1 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3 PR tree-optimization/83605
4 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
8 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
10 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
11 * config/epiphany/rtems.h: New file.
13 2018-01-04 Jakub Jelinek <jakub@redhat.com>
14 Uros Bizjak <ubizjak@gmail.com>
17 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
18 QIreg_operand instead of register_operand predicate.
19 * config/i386/i386.c (ix86_rop_should_change_byte_p,
20 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
21 comments instead of -fmitigate[-_]rop.
23 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
26 * cgraphunit.c (symbol_table::compile): Switch to text_section
27 before calling assembly_start debug hook.
28 * run-rtl-passes.c (run_rtl_passes): Likewise.
31 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
33 * tree-vrp.c (extract_range_from_binary_expr_1): Check
34 range_int_cst_p rather than !symbolic_range_p before calling
35 extract_range_from_multiplicative_op_1.
37 2017-01-04 Jeff Law <law@redhat.com>
39 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
40 redundant test in assertion.
42 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
44 * doc/rtl.texi: Document machine_mode wrapper classes.
46 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
48 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
51 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
53 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
54 the VEC_PERM_EXPR fold to fail.
56 2018-01-04 Jakub Jelinek <jakub@redhat.com>
59 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
62 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
65 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
68 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
71 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
72 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
74 2018-01-04 Jakub Jelinek <jakub@redhat.com>
77 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
78 is BLKmode and bitpos not zero or mode change is needed.
80 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
83 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
86 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
89 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
90 instead of MULT rtx. Update all corresponding splitters.
92 (*ssub<modesuffix>): Ditto.
94 (*cmp_sadd_di): Update split patterns.
95 (*cmp_sadd_si): Ditto.
96 (*cmp_sadd_sidi): Ditto.
97 (*cmp_ssub_di): Ditto.
98 (*cmp_ssub_si): Ditto.
99 (*cmp_ssub_sidi): Ditto.
100 * config/alpha/predicates.md (const23_operand): New predicate.
101 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
102 Look for ASHIFT, not MULT inner operand.
103 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
105 2018-01-04 Martin Liska <mliska@suse.cz>
107 PR gcov-profile/83669
108 * gcov.c (output_intermediate_file): Add version to intermediate
110 * doc/gcov.texi: Document new field 'version' in intermediate
111 file format. Fix location of '-k' option of gcov command.
113 2018-01-04 Martin Liska <mliska@suse.cz>
116 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
118 2018-01-04 Jakub Jelinek <jakub@redhat.com>
120 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
122 2018-01-03 Martin Sebor <msebor@redhat.com>
124 PR tree-optimization/83655
125 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
126 checking calls with invalid arguments.
128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
130 * tree-vect-stmts.c (vect_get_store_rhs): New function.
131 (vectorizable_mask_load_store): Delete.
132 (vectorizable_call): Return false for masked loads and stores.
133 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
134 instead of gimple_assign_rhs1.
135 (vectorizable_load): Handle IFN_MASK_LOAD.
136 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
140 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
142 (vectorizable_mask_load_store): ...here.
143 (vectorizable_load): ...and here.
145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
147 * tree-vect-stmts.c (vect_build_all_ones_mask)
148 (vect_build_zero_merge_argument): New functions, split out from...
149 (vectorizable_load): ...here.
151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
153 * tree-vect-stmts.c (vect_check_store_rhs): New function,
155 (vectorizable_mask_load_store): ...here.
156 (vectorizable_store): ...and here.
158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
160 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
162 (vectorizable_mask_load_store): ...here.
164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
166 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
167 (vect_model_store_cost): Take a vec_load_store_type instead of a
169 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
170 (vect_model_store_cost): Take a vec_load_store_type instead of a
172 (vectorizable_mask_load_store): Update accordingly.
173 (vectorizable_store): Likewise.
174 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
178 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
179 IFN_MASK_LOAD calls here rather than...
180 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
182 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
183 Alan Hayward <alan.hayward@arm.com>
184 David Sherwood <david.sherwood@arm.com>
186 * expmed.c (extract_bit_field_1): For vector extracts,
187 fall back to extract_bit_field_as_subreg if vec_extract
190 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
191 Alan Hayward <alan.hayward@arm.com>
192 David Sherwood <david.sherwood@arm.com>
194 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
195 they are variable or constant sized.
196 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
197 slots for constant-sized data.
199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
200 Alan Hayward <alan.hayward@arm.com>
201 David Sherwood <david.sherwood@arm.com>
203 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
204 handling COND_EXPRs with boolean comparisons, try to find a better
205 basis for the mask type than the boolean itself.
207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
209 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
210 is calculated and how it can be overridden.
211 * genmodes.c (max_bitsize_mode_any_mode): New variable.
212 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
214 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
218 Alan Hayward <alan.hayward@arm.com>
219 David Sherwood <david.sherwood@arm.com>
221 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
222 Remove the mode argument.
223 (aarch64_simd_valid_immediate): Remove the mode and inverse
225 * config/aarch64/iterators.md (bitsize): New iterator.
226 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
227 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
228 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
229 aarch64_simd_valid_immediate.
230 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
231 (aarch64_reg_or_bic_imm): Likewise.
232 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
233 with an insn_type enum and msl with a modifier_type enum.
234 Replace element_width with a scalar_mode. Change the shift
235 to unsigned int. Add constructors for scalar_float_mode and
236 scalar_int_mode elements.
237 (aarch64_vect_float_const_representable_p): Delete.
238 (aarch64_can_const_movi_rtx_p)
239 (aarch64_simd_scalar_immediate_valid_for_move)
240 (aarch64_simd_make_constant): Update call to
241 aarch64_simd_valid_immediate.
242 (aarch64_advsimd_valid_immediate_hs): New function.
243 (aarch64_advsimd_valid_immediate): Likewise.
244 (aarch64_simd_valid_immediate): Remove mode and inverse
245 arguments. Rewrite to use the above. Use const_vec_duplicate_p
246 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
247 and aarch64_float_const_representable_p on the result.
248 (aarch64_output_simd_mov_immediate): Remove mode argument.
249 Update call to aarch64_simd_valid_immediate and use of
251 (aarch64_output_scalar_simd_mov_immediate): Update call
254 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
255 Alan Hayward <alan.hayward@arm.com>
256 David Sherwood <david.sherwood@arm.com>
258 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
259 (mode_nunits): Likewise CONST_MODE_NUNITS.
260 * machmode.def (ADJUST_NUNITS): Document.
261 * genmodes.c (mode_data::need_nunits_adj): New field.
262 (blank_mode): Update accordingly.
263 (adj_nunits): New variable.
264 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
266 (emit_mode_size_inline): Set need_bytesize_adj for all modes
267 listed in adj_nunits.
