1 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4 punt for user-aligned variables.
6 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
8 * tree-chrec.c (chrec_contains_symbols): Return true for
11 2018-01-05 Sudakshina Das <sudi.das@arm.com>
14 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
15 of (x|y) == x for BICS pattern.
17 2018-01-05 Jakub Jelinek <jakub@redhat.com>
19 PR tree-optimization/83605
20 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
21 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
24 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
26 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
27 * config/epiphany/rtems.h: New file.
29 2018-01-04 Jakub Jelinek <jakub@redhat.com>
30 Uros Bizjak <ubizjak@gmail.com>
33 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
34 QIreg_operand instead of register_operand predicate.
35 * config/i386/i386.c (ix86_rop_should_change_byte_p,
36 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
37 comments instead of -fmitigate[-_]rop.
39 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
42 * cgraphunit.c (symbol_table::compile): Switch to text_section
43 before calling assembly_start debug hook.
44 * run-rtl-passes.c (run_rtl_passes): Likewise.
47 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
49 * tree-vrp.c (extract_range_from_binary_expr_1): Check
50 range_int_cst_p rather than !symbolic_range_p before calling
51 extract_range_from_multiplicative_op_1.
53 2017-01-04 Jeff Law <law@redhat.com>
55 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
56 redundant test in assertion.
58 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
60 * doc/rtl.texi: Document machine_mode wrapper classes.
62 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
64 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
67 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
69 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
70 the VEC_PERM_EXPR fold to fail.
72 2018-01-04 Jakub Jelinek <jakub@redhat.com>
75 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
78 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
81 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
84 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
87 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
88 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
90 2018-01-04 Jakub Jelinek <jakub@redhat.com>
93 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
94 is BLKmode and bitpos not zero or mode change is needed.
96 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
99 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
102 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
105 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
106 instead of MULT rtx. Update all corresponding splitters.
108 (*ssub<modesuffix>): Ditto.
110 (*cmp_sadd_di): Update split patterns.
111 (*cmp_sadd_si): Ditto.
112 (*cmp_sadd_sidi): Ditto.
113 (*cmp_ssub_di): Ditto.
114 (*cmp_ssub_si): Ditto.
115 (*cmp_ssub_sidi): Ditto.
116 * config/alpha/predicates.md (const23_operand): New predicate.
117 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
118 Look for ASHIFT, not MULT inner operand.
119 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
121 2018-01-04 Martin Liska <mliska@suse.cz>
123 PR gcov-profile/83669
124 * gcov.c (output_intermediate_file): Add version to intermediate
126 * doc/gcov.texi: Document new field 'version' in intermediate
127 file format. Fix location of '-k' option of gcov command.
129 2018-01-04 Martin Liska <mliska@suse.cz>
132 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
134 2018-01-04 Jakub Jelinek <jakub@redhat.com>
136 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
138 2018-01-03 Martin Sebor <msebor@redhat.com>
140 PR tree-optimization/83655
141 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
142 checking calls with invalid arguments.
144 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
146 * tree-vect-stmts.c (vect_get_store_rhs): New function.
147 (vectorizable_mask_load_store): Delete.
148 (vectorizable_call): Return false for masked loads and stores.
149 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
150 instead of gimple_assign_rhs1.
151 (vectorizable_load): Handle IFN_MASK_LOAD.
152 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
156 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
158 (vectorizable_mask_load_store): ...here.
159 (vectorizable_load): ...and here.
161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
163 * tree-vect-stmts.c (vect_build_all_ones_mask)
164 (vect_build_zero_merge_argument): New functions, split out from...
165 (vectorizable_load): ...here.
167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
169 * tree-vect-stmts.c (vect_check_store_rhs): New function,
171 (vectorizable_mask_load_store): ...here.
172 (vectorizable_store): ...and here.
174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
176 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
178 (vectorizable_mask_load_store): ...here.
180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
182 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
183 (vect_model_store_cost): Take a vec_load_store_type instead of a
185 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
186 (vect_model_store_cost): Take a vec_load_store_type instead of a
188 (vectorizable_mask_load_store): Update accordingly.
189 (vectorizable_store): Likewise.
190 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
192 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
194 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
195 IFN_MASK_LOAD calls here rather than...
196 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
198 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
199 Alan Hayward <alan.hayward@arm.com>
200 David Sherwood <david.sherwood@arm.com>
202 * expmed.c (extract_bit_field_1): For vector extracts,
203 fall back to extract_bit_field_as_subreg if vec_extract
206 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
207 Alan Hayward <alan.hayward@arm.com>
208 David Sherwood <david.sherwood@arm.com>
210 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
211 they are variable or constant sized.
212 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
213 slots for constant-sized data.
215 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
216 Alan Hayward <alan.hayward@arm.com>
217 David Sherwood <david.sherwood@arm.com>
219 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
220 handling COND_EXPRs with boolean comparisons, try to find a better
221 basis for the mask type than the boolean itself.
223 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
225 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
226 is calculated and how it can be overridden.
227 * genmodes.c (max_bitsize_mode_any_mode): New variable.
228 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
230 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
233 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
234 Alan Hayward <alan.hayward@arm.com>
235 David Sherwood <david.sherwood@arm.com>
237 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
238 Remove the mode argument.
239 (aarch64_simd_valid_immediate): Remove the mode and inverse
241 * config/aarch64/iterators.md (bitsize): New iterator.
242 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
243 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
244 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
245 aarch64_simd_valid_immediate.
246 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
247 (aarch64_reg_or_bic_imm): Likewise.
248 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
249 with an insn_type enum and msl with a modifier_type enum.
250 Replace element_width with a scalar_mode. Change the shift
251 to unsigned int. Add constructors for scalar_float_mode and
252 scalar_int_mode elements.
253 (aarch64_vect_float_const_representable_p): Delete.
254 (aarch64_can_const_movi_rtx_p)
255 (aarch64_simd_scalar_immediate_valid_for_move)
256 (aarch64_simd_make_constant): Update call to
257 aarch64_simd_valid_immediate.
258 (aarch64_advsimd_valid_immediate_hs): New function.
259 (aarch64_advsimd_valid_immediate): Likewise.
260 (aarch64_simd_valid_immediate): Remove mode and inverse
261 arguments. Rewrite to use the above. Use const_vec_duplicate_p
262 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
263 and aarch64_float_const_representable_p on the result.
264 (aarch64_output_simd_mov_immediate): Remove mode argument.
265 Update call to aarch64_simd_valid_immediate and use of
267 (aarch64_output_scalar_simd_mov_immediate): Update call
270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
271 Alan Hayward <alan.hayward@arm.com>
272 David Sherwood <david.sherwood@arm.com>
274 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
275 (mode_nunits): Likewise CONST_MODE_NUNITS.
