge intructions.
1601 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1602 an element count and rtx array.
1603 (rs6000_expand_extract_even): Update call accordingly.
1604 (rs6000_expand_interleave): Likewise.
1605 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1606 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1608 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1610 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1612 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1613 vector mode and that that mode matches the mode of the data
1615 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1616 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1617 directly using expand_vec_perm_1 when forcing selectors into
1619 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1621 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1623 * optabs-query.h (can_vec_perm_p): Delete.
1624 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1625 * optabs-query.c (can_vec_perm_p): Split into...
1626 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1627 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1628 particular selector is valid.
1629 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1630 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1631 (vect_grouped_load_supported): Likewise.
1632 (vect_shift_permute_load_chain): Likewise.
1633 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1634 (vect_transform_slp_perm_load): Likewise.
1635 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1636 (vectorizable_bswap): Likewise.
1637 (vect_gen_perm_mask_checked): Likewise.
1638 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1639 implementations of variable permutation vectors into account
1640 when deciding which selector to use.
1641 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1642 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1643 with a false third argument.
1644 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1645 to test whether the constant selector is valid and can_vec_perm_var_p
1646 to test whether a variable selector is valid.
1648 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1650 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1651 * optabs-query.c (can_vec_perm_p): Likewise.
1652 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1653 instead of vec_perm_indices.
1654 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1655 (vect_gen_perm_mask_checked): Likewise,
1656 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1657 (vect_gen_perm_mask_checked): Likewise,
1659 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1661 * optabs-query.h (qimode_for_vec_perm): Declare.
1662 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1663 (qimode_for_vec_perm): ...this new function.
1664 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1666 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1668 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1669 does not have a conditional at the top.
1671 2018-01-02 Richard Biener <rguenther@suse.de>
1673 * ipa-inline.c (big_speedup_p): Fix expression.
1675 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1678 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1681 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1685 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1686 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1687 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1688 cond_taken_branch_cost 3->4.
1690 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1692 PR tree-optimization/83581
1693 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1694 TODO_cleanup_cfg if any changes have been made.
1697 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1698 convert_modes if target mode has the right side, but different mode
1702 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1703 last argument when extracting from CONCAT. If either from_real or
1704 from_imag is NULL, use expansion through memory. If result is not
1705 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1706 the parts directly to inner mode, if even that fails, use expansion
1710 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1711 check for bswap in mode rather than HImode and use that in expand_unop
1714 Copyright (C) 2018 Free Software Foundation, Inc.
1716 Copying and distribution of this file, with or without modification,
1717 are permitted in any medium without royalty provided the copyright
1718 notice and this notice are preserved.