268 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
269 listed in adj_nunits. Don't emit case statements for such modes.
270 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
271 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
272 nothing if adj_nunits is nonnull.
273 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
274 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
275 (emit_mode_fbit): Update use of print_maybe_const_decl.
276 (emit_move_size): Likewise. Treat the array as non-const
278 (emit_mode_adjustments): Handle adj_nunits.
280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
282 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
283 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
284 (VECTOR_MODES): Use it.
285 (make_vector_modes): Take the prefix as an argument.
287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
288 Alan Hayward <alan.hayward@arm.com>
289 David Sherwood <david.sherwood@arm.com>
291 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
292 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
293 for MODE_VECTOR_BOOL.
294 * machmode.def (VECTOR_BOOL_MODE): Document.
295 * genmodes.c (VECTOR_BOOL_MODE): New macro.
296 (make_vector_bool_mode): New function.
297 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
299 * lto-streamer-in.c (lto_input_mode_table): Likewise.
300 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
302 * stor-layout.c (int_mode_for_mode): Likewise.
303 * tree.c (build_vector_type_for_mode): Likewise.
304 * varasm.c (output_constant_pool_2): Likewise.
305 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
306 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
307 for MODE_VECTOR_BOOL.
308 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
309 of mode class checks.
310 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
311 instead of a list of mode class checks.
312 (expand_vector_scalar_condition): Likewise.
313 (type_for_widest_vector_mode): Handle BImode as an inner mode.
315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
316 Alan Hayward <alan.hayward@arm.com>
317 David Sherwood <david.sherwood@arm.com>
319 * machmode.h (mode_size): Change from unsigned short to
321 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
322 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
323 or if measurement_type is not polynomial.
324 (fixed_size_mode::includes_p): Check for constant-sized modes.
325 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
326 return a poly_uint16 rather than an unsigned short.
327 (emit_mode_size): Change the type of mode_size from unsigned short
328 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
329 (emit_mode_adjustments): Cope with polynomial vector sizes.
330 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
332 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
334 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
335 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
336 * caller-save.c (setup_save_areas): Likewise.
337 (replace_reg_with_saved_mem): Likewise.
338 * calls.c (emit_library_call_value_1): Likewise.
339 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
340 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
341 (gen_lowpart_for_combine): Likewise.
342 * convert.c (convert_to_integer_1): Likewise.
343 * cse.c (equiv_constant, cse_insn): Likewise.
344 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
345 (cselib_subst_to_values): Likewise.
346 * dce.c (word_dce_process_block): Likewise.
347 * df-problems.c (df_word_lr_mark_ref): Likewise.
348 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
349 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
350 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
351 (rtl_for_decl_location): Likewise.
352 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
353 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
354 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
355 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
356 (expand_expr_real_1): Likewise.
357 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
358 (pad_below): Likewise.
359 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
360 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
361 * ira.c (get_subreg_tracking_sizes): Likewise.
362 * ira-build.c (ira_create_allocno_objects): Likewise.
363 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
364 (ira_sort_regnos_for_alter_reg): Likewise.
365 * ira-costs.c (record_operand_costs): Likewise.
366 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
367 (resolve_simple_move): Likewise.
368 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
369 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
370 (lra_constraints): Likewise.
371 (CONST_POOL_OK_P): Reject variable-sized modes.
372 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
373 (add_pseudo_to_slot, lra_spill): Likewise.
374 * omp-low.c (omp_clause_aligned_alignment): Likewise.
375 * optabs-query.c (get_best_extraction_insn): Likewise.
376 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
377 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
378 (expand_mult_highpart, valid_multiword_target_p): Likewise.
379 * recog.c (offsettable_address_addr_space_p): Likewise.
380 * regcprop.c (maybe_mode_change): Likewise.
381 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
382 * regrename.c (build_def_use): Likewise.
383 * regstat.c (dump_reg_info): Likewise.
384 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
385 (find_reloads, find_reloads_subreg_address): Likewise.
386 * reload1.c (eliminate_regs_1): Likewise.
387 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
388 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
389 (simplify_binary_operation_1, simplify_subreg): Likewise.
390 * targhooks.c (default_function_arg_padding): Likewise.
391 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
392 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
393 (verify_gimple_assign_ternary): Likewise.
394 * tree-inline.c (estimate_move_cost): Likewise.
395 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
396 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
397 (get_address_cost_ainc): Likewise.
398 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
399 (vect_supportable_dr_alignment): Likewise.
400 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
401 (vectorizable_reduction): Likewise.
402 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
403 (vectorizable_operation, vectorizable_load): Likewise.
404 * tree.c (build_same_sized_truth_vector_type): Likewise.
405 * valtrack.c (cleanup_auto_inc_dec): Likewise.
406 * var-tracking.c (emit_note_insn_var_location): Likewise.
407 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
408 (ADDR_VEC_ALIGN): Likewise.
410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
411 Alan Hayward <alan.hayward@arm.com>
412 David Sherwood <david.sherwood@arm.com>
414 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
416 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
417 or if measurement_type is polynomial.
418 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
419 * combine.c (make_extraction): Likewise.
420 * dse.c (find_shift_sequence): Likewise.
421 * dwarf2out.c (mem_loc_descriptor): Likewise.
422 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
423 (extract_bit_field, extract_low_bits): Likewise.
424 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
425 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
426 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
427 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
428 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
429 * reload.c (find_reloads): Likewise.
430 * reload1.c (alter_reg): Likewise.
431 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
432 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
433 * tree-if-conv.c (predicate_mem_writes): Likewise.
434 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
435 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
436 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
437 * valtrack.c (dead_debug_insert_temp): Likewise.
438 * varasm.c (mergeable_constant_section): Likewise.
439 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
442 Alan Hayward <alan.hayward@arm.com>
443 David Sherwood <david.sherwood@arm.com>
445 * expr.c (expand_assignment): Cope with polynomial mode sizes
446 when assigning to a CONCAT.
448 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
449 Alan Hayward <alan.hayward@arm.com>
450 David Sherwood <david.sherwood@arm.com>
452 * machmode.h (mode_precision): Change from unsigned short to
454 (mode_to_precision): Return a poly_uint16 rather than an unsigned
456 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
457 or if measurement_type is not polynomial.
458 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
459 in which the mode is already known to be a scalar_int_mode.
460 * genmodes.c (emit_mode_precision): Change the type of mode_precision
461 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
463 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
464 for GET_MODE_PRECISION.
465 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
466 for GET_MODE_PRECISION.
467 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
469 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
470 (expand_field_assignment, make_extraction): Likewise.
471 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
472 (get_last_value): Likewise.
473 * convert.c (convert_to_integer_1): Likewise.
474 * cse.c (cse_insn): Likewise.
475 * expr.c (expand_expr_real_1): Likewise.