276 * machmode.def (ADJUST_NUNITS): Document.
277 * genmodes.c (mode_data::need_nunits_adj): New field.
278 (blank_mode): Update accordingly.
279 (adj_nunits): New variable.
280 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
282 (emit_mode_size_inline): Set need_bytesize_adj for all modes
283 listed in adj_nunits.
284 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
285 listed in adj_nunits. Don't emit case statements for such modes.
286 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
287 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
288 nothing if adj_nunits is nonnull.
289 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
290 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
291 (emit_mode_fbit): Update use of print_maybe_const_decl.
292 (emit_move_size): Likewise. Treat the array as non-const
294 (emit_mode_adjustments): Handle adj_nunits.
296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
298 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
299 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
300 (VECTOR_MODES): Use it.
301 (make_vector_modes): Take the prefix as an argument.
303 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
304 Alan Hayward <alan.hayward@arm.com>
305 David Sherwood <david.sherwood@arm.com>
307 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
308 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
309 for MODE_VECTOR_BOOL.
310 * machmode.def (VECTOR_BOOL_MODE): Document.
311 * genmodes.c (VECTOR_BOOL_MODE): New macro.
312 (make_vector_bool_mode): New function.
313 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
315 * lto-streamer-in.c (lto_input_mode_table): Likewise.
316 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
318 * stor-layout.c (int_mode_for_mode): Likewise.
319 * tree.c (build_vector_type_for_mode): Likewise.
320 * varasm.c (output_constant_pool_2): Likewise.
321 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
322 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
323 for MODE_VECTOR_BOOL.
324 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
325 of mode class checks.
326 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
327 instead of a list of mode class checks.
328 (expand_vector_scalar_condition): Likewise.
329 (type_for_widest_vector_mode): Handle BImode as an inner mode.
331 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
332 Alan Hayward <alan.hayward@arm.com>
333 David Sherwood <david.sherwood@arm.com>
335 * machmode.h (mode_size): Change from unsigned short to
337 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
338 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
339 or if measurement_type is not polynomial.
340 (fixed_size_mode::includes_p): Check for constant-sized modes.
341 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
342 return a poly_uint16 rather than an unsigned short.
343 (emit_mode_size): Change the type of mode_size from unsigned short
344 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
345 (emit_mode_adjustments): Cope with polynomial vector sizes.
346 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
348 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
350 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
351 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
352 * caller-save.c (setup_save_areas): Likewise.
353 (replace_reg_with_saved_mem): Likewise.
354 * calls.c (emit_library_call_value_1): Likewise.
355 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
356 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
357 (gen_lowpart_for_combine): Likewise.
358 * convert.c (convert_to_integer_1): Likewise.
359 * cse.c (equiv_constant, cse_insn): Likewise.
360 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
361 (cselib_subst_to_values): Likewise.
362 * dce.c (word_dce_process_block): Likewise.
363 * df-problems.c (df_word_lr_mark_ref): Likewise.
364 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
365 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
366 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
367 (rtl_for_decl_location): Likewise.
368 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
369 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
370 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
371 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
372 (expand_expr_real_1): Likewise.
373 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
374 (pad_below): Likewise.
375 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
376 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
377 * ira.c (get_subreg_tracking_sizes): Likewise.
378 * ira-build.c (ira_create_allocno_objects): Likewise.
379 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
380 (ira_sort_regnos_for_alter_reg): Likewise.
381 * ira-costs.c (record_operand_costs): Likewise.
382 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
383 (resolve_simple_move): Likewise.
384 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
385 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
386 (lra_constraints): Likewise.
387 (CONST_POOL_OK_P): Reject variable-sized modes.
388 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
389 (add_pseudo_to_slot, lra_spill): Likewise.
390 * omp-low.c (omp_clause_aligned_alignment): Likewise.
391 * optabs-query.c (get_best_extraction_insn): Likewise.
392 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
393 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
394 (expand_mult_highpart, valid_multiword_target_p): Likewise.
395 * recog.c (offsettable_address_addr_space_p): Likewise.
396 * regcprop.c (maybe_mode_change): Likewise.
397 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
398 * regrename.c (build_def_use): Likewise.
399 * regstat.c (dump_reg_info): Likewise.
400 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
401 (find_reloads, find_reloads_subreg_address): Likewise.
402 * reload1.c (eliminate_regs_1): Likewise.
403 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
404 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
405 (simplify_binary_operation_1, simplify_subreg): Likewise.
406 * targhooks.c (default_function_arg_padding): Likewise.
407 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
408 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
409 (verify_gimple_assign_ternary): Likewise.
410 * tree-inline.c (estimate_move_cost): Likewise.
411 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
412 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
413 (get_address_cost_ainc): Likewise.
414 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
415 (vect_supportable_dr_alignment): Likewise.
416 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
417 (vectorizable_reduction): Likewise.
418 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
419 (vectorizable_operation, vectorizable_load): Likewise.
420 * tree.c (build_same_sized_truth_vector_type): Likewise.
421 * valtrack.c (cleanup_auto_inc_dec): Likewise.
422 * var-tracking.c (emit_note_insn_var_location): Likewise.
423 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
424 (ADDR_VEC_ALIGN): Likewise.
426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
427 Alan Hayward <alan.hayward@arm.com>
428 David Sherwood <david.sherwood@arm.com>
430 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
432 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
433 or if measurement_type is polynomial.
434 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
435 * combine.c (make_extraction): Likewise.
436 * dse.c (find_shift_sequence): Likewise.
437 * dwarf2out.c (mem_loc_descriptor): Likewise.
438 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
439 (extract_bit_field, extract_low_bits): Likewise.
440 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
441 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
442 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
443 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
444 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
445 * reload.c (find_reloads): Likewise.
446 * reload1.c (alter_reg): Likewise.
447 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
448 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
449 * tree-if-conv.c (predicate_mem_writes): Likewise.
450 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
451 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
452 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
453 * valtrack.c (dead_debug_insert_temp): Likewise.
454 * varasm.c (mergeable_constant_section): Likewise.
455 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
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 * expr.c (expand_assignment): Cope with polynomial mode sizes
462 when assigning to a CONCAT.
464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
465 Alan Hayward <alan.hayward@arm.com>
466 David Sherwood <david.sherwood@arm.com>
468 * machmode.h (mode_precision): Change from unsigned short to
470 (mode_to_precision): Return a poly_uint16 rather than an unsigned
472 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
473 or if measurement_type is not polynomial.
474 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
475 in which the mode is already known to be a scalar_int_mode.
476 * genmodes.c (emit_mode_precision): Change the type of mode_precision
477 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
479 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
480 for GET_MODE_PRECISION.
481 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
482 for GET_MODE_PRECISION.
483 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
485 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
486 (expand_field_assignment, make_extraction): Likewise.
