1 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4 (record_modified): Skip clobbers; add debug output.
5 (param_change_prob): Use sreal frequencies.
7 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
9 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
10 punt for user-aligned variables.
12 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
14 * tree-chrec.c (chrec_contains_symbols): Return true for
17 2018-01-05 Sudakshina Das <sudi.das@arm.com>
20 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
21 of (x|y) == x for BICS pattern.
23 2018-01-05 Jakub Jelinek <jakub@redhat.com>
25 PR tree-optimization/83605
26 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
27 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
30 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
32 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
33 * config/epiphany/rtems.h: New file.
35 2018-01-04 Jakub Jelinek <jakub@redhat.com>
36 Uros Bizjak <ubizjak@gmail.com>
39 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
40 QIreg_operand instead of register_operand predicate.
41 * config/i386/i386.c (ix86_rop_should_change_byte_p,
42 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
43 comments instead of -fmitigate[-_]rop.
45 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
48 * cgraphunit.c (symbol_table::compile): Switch to text_section
49 before calling assembly_start debug hook.
50 * run-rtl-passes.c (run_rtl_passes): Likewise.
53 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
55 * tree-vrp.c (extract_range_from_binary_expr_1): Check
56 range_int_cst_p rather than !symbolic_range_p before calling
57 extract_range_from_multiplicative_op_1.
59 2017-01-04 Jeff Law <law@redhat.com>
61 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
62 redundant test in assertion.
64 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
66 * doc/rtl.texi: Document machine_mode wrapper classes.
68 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
70 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
73 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
75 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
76 the VEC_PERM_EXPR fold to fail.
78 2018-01-04 Jakub Jelinek <jakub@redhat.com>
81 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
84 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
87 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
90 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
93 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
94 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
96 2018-01-04 Jakub Jelinek <jakub@redhat.com>
99 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
100 is BLKmode and bitpos not zero or mode change is needed.
102 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
105 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
108 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
111 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
112 instead of MULT rtx. Update all corresponding splitters.
114 (*ssub<modesuffix>): Ditto.
116 (*cmp_sadd_di): Update split patterns.
117 (*cmp_sadd_si): Ditto.
118 (*cmp_sadd_sidi): Ditto.
119 (*cmp_ssub_di): Ditto.
120 (*cmp_ssub_si): Ditto.
121 (*cmp_ssub_sidi): Ditto.
122 * config/alpha/predicates.md (const23_operand): New predicate.
123 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
124 Look for ASHIFT, not MULT inner operand.
125 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
127 2018-01-04 Martin Liska <mliska@suse.cz>
129 PR gcov-profile/83669
130 * gcov.c (output_intermediate_file): Add version to intermediate
132 * doc/gcov.texi: Document new field 'version' in intermediate
133 file format. Fix location of '-k' option of gcov command.
135 2018-01-04 Martin Liska <mliska@suse.cz>
138 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
140 2018-01-04 Jakub Jelinek <jakub@redhat.com>
142 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
144 2018-01-03 Martin Sebor <msebor@redhat.com>
146 PR tree-optimization/83655
147 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
148 checking calls with invalid arguments.
150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
152 * tree-vect-stmts.c (vect_get_store_rhs): New function.
153 (vectorizable_mask_load_store): Delete.
154 (vectorizable_call): Return false for masked loads and stores.
155 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
156 instead of gimple_assign_rhs1.
157 (vectorizable_load): Handle IFN_MASK_LOAD.
158 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
162 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
164 (vectorizable_mask_load_store): ...here.
165 (vectorizable_load): ...and here.
167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
169 * tree-vect-stmts.c (vect_build_all_ones_mask)
170 (vect_build_zero_merge_argument): New functions, split out from...
171 (vectorizable_load): ...here.
173 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
175 * tree-vect-stmts.c (vect_check_store_rhs): New function,
177 (vectorizable_mask_load_store): ...here.
178 (vectorizable_store): ...and here.
180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
182 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
184 (vectorizable_mask_load_store): ...here.
186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
188 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
189 (vect_model_store_cost): Take a vec_load_store_type instead of a
191 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
192 (vect_model_store_cost): Take a vec_load_store_type instead of a
194 (vectorizable_mask_load_store): Update accordingly.
195 (vectorizable_store): Likewise.
196 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
198 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
200 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
201 IFN_MASK_LOAD calls here rather than...
202 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
205 Alan Hayward <alan.hayward@arm.com>
206 David Sherwood <david.sherwood@arm.com>
208 * expmed.c (extract_bit_field_1): For vector extracts,
209 fall back to extract_bit_field_as_subreg if vec_extract
212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
213 Alan Hayward <alan.hayward@arm.com>
214 David Sherwood <david.sherwood@arm.com>
216 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
217 they are variable or constant sized.
218 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
219 slots for constant-sized data.
221 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
222 Alan Hayward <alan.hayward@arm.com>
223 David Sherwood <david.sherwood@arm.com>
225 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
226 handling COND_EXPRs with boolean comparisons, try to find a better
227 basis for the mask type than the boolean itself.
229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
231 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
232 is calculated and how it can be overridden.
233 * genmodes.c (max_bitsize_mode_any_mode): New variable.
234 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
236 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
239 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
240 Alan Hayward <alan.hayward@arm.com>
241 David Sherwood <david.sherwood@arm.com>
243 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
244 Remove the mode argument.
245 (aarch64_simd_valid_immediate): Remove the mode and inverse
247 * config/aarch64/iterators.md (bitsize): New iterator.
248 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
249 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
250 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
251 aarch64_simd_valid_immediate.
252 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
253 (aarch64_reg_or_bic_imm): Likewise.
254 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
255 with an insn_type enum and msl with a modifier_type enum.
256 Replace element_width with a scalar_mode. Change the shift
257 to unsigned int. Add constructors for scalar_float_mode and
258 scalar_int_mode elements.
259 (aarch64_vect_float_const_representable_p): Delete.
260 (aarch64_can_const_movi_rtx_p)
261 (aarch64_simd_scalar_immediate_valid_for_move)
262 (aarch64_simd_make_constant): Update call to
263 aarch64_simd_valid_immediate.
264 (aarch64_advsimd_valid_immediate_hs): New function.
265 (aarch64_advsimd_valid_immediate): Likewise.
266 (aarch64_simd_valid_immediate): Remove mode and inverse
267 arguments. Rewrite to use the above. Use const_vec_duplicate_p
268 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
269 and aarch64_float_const_representable_p on the result.
270 (aarch64_output_simd_mov_immediate): Remove mode argument.
271 Update call to aarch64_simd_valid_immediate and use of
273 (aarch64_output_scalar_simd_mov_immediate): Update call
276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
277 Alan Hayward <alan.hayward@arm.com>
278 David Sherwood <david.sherwood@arm.com>
280 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
281 (mode_nunits): Likewise CONST_MODE_NUNITS.