476 * lra-constraints.c (simplify_operand_subreg): Likewise.
477 * optabs-query.c (can_atomic_load_p): Likewise.
478 * optabs.c (expand_atomic_load): Likewise.
479 (expand_atomic_store): Likewise.
480 * ree.c (combine_reaching_defs): Likewise.
481 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
482 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
483 * tree.h (type_has_mode_precision_p): Likewise.
484 * ubsan.c (instrument_si_overflow): Likewise.
486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
487 Alan Hayward <alan.hayward@arm.com>
488 David Sherwood <david.sherwood@arm.com>
490 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
491 polynomial numbers of units.
492 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
493 (valid_vector_subparts_p): New function.
494 (build_vector_type): Remove temporary shim and take the number
495 of units as a poly_uint64 rather than an int.
496 (build_opaque_vector_type): Take the number of units as a
497 poly_uint64 rather than an int.
498 * tree.c (build_vector_from_ctor): Handle polynomial
499 TYPE_VECTOR_SUBPARTS.
500 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
501 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
502 (build_vector_from_val): If the number of units is variable,
503 use build_vec_duplicate_cst for constant operands and
504 VEC_DUPLICATE_EXPR otherwise.
505 (make_vector_type): Remove temporary is_constant ().
506 (build_vector_type, build_opaque_vector_type): Take the number of
507 units as a poly_uint64 rather than an int.
508 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
510 * cfgexpand.c (expand_debug_expr): Likewise.
511 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
512 (store_constructor, expand_expr_real_1): Likewise.
513 (const_scalar_mask_from_tree): Likewise.
514 * fold-const-call.c (fold_const_reduction): Likewise.
515 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
516 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
517 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
518 (fold_relational_const): Likewise.
519 (native_interpret_vector): Likewise. Change the size from an
520 int to an unsigned int.
521 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
522 TYPE_VECTOR_SUBPARTS.
523 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
524 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
525 duplicating a non-constant operand into a variable-length vector.
526 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
527 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
528 * ipa-icf.c (sem_variable::equals): Likewise.
529 * match.pd: Likewise.
530 * omp-simd-clone.c (simd_clone_subparts): Likewise.
531 * print-tree.c (print_node): Likewise.
532 * stor-layout.c (layout_type): Likewise.
533 * targhooks.c (default_builtin_vectorization_cost): Likewise.
534 * tree-cfg.c (verify_gimple_comparison): Likewise.
535 (verify_gimple_assign_binary): Likewise.
536 (verify_gimple_assign_ternary): Likewise.
537 (verify_gimple_assign_single): Likewise.
538 * tree-pretty-print.c (dump_generic_node): Likewise.
539 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
540 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
541 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
542 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
543 (vect_shift_permute_load_chain): Likewise.
544 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
545 (expand_vector_condition, optimize_vector_constructor): Likewise.
546 (lower_vec_perm, get_compute_type): Likewise.
547 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
548 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
549 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
550 (vect_recog_mask_conversion_pattern): Likewise.
551 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
552 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
553 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
554 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
555 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
556 (vectorizable_shift, vectorizable_operation, vectorizable_store)
557 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
558 (supportable_widening_operation): Likewise.
559 (supportable_narrowing_operation): Likewise.
560 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
562 * varasm.c (output_constant): Likewise.
564 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
565 Alan Hayward <alan.hayward@arm.com>
566 David Sherwood <david.sherwood@arm.com>
568 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
569 so that both the length == 3 and length != 3 cases set up their
570 own permute vectors. Add comments explaining why we know the
571 number of elements is constant.
572 (vect_permute_load_chain): Likewise.
574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
575 Alan Hayward <alan.hayward@arm.com>
576 David Sherwood <david.sherwood@arm.com>
578 * machmode.h (mode_nunits): Change from unsigned char to
580 (ONLY_FIXED_SIZE_MODES): New macro.
581 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
582 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
583 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
585 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
586 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
587 or if measurement_type is not polynomial.
588 * genmodes.c (ZERO_COEFFS): New macro.
589 (emit_mode_nunits_inline): Make mode_nunits_inline return a
591 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
592 Use ZERO_COEFFS when emitting initializers.
593 * data-streamer.h (bp_pack_poly_value): New function.
594 (bp_unpack_poly_value): Likewise.
595 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
597 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
599 * tree.c (make_vector_type): Remove temporary shim and make
600 the real function take the number of units as a poly_uint64
602 (build_vector_type_for_mode): Handle polynomial nunits.
603 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
604 * emit-rtl.c (const_vec_series_p_1): Likewise.
605 (gen_rtx_CONST_VECTOR): Likewise.
606 * fold-const.c (test_vec_duplicate_folding): Likewise.
607 * genrecog.c (validate_pattern): Likewise.
608 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
609 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
610 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
611 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
612 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
613 * rtlanal.c (subreg_get_info): Likewise.
614 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
615 (vect_grouped_load_supported): Likewise.
616 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
617 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
618 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
619 (simplify_const_unary_operation, simplify_binary_operation_1)
620 (simplify_const_binary_operation, simplify_ternary_operation)
621 (test_vector_ops_duplicate, test_vector_ops): Likewise.
622 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
623 instead of CONST_VECTOR_NUNITS.
624 * varasm.c (output_constant_pool_2): Likewise.
625 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
626 explicit-encoded elements in the XVEC for variable-length vectors.
628 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
630 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
632 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
633 Alan Hayward <alan.hayward@arm.com>
634 David Sherwood <david.sherwood@arm.com>
636 * coretypes.h (fixed_size_mode): Declare.
637 (fixed_size_mode_pod): New typedef.
638 * builtins.h (target_builtins::x_apply_args_mode)
639 (target_builtins::x_apply_result_mode): Change type to
641 * builtins.c (apply_args_size, apply_result_size, result_vector)
642 (expand_builtin_apply_args_1, expand_builtin_apply)
643 (expand_builtin_return): Update accordingly.
645 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
647 * cse.c (hash_rtx_cb): Hash only the encoded elements.
648 * cselib.c (cselib_hash_rtx): Likewise.
649 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
650 CONST_VECTOR encoding.
652 2017-01-03 Jakub Jelinek <jakub@redhat.com>
653 Jeff Law <law@redhat.com>
656 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
657 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
658 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
659 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
662 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
663 explicitly probe *sp in a noreturn function if there were any callee
664 register saves or frame pointer is needed.
666 2018-01-03 Jakub Jelinek <jakub@redhat.com>
669 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
670 BLKmode for ternary, binary or unary expressions.
673 * var-tracking.c (delete_vta_debug_insn): New inline function.
674 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
675 insns from get_insns () to NULL instead of each bb separately.