487 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
488 (get_last_value): Likewise.
489 * convert.c (convert_to_integer_1): Likewise.
490 * cse.c (cse_insn): Likewise.
491 * expr.c (expand_expr_real_1): Likewise.
492 * lra-constraints.c (simplify_operand_subreg): Likewise.
493 * optabs-query.c (can_atomic_load_p): Likewise.
494 * optabs.c (expand_atomic_load): Likewise.
495 (expand_atomic_store): Likewise.
496 * ree.c (combine_reaching_defs): Likewise.
497 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
498 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
499 * tree.h (type_has_mode_precision_p): Likewise.
500 * ubsan.c (instrument_si_overflow): Likewise.
502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
503 Alan Hayward <alan.hayward@arm.com>
504 David Sherwood <david.sherwood@arm.com>
506 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
507 polynomial numbers of units.
508 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
509 (valid_vector_subparts_p): New function.
510 (build_vector_type): Remove temporary shim and take the number
511 of units as a poly_uint64 rather than an int.
512 (build_opaque_vector_type): Take the number of units as a
513 poly_uint64 rather than an int.
514 * tree.c (build_vector_from_ctor): Handle polynomial
515 TYPE_VECTOR_SUBPARTS.
516 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
517 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
518 (build_vector_from_val): If the number of units is variable,
519 use build_vec_duplicate_cst for constant operands and
520 VEC_DUPLICATE_EXPR otherwise.
521 (make_vector_type): Remove temporary is_constant ().
522 (build_vector_type, build_opaque_vector_type): Take the number of
523 units as a poly_uint64 rather than an int.
524 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
526 * cfgexpand.c (expand_debug_expr): Likewise.
527 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
528 (store_constructor, expand_expr_real_1): Likewise.
529 (const_scalar_mask_from_tree): Likewise.
530 * fold-const-call.c (fold_const_reduction): Likewise.
531 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
532 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
533 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
534 (fold_relational_const): Likewise.
535 (native_interpret_vector): Likewise. Change the size from an
536 int to an unsigned int.
537 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
538 TYPE_VECTOR_SUBPARTS.
539 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
540 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
541 duplicating a non-constant operand into a variable-length vector.
542 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
543 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
544 * ipa-icf.c (sem_variable::equals): Likewise.
545 * match.pd: Likewise.
546 * omp-simd-clone.c (simd_clone_subparts): Likewise.
547 * print-tree.c (print_node): Likewise.
548 * stor-layout.c (layout_type): Likewise.
549 * targhooks.c (default_builtin_vectorization_cost): Likewise.
550 * tree-cfg.c (verify_gimple_comparison): Likewise.
551 (verify_gimple_assign_binary): Likewise.
552 (verify_gimple_assign_ternary): Likewise.
553 (verify_gimple_assign_single): Likewise.
554 * tree-pretty-print.c (dump_generic_node): Likewise.
555 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
556 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
557 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
558 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
559 (vect_shift_permute_load_chain): Likewise.
560 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
561 (expand_vector_condition, optimize_vector_constructor): Likewise.
562 (lower_vec_perm, get_compute_type): Likewise.
563 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
564 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
565 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
566 (vect_recog_mask_conversion_pattern): Likewise.
567 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
568 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
569 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
570 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
571 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
572 (vectorizable_shift, vectorizable_operation, vectorizable_store)
573 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
574 (supportable_widening_operation): Likewise.
575 (supportable_narrowing_operation): Likewise.
576 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
578 * varasm.c (output_constant): Likewise.
580 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
581 Alan Hayward <alan.hayward@arm.com>
582 David Sherwood <david.sherwood@arm.com>
584 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
585 so that both the length == 3 and length != 3 cases set up their
586 own permute vectors. Add comments explaining why we know the
587 number of elements is constant.
588 (vect_permute_load_chain): Likewise.
590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
591 Alan Hayward <alan.hayward@arm.com>
592 David Sherwood <david.sherwood@arm.com>
594 * machmode.h (mode_nunits): Change from unsigned char to
596 (ONLY_FIXED_SIZE_MODES): New macro.
597 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
598 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
599 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
601 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
602 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
603 or if measurement_type is not polynomial.
604 * genmodes.c (ZERO_COEFFS): New macro.
605 (emit_mode_nunits_inline): Make mode_nunits_inline return a
607 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
608 Use ZERO_COEFFS when emitting initializers.
609 * data-streamer.h (bp_pack_poly_value): New function.
610 (bp_unpack_poly_value): Likewise.
611 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
613 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
615 * tree.c (make_vector_type): Remove temporary shim and make
616 the real function take the number of units as a poly_uint64
618 (build_vector_type_for_mode): Handle polynomial nunits.
619 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
620 * emit-rtl.c (const_vec_series_p_1): Likewise.
621 (gen_rtx_CONST_VECTOR): Likewise.
622 * fold-const.c (test_vec_duplicate_folding): Likewise.
623 * genrecog.c (validate_pattern): Likewise.
624 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
625 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
626 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
627 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
628 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
629 * rtlanal.c (subreg_get_info): Likewise.
630 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
631 (vect_grouped_load_supported): Likewise.
632 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
633 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
634 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
635 (simplify_const_unary_operation, simplify_binary_operation_1)
636 (simplify_const_binary_operation, simplify_ternary_operation)
637 (test_vector_ops_duplicate, test_vector_ops): Likewise.
638 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
639 instead of CONST_VECTOR_NUNITS.
640 * varasm.c (output_constant_pool_2): Likewise.
641 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
642 explicit-encoded elements in the XVEC for variable-length vectors.
644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
646 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
649 Alan Hayward <alan.hayward@arm.com>
650 David Sherwood <david.sherwood@arm.com>
652 * coretypes.h (fixed_size_mode): Declare.
653 (fixed_size_mode_pod): New typedef.
654 * builtins.h (target_builtins::x_apply_args_mode)
655 (target_builtins::x_apply_result_mode): Change type to
657 * builtins.c (apply_args_size, apply_result_size, result_vector)
658 (expand_builtin_apply_args_1, expand_builtin_apply)
659 (expand_builtin_return): Update accordingly.
661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
663 * cse.c (hash_rtx_cb): Hash only the encoded elements.
664 * cselib.c (cselib_hash_rtx): Likewise.
665 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
666 CONST_VECTOR encoding.
668 2017-01-03 Jakub Jelinek <jakub@redhat.com>
669 Jeff Law <law@redhat.com>
672 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
673 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
674 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
675 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
678 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
679 explicitly probe *sp in a noreturn function if there were any callee
680 register saves or frame pointer is needed.
682 2018-01-03 Jakub Jelinek <jakub@redhat.com>
685 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
686 BLKmode for ternary, binary or unary expressions.
689 * var-tracking.c (delete_vta_debug_insn): New inline function.