282 * machmode.def (ADJUST_NUNITS): Document.
283 * genmodes.c (mode_data::need_nunits_adj): New field.
284 (blank_mode): Update accordingly.
285 (adj_nunits): New variable.
286 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
288 (emit_mode_size_inline): Set need_bytesize_adj for all modes
289 listed in adj_nunits.
290 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
291 listed in adj_nunits. Don't emit case statements for such modes.
292 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
293 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
294 nothing if adj_nunits is nonnull.
295 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
296 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
297 (emit_mode_fbit): Update use of print_maybe_const_decl.
298 (emit_move_size): Likewise. Treat the array as non-const
300 (emit_mode_adjustments): Handle adj_nunits.
302 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
304 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
305 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
306 (VECTOR_MODES): Use it.
307 (make_vector_modes): Take the prefix as an argument.
309 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
310 Alan Hayward <alan.hayward@arm.com>
311 David Sherwood <david.sherwood@arm.com>
313 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
314 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
315 for MODE_VECTOR_BOOL.
316 * machmode.def (VECTOR_BOOL_MODE): Document.
317 * genmodes.c (VECTOR_BOOL_MODE): New macro.
318 (make_vector_bool_mode): New function.
319 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
321 * lto-streamer-in.c (lto_input_mode_table): Likewise.
322 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
324 * stor-layout.c (int_mode_for_mode): Likewise.
325 * tree.c (build_vector_type_for_mode): Likewise.
326 * varasm.c (output_constant_pool_2): Likewise.
327 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
328 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
329 for MODE_VECTOR_BOOL.
330 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
331 of mode class checks.
332 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
333 instead of a list of mode class checks.
334 (expand_vector_scalar_condition): Likewise.
335 (type_for_widest_vector_mode): Handle BImode as an inner mode.
337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
338 Alan Hayward <alan.hayward@arm.com>
339 David Sherwood <david.sherwood@arm.com>
341 * machmode.h (mode_size): Change from unsigned short to
343 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
344 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
345 or if measurement_type is not polynomial.
346 (fixed_size_mode::includes_p): Check for constant-sized modes.
347 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
348 return a poly_uint16 rather than an unsigned short.
349 (emit_mode_size): Change the type of mode_size from unsigned short
350 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
351 (emit_mode_adjustments): Cope with polynomial vector sizes.
352 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
354 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
356 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
357 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
358 * caller-save.c (setup_save_areas): Likewise.
359 (replace_reg_with_saved_mem): Likewise.
360 * calls.c (emit_library_call_value_1): Likewise.
361 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
362 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
363 (gen_lowpart_for_combine): Likewise.
364 * convert.c (convert_to_integer_1): Likewise.
365 * cse.c (equiv_constant, cse_insn): Likewise.
366 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
367 (cselib_subst_to_values): Likewise.
368 * dce.c (word_dce_process_block): Likewise.
369 * df-problems.c (df_word_lr_mark_ref): Likewise.
370 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
371 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
372 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
373 (rtl_for_decl_location): Likewise.
374 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
375 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
376 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
377 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
378 (expand_expr_real_1): Likewise.
379 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
380 (pad_below): Likewise.
381 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
382 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
383 * ira.c (get_subreg_tracking_sizes): Likewise.
384 * ira-build.c (ira_create_allocno_objects): Likewise.
385 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
386 (ira_sort_regnos_for_alter_reg): Likewise.
387 * ira-costs.c (record_operand_costs): Likewise.
388 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
389 (resolve_simple_move): Likewise.
390 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
391 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
392 (lra_constraints): Likewise.
393 (CONST_POOL_OK_P): Reject variable-sized modes.
394 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
395 (add_pseudo_to_slot, lra_spill): Likewise.
396 * omp-low.c (omp_clause_aligned_alignment): Likewise.
397 * optabs-query.c (get_best_extraction_insn): Likewise.
398 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
399 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
400 (expand_mult_highpart, valid_multiword_target_p): Likewise.
401 * recog.c (offsettable_address_addr_space_p): Likewise.
402 * regcprop.c (maybe_mode_change): Likewise.
403 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
404 * regrename.c (build_def_use): Likewise.
405 * regstat.c (dump_reg_info): Likewise.
406 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
407 (find_reloads, find_reloads_subreg_address): Likewise.
408 * reload1.c (eliminate_regs_1): Likewise.
409 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
410 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
411 (simplify_binary_operation_1, simplify_subreg): Likewise.
412 * targhooks.c (default_function_arg_padding): Likewise.
413 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
414 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
415 (verify_gimple_assign_ternary): Likewise.
416 * tree-inline.c (estimate_move_cost): Likewise.
417 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
418 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
419 (get_address_cost_ainc): Likewise.
420 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
421 (vect_supportable_dr_alignment): Likewise.
422 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
423 (vectorizable_reduction): Likewise.
424 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
425 (vectorizable_operation, vectorizable_load): Likewise.
426 * tree.c (build_same_sized_truth_vector_type): Likewise.
427 * valtrack.c (cleanup_auto_inc_dec): Likewise.
428 * var-tracking.c (emit_note_insn_var_location): Likewise.
429 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
430 (ADDR_VEC_ALIGN): Likewise.
432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
433 Alan Hayward <alan.hayward@arm.com>
434 David Sherwood <david.sherwood@arm.com>
436 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
438 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
439 or if measurement_type is polynomial.
440 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
441 * combine.c (make_extraction): Likewise.
442 * dse.c (find_shift_sequence): Likewise.
443 * dwarf2out.c (mem_loc_descriptor): Likewise.
444 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
445 (extract_bit_field, extract_low_bits): Likewise.
446 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
447 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
448 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
449 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
450 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
451 * reload.c (find_reloads): Likewise.
452 * reload1.c (alter_reg): Likewise.
453 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
454 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
455 * tree-if-conv.c (predicate_mem_writes): Likewise.
456 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
457 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
458 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
459 * valtrack.c (dead_debug_insert_temp): Likewise.
460 * varasm.c (mergeable_constant_section): Likewise.
461 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
464 Alan Hayward <alan.hayward@arm.com>
465 David Sherwood <david.sherwood@arm.com>
467 * expr.c (expand_assignment): Cope with polynomial mode sizes
468 when assigning to a CONCAT.
470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
471 Alan Hayward <alan.hayward@arm.com>
472 David Sherwood <david.sherwood@arm.com>
474 * machmode.h (mode_precision): Change from unsigned short to
476 (mode_to_precision): Return a poly_uint16 rather than an unsigned
478 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
479 or if measurement_type is not polynomial.
480 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
481 in which the mode is already known to be a scalar_int_mode.