676 Use delete_vta_debug_insn. No longer static.
677 (vt_debug_insns_local, variable_tracking_main_1): Adjust
678 delete_vta_debug_insns callers.
679 * rtl.h (delete_vta_debug_insns): Declare.
680 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
681 instead of variable_tracking_main.
683 2018-01-03 Martin Sebor <msebor@redhat.com>
685 PR tree-optimization/83603
686 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
687 arguments past the endof the argument list in functions declared
689 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
690 Avoid checking when arguments are null.
692 2018-01-03 Martin Sebor <msebor@redhat.com>
695 * doc/extend.texi (attribute const): Fix a typo.
696 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
697 issuing -Wsuggest-attribute for void functions.
699 2018-01-03 Martin Sebor <msebor@redhat.com>
701 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
702 offset_int::from instead of wide_int::to_shwi.
703 (maybe_diag_overlap): Remove assertion.
704 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
705 * gimple-ssa-sprintf.c (format_directive): Same.
706 (parse_directive): Same.
707 (sprintf_dom_walker::compute_format_length): Same.
708 (try_substitute_return_value): Same.
710 2017-01-03 Jeff Law <law@redhat.com>
713 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
714 non-constant residual for zero at runtime and avoid probing in
715 that case. Reorganize code for trailing problem to mirror handling
718 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
720 PR tree-optimization/83501
721 * tree-ssa-strlen.c (get_string_cst): New.
722 (handle_char_store): Call get_string_cst.
724 2018-01-03 Martin Liska <mliska@suse.cz>
726 PR tree-optimization/83593
727 * tree-ssa-strlen.c: Include tree-cfg.h.
728 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
729 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
730 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
732 (strlen_dom_walker::before_dom_children): Call
733 gimple_purge_dead_eh_edges. Dump tranformation with details
735 (strlen_dom_walker::before_dom_children): Update call by adding
736 new argument cleanup_eh.
737 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
739 2018-01-03 Martin Liska <mliska@suse.cz>
742 * cif-code.def (VARIADIC_THUNK): New enum value.
743 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
746 2018-01-03 Jan Beulich <jbeulich@suse.com>
748 * sse.md (mov<mode>_internal): Tighten condition for when to use
749 vmovdqu<ssescalarsize> for TI and OI modes.
751 2018-01-03 Jakub Jelinek <jakub@redhat.com>
753 Update copyright years.
755 2018-01-03 Martin Liska <mliska@suse.cz>
758 * ipa-visibility.c (function_and_variable_visibility): Skip
759 functions with noipa attribure.
761 2018-01-03 Jakub Jelinek <jakub@redhat.com>
763 * gcc.c (process_command): Update copyright notice dates.
764 * gcov-dump.c (print_version): Ditto.
765 * gcov.c (print_version): Ditto.
766 * gcov-tool.c (print_version): Ditto.
767 * gengtype.c (create_file): Ditto.
768 * doc/cpp.texi: Bump @copying's copyright year.
769 * doc/cppinternals.texi: Ditto.
770 * doc/gcc.texi: Ditto.
771 * doc/gccint.texi: Ditto.
772 * doc/gcov.texi: Ditto.
773 * doc/install.texi: Ditto.
774 * doc/invoke.texi: Ditto.
776 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
778 * vector-builder.h (vector_builder::m_full_nelts): Change from
779 unsigned int to poly_uint64.
780 (vector_builder::full_nelts): Update prototype accordingly.
781 (vector_builder::new_vector): Likewise.
782 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
783 (vector_builder::operator ==): Likewise.
784 (vector_builder::finalize): Likewise.
785 * int-vector-builder.h (int_vector_builder::int_vector_builder):
786 Take the number of elements as a poly_uint64 rather than an
788 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
789 from unsigned int to poly_uint64.
790 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
791 (vec_perm_indices::new_vector): Likewise.
792 (vec_perm_indices::length): Likewise.
793 (vec_perm_indices::nelts_per_input): Likewise.
794 (vec_perm_indices::input_nelts): Likewise.
795 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
796 number of elements per input as a poly_uint64 rather than an
797 unsigned int. Use the original encoding for variable-length
798 vectors, rather than clamping each individual element.
799 For the second and subsequent elements in each pattern,
800 clamp the step and base before clamping their sum.
801 (vec_perm_indices::series_p): Handle polynomial element counts.
802 (vec_perm_indices::all_in_range_p): Likewise.
803 (vec_perm_indices_to_tree): Likewise.
804 (vec_perm_indices_to_rtx): Likewise.
805 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
806 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
807 (tree_vector_builder::new_binary_operation): Handle polynomial
808 element counts. Return false if we need to know the number
809 of elements at compile time.
810 * fold-const.c (fold_vec_perm): Punt if the number of elements
811 isn't known at compile time.
813 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
815 * vec-perm-indices.h (vec_perm_builder): Change element type
816 from HOST_WIDE_INT to poly_int64.
817 (vec_perm_indices::element_type): Update accordingly.
818 (vec_perm_indices::clamp): Handle polynomial element_types.
819 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
820 (vec_perm_indices::all_in_range_p): Likewise.
821 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
823 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
824 polynomial vec_perm_indices element types.
825 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
826 * fold-const.c (fold_vec_perm): Likewise.
827 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
828 * tree-vect-generic.c (lower_vec_perm): Likewise.
829 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
830 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
831 element type to HOST_WIDE_INT.
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 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
838 rather than an int. Use plus_constant.
839 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
840 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
843 Alan Hayward <alan.hayward@arm.com>
844 David Sherwood <david.sherwood@arm.com>
846 * calls.c (emit_call_1, expand_call): Change struct_value_size from
847 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
854 mode sizes. Require a constant size for BLKmode parameters
855 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
856 forces a parameter to be padded at the lsb end in order to
857 fill a complete number of words, require the parameter size
858 to be ordered wrt UNITS_PER_WORD.
860 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
861 Alan Hayward <alan.hayward@arm.com>
862 David Sherwood <david.sherwood@arm.com>
864 * reload1.c (spill_stack_slot_width): Change element type
865 from unsigned int to poly_uint64_pod.
866 (alter_reg): Treat mode sizes as polynomial.
868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
869 Alan Hayward <alan.hayward@arm.com>
870 David Sherwood <david.sherwood@arm.com>
872 * reload.c (complex_word_subreg_p): New function.
873 (reload_inner_reg_of_subreg, push_reload): Use it.
875 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
876 Alan Hayward <alan.hayward@arm.com>
877 David Sherwood <david.sherwood@arm.com>
879 * lra-constraints.c (process_alt_operands): Reject matched
880 operands whose sizes aren't ordered.
881 (match_reload): Refer to this check here.
883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
884 Alan Hayward <alan.hayward@arm.com>
885 David Sherwood <david.sherwood@arm.com>
887 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
888 that the mode size is in the set {1, 2, 4, 8, 16}.