690 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
691 insns from get_insns () to NULL instead of each bb separately.
692 Use delete_vta_debug_insn. No longer static.
693 (vt_debug_insns_local, variable_tracking_main_1): Adjust
694 delete_vta_debug_insns callers.
695 * rtl.h (delete_vta_debug_insns): Declare.
696 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
697 instead of variable_tracking_main.
699 2018-01-03 Martin Sebor <msebor@redhat.com>
701 PR tree-optimization/83603
702 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
703 arguments past the endof the argument list in functions declared
705 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
706 Avoid checking when arguments are null.
708 2018-01-03 Martin Sebor <msebor@redhat.com>
711 * doc/extend.texi (attribute const): Fix a typo.
712 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
713 issuing -Wsuggest-attribute for void functions.
715 2018-01-03 Martin Sebor <msebor@redhat.com>
717 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
718 offset_int::from instead of wide_int::to_shwi.
719 (maybe_diag_overlap): Remove assertion.
720 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
721 * gimple-ssa-sprintf.c (format_directive): Same.
722 (parse_directive): Same.
723 (sprintf_dom_walker::compute_format_length): Same.
724 (try_substitute_return_value): Same.
726 2017-01-03 Jeff Law <law@redhat.com>
729 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
730 non-constant residual for zero at runtime and avoid probing in
731 that case. Reorganize code for trailing problem to mirror handling
734 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
736 PR tree-optimization/83501
737 * tree-ssa-strlen.c (get_string_cst): New.
738 (handle_char_store): Call get_string_cst.
740 2018-01-03 Martin Liska <mliska@suse.cz>
742 PR tree-optimization/83593
743 * tree-ssa-strlen.c: Include tree-cfg.h.
744 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
745 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
746 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
748 (strlen_dom_walker::before_dom_children): Call
749 gimple_purge_dead_eh_edges. Dump tranformation with details
751 (strlen_dom_walker::before_dom_children): Update call by adding
752 new argument cleanup_eh.
753 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
755 2018-01-03 Martin Liska <mliska@suse.cz>
758 * cif-code.def (VARIADIC_THUNK): New enum value.
759 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
762 2018-01-03 Jan Beulich <jbeulich@suse.com>
764 * sse.md (mov<mode>_internal): Tighten condition for when to use
765 vmovdqu<ssescalarsize> for TI and OI modes.
767 2018-01-03 Jakub Jelinek <jakub@redhat.com>
769 Update copyright years.
771 2018-01-03 Martin Liska <mliska@suse.cz>
774 * ipa-visibility.c (function_and_variable_visibility): Skip
775 functions with noipa attribure.
777 2018-01-03 Jakub Jelinek <jakub@redhat.com>
779 * gcc.c (process_command): Update copyright notice dates.
780 * gcov-dump.c (print_version): Ditto.
781 * gcov.c (print_version): Ditto.
782 * gcov-tool.c (print_version): Ditto.
783 * gengtype.c (create_file): Ditto.
784 * doc/cpp.texi: Bump @copying's copyright year.
785 * doc/cppinternals.texi: Ditto.
786 * doc/gcc.texi: Ditto.
787 * doc/gccint.texi: Ditto.
788 * doc/gcov.texi: Ditto.
789 * doc/install.texi: Ditto.
790 * doc/invoke.texi: Ditto.
792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
794 * vector-builder.h (vector_builder::m_full_nelts): Change from
795 unsigned int to poly_uint64.
796 (vector_builder::full_nelts): Update prototype accordingly.
797 (vector_builder::new_vector): Likewise.
798 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
799 (vector_builder::operator ==): Likewise.
800 (vector_builder::finalize): Likewise.
801 * int-vector-builder.h (int_vector_builder::int_vector_builder):
802 Take the number of elements as a poly_uint64 rather than an
804 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
805 from unsigned int to poly_uint64.
806 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
807 (vec_perm_indices::new_vector): Likewise.
808 (vec_perm_indices::length): Likewise.
809 (vec_perm_indices::nelts_per_input): Likewise.
810 (vec_perm_indices::input_nelts): Likewise.
811 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
812 number of elements per input as a poly_uint64 rather than an
813 unsigned int. Use the original encoding for variable-length
814 vectors, rather than clamping each individual element.
815 For the second and subsequent elements in each pattern,
816 clamp the step and base before clamping their sum.
817 (vec_perm_indices::series_p): Handle polynomial element counts.
818 (vec_perm_indices::all_in_range_p): Likewise.
819 (vec_perm_indices_to_tree): Likewise.
820 (vec_perm_indices_to_rtx): Likewise.
821 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
822 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
823 (tree_vector_builder::new_binary_operation): Handle polynomial
824 element counts. Return false if we need to know the number
825 of elements at compile time.
826 * fold-const.c (fold_vec_perm): Punt if the number of elements
827 isn't known at compile time.
829 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
831 * vec-perm-indices.h (vec_perm_builder): Change element type
832 from HOST_WIDE_INT to poly_int64.
833 (vec_perm_indices::element_type): Update accordingly.
834 (vec_perm_indices::clamp): Handle polynomial element_types.
835 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
836 (vec_perm_indices::all_in_range_p): Likewise.
837 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
839 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
840 polynomial vec_perm_indices element types.
841 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
842 * fold-const.c (fold_vec_perm): Likewise.
843 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
844 * tree-vect-generic.c (lower_vec_perm): Likewise.
845 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
846 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
847 element type to HOST_WIDE_INT.
849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
850 Alan Hayward <alan.hayward@arm.com>
851 David Sherwood <david.sherwood@arm.com>
853 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
854 rather than an int. Use plus_constant.
855 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
856 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
859 Alan Hayward <alan.hayward@arm.com>
860 David Sherwood <david.sherwood@arm.com>
862 * calls.c (emit_call_1, expand_call): Change struct_value_size from
863 a HOST_WIDE_INT to a poly_int64.
865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
866 Alan Hayward <alan.hayward@arm.com>
867 David Sherwood <david.sherwood@arm.com>
869 * calls.c (load_register_parameters): Cope with polynomial
870 mode sizes. Require a constant size for BLKmode parameters
871 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
872 forces a parameter to be padded at the lsb end in order to
873 fill a complete number of words, require the parameter size
874 to be ordered wrt UNITS_PER_WORD.
876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
877 Alan Hayward <alan.hayward@arm.com>
878 David Sherwood <david.sherwood@arm.com>
880 * reload1.c (spill_stack_slot_width): Change element type
881 from unsigned int to poly_uint64_pod.
882 (alter_reg): Treat mode sizes as polynomial.
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 * reload.c (complex_word_subreg_p): New function.
889 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
896 operands whose sizes aren't ordered.