482 * genmodes.c (emit_mode_precision): Change the type of mode_precision
483 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
485 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
486 for GET_MODE_PRECISION.
487 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
488 for GET_MODE_PRECISION.
489 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
491 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
492 (expand_field_assignment, make_extraction): Likewise.
493 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
494 (get_last_value): Likewise.
495 * convert.c (convert_to_integer_1): Likewise.
496 * cse.c (cse_insn): Likewise.
497 * expr.c (expand_expr_real_1): Likewise.
498 * lra-constraints.c (simplify_operand_subreg): Likewise.
499 * optabs-query.c (can_atomic_load_p): Likewise.
500 * optabs.c (expand_atomic_load): Likewise.
501 (expand_atomic_store): Likewise.
502 * ree.c (combine_reaching_defs): Likewise.
503 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
504 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
505 * tree.h (type_has_mode_precision_p): Likewise.
506 * ubsan.c (instrument_si_overflow): Likewise.
508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
509 Alan Hayward <alan.hayward@arm.com>
510 David Sherwood <david.sherwood@arm.com>
512 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
513 polynomial numbers of units.
514 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
515 (valid_vector_subparts_p): New function.
516 (build_vector_type): Remove temporary shim and take the number
517 of units as a poly_uint64 rather than an int.
518 (build_opaque_vector_type): Take the number of units as a
519 poly_uint64 rather than an int.
520 * tree.c (build_vector_from_ctor): Handle polynomial
521 TYPE_VECTOR_SUBPARTS.
522 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
523 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
524 (build_vector_from_val): If the number of units is variable,
525 use build_vec_duplicate_cst for constant operands and
526 VEC_DUPLICATE_EXPR otherwise.
527 (make_vector_type): Remove temporary is_constant ().
528 (build_vector_type, build_opaque_vector_type): Take the number of
529 units as a poly_uint64 rather than an int.
530 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
532 * cfgexpand.c (expand_debug_expr): Likewise.
533 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
534 (store_constructor, expand_expr_real_1): Likewise.
535 (const_scalar_mask_from_tree): Likewise.
536 * fold-const-call.c (fold_const_reduction): Likewise.
537 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
538 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
539 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
540 (fold_relational_const): Likewise.
541 (native_interpret_vector): Likewise. Change the size from an
542 int to an unsigned int.
543 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
544 TYPE_VECTOR_SUBPARTS.
545 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
546 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
547 duplicating a non-constant operand into a variable-length vector.
548 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
549 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
550 * ipa-icf.c (sem_variable::equals): Likewise.
551 * match.pd: Likewise.
552 * omp-simd-clone.c (simd_clone_subparts): Likewise.
553 * print-tree.c (print_node): Likewise.
554 * stor-layout.c (layout_type): Likewise.
555 * targhooks.c (default_builtin_vectorization_cost): Likewise.
556 * tree-cfg.c (verify_gimple_comparison): Likewise.
557 (verify_gimple_assign_binary): Likewise.
558 (verify_gimple_assign_ternary): Likewise.
559 (verify_gimple_assign_single): Likewise.
560 * tree-pretty-print.c (dump_generic_node): Likewise.
561 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
562 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
563 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
564 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
565 (vect_shift_permute_load_chain): Likewise.
566 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
567 (expand_vector_condition, optimize_vector_constructor): Likewise.
568 (lower_vec_perm, get_compute_type): Likewise.
569 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
570 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
571 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
572 (vect_recog_mask_conversion_pattern): Likewise.
573 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
574 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
575 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
576 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
577 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
578 (vectorizable_shift, vectorizable_operation, vectorizable_store)
579 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
580 (supportable_widening_operation): Likewise.
581 (supportable_narrowing_operation): Likewise.
582 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
584 * varasm.c (output_constant): Likewise.
586 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
587 Alan Hayward <alan.hayward@arm.com>
588 David Sherwood <david.sherwood@arm.com>
590 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
591 so that both the length == 3 and length != 3 cases set up their
592 own permute vectors. Add comments explaining why we know the
593 number of elements is constant.
594 (vect_permute_load_chain): Likewise.
596 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
597 Alan Hayward <alan.hayward@arm.com>
598 David Sherwood <david.sherwood@arm.com>
600 * machmode.h (mode_nunits): Change from unsigned char to
602 (ONLY_FIXED_SIZE_MODES): New macro.
603 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
604 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
605 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
607 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
608 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
609 or if measurement_type is not polynomial.
610 * genmodes.c (ZERO_COEFFS): New macro.
611 (emit_mode_nunits_inline): Make mode_nunits_inline return a
613 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
614 Use ZERO_COEFFS when emitting initializers.
615 * data-streamer.h (bp_pack_poly_value): New function.
616 (bp_unpack_poly_value): Likewise.
617 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
619 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
621 * tree.c (make_vector_type): Remove temporary shim and make
622 the real function take the number of units as a poly_uint64
624 (build_vector_type_for_mode): Handle polynomial nunits.
625 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
626 * emit-rtl.c (const_vec_series_p_1): Likewise.
627 (gen_rtx_CONST_VECTOR): Likewise.
628 * fold-const.c (test_vec_duplicate_folding): Likewise.
629 * genrecog.c (validate_pattern): Likewise.
630 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
631 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
632 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
633 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
634 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
635 * rtlanal.c (subreg_get_info): Likewise.
636 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
637 (vect_grouped_load_supported): Likewise.
638 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
639 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
640 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
641 (simplify_const_unary_operation, simplify_binary_operation_1)
642 (simplify_const_binary_operation, simplify_ternary_operation)
643 (test_vector_ops_duplicate, test_vector_ops): Likewise.
644 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
645 instead of CONST_VECTOR_NUNITS.
646 * varasm.c (output_constant_pool_2): Likewise.
647 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
648 explicit-encoded elements in the XVEC for variable-length vectors.
650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
652 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
654 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
655 Alan Hayward <alan.hayward@arm.com>
656 David Sherwood <david.sherwood@arm.com>
658 * coretypes.h (fixed_size_mode): Declare.
659 (fixed_size_mode_pod): New typedef.
660 * builtins.h (target_builtins::x_apply_args_mode)
661 (target_builtins::x_apply_result_mode): Change type to
663 * builtins.c (apply_args_size, apply_result_size, result_vector)
664 (expand_builtin_apply_args_1, expand_builtin_apply)
665 (expand_builtin_return): Update accordingly.
667 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
669 * cse.c (hash_rtx_cb): Hash only the encoded elements.
670 * cselib.c (cselib_hash_rtx): Likewise.
671 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
672 CONST_VECTOR encoding.
674 2017-01-03 Jakub Jelinek <jakub@redhat.com>
675 Jeff Law <law@redhat.com>
678 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
679 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
680 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
681 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
684 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
685 explicitly probe *sp in a noreturn function if there were any callee
686 register saves or frame pointer is needed.
