1 2018-02-01 Jakub Jelinek <jakub@redhat.com>
3 PR tree-optimization/81661
4 PR tree-optimization/84117
5 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
6 * tree-eh.c: Include gimplify.h.
7 (find_trapping_overflow, replace_trapping_overflow,
8 rewrite_to_non_trapping_overflow): New functions.
9 * tree-vect-loop.c: Include tree-eh.h.
10 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
11 * tree-data-ref.c: Include tree-eh.h.
12 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
14 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
16 PR rtl-optimization/84123
17 * combine.c (change_zero_ext): Check if hard register satisfies
18 can_change_dest_mode before calling gen_lowpart_SUBREG.
20 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
23 * ira.c (ira_init_register_move_cost): Remove assert.
25 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
27 PR rtl-optimization/84071
28 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
29 * doc/tm.