897 (match_reload): Refer to this check here.
899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
900 Alan Hayward <alan.hayward@arm.com>
901 David Sherwood <david.sherwood@arm.com>
903 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
904 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
911 Use plus_constant instead of gen_rtx_PLUS.
913 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
914 Alan Hayward <alan.hayward@arm.com>
915 David Sherwood <david.sherwood@arm.com>
917 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
918 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
919 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
920 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
921 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
922 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
923 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
924 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
925 * config/i386/i386.c (ix86_push_rounding): ...this new function.
926 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
928 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
929 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
930 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
931 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
932 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
933 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
934 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
935 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
936 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
937 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
939 * expr.c (emit_move_resolve_push): Treat the input and result
940 of PUSH_ROUNDING as a poly_int64.
941 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
942 (emit_push_insn): Likewise.
943 * lra-eliminations.c (mark_not_eliminable): Likewise.
944 * recog.c (push_operand): Likewise.
945 * reload1.c (elimination_effects): Likewise.
946 * rtlanal.c (nonzero_bits1): Likewise.
947 * calls.c (store_one_arg): Likewise. Require the padding to be
948 known at compile time.
950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
951 Alan Hayward <alan.hayward@arm.com>
952 David Sherwood <david.sherwood@arm.com>
954 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
955 Use plus_constant instead of gen_rtx_PLUS.
957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
958 Alan Hayward <alan.hayward@arm.com>
959 David Sherwood <david.sherwood@arm.com>
961 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
964 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
965 Alan Hayward <alan.hayward@arm.com>
966 David Sherwood <david.sherwood@arm.com>
968 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
969 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
970 via stack temporaries. Treat the mode size as polynomial too.
972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
973 Alan Hayward <alan.hayward@arm.com>
974 David Sherwood <david.sherwood@arm.com>
976 * expr.c (expand_expr_real_2): When handling conversions involving
977 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
978 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
979 as a poly_uint64 too.
981 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
982 Alan Hayward <alan.hayward@arm.com>
983 David Sherwood <david.sherwood@arm.com>
985 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
987 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
988 Alan Hayward <alan.hayward@arm.com>
989 David Sherwood <david.sherwood@arm.com>
991 * combine.c (can_change_dest_mode): Handle polynomial
992 REGMODE_NATURAL_SIZE.
993 * expmed.c (store_bit_field_1): Likewise.
994 * expr.c (store_constructor): Likewise.
995 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
996 and polynomial REGMODE_NATURAL_SIZE.
997 (gen_lowpart_common): Likewise.
998 * reginfo.c (record_subregs_of_mode): Likewise.
999 * rtlanal.c (read_modify_subreg_p): Likewise.
1001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1002 Alan Hayward <alan.hayward@arm.com>
1003 David Sherwood <david.sherwood@arm.com>
1005 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
1006 numbers of elements.
1008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1009 Alan Hayward <alan.hayward@arm.com>
1010 David Sherwood <david.sherwood@arm.com>
1012 * match.pd: Cope with polynomial numbers of vector elements.
1014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1015 Alan Hayward <alan.hayward@arm.com>
1016 David Sherwood <david.sherwood@arm.com>
1018 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1019 in a POINTER_PLUS_EXPR.
1021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1022 Alan Hayward <alan.hayward@arm.com>
1023 David Sherwood <david.sherwood@arm.com>
1025 * omp-simd-clone.c (simd_clone_subparts): New function.
1026 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1027 (ipa_simd_modify_function_body): Likewise.
1029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1030 Alan Hayward <alan.hayward@arm.com>
1031 David Sherwood <david.sherwood@arm.com>
1033 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1034 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1035 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1036 (expand_vector_condition, vector_element): Likewise.
1037 (subparts_gt): New function.
1038 (get_compute_type): Use subparts_gt.
1039 (count_type_subparts): Delete.
1040 (expand_vector_operations_1): Use subparts_gt instead of
1041 count_type_subparts.
1043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1044 Alan Hayward <alan.hayward@arm.com>
1045 David Sherwood <david.sherwood@arm.com>
1047 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1048 (vect_compile_time_alias): ...this new function. Do the calculation
1049 on poly_ints rather than trees.
1050 (vect_prune_runtime_alias_test_list): Update call accordingly.
1052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1053 Alan Hayward <alan.hayward@arm.com>
1054 David Sherwood <david.sherwood@arm.com>
1056 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1058 (vect_schedule_slp_instance): Likewise.
1060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1061 Alan Hayward <alan.hayward@arm.com>
1062 David Sherwood <david.sherwood@arm.com>
1064 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1065 constant and extern definitions for variable-length vectors.
1066 (vect_get_constant_vectors): Note that the number of units
1067 is known to be constant.
1069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1070 Alan Hayward <alan.hayward@arm.com>
1071 David Sherwood <david.sherwood@arm.com>
1073 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1074 of units as polynomial. Choose between WIDE and NARROW based
1077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1078 Alan Hayward <alan.hayward@arm.com>
1079 David Sherwood <david.sherwood@arm.com>
1081 * tree-vect-stmts.c (simd_clone_subparts): New function.
1082 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1085 Alan Hayward <alan.hayward@arm.com>
1086 David Sherwood <david.sherwood@arm.com>
1088 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1089 vectors as polynomial. Use build_index_vector for
1092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1093 Alan Hayward <alan.hayward@arm.com>
1094 David Sherwood <david.sherwood@arm.com>
1096 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1097 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1098 for variable-length vectors.
1099 (vectorizable_mask_load_store): Treat the number of units as
1100 polynomial, asserting that it is constant if the condition has
1101 already been enforced.
1102 (vectorizable_store, vectorizable_load): Likewise.
1104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1105 Alan Hayward <alan.hayward@arm.com>
1106 David Sherwood <david.sherwood@arm.com>
1108 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1109 of units as polynomial. Punt if we can't tell at compile time
1110 which vector contains the final result.
1112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1113 Alan Hayward <alan.hayward@arm.com>
1114 David Sherwood <david.sherwood@arm.com>
1116 * tree-vect-loop.c (vectorizable_induction): Treat the number
1117 of units as polynomial. Punt on SLP inductions. Use an integer
1118 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1119 cast of such a series for variable-length floating-point
1122 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1123 Alan Hayward <alan.hayward@arm.com>
1124 David Sherwood <david.sherwood@arm.com>
1126 * tree.h (build_index_vector): Declare.
1127 * tree.c (build_index_vector): New function.
1128 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1129 of units as polynomial, forcibly converting it to a constant if
1130 vectorizable_reduction has already enforced the condition.
1131 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1132 to create a {1,2,3,...} vector.
1133 (vectorizable_reduction): Treat the number of units as polynomial.