890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
891 Alan Hayward <alan.hayward@arm.com>
892 David Sherwood <david.sherwood@arm.com>
894 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
895 Use plus_constant instead of gen_rtx_PLUS.
897 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
898 Alan Hayward <alan.hayward@arm.com>
899 David Sherwood <david.sherwood@arm.com>
901 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
902 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
903 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
904 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
905 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
906 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
907 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
908 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
909 * config/i386/i386.c (ix86_push_rounding): ...this new function.
910 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
912 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
913 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
914 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
915 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
916 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
917 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
918 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
919 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
920 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
921 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
923 * expr.c (emit_move_resolve_push): Treat the input and result
924 of PUSH_ROUNDING as a poly_int64.
925 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
926 (emit_push_insn): Likewise.
927 * lra-eliminations.c (mark_not_eliminable): Likewise.
928 * recog.c (push_operand): Likewise.
929 * reload1.c (elimination_effects): Likewise.
930 * rtlanal.c (nonzero_bits1): Likewise.
931 * calls.c (store_one_arg): Likewise. Require the padding to be
932 known at compile time.
934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
935 Alan Hayward <alan.hayward@arm.com>
936 David Sherwood <david.sherwood@arm.com>
938 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
939 Use plus_constant instead of gen_rtx_PLUS.
941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
942 Alan Hayward <alan.hayward@arm.com>
943 David Sherwood <david.sherwood@arm.com>
945 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
953 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
954 via stack temporaries. Treat the mode size as polynomial too.
956 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
957 Alan Hayward <alan.hayward@arm.com>
958 David Sherwood <david.sherwood@arm.com>
960 * expr.c (expand_expr_real_2): When handling conversions involving
961 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
962 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
963 as a poly_uint64 too.
965 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
966 Alan Hayward <alan.hayward@arm.com>
967 David Sherwood <david.sherwood@arm.com>
969 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
972 Alan Hayward <alan.hayward@arm.com>
973 David Sherwood <david.sherwood@arm.com>
975 * combine.c (can_change_dest_mode): Handle polynomial
976 REGMODE_NATURAL_SIZE.
977 * expmed.c (store_bit_field_1): Likewise.
978 * expr.c (store_constructor): Likewise.
979 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
980 and polynomial REGMODE_NATURAL_SIZE.
981 (gen_lowpart_common): Likewise.
982 * reginfo.c (record_subregs_of_mode): Likewise.
983 * rtlanal.c (read_modify_subreg_p): Likewise.
985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
986 Alan Hayward <alan.hayward@arm.com>
987 David Sherwood <david.sherwood@arm.com>
989 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
993 Alan Hayward <alan.hayward@arm.com>
994 David Sherwood <david.sherwood@arm.com>
996 * match.pd: Cope with polynomial numbers of vector elements.
998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
999 Alan Hayward <alan.hayward@arm.com>
1000 David Sherwood <david.sherwood@arm.com>
1002 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1003 in a POINTER_PLUS_EXPR.
1005 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1006 Alan Hayward <alan.hayward@arm.com>
1007 David Sherwood <david.sherwood@arm.com>
1009 * omp-simd-clone.c (simd_clone_subparts): New function.
1010 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1011 (ipa_simd_modify_function_body): Likewise.
1013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1014 Alan Hayward <alan.hayward@arm.com>
1015 David Sherwood <david.sherwood@arm.com>
1017 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1018 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1019 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1020 (expand_vector_condition, vector_element): Likewise.
1021 (subparts_gt): New function.
1022 (get_compute_type): Use subparts_gt.
1023 (count_type_subparts): Delete.
1024 (expand_vector_operations_1): Use subparts_gt instead of
1025 count_type_subparts.
1027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1028 Alan Hayward <alan.hayward@arm.com>
1029 David Sherwood <david.sherwood@arm.com>
1031 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1032 (vect_compile_time_alias): ...this new function. Do the calculation
1033 on poly_ints rather than trees.
1034 (vect_prune_runtime_alias_test_list): Update call accordingly.
1036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1037 Alan Hayward <alan.hayward@arm.com>
1038 David Sherwood <david.sherwood@arm.com>
1040 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1042 (vect_schedule_slp_instance): Likewise.
1044 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1045 Alan Hayward <alan.hayward@arm.com>
1046 David Sherwood <david.sherwood@arm.com>
1048 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1049 constant and extern definitions for variable-length vectors.
1050 (vect_get_constant_vectors): Note that the number of units
1051 is known to be constant.
1053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1054 Alan Hayward <alan.hayward@arm.com>
1055 David Sherwood <david.sherwood@arm.com>
1057 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1058 of units as polynomial. Choose between WIDE and NARROW based
1061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1062 Alan Hayward <alan.hayward@arm.com>
1063 David Sherwood <david.sherwood@arm.com>
1065 * tree-vect-stmts.c (simd_clone_subparts): New function.
1066 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1069 Alan Hayward <alan.hayward@arm.com>
1070 David Sherwood <david.sherwood@arm.com>
1072 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1073 vectors as polynomial. Use build_index_vector for
1076 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1077 Alan Hayward <alan.hayward@arm.com>
1078 David Sherwood <david.sherwood@arm.com>
1080 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1081 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1082 for variable-length vectors.
1083 (vectorizable_mask_load_store): Treat the number of units as
1084 polynomial, asserting that it is constant if the condition has
1085 already been enforced.
1086 (vectorizable_store, vectorizable_load): Likewise.
1088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1089 Alan Hayward <alan.hayward@arm.com>
1090 David Sherwood <david.sherwood@arm.com>
1092 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1093 of units as polynomial. Punt if we can't tell at compile time
1094 which vector contains the final result.
1096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1097 Alan Hayward <alan.hayward@arm.com>
1098 David Sherwood <david.sherwood@arm.com>
1100 * tree-vect-loop.c (vectorizable_induction): Treat the number
1101 of units as polynomial. Punt on SLP inductions. Use an integer
1102 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1103 cast of such a series for variable-length floating-point
1106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1107 Alan Hayward <alan.hayward@arm.com>
1108 David Sherwood <david.sherwood@arm.com>
1110 * tree.h (build_index_vector): Declare.
1111 * tree.c (build_index_vector): New function.
1112 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1113 of units as polynomial, forcibly converting it to a constant if
1114 vectorizable_reduction has already enforced the condition.
1115 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1116 to create a {1,2,3,...} vector.
1117 (vectorizable_reduction): Treat the number of units as polynomial.
1118 Choose vectype_in based on the largest scalar element size rather
1119 than the smallest number of units. Enforce the restrictions
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-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1127 number of units as polynomial.