688 2018-01-03 Jakub Jelinek <jakub@redhat.com>
691 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
692 BLKmode for ternary, binary or unary expressions.
695 * var-tracking.c (delete_vta_debug_insn): New inline function.
696 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
697 insns from get_insns () to NULL instead of each bb separately.
698 Use delete_vta_debug_insn. No longer static.
699 (vt_debug_insns_local, variable_tracking_main_1): Adjust
700 delete_vta_debug_insns callers.
701 * rtl.h (delete_vta_debug_insns): Declare.
702 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
703 instead of variable_tracking_main.
705 2018-01-03 Martin Sebor <msebor@redhat.com>
707 PR tree-optimization/83603
708 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
709 arguments past the endof the argument list in functions declared
711 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
712 Avoid checking when arguments are null.
714 2018-01-03 Martin Sebor <msebor@redhat.com>
717 * doc/extend.texi (attribute const): Fix a typo.
718 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
719 issuing -Wsuggest-attribute for void functions.
721 2018-01-03 Martin Sebor <msebor@redhat.com>
723 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
724 offset_int::from instead of wide_int::to_shwi.
725 (maybe_diag_overlap): Remove assertion.
726 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
727 * gimple-ssa-sprintf.c (format_directive): Same.
728 (parse_directive): Same.
729 (sprintf_dom_walker::compute_format_length): Same.
730 (try_substitute_return_value): Same.
732 2017-01-03 Jeff Law <law@redhat.com>
735 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
736 non-constant residual for zero at runtime and avoid probing in
737 that case. Reorganize code for trailing problem to mirror handling
740 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
742 PR tree-optimization/83501
743 * tree-ssa-strlen.c (get_string_cst): New.
744 (handle_char_store): Call get_string_cst.
746 2018-01-03 Martin Liska <mliska@suse.cz>
748 PR tree-optimization/83593
749 * tree-ssa-strlen.c: Include tree-cfg.h.
750 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
751 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
752 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
754 (strlen_dom_walker::before_dom_children): Call
755 gimple_purge_dead_eh_edges. Dump tranformation with details
757 (strlen_dom_walker::before_dom_children): Update call by adding
758 new argument cleanup_eh.
759 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
761 2018-01-03 Martin Liska <mliska@suse.cz>
764 * cif-code.def (VARIADIC_THUNK): New enum value.
765 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
768 2018-01-03 Jan Beulich <jbeulich@suse.com>
770 * sse.md (mov<mode>_internal): Tighten condition for when to use
771 vmovdqu<ssescalarsize> for TI and OI modes.
773 2018-01-03 Jakub Jelinek <jakub@redhat.com>
775 Update copyright years.
777 2018-01-03 Martin Liska <mliska@suse.cz>
780 * ipa-visibility.c (function_and_variable_visibility): Skip
781 functions with noipa attribure.
783 2018-01-03 Jakub Jelinek <jakub@redhat.com>
785 * gcc.c (process_command): Update copyright notice dates.
786 * gcov-dump.c (print_version): Ditto.
787 * gcov.c (print_version): Ditto.
788 * gcov-tool.c (print_version): Ditto.
789 * gengtype.c (create_file): Ditto.
790 * doc/cpp.texi: Bump @copying's copyright year.
791 * doc/cppinternals.texi: Ditto.
792 * doc/gcc.texi: Ditto.
793 * doc/gccint.texi: Ditto.
794 * doc/gcov.texi: Ditto.
795 * doc/install.texi: Ditto.
796 * doc/invoke.texi: Ditto.
798 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
800 * vector-builder.h (vector_builder::m_full_nelts): Change from
801 unsigned int to poly_uint64.
802 (vector_builder::full_nelts): Update prototype accordingly.
803 (vector_builder::new_vector): Likewise.
804 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
805 (vector_builder::operator ==): Likewise.
806 (vector_builder::finalize): Likewise.
807 * int-vector-builder.h (int_vector_builder::int_vector_builder):
808 Take the number of elements as a poly_uint64 rather than an
810 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
811 from unsigned int to poly_uint64.
812 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
813 (vec_perm_indices::new_vector): Likewise.
814 (vec_perm_indices::length): Likewise.
815 (vec_perm_indices::nelts_per_input): Likewise.
816 (vec_perm_indices::input_nelts): Likewise.
817 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
818 number of elements per input as a poly_uint64 rather than an
819 unsigned int. Use the original encoding for variable-length
820 vectors, rather than clamping each individual element.
821 For the second and subsequent elements in each pattern,
822 clamp the step and base before clamping their sum.
823 (vec_perm_indices::series_p): Handle polynomial element counts.
824 (vec_perm_indices::all_in_range_p): Likewise.
825 (vec_perm_indices_to_tree): Likewise.
826 (vec_perm_indices_to_rtx): Likewise.
827 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
828 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
829 (tree_vector_builder::new_binary_operation): Handle polynomial
830 element counts. Return false if we need to know the number
831 of elements at compile time.
832 * fold-const.c (fold_vec_perm): Punt if the number of elements
833 isn't known at compile time.
835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
837 * vec-perm-indices.h (vec_perm_builder): Change element type
838 from HOST_WIDE_INT to poly_int64.
839 (vec_perm_indices::element_type): Update accordingly.
840 (vec_perm_indices::clamp): Handle polynomial element_types.
841 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
842 (vec_perm_indices::all_in_range_p): Likewise.
843 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
845 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
846 polynomial vec_perm_indices element types.
847 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
848 * fold-const.c (fold_vec_perm): Likewise.
849 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
850 * tree-vect-generic.c (lower_vec_perm): Likewise.
851 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
852 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
853 element type to HOST_WIDE_INT.
855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
856 Alan Hayward <alan.hayward@arm.com>
857 David Sherwood <david.sherwood@arm.com>
859 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
860 rather than an int. Use plus_constant.
861 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
862 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
865 Alan Hayward <alan.hayward@arm.com>
866 David Sherwood <david.sherwood@arm.com>
868 * calls.c (emit_call_1, expand_call): Change struct_value_size from
869 a HOST_WIDE_INT to a poly_int64.
871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
872 Alan Hayward <alan.hayward@arm.com>
873 David Sherwood <david.sherwood@arm.com>
875 * calls.c (load_register_parameters): Cope with polynomial
876 mode sizes. Require a constant size for BLKmode parameters
877 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
878 forces a parameter to be padded at the lsb end in order to
879 fill a complete number of words, require the parameter size
880 to be ordered wrt UNITS_PER_WORD.
882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
883 Alan Hayward <alan.hayward@arm.com>
884 David Sherwood <david.sherwood@arm.com>
886 * reload1.c (spill_stack_slot_width): Change element type
887 from unsigned int to poly_uint64_pod.