1134 Choose vectype_in based on the largest scalar element size rather
1135 than the smallest number of units. Enforce the restrictions
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-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1143 number of units as polynomial.
1145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1146 Alan Hayward <alan.hayward@arm.com>
1147 David Sherwood <david.sherwood@arm.com>
1149 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1150 * target.def (autovectorize_vector_sizes): Return the vector sizes
1151 by pointer, using vector_sizes rather than a bitmask.
1152 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1153 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1154 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1156 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1157 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1158 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1159 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1160 * omp-general.c (omp_max_vf): Likewise.
1161 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1162 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1163 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1164 * tree-vect-slp.c (vect_slp_bb): Likewise.
1165 * doc/tm.texi: Regenerate.
1166 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1168 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1169 the vector size as a poly_uint64 rather than an unsigned int.
1170 (current_vector_size): Change from an unsigned int to a poly_uint64.
1171 (get_vectype_for_scalar_type): Update accordingly.
1172 * tree.h (build_truth_vector_type): Take the size and number of
1173 units as a poly_uint64 rather than an unsigned int.
1174 (build_vector_type): Add a temporary overload that takes
1175 the number of units as a poly_uint64 rather than an unsigned int.
1176 * tree.c (make_vector_type): Likewise.
1177 (build_truth_vector_type): Take the number of units as a poly_uint64
1178 rather than an unsigned int.
1180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1181 Alan Hayward <alan.hayward@arm.com>
1182 David Sherwood <david.sherwood@arm.com>
1184 * target.def (get_mask_mode): Take the number of units and length
1185 as poly_uint64s rather than unsigned ints.
1186 * targhooks.h (default_get_mask_mode): Update accordingly.
1187 * targhooks.c (default_get_mask_mode): Likewise.
1188 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1189 * doc/tm.texi: Regenerate.
1191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1192 Alan Hayward <alan.hayward@arm.com>
1193 David Sherwood <david.sherwood@arm.com>
1195 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1196 * omp-general.c (omp_max_vf): Likewise.
1197 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1198 (expand_omp_simd): Handle polynomial safelen.
1199 * omp-low.c (omplow_simd_context): Add a default constructor.
1200 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1201 (lower_rec_simd_input_clauses): Update accordingly.
1202 (lower_rec_input_clauses): Likewise.
1204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1205 Alan Hayward <alan.hayward@arm.com>
1206 David Sherwood <david.sherwood@arm.com>
1208 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1209 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1210 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1211 (vect_analyze_slp_cost): Likewise.
1212 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1213 (vect_model_load_cost): Likewise.
1215 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1216 Alan Hayward <alan.hayward@arm.com>
1217 David Sherwood <david.sherwood@arm.com>
1219 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1220 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1221 from an unsigned int * to a poly_uint64_pod *.
1222 (calculate_unrolling_factor): New function.
1223 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
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 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1230 from an unsigned int to a poly_uint64.
1231 (_loop_vec_info::slp_unrolling_factor): Likewise.
1232 (_loop_vec_info::vectorization_factor): Change from an int
1234 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1235 (vect_get_num_vectors): New function.
1236 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1237 (vect_get_num_copies): Use vect_get_num_vectors.
1238 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1239 to an unsigned int *.
1240 (vect_analyze_data_refs): Change min_vf from an int * to a
1242 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1243 than an unsigned HOST_WIDE_INT.
1244 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1245 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1246 to an unsigned int *.
1247 (vect_analyze_data_ref_dependences): Likewise.
1248 (vect_compute_data_ref_alignment): Handle polynomial vf.
1249 (vect_enhance_data_refs_alignment): Likewise.
1250 (vect_prune_runtime_alias_test_list): Likewise.
1251 (vect_shift_permute_load_chain): Likewise.
1252 (vect_supportable_dr_alignment): Likewise.
1253 (dependence_distance_ge_vf): Take the vectorization factor as a
1254 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1255 (vect_analyze_data_refs): Change min_vf from an int * to a
1257 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1258 vfm1 as a poly_uint64 rather than an int. Make the same change
1259 for the returned bound_scalar.
1260 (vect_gen_vector_loop_niters): Handle polynomial vf.
1261 (vect_do_peeling): Likewise. Update call to
1262 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1263 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1265 * tree-vect-loop.c (vect_determine_vectorization_factor)
1266 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1267 (vect_get_known_peeling_cost): Likewise.
1268 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1269 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1270 (vect_transform_loop): Likewise. Use the lowest possible VF when
1271 updating the upper bounds of the loop.
1272 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1274 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1275 polynomial unroll factors.
1276 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1277 (vect_make_slp_decision): Likewise.
1278 (vect_supported_load_permutation_p): Likewise, and polynomial
1280 (vect_analyze_slp_cost): Handle polynomial vf.
1281 (vect_slp_analyze_node_operations): Likewise.
1282 (vect_slp_analyze_bb_1): Likewise.
1283 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1284 than an unsigned HOST_WIDE_INT.
1285 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1286 (vectorizable_load): Handle polynomial vf.
1287 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1289 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1292 Alan Hayward <alan.hayward@arm.com>
1293 David Sherwood <david.sherwood@arm.com>
1295 * match.pd: Handle bit operations involving three constants
1296 and try to fold one pair.
1298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1300 * tree-vect-loop-manip.c: Include gimple-fold.h.
1301 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1302 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1303 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1304 Add a path that uses a step of VF instead of 1, but disable it
1306 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1307 and niters_no_overflow parameters. Update calls to
1308 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1309 Create a new SSA name if the latter choses to use a ste other
1310 than zero, and return it via niters_vector_mult_vf_var.
1311 * tree-vect-loop.c (vect_transform_loop): Update calls to
1312 vect_do_peeling, vect_gen_vector_loop_niters and
1313 slpeel_make_loop_iterate_ntimes.
1314 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1315 (vect_gen_vector_loop_niters): Update declarations after above changes.
1317 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1319 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1320 128-bit round to integer instructions.
1321 (ceil<mode>2): Likewise.
1322 (btrunc<mode>2): Likewise.
1323 (round<mode>2): Likewise.
1325 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1327 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1328 unaligned VSX load/store on P8/P9.
1329 (expand_block_clear): Allow the use of unaligned VSX
1330 load/store on P8/P9.
1332 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1334 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1336 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1337 swap associated with both a load and a store.
1339 2018-01-02 Andrew Waterman <andrew@sifive.com>
1341 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1342 * config/riscv/riscv.md (clear_cache): Use it.
1344 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1346 * web.c: Remove out-of-date comment.
1348 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1350 * expr.c (fixup_args_size_notes): Check that any existing
1351 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1352 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1353 (emit_single_push_insn): ...here.
1355 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1357 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1358 (const_vector_encoded_nelts): New function.
1359 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1360 (const_vector_int_elt, const_vector_elt): Declare.
1361 * emit-rtl.c (const_vector_int_elt_1): New function.
1362 (const_vector_elt): Likewise.
1363 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1364 of CONST_VECTOR_ELT.
1366 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1368 * expr.c: Include rtx-vector-builder.h.
1369 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1370 directly on the tree encoding.
1371 (const_vector_from_tree): Likewise.
1372 * optabs.c: Include rtx-vector-builder.h.
1373 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1374 sequence of "u" values.
1375 * vec-perm-indices.c: Include rtx-vector-builder.h.
1376 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1377 directly on the vec_perm_indices encoding.
1379 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1381 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1382 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1383 * rtx-vector-builder.h: New file.
1384 * rtx-vector-builder.c: Likewise.
1385 * rtl.h (rtx_def::u2): Add a const_vector field.
1386 (CONST_VECTOR_NPATTERNS): New macro.
1387 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1388 (CONST_VECTOR_DUPLICATE_P): Likewise.
1389 (CONST_VECTOR_STEPPED_P): Likewise.
1390 (CONST_VECTOR_ENCODED_ELT): Likewise.
1391 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1392 (unwrap_const_vec_duplicate): Likewise.
1393 (const_vec_series_p): Check for a non-duplicated vector encoding.
1394 Say that the function only returns true for integer vectors.
1395 * emit-rtl.c: Include rtx-vector-builder.h.
1396 (gen_const_vec_duplicate_1): Delete.
1397 (gen_const_vector): Call gen_const_vec_duplicate instead of
1398 gen_const_vec_duplicate_1.
1399 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1400 (gen_const_vec_duplicate): Use rtx_vector_builder.
1401 (gen_const_vec_series): Likewise.
1402 (gen_rtx_CONST_VECTOR): Likewise.
1403 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1404 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1405 Build a new vector rather than modifying a CONST_VECTOR in-place.
1406 (handle_special_swappables): Update call accordingly.
1407 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1408 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1409 Build a new vector rather than modifying a CONST_VECTOR in-place.
1410 (handle_special_swappables): Update call accordingly.
1412 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1414 * simplify-rtx.c (simplify_const_binary_operation): Use
1415 CONST_VECTOR_ELT instead of XVECEXP.
1417 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1419 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1420 the selector elements to be different from the data elements
1421 if the selector is a VECTOR_CST.
1422 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1423 ssizetype for the selector.
1425 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1427 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1428 before testing each element individually.
1429 * tree-vect-generic.c (lower_vec_perm): Likewise.
1431 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1433 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1434 * selftest-run-tests.c (selftest::run_tests): Call it.
1435 * vector-builder.h (vector_builder::operator ==): New function.
1436 (vector_builder::operator !=): Likewise.
1437 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1438 (vec_perm_indices::all_from_input_p): New function.
1439 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1440 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1441 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1442 instead of reading the VECTOR_CST directly. Detect whether both
1443 vector inputs are the same before constructing the vec_perm_indices,
1444 and update the number of inputs argument accordingly. Use the
1445 utility functions added above. Only construct sel2 if we need to.
1447 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1449 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1450 the broadcast of the low byte.
1451 (expand_mult_highpart): Use an explicit encoding for the permutes.
1452 * optabs-query.c (can_mult_highpart_p): Likewise.
1453 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1454 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1455 (vectorizable_bswap): Likewise.
1456 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1457 explicit encoding for the power-of-2 permutes.
1458 (vect_permute_store_chain): Likewise.
1459 (vect_grouped_load_supported): Likewise.
1460 (vect_permute_load_chain): Likewise.
1462 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1464 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1465 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1466 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1467 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1468 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1469 (vect_gen_perm_mask_any): Likewise.
1471 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1473 * int-vector-builder.h: New file.
1474 * vec-perm-indices.h: Include int-vector-builder.h.
1475 (vec_perm_indices): Redefine as an int_vector_builder.
1476 (auto_vec_perm_indices): Delete.
1477 (vec_perm_builder): Redefine as a stand-alone class.
1478 (vec_perm_indices::vec_perm_indices): New function.
1479 (vec_perm_indices::clamp): Likewise.
1480 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1481 (vec_perm_indices::new_vector): New function.
1482 (vec_perm_indices::new_expanded_vector): Update for new
1483 vec_perm_indices class.
1484 (vec_perm_indices::rotate_inputs): New function.
1485 (vec_perm_indices::all_in_range_p): Operate directly on the
1486 encoded form, without computing elided elements.
1487 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1488 encoding. Update for new vec_perm_indices class.
1489 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1490 the given vec_perm_builder.
1491 (expand_vec_perm_var): Update vec_perm_builder constructor.
1492 (expand_mult_highpart): Use vec_perm_builder instead of
1493 auto_vec_perm_indices.
1494 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1495 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1496 or double series encoding as appropriate.
1497 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1498 vec_perm_indices instead of auto_vec_perm_indices.
1499 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1500 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1501 (vect_permute_store_chain): Likewise.
1502 (vect_grouped_load_supported): Likewise.
1503 (vect_permute_load_chain): Likewise.
1504 (vect_shift_permute_load_chain): Likewise.
1505 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1506 (vect_transform_slp_perm_load): Likewise.
1507 (vect_schedule_slp_instance): Likewise.
1508 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1509 (vectorizable_mask_load_store): Likewise.
1510 (vectorizable_bswap): Likewise.
1511 (vectorizable_store): Likewise.
1512 (vectorizable_load): Likewise.
1513 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1514 vec_perm_indices instead of auto_vec_perm_indices. Use
1515 tree_to_vec_perm_builder to read the vector from a tree.
1516 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1517 vec_perm_builder instead of a vec_perm_indices.
1518 (have_whole_vector_shift): Use vec_perm_builder and
1519 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1520 truncation to calc_vec_perm_mask_for_shift.
1521 (vect_create_epilog_for_reduction): Likewise.
1522 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1523 from auto_vec_perm_indices to vec_perm_indices.
1524 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1525 instead of changing individual elements.
1526 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1527 the vector in d.perm.
1528 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1529 from auto_vec_perm_indices to vec_perm_indices.
1530 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1531 instead of changing individual elements.
1532 (arm_vectorize_vec_perm_const): Use new_vector to install
1533 the vector in d.perm.
1534 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1535 Update vec_perm_builder constructor.
1536 (rs6000_expand_interleave): Likewise.
1537 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1538 (rs6000_expand_interleave): Likewise.
1540 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1542 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1543 to qimode could truncate the indices.
1544 * optabs.c (expand_vec_perm_var): Likewise.
1546 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1548 * Makefile.in (OBJS): Add vec-perm-indices.o.
1549 * vec-perm-indices.h: New file.
1550 * vec-perm-indices.c: Likewise.
1551 * target.h (vec_perm_indices): Replace with a forward class
1553 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1554 * optabs.h: Include vec-perm-indices.h.
1555 (expand_vec_perm): Delete.
1556 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1557 (expand_vec_perm_const): Declare.
1558 * target.def (vec_perm_const_ok): Replace with...
1559 (vec_perm_const): ...this new hook.
1560 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1561 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1562 * doc/tm.texi: Regenerate.
1563 * optabs.def (vec_perm_const): Delete.
1564 * doc/md.texi (vec_perm_const): Likewise.
1565 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1566 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1567 expand_vec_perm for constant permutation vectors. Assert that
1568 the mode of variable permutation vectors is the integer equivalent
1569 of the mode that is being permuted.
1570 * optabs-query.h (selector_fits_mode_p): Declare.
1571 * optabs-query.c: Include vec-perm-indices.h.
1572 (selector_fits_mode_p): New function.
1573 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1574 is defined, instead of checking whether the vec_perm_const_optab
1575 exists. Use targetm.vectorize.vec_perm_const instead of
1576 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1577 fit in the vector mode before using a variable permute.
1578 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1579 vec_perm_indices instead of an rtx.
1580 (expand_vec_perm): Replace with...
1581 (expand_vec_perm_const): ...this new function. Take the selector
1582 as a vec_perm_indices rather than an rtx. Also take the mode of
1583 the selector. Update call to shift_amt_for_vec_perm_mask.
1584 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1585 Use vec_perm_indices::new_expanded_vector to expand the original
1586 selector into bytes. Check whether the indices fit in the vector
1587 mode before using a variable permute.
1588 (expand_vec_perm_var): Make global.
1589 (expand_mult_highpart): Use expand_vec_perm_const.
1590 * fold-const.c: Includes vec-perm-indices.h.
1591 * tree-ssa-forwprop.c: Likewise.
1592 * tree-vect-data-refs.c: Likewise.
1593 * tree-vect-generic.c: Likewise.
1594 * tree-vect-loop.c: Likewise.
1595 * tree-vect-slp.c: Likewise.
1596 * tree-vect-stmts.c: Likewise.
1597 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1599 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1600 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1601 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1602 (aarch64_vectorize_vec_perm_const): ...this new function.
1603 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1604 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1605 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1606 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1607 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1608 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1609 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1611 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1612 check for NEON modes.
1613 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1614 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1615 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1616 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1618 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1619 the old VEC_PERM_CONST conditions.
1620 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1621 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1622 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1623 (ia64_vectorize_vec_perm_const_ok): Merge into...
1624 (ia64_vectorize_vec_perm_const): ...this new function.
1625 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1626 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1627 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1628 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1629 * config/mips/mips.c (mips_expand_vec_perm_const)
1630 (mips_vectorize_vec_perm_const_ok): Merge into...
1631 (mips_vectorize_vec_perm_const): ...this new function.
1632 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1633 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1634 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1635 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1636 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1637 (rs6000_expand_vec_perm_const): Delete.
1638 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1640 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1641 (altivec_expand_vec_perm_const_le): Take each operand individually.
1642 Operate on constant selectors rather than rtxes.
1643 (altivec_expand_vec_perm_const): Likewise. Update call to
1644 altivec_expand_vec_perm_const_le.
1645 (rs6000_expand_vec_perm_const): Delete.
1646 (rs6000_vectorize_vec_perm_const_ok): Delete.
1647 (rs6000_vectorize_vec_perm_const): New function.
1648 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1649 an element count and rtx array.
1650 (rs6000_expand_extract_even): Update call accordingly.
1651 (rs6000_expand_interleave): Likewise.
1652 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1653 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1654 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1655 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1656 (rs6000_expand_vec_perm_const): Delete.
1657 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1658 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1659 (altivec_expand_vec_perm_const_le): Take each operand individually.
1660 Operate on constant selectors rather than rtxes.
1661 (altivec_expand_vec_perm_const): Likewise. Update call to
1662 altivec_expand_vec_perm_const_le.
1663 (rs6000_expand_vec_perm_const): Delete.
1664 (rs6000_vectorize_vec_perm_const_ok): Delete.
1665 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1666 reference to the SPE evmerge intructions.
1667 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1668 an element count and rtx array.
1669 (rs6000_expand_extract_even): Update call accordingly.
1670 (rs6000_expand_interleave): Likewise.
1671 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1672 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1674 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1676 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1678 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1679 vector mode and that that mode matches the mode of the data
1681 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1682 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1683 directly using expand_vec_perm_1 when forcing selectors into
1685 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1687 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1689 * optabs-query.h (can_vec_perm_p): Delete.
1690 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1691 * optabs-query.c (can_vec_perm_p): Split into...
1692 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1693 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1694 particular selector is valid.
1695 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1696 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1697 (vect_grouped_load_supported): Likewise.
1698 (vect_shift_permute_load_chain): Likewise.
1699 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1700 (vect_transform_slp_perm_load): Likewise.
1701 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1702 (vectorizable_bswap): Likewise.
1703 (vect_gen_perm_mask_checked): Likewise.
1704 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1705 implementations of variable permutation vectors into account
1706 when deciding which selector to use.
1707 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1708 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1709 with a false third argument.
1710 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1711 to test whether the constant selector is valid and can_vec_perm_var_p
1712 to test whether a variable selector is valid.
1714 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1716 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1717 * optabs-query.c (can_vec_perm_p): Likewise.
1718 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1719 instead of vec_perm_indices.
1720 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1721 (vect_gen_perm_mask_checked): Likewise,
1722 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1723 (vect_gen_perm_mask_checked): Likewise,
1725 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1727 * optabs-query.h (qimode_for_vec_perm): Declare.
1728 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1729 (qimode_for_vec_perm): ...this new function.
1730 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1732 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1734 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1735 does not have a conditional at the top.
1737 2018-01-02 Richard Biener <rguenther@suse.de>
1739 * ipa-inline.c (big_speedup_p): Fix expression.
1741 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1744 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1747 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1751 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1752 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1753 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1754 cond_taken_branch_cost 3->4.
1756 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1758 PR tree-optimization/83581
1759 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1760 TODO_cleanup_cfg if any changes have been made.
1763 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1764 convert_modes if target mode has the right side, but different mode
1768 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1769 last argument when extracting from CONCAT. If either from_real or
1770 from_imag is NULL, use expansion through memory. If result is not
1771 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1772 the parts directly to inner mode, if even that fails, use expansion
1776 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1777 check for bswap in mode rather than HImode and use that in expand_unop
1780 Copyright (C) 2018 Free Software Foundation, Inc.
1782 Copying and distribution of this file, with or without modification,
1783 are permitted in any medium without royalty provided the copyright
1784 notice and this notice are preserved.