1129 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1130 Alan Hayward <alan.hayward@arm.com>
1131 David Sherwood <david.sherwood@arm.com>
1133 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1134 * target.def (autovectorize_vector_sizes): Return the vector sizes
1135 by pointer, using vector_sizes rather than a bitmask.
1136 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1137 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1138 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1140 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1141 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1142 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1143 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1144 * omp-general.c (omp_max_vf): Likewise.
1145 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1146 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1147 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1148 * tree-vect-slp.c (vect_slp_bb): Likewise.
1149 * doc/tm.texi: Regenerate.
1150 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1152 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1153 the vector size as a poly_uint64 rather than an unsigned int.
1154 (current_vector_size): Change from an unsigned int to a poly_uint64.
1155 (get_vectype_for_scalar_type): Update accordingly.
1156 * tree.h (build_truth_vector_type): Take the size and number of
1157 units as a poly_uint64 rather than an unsigned int.
1158 (build_vector_type): Add a temporary overload that takes
1159 the number of units as a poly_uint64 rather than an unsigned int.
1160 * tree.c (make_vector_type): Likewise.
1161 (build_truth_vector_type): Take the number of units as a poly_uint64
1162 rather than an unsigned int.
1164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1165 Alan Hayward <alan.hayward@arm.com>
1166 David Sherwood <david.sherwood@arm.com>
1168 * target.def (get_mask_mode): Take the number of units and length
1169 as poly_uint64s rather than unsigned ints.
1170 * targhooks.h (default_get_mask_mode): Update accordingly.
1171 * targhooks.c (default_get_mask_mode): Likewise.
1172 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1173 * doc/tm.texi: Regenerate.
1175 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1176 Alan Hayward <alan.hayward@arm.com>
1177 David Sherwood <david.sherwood@arm.com>
1179 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1180 * omp-general.c (omp_max_vf): Likewise.
1181 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1182 (expand_omp_simd): Handle polynomial safelen.
1183 * omp-low.c (omplow_simd_context): Add a default constructor.
1184 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1185 (lower_rec_simd_input_clauses): Update accordingly.
1186 (lower_rec_input_clauses): Likewise.
1188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1189 Alan Hayward <alan.hayward@arm.com>
1190 David Sherwood <david.sherwood@arm.com>
1192 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1193 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1194 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1195 (vect_analyze_slp_cost): Likewise.
1196 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1197 (vect_model_load_cost): Likewise.
1199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1200 Alan Hayward <alan.hayward@arm.com>
1201 David Sherwood <david.sherwood@arm.com>
1203 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1204 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1205 from an unsigned int * to a poly_uint64_pod *.
1206 (calculate_unrolling_factor): New function.
1207 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1210 Alan Hayward <alan.hayward@arm.com>
1211 David Sherwood <david.sherwood@arm.com>
1213 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1214 from an unsigned int to a poly_uint64.
1215 (_loop_vec_info::slp_unrolling_factor): Likewise.
1216 (_loop_vec_info::vectorization_factor): Change from an int
1218 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1219 (vect_get_num_vectors): New function.
1220 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1221 (vect_get_num_copies): Use vect_get_num_vectors.
1222 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1223 to an unsigned int *.
1224 (vect_analyze_data_refs): Change min_vf from an int * to a
1226 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1227 than an unsigned HOST_WIDE_INT.
1228 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1229 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1230 to an unsigned int *.
1231 (vect_analyze_data_ref_dependences): Likewise.
1232 (vect_compute_data_ref_alignment): Handle polynomial vf.
1233 (vect_enhance_data_refs_alignment): Likewise.
1234 (vect_prune_runtime_alias_test_list): Likewise.
1235 (vect_shift_permute_load_chain): Likewise.
1236 (vect_supportable_dr_alignment): Likewise.
1237 (dependence_distance_ge_vf): Take the vectorization factor as a
1238 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1239 (vect_analyze_data_refs): Change min_vf from an int * to a
1241 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1242 vfm1 as a poly_uint64 rather than an int. Make the same change
1243 for the returned bound_scalar.
1244 (vect_gen_vector_loop_niters): Handle polynomial vf.
1245 (vect_do_peeling): Likewise. Update call to
1246 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1247 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1249 * tree-vect-loop.c (vect_determine_vectorization_factor)
1250 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1251 (vect_get_known_peeling_cost): Likewise.
1252 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1253 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1254 (vect_transform_loop): Likewise. Use the lowest possible VF when
1255 updating the upper bounds of the loop.
1256 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1258 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1259 polynomial unroll factors.
1260 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1261 (vect_make_slp_decision): Likewise.
1262 (vect_supported_load_permutation_p): Likewise, and polynomial
1264 (vect_analyze_slp_cost): Handle polynomial vf.
1265 (vect_slp_analyze_node_operations): Likewise.
1266 (vect_slp_analyze_bb_1): Likewise.
1267 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1268 than an unsigned HOST_WIDE_INT.
1269 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1270 (vectorizable_load): Handle polynomial vf.
1271 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1273 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1275 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1276 Alan Hayward <alan.hayward@arm.com>
1277 David Sherwood <david.sherwood@arm.com>
1279 * match.pd: Handle bit operations involving three constants
1280 and try to fold one pair.
1282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1284 * tree-vect-loop-manip.c: Include gimple-fold.h.
1285 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1286 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1287 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1288 Add a path that uses a step of VF instead of 1, but disable it
1290 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1291 and niters_no_overflow parameters. Update calls to
1292 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1293 Create a new SSA name if the latter choses to use a ste other
1294 than zero, and return it via niters_vector_mult_vf_var.
1295 * tree-vect-loop.c (vect_transform_loop): Update calls to
1296 vect_do_peeling, vect_gen_vector_loop_niters and
1297 slpeel_make_loop_iterate_ntimes.
1298 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1299 (vect_gen_vector_loop_niters): Update declarations after above changes.
1301 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1303 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1304 128-bit round to integer instructions.
1305 (ceil<mode>2): Likewise.
1306 (btrunc<mode>2): Likewise.
1307 (round<mode>2): Likewise.
1309 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1311 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1312 unaligned VSX load/store on P8/P9.
1313 (expand_block_clear): Allow the use of unaligned VSX
1314 load/store on P8/P9.
1316 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1318 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1320 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1321 swap associated with both a load and a store.
1323 2018-01-02 Andrew Waterman <andrew@sifive.com>
1325 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1326 * config/riscv/riscv.md (clear_cache): Use it.
1328 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1330 * web.c: Remove out-of-date comment.
1332 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1334 * expr.c (fixup_args_size_notes): Check that any existing
1335 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1336 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1337 (emit_single_push_insn): ...here.
1339 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1341 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1342 (const_vector_encoded_nelts): New function.
1343 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1344 (const_vector_int_elt, const_vector_elt): Declare.