888 (alter_reg): Treat mode sizes as polynomial.
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 * reload.c (complex_word_subreg_p): New function.
895 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
902 operands whose sizes aren't ordered.
903 (match_reload): Refer to this check here.
905 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
906 Alan Hayward <alan.hayward@arm.com>
907 David Sherwood <david.sherwood@arm.com>
909 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
910 that the mode size is in the set {1, 2, 4, 8, 16}.
912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
913 Alan Hayward <alan.hayward@arm.com>
914 David Sherwood <david.sherwood@arm.com>
916 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
917 Use plus_constant instead of gen_rtx_PLUS.
919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
920 Alan Hayward <alan.hayward@arm.com>
921 David Sherwood <david.sherwood@arm.com>
923 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
924 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
925 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
926 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
927 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
928 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
929 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
930 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
931 * config/i386/i386.c (ix86_push_rounding): ...this new function.
932 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
934 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
935 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
936 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
937 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
938 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
939 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
940 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
941 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
942 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
943 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
945 * expr.c (emit_move_resolve_push): Treat the input and result
946 of PUSH_ROUNDING as a poly_int64.
947 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
948 (emit_push_insn): Likewise.
949 * lra-eliminations.c (mark_not_eliminable): Likewise.
950 * recog.c (push_operand): Likewise.
951 * reload1.c (elimination_effects): Likewise.
952 * rtlanal.c (nonzero_bits1): Likewise.
953 * calls.c (store_one_arg): Likewise. Require the padding to be
954 known at compile time.
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 (emit_single_push_insn_1): Treat mode sizes as polynomial.
961 Use plus_constant instead of gen_rtx_PLUS.
963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
964 Alan Hayward <alan.hayward@arm.com>
965 David Sherwood <david.sherwood@arm.com>
967 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
971 Alan Hayward <alan.hayward@arm.com>
972 David Sherwood <david.sherwood@arm.com>
974 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
975 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
976 via stack temporaries. Treat the mode size as polynomial too.
978 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
979 Alan Hayward <alan.hayward@arm.com>
980 David Sherwood <david.sherwood@arm.com>
982 * expr.c (expand_expr_real_2): When handling conversions involving
983 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
984 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
985 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
993 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
994 Alan Hayward <alan.hayward@arm.com>
995 David Sherwood <david.sherwood@arm.com>
997 * combine.c (can_change_dest_mode): Handle polynomial
998 REGMODE_NATURAL_SIZE.
999 * expmed.c (store_bit_field_1): Likewise.
1000 * expr.c (store_constructor): Likewise.
1001 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
1002 and polynomial REGMODE_NATURAL_SIZE.
1003 (gen_lowpart_common): Likewise.
1004 * reginfo.c (record_subregs_of_mode): Likewise.
1005 * rtlanal.c (read_modify_subreg_p): Likewise.
1007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1008 Alan Hayward <alan.hayward@arm.com>
1009 David Sherwood <david.sherwood@arm.com>
1011 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
1012 numbers of 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 * match.pd: Cope with polynomial numbers of vector elements.
1020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1021 Alan Hayward <alan.hayward@arm.com>
1022 David Sherwood <david.sherwood@arm.com>
1024 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1025 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
1032 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1033 (ipa_simd_modify_function_body): Likewise.
1035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1036 Alan Hayward <alan.hayward@arm.com>
1037 David Sherwood <david.sherwood@arm.com>
1039 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1040 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1041 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1042 (expand_vector_condition, vector_element): Likewise.
1043 (subparts_gt): New function.
1044 (get_compute_type): Use subparts_gt.
1045 (count_type_subparts): Delete.
1046 (expand_vector_operations_1): Use subparts_gt instead of
1047 count_type_subparts.
1049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1050 Alan Hayward <alan.hayward@arm.com>
1051 David Sherwood <david.sherwood@arm.com>
1053 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1054 (vect_compile_time_alias): ...this new function. Do the calculation
1055 on poly_ints rather than trees.
1056 (vect_prune_runtime_alias_test_list): Update call accordingly.
1058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1059 Alan Hayward <alan.hayward@arm.com>
1060 David Sherwood <david.sherwood@arm.com>
1062 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1064 (vect_schedule_slp_instance): Likewise.
1066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1067 Alan Hayward <alan.hayward@arm.com>
1068 David Sherwood <david.sherwood@arm.com>
1070 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1071 constant and extern definitions for variable-length vectors.
1072 (vect_get_constant_vectors): Note that the number of units
1073 is known to be constant.
1075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1076 Alan Hayward <alan.hayward@arm.com>
1077 David Sherwood <david.sherwood@arm.com>
1079 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1080 of units as polynomial. Choose between WIDE and NARROW based
1083 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1084 Alan Hayward <alan.hayward@arm.com>
1085 David Sherwood <david.sherwood@arm.com>
1087 * tree-vect-stmts.c (simd_clone_subparts): New function.
1088 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1090 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1091 Alan Hayward <alan.hayward@arm.com>
1092 David Sherwood <david.sherwood@arm.com>
1094 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1095 vectors as polynomial. Use build_index_vector for
1098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1099 Alan Hayward <alan.hayward@arm.com>
1100 David Sherwood <david.sherwood@arm.com>
1102 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1103 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1104 for variable-length vectors.
1105 (vectorizable_mask_load_store): Treat the number of units as
1106 polynomial, asserting that it is constant if the condition has
1107 already been enforced.
1108 (vectorizable_store, vectorizable_load): Likewise.
1110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1111 Alan Hayward <alan.hayward@arm.com>
1112 David Sherwood <david.sherwood@arm.com>
1114 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1115 of units as polynomial. Punt if we can't tell at compile time
1116 which vector contains the final result.
1118 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1119 Alan Hayward <alan.hayward@arm.com>
1120 David Sherwood <david.sherwood@arm.com>
1122 * tree-vect-loop.c (vectorizable_induction): Treat the number
1123 of units as polynomial. Punt on SLP inductions. Use an integer
1124 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1125 cast of such a series for variable-length floating-point
1128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1129 Alan Hayward <alan.hayward@arm.com>
1130 David Sherwood <david.sherwood@arm.com>
1132 * tree.h (build_index_vector): Declare.
1133 * tree.c (build_index_vector): New function.
1134 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1135 of units as polynomial, forcibly converting it to a constant if
1136 vectorizable_reduction has already enforced the condition.
1137 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1138 to create a {1,2,3,...} vector.
1139 (vectorizable_reduction): Treat the number of units as polynomial.
1140 Choose vectype_in based on the largest scalar element size rather
1141 than the smallest number of units. Enforce the restrictions
1144 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1145 Alan Hayward <alan.hayward@arm.com>
1146 David Sherwood <david.sherwood@arm.com>
1148 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1149 number of units as polynomial.
1151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1152 Alan Hayward <alan.hayward@arm.com>
1153 David Sherwood <david.sherwood@arm.com>
1155 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1156 * target.def (autovectorize_vector_sizes): Return the vector sizes
1157 by pointer, using vector_sizes rather than a bitmask.
1158 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1159 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1160 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1162 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1163 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1164 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1165 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1166 * omp-general.c (omp_max_vf): Likewise.
1167 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1168 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1169 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1170 * tree-vect-slp.c (vect_slp_bb): Likewise.
1171 * doc/tm.texi: Regenerate.
1172 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1174 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1175 the vector size as a poly_uint64 rather than an unsigned int.
1176 (current_vector_size): Change from an unsigned int to a poly_uint64.
1177 (get_vectype_for_scalar_type): Update accordingly.
1178 * tree.h (build_truth_vector_type): Take the size and number of
1179 units as a poly_uint64 rather than an unsigned int.
1180 (build_vector_type): Add a temporary overload that takes
1181 the number of units as a poly_uint64 rather than an unsigned int.
1182 * tree.c (make_vector_type): Likewise.
1183 (build_truth_vector_type): Take the number of units as a poly_uint64
1184 rather than an unsigned int.
1186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1187 Alan Hayward <alan.hayward@arm.com>
1188 David Sherwood <david.sherwood@arm.com>
1190 * target.def (get_mask_mode): Take the number of units and length
1191 as poly_uint64s rather than unsigned ints.
1192 * targhooks.h (default_get_mask_mode): Update accordingly.
1193 * targhooks.c (default_get_mask_mode): Likewise.
1194 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1195 * doc/tm.texi: Regenerate.
1197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1198 Alan Hayward <alan.hayward@arm.com>
1199 David Sherwood <david.sherwood@arm.com>
1201 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1202 * omp-general.c (omp_max_vf): Likewise.
1203 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1204 (expand_omp_simd): Handle polynomial safelen.
1205 * omp-low.c (omplow_simd_context): Add a default constructor.
1206 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1207 (lower_rec_simd_input_clauses): Update accordingly.
1208 (lower_rec_input_clauses): Likewise.
1210 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1211 Alan Hayward <alan.hayward@arm.com>
1212 David Sherwood <david.sherwood@arm.com>
1214 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1215 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1216 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1217 (vect_analyze_slp_cost): Likewise.
1218 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1219 (vect_model_load_cost): Likewise.
1221 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1222 Alan Hayward <alan.hayward@arm.com>
1223 David Sherwood <david.sherwood@arm.com>
1225 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1226 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1227 from an unsigned int * to a poly_uint64_pod *.
1228 (calculate_unrolling_factor): New function.
1229 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1232 Alan Hayward <alan.hayward@arm.com>
1233 David Sherwood <david.sherwood@arm.com>
1235 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1236 from an unsigned int to a poly_uint64.
1237 (_loop_vec_info::slp_unrolling_factor): Likewise.
1238 (_loop_vec_info::vectorization_factor): Change from an int
1240 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1241 (vect_get_num_vectors): New function.
1242 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1243 (vect_get_num_copies): Use vect_get_num_vectors.
1244 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1245 to an unsigned int *.
1246 (vect_analyze_data_refs): Change min_vf from an int * to a
1248 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1249 than an unsigned HOST_WIDE_INT.
1250 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1251 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1252 to an unsigned int *.
1253 (vect_analyze_data_ref_dependences): Likewise.
1254 (vect_compute_data_ref_alignment): Handle polynomial vf.
1255 (vect_enhance_data_refs_alignment): Likewise.
1256 (vect_prune_runtime_alias_test_list): Likewise.
1257 (vect_shift_permute_load_chain): Likewise.
1258 (vect_supportable_dr_alignment): Likewise.
1259 (dependence_distance_ge_vf): Take the vectorization factor as a
1260 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1261 (vect_analyze_data_refs): Change min_vf from an int * to a
1263 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1264 vfm1 as a poly_uint64 rather than an int. Make the same change
1265 for the returned bound_scalar.
1266 (vect_gen_vector_loop_niters): Handle polynomial vf.
1267 (vect_do_peeling): Likewise. Update call to
1268 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1269 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1271 * tree-vect-loop.c (vect_determine_vectorization_factor)
1272 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1273 (vect_get_known_peeling_cost): Likewise.
1274 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1275 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1276 (vect_transform_loop): Likewise. Use the lowest possible VF when
1277 updating the upper bounds of the loop.
1278 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1280 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1281 polynomial unroll factors.
1282 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1283 (vect_make_slp_decision): Likewise.
1284 (vect_supported_load_permutation_p): Likewise, and polynomial
1286 (vect_analyze_slp_cost): Handle polynomial vf.
1287 (vect_slp_analyze_node_operations): Likewise.
1288 (vect_slp_analyze_bb_1): Likewise.
1289 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1290 than an unsigned HOST_WIDE_INT.
1291 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1292 (vectorizable_load): Handle polynomial vf.
1293 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1295 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1298 Alan Hayward <alan.hayward@arm.com>
1299 David Sherwood <david.sherwood@arm.com>
1301 * match.pd: Handle bit operations involving three constants
1302 and try to fold one pair.
1304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1306 * tree-vect-loop-manip.c: Include gimple-fold.h.
1307 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1308 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1309 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1310 Add a path that uses a step of VF instead of 1, but disable it
1312 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1313 and niters_no_overflow parameters. Update calls to
1314 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1315 Create a new SSA name if the latter choses to use a ste other
1316 than zero, and return it via niters_vector_mult_vf_var.
1317 * tree-vect-loop.c (vect_transform_loop): Update calls to
1318 vect_do_peeling, vect_gen_vector_loop_niters and
1319 slpeel_make_loop_iterate_ntimes.
1320 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1321 (vect_gen_vector_loop_niters): Update declarations after above changes.
1323 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1325 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1326 128-bit round to integer instructions.
1327 (ceil<mode>2): Likewise.
1328 (btrunc<mode>2): Likewise.
1329 (round<mode>2): Likewise.
1331 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1333 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1334 unaligned VSX load/store on P8/P9.
1335 (expand_block_clear): Allow the use of unaligned VSX
1336 load/store on P8/P9.
1338 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1340 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1342 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1343 swap associated with both a load and a store.
1345 2018-01-02 Andrew Waterman <andrew@sifive.com>
1347 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1348 * config/riscv/riscv.md (clear_cache): Use it.
1350 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1352 * web.c: Remove out-of-date comment.
1354 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1356 * expr.c (fixup_args_size_notes): Check that any existing
1357 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1358 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1359 (emit_single_push_insn): ...here.
1361 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1363 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1364 (const_vector_encoded_nelts): New function.
1365 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1366 (const_vector_int_elt, const_vector_elt): Declare.
1367 * emit-rtl.c (const_vector_int_elt_1): New function.
1368 (const_vector_elt): Likewise.
1369 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1370 of CONST_VECTOR_ELT.
1372 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1374 * expr.c: Include rtx-vector-builder.h.
1375 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1376 directly on the tree encoding.
1377 (const_vector_from_tree): Likewise.
1378 * optabs.c: Include rtx-vector-builder.h.
1379 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1380 sequence of "u" values.
1381 * vec-perm-indices.c: Include rtx-vector-builder.h.
1382 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1383 directly on the vec_perm_indices encoding.
1385 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1387 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1388 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1389 * rtx-vector-builder.h: New file.
1390 * rtx-vector-builder.c: Likewise.
1391 * rtl.h (rtx_def::u2): Add a const_vector field.
1392 (CONST_VECTOR_NPATTERNS): New macro.
1393 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1394 (CONST_VECTOR_DUPLICATE_P): Likewise.
1395 (CONST_VECTOR_STEPPED_P): Likewise.
1396 (CONST_VECTOR_ENCODED_ELT): Likewise.
1397 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1398 (unwrap_const_vec_duplicate): Likewise.
1399 (const_vec_series_p): Check for a non-duplicated vector encoding.
1400 Say that the function only returns true for integer vectors.
1401 * emit-rtl.c: Include rtx-vector-builder.h.
1402 (gen_const_vec_duplicate_1): Delete.
1403 (gen_const_vector): Call gen_const_vec_duplicate instead of
1404 gen_const_vec_duplicate_1.
1405 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1406 (gen_const_vec_duplicate): Use rtx_vector_builder.
1407 (gen_const_vec_series): Likewise.
1408 (gen_rtx_CONST_VECTOR): Likewise.
1409 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1410 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1411 Build a new vector rather than modifying a CONST_VECTOR in-place.
1412 (handle_special_swappables): Update call accordingly.
1413 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1414 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1415 Build a new vector rather than modifying a CONST_VECTOR in-place.
1416 (handle_special_swappables): Update call accordingly.
1418 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1420 * simplify-rtx.c (simplify_const_binary_operation): Use
1421 CONST_VECTOR_ELT instead of XVECEXP.
1423 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1425 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1426 the selector elements to be different from the data elements
1427 if the selector is a VECTOR_CST.
1428 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1429 ssizetype for the selector.
1431 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1433 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1434 before testing each element individually.
1435 * tree-vect-generic.c (lower_vec_perm): Likewise.
1437 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1439 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1440 * selftest-run-tests.c (selftest::run_tests): Call it.
1441 * vector-builder.h (vector_builder::operator ==): New function.
1442 (vector_builder::operator !=): Likewise.
1443 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1444 (vec_perm_indices::all_from_input_p): New function.
1445 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1446 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1447 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1448 instead of reading the VECTOR_CST directly. Detect whether both
1449 vector inputs are the same before constructing the vec_perm_indices,
1450 and update the number of inputs argument accordingly. Use the
1451 utility functions added above. Only construct sel2 if we need to.
1453 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1455 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1456 the broadcast of the low byte.
1457 (expand_mult_highpart): Use an explicit encoding for the permutes.
1458 * optabs-query.c (can_mult_highpart_p): Likewise.
1459 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1460 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1461 (vectorizable_bswap): Likewise.
1462 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1463 explicit encoding for the power-of-2 permutes.
1464 (vect_permute_store_chain): Likewise.
1465 (vect_grouped_load_supported): Likewise.
1466 (vect_permute_load_chain): Likewise.
1468 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1470 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1471 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1472 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1473 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1474 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1475 (vect_gen_perm_mask_any): Likewise.
1477 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1479 * int-vector-builder.h: New file.
1480 * vec-perm-indices.h: Include int-vector-builder.h.
1481 (vec_perm_indices): Redefine as an int_vector_builder.
1482 (auto_vec_perm_indices): Delete.
1483 (vec_perm_builder): Redefine as a stand-alone class.
1484 (vec_perm_indices::vec_perm_indices): New function.
1485 (vec_perm_indices::clamp): Likewise.
1486 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1487 (vec_perm_indices::new_vector): New function.
1488 (vec_perm_indices::new_expanded_vector): Update for new
1489 vec_perm_indices class.
1490 (vec_perm_indices::rotate_inputs): New function.
1491 (vec_perm_indices::all_in_range_p): Operate directly on the
1492 encoded form, without computing elided elements.
1493 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1494 encoding. Update for new vec_perm_indices class.
1495 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1496 the given vec_perm_builder.
1497 (expand_vec_perm_var): Update vec_perm_builder constructor.
1498 (expand_mult_highpart): Use vec_perm_builder instead of
1499 auto_vec_perm_indices.
1500 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1501 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1502 or double series encoding as appropriate.
1503 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1504 vec_perm_indices instead of auto_vec_perm_indices.
1505 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1506 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1507 (vect_permute_store_chain): Likewise.
1508 (vect_grouped_load_supported): Likewise.
1509 (vect_permute_load_chain): Likewise.
1510 (vect_shift_permute_load_chain): Likewise.
1511 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1512 (vect_transform_slp_perm_load): Likewise.
1513 (vect_schedule_slp_instance): Likewise.
1514 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1515 (vectorizable_mask_load_store): Likewise.
1516 (vectorizable_bswap): Likewise.
1517 (vectorizable_store): Likewise.
1518 (vectorizable_load): Likewise.
1519 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1520 vec_perm_indices instead of auto_vec_perm_indices. Use
1521 tree_to_vec_perm_builder to read the vector from a tree.
1522 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1523 vec_perm_builder instead of a vec_perm_indices.
1524 (have_whole_vector_shift): Use vec_perm_builder and
1525 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1526 truncation to calc_vec_perm_mask_for_shift.
1527 (vect_create_epilog_for_reduction): Likewise.
1528 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1529 from auto_vec_perm_indices to vec_perm_indices.
1530 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1531 instead of changing individual elements.
1532 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1533 the vector in d.perm.
1534 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1535 from auto_vec_perm_indices to vec_perm_indices.
1536 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1537 instead of changing individual elements.
1538 (arm_vectorize_vec_perm_const): Use new_vector to install
1539 the vector in d.perm.
1540 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1541 Update vec_perm_builder constructor.
1542 (rs6000_expand_interleave): Likewise.
1543 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1544 (rs6000_expand_interleave): Likewise.
1546 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1548 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1549 to qimode could truncate the indices.
1550 * optabs.c (expand_vec_perm_var): Likewise.
1552 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1554 * Makefile.in (OBJS): Add vec-perm-indices.o.
1555 * vec-perm-indices.h: New file.
1556 * vec-perm-indices.c: Likewise.
1557 * target.h (vec_perm_indices): Replace with a forward class
1559 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1560 * optabs.h: Include vec-perm-indices.h.
1561 (expand_vec_perm): Delete.
1562 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1563 (expand_vec_perm_const): Declare.
1564 * target.def (vec_perm_const_ok): Replace with...
1565 (vec_perm_const): ...this new hook.
1566 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1567 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1568 * doc/tm.texi: Regenerate.
1569 * optabs.def (vec_perm_const): Delete.
1570 * doc/md.texi (vec_perm_const): Likewise.
1571 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1572 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1573 expand_vec_perm for constant permutation vectors. Assert that
1574 the mode of variable permutation vectors is the integer equivalent
1575 of the mode that is being permuted.
1576 * optabs-query.h (selector_fits_mode_p): Declare.
1577 * optabs-query.c: Include vec-perm-indices.h.
1578 (selector_fits_mode_p): New function.
1579 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1580 is defined, instead of checking whether the vec_perm_const_optab
1581 exists. Use targetm.vectorize.vec_perm_const instead of
1582 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1583 fit in the vector mode before using a variable permute.
1584 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1585 vec_perm_indices instead of an rtx.
1586 (expand_vec_perm): Replace with...
1587 (expand_vec_perm_const): ...this new function. Take the selector
1588 as a vec_perm_indices rather than an rtx. Also take the mode of
1589 the selector. Update call to shift_amt_for_vec_perm_mask.
1590 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1591 Use vec_perm_indices::new_expanded_vector to expand the original
1592 selector into bytes. Check whether the indices fit in the vector
1593 mode before using a variable permute.
1594 (expand_vec_perm_var): Make global.
1595 (expand_mult_highpart): Use expand_vec_perm_const.
1596 * fold-const.c: Includes vec-perm-indices.h.
1597 * tree-ssa-forwprop.c: Likewise.
1598 * tree-vect-data-refs.c: Likewise.
1599 * tree-vect-generic.c: Likewise.
1600 * tree-vect-loop.c: Likewise.
1601 * tree-vect-slp.c: Likewise.
1602 * tree-vect-stmts.c: Likewise.
1603 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1605 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1606 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1607 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1608 (aarch64_vectorize_vec_perm_const): ...this new function.
1609 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1610 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1611 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1612 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1613 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1614 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1615 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1617 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1618 check for NEON modes.
1619 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1620 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1621 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1622 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1624 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1625 the old VEC_PERM_CONST conditions.
1626 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1627 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1628 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1629 (ia64_vectorize_vec_perm_const_ok): Merge into...
1630 (ia64_vectorize_vec_perm_const): ...this new function.
1631 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1632 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1633 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1634 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1635 * config/mips/mips.c (mips_expand_vec_perm_const)
1636 (mips_vectorize_vec_perm_const_ok): Merge into...
1637 (mips_vectorize_vec_perm_const): ...this new function.
1638 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1639 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1640 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1641 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1642 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1643 (rs6000_expand_vec_perm_const): Delete.
1644 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1646 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1647 (altivec_expand_vec_perm_const_le): Take each operand individually.
1648 Operate on constant selectors rather than rtxes.
1649 (altivec_expand_vec_perm_const): Likewise. Update call to
1650 altivec_expand_vec_perm_const_le.
1651 (rs6000_expand_vec_perm_const): Delete.
1652 (rs6000_vectorize_vec_perm_const_ok): Delete.
1653 (rs6000_vectorize_vec_perm_const): New function.
1654 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1655 an element count and rtx array.
1656 (rs6000_expand_extract_even): Update call accordingly.
1657 (rs6000_expand_interleave): Likewise.
1658 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1659 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1660 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1661 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1662 (rs6000_expand_vec_perm_const): Delete.
1663 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1664 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1665 (altivec_expand_vec_perm_const_le): Take each operand individually.
1666 Operate on constant selectors rather than rtxes.
1667 (altivec_expand_vec_perm_const): Likewise. Update call to
1668 altivec_expand_vec_perm_const_le.
1669 (rs6000_expand_vec_perm_const): Delete.
1670 (rs6000_vectorize_vec_perm_const_ok): Delete.
1671 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1672 reference to the SPE evmerge intructions.
1673 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1674 an element count and rtx array.
1675 (rs6000_expand_extract_even): Update call accordingly.
1676 (rs6000_expand_interleave): Likewise.
1677 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1678 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1680 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1682 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1684 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1685 vector mode and that that mode matches the mode of the data
1687 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1688 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1689 directly using expand_vec_perm_1 when forcing selectors into
1691 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1693 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1695 * optabs-query.h (can_vec_perm_p): Delete.
1696 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1697 * optabs-query.c (can_vec_perm_p): Split into...
1698 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1699 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1700 particular selector is valid.
1701 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1702 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1703 (vect_grouped_load_supported): Likewise.
1704 (vect_shift_permute_load_chain): Likewise.
1705 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1706 (vect_transform_slp_perm_load): Likewise.
1707 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1708 (vectorizable_bswap): Likewise.
1709 (vect_gen_perm_mask_checked): Likewise.
1710 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1711 implementations of variable permutation vectors into account
1712 when deciding which selector to use.
1713 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1714 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1715 with a false third argument.
1716 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1717 to test whether the constant selector is valid and can_vec_perm_var_p
1718 to test whether a variable selector is valid.
1720 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1722 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1723 * optabs-query.c (can_vec_perm_p): Likewise.
1724 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1725 instead of vec_perm_indices.
1726 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1727 (vect_gen_perm_mask_checked): Likewise,
1728 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1729 (vect_gen_perm_mask_checked): Likewise,
1731 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1733 * optabs-query.h (qimode_for_vec_perm): Declare.
1734 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1735 (qimode_for_vec_perm): ...this new function.
1736 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1738 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1740 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1741 does not have a conditional at the top.
1743 2018-01-02 Richard Biener <rguenther@suse.de>
1745 * ipa-inline.c (big_speedup_p): Fix expression.
1747 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1750 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1753 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1757 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1758 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1759 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1760 cond_taken_branch_cost 3->4.
1762 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1764 PR tree-optimization/83581
1765 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1766 TODO_cleanup_cfg if any changes have been made.
1769 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1770 convert_modes if target mode has the right side, but different mode
1774 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1775 last argument when extracting from CONCAT. If either from_real or
1776 from_imag is NULL, use expansion through memory. If result is not
1777 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1778 the parts directly to inner mode, if even that fails, use expansion
1782 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1783 check for bswap in mode rather than HImode and use that in expand_unop
1786 Copyright (C) 2018 Free Software Foundation, Inc.
1788 Copying and distribution of this file, with or without modification,
1789 are permitted in any medium without royalty provided the copyright
1790 notice and this notice are preserved.