1345 * emit-rtl.c (const_vector_int_elt_1): New function.
1346 (const_vector_elt): Likewise.
1347 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1348 of CONST_VECTOR_ELT.
1350 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1352 * expr.c: Include rtx-vector-builder.h.
1353 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1354 directly on the tree encoding.
1355 (const_vector_from_tree): Likewise.
1356 * optabs.c: Include rtx-vector-builder.h.
1357 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1358 sequence of "u" values.
1359 * vec-perm-indices.c: Include rtx-vector-builder.h.
1360 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1361 directly on the vec_perm_indices encoding.
1363 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1365 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1366 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1367 * rtx-vector-builder.h: New file.
1368 * rtx-vector-builder.c: Likewise.
1369 * rtl.h (rtx_def::u2): Add a const_vector field.
1370 (CONST_VECTOR_NPATTERNS): New macro.
1371 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1372 (CONST_VECTOR_DUPLICATE_P): Likewise.
1373 (CONST_VECTOR_STEPPED_P): Likewise.
1374 (CONST_VECTOR_ENCODED_ELT): Likewise.
1375 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1376 (unwrap_const_vec_duplicate): Likewise.
1377 (const_vec_series_p): Check for a non-duplicated vector encoding.
1378 Say that the function only returns true for integer vectors.
1379 * emit-rtl.c: Include rtx-vector-builder.h.
1380 (gen_const_vec_duplicate_1): Delete.
1381 (gen_const_vector): Call gen_const_vec_duplicate instead of
1382 gen_const_vec_duplicate_1.
1383 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1384 (gen_const_vec_duplicate): Use rtx_vector_builder.
1385 (gen_const_vec_series): Likewise.
1386 (gen_rtx_CONST_VECTOR): Likewise.
1387 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1388 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1389 Build a new vector rather than modifying a CONST_VECTOR in-place.
1390 (handle_special_swappables): Update call accordingly.
1391 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1392 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1393 Build a new vector rather than modifying a CONST_VECTOR in-place.
1394 (handle_special_swappables): Update call accordingly.
1396 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1398 * simplify-rtx.c (simplify_const_binary_operation): Use
1399 CONST_VECTOR_ELT instead of XVECEXP.
1401 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1403 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1404 the selector elements to be different from the data elements
1405 if the selector is a VECTOR_CST.
1406 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1407 ssizetype for the selector.
1409 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1411 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1412 before testing each element individually.
1413 * tree-vect-generic.c (lower_vec_perm): Likewise.
1415 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1417 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1418 * selftest-run-tests.c (selftest::run_tests): Call it.
1419 * vector-builder.h (vector_builder::operator ==): New function.
1420 (vector_builder::operator !=): Likewise.
1421 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1422 (vec_perm_indices::all_from_input_p): New function.
1423 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1424 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1425 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1426 instead of reading the VECTOR_CST directly. Detect whether both
1427 vector inputs are the same before constructing the vec_perm_indices,
1428 and update the number of inputs argument accordingly. Use the
1429 utility functions added above. Only construct sel2 if we need to.
1431 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1433 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1434 the broadcast of the low byte.
1435 (expand_mult_highpart): Use an explicit encoding for the permutes.
1436 * optabs-query.c (can_mult_highpart_p): Likewise.
1437 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1438 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1439 (vectorizable_bswap): Likewise.
1440 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1441 explicit encoding for the power-of-2 permutes.
1442 (vect_permute_store_chain): Likewise.
1443 (vect_grouped_load_supported): Likewise.
1444 (vect_permute_load_chain): Likewise.
1446 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1448 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1449 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1450 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1451 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1452 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1453 (vect_gen_perm_mask_any): Likewise.
1455 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1457 * int-vector-builder.h: New file.
1458 * vec-perm-indices.h: Include int-vector-builder.h.
1459 (vec_perm_indices): Redefine as an int_vector_builder.
1460 (auto_vec_perm_indices): Delete.
1461 (vec_perm_builder): Redefine as a stand-alone class.
1462 (vec_perm_indices::vec_perm_indices): New function.
1463 (vec_perm_indices::clamp): Likewise.
1464 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1465 (vec_perm_indices::new_vector): New function.
1466 (vec_perm_indices::new_expanded_vector): Update for new
1467 vec_perm_indices class.
1468 (vec_perm_indices::rotate_inputs): New function.
1469 (vec_perm_indices::all_in_range_p): Operate directly on the
1470 encoded form, without computing elided elements.
1471 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1472 encoding. Update for new vec_perm_indices class.
1473 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1474 the given vec_perm_builder.
1475 (expand_vec_perm_var): Update vec_perm_builder constructor.
1476 (expand_mult_highpart): Use vec_perm_builder instead of
1477 auto_vec_perm_indices.
1478 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1479 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1480 or double series encoding as appropriate.
1481 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1482 vec_perm_indices instead of auto_vec_perm_indices.
1483 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1484 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1485 (vect_permute_store_chain): Likewise.
1486 (vect_grouped_load_supported): Likewise.
1487 (vect_permute_load_chain): Likewise.
1488 (vect_shift_permute_load_chain): Likewise.
1489 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1490 (vect_transform_slp_perm_load): Likewise.
1491 (vect_schedule_slp_instance): Likewise.
1492 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1493 (vectorizable_mask_load_store): Likewise.
1494 (vectorizable_bswap): Likewise.
1495 (vectorizable_store): Likewise.
1496 (vectorizable_load): Likewise.
1497 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1498 vec_perm_indices instead of auto_vec_perm_indices. Use
1499 tree_to_vec_perm_builder to read the vector from a tree.
1500 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1501 vec_perm_builder instead of a vec_perm_indices.
1502 (have_whole_vector_shift): Use vec_perm_builder and
1503 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1504 truncation to calc_vec_perm_mask_for_shift.
1505 (vect_create_epilog_for_reduction): Likewise.
1506 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1507 from auto_vec_perm_indices to vec_perm_indices.
1508 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1509 instead of changing individual elements.
1510 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1511 the vector in d.perm.
1512 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1513 from auto_vec_perm_indices to vec_perm_indices.
1514 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1515 instead of changing individual elements.
1516 (arm_vectorize_vec_perm_const): Use new_vector to install
1517 the vector in d.perm.
1518 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1519 Update vec_perm_builder constructor.
1520 (rs6000_expand_interleave): Likewise.
1521 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1522 (rs6000_expand_interleave): Likewise.
1524 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1526 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1527 to qimode could truncate the indices.
1528 * optabs.c (expand_vec_perm_var): Likewise.
1530 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1532 * Makefile.in (OBJS): Add vec-perm-indices.o.
1533 * vec-perm-indices.h: New file.
1534 * vec-perm-indices.c: Likewise.
1535 * target.h (vec_perm_indices): Replace with a forward class
1537 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1538 * optabs.h: Include vec-perm-indices.h.
1539 (expand_vec_perm): Delete.
1540 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1541 (expand_vec_perm_const): Declare.
1542 * target.def (vec_perm_const_ok): Replace with...
1543 (vec_perm_const): ...this new hook.
1544 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1545 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1546 * doc/tm.texi: Regenerate.
1547 * optabs.def (vec_perm_const): Delete.
1548 * doc/md.texi (vec_perm_const): Likewise.
1549 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1550 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1551 expand_vec_perm for constant permutation vectors. Assert that
1552 the mode of variable permutation vectors is the integer equivalent
1553 of the mode that is being permuted.
1554 * optabs-query.h (selector_fits_mode_p): Declare.
1555 * optabs-query.c: Include vec-perm-indices.h.
1556 (selector_fits_mode_p): New function.
1557 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1558 is defined, instead of checking whether the vec_perm_const_optab
1559 exists. Use targetm.vectorize.vec_perm_const instead of
1560 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1561 fit in the vector mode before using a variable permute.
1562 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1563 vec_perm_indices instead of an rtx.
1564 (expand_vec_perm): Replace with...
1565 (expand_vec_perm_const): ...this new function. Take the selector
1566 as a vec_perm_indices rather than an rtx. Also take the mode of
1567 the selector. Update call to shift_amt_for_vec_perm_mask.
1568 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1569 Use vec_perm_indices::new_expanded_vector to expand the original
1570 selector into bytes. Check whether the indices fit in the vector
1571 mode before using a variable permute.
1572 (expand_vec_perm_var): Make global.
1573 (expand_mult_highpart): Use expand_vec_perm_const.
1574 * fold-const.c: Includes vec-perm-indices.h.
1575 * tree-ssa-forwprop.c: Likewise.
1576 * tree-vect-data-refs.c: Likewise.
1577 * tree-vect-generic.c: Likewise.
1578 * tree-vect-loop.c: Likewise.
1579 * tree-vect-slp.c: Likewise.
1580 * tree-vect-stmts.c: Likewise.
1581 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1583 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1584 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1585 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1586 (aarch64_vectorize_vec_perm_const): ...this new function.
1587 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1588 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1589 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1590 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1591 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1592 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1593 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1595 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1596 check for NEON modes.
1597 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1598 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1599 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1600 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1602 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1603 the old VEC_PERM_CONST conditions.
1604 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1605 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1606 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1607 (ia64_vectorize_vec_perm_const_ok): Merge into...
1608 (ia64_vectorize_vec_perm_const): ...this new function.
1609 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1610 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1611 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1612 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1613 * config/mips/mips.c (mips_expand_vec_perm_const)
1614 (mips_vectorize_vec_perm_const_ok): Merge into...
1615 (mips_vectorize_vec_perm_const): ...this new function.
1616 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1617 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1618 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1619 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1620 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1621 (rs6000_expand_vec_perm_const): Delete.
1622 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1624 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1625 (altivec_expand_vec_perm_const_le): Take each operand individually.
1626 Operate on constant selectors rather than rtxes.
1627 (altivec_expand_vec_perm_const): Likewise. Update call to
1628 altivec_expand_vec_perm_const_le.
1629 (rs6000_expand_vec_perm_const): Delete.
1630 (rs6000_vectorize_vec_perm_const_ok): Delete.
1631 (rs6000_vectorize_vec_perm_const): New function.
1632 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1633 an element count and rtx array.
1634 (rs6000_expand_extract_even): Update call accordingly.
1635 (rs6000_expand_interleave): Likewise.
1636 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1637 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1638 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1639 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1640 (rs6000_expand_vec_perm_const): Delete.
1641 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1642 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1643 (altivec_expand_vec_perm_const_le): Take each operand individually.
1644 Operate on constant selectors rather than rtxes.
1645 (altivec_expand_vec_perm_const): Likewise. Update call to
1646 altivec_expand_vec_perm_const_le.
1647 (rs6000_expand_vec_perm_const): Delete.
1648 (rs6000_vectorize_vec_perm_const_ok): Delete.
1649 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1650 reference to the SPE evmerge intructions.
1651 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1652 an element count and rtx array.
1653 (rs6000_expand_extract_even): Update call accordingly.
1654 (rs6000_expand_interleave): Likewise.
1655 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1656 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1658 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1660 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1662 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1663 vector mode and that that mode matches the mode of the data
1665 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1666 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1667 directly using expand_vec_perm_1 when forcing selectors into
1669 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1671 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1673 * optabs-query.h (can_vec_perm_p): Delete.
1674 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1675 * optabs-query.c (can_vec_perm_p): Split into...
1676 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1677 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1678 particular selector is valid.
1679 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1680 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1681 (vect_grouped_load_supported): Likewise.
1682 (vect_shift_permute_load_chain): Likewise.
1683 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1684 (vect_transform_slp_perm_load): Likewise.
1685 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1686 (vectorizable_bswap): Likewise.
1687 (vect_gen_perm_mask_checked): Likewise.
1688 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1689 implementations of variable permutation vectors into account
1690 when deciding which selector to use.
1691 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1692 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1693 with a false third argument.
1694 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1695 to test whether the constant selector is valid and can_vec_perm_var_p
1696 to test whether a variable selector is valid.
1698 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1700 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1701 * optabs-query.c (can_vec_perm_p): Likewise.
1702 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1703 instead of vec_perm_indices.
1704 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1705 (vect_gen_perm_mask_checked): Likewise,
1706 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1707 (vect_gen_perm_mask_checked): Likewise,
1709 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1711 * optabs-query.h (qimode_for_vec_perm): Declare.
1712 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1713 (qimode_for_vec_perm): ...this new function.
1714 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1716 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1718 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1719 does not have a conditional at the top.
1721 2018-01-02 Richard Biener <rguenther@suse.de>
1723 * ipa-inline.c (big_speedup_p): Fix expression.
1725 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1728 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1731 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1735 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1736 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1737 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1738 cond_taken_branch_cost 3->4.
1740 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1742 PR tree-optimization/83581
1743 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1744 TODO_cleanup_cfg if any changes have been made.
1747 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1748 convert_modes if target mode has the right side, but different mode
1752 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1753 last argument when extracting from CONCAT. If either from_real or
1754 from_imag is NULL, use expansion through memory. If result is not
1755 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1756 the parts directly to inner mode, if even that fails, use expansion
1760 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1761 check for bswap in mode rather than HImode and use that in expand_unop
1764 Copyright (C) 2018 Free Software Foundation, Inc.
1766 Copying and distribution of this file, with or without modification,
1767 are permitted in any medium without royalty provided the copyright
1768 notice and this notice are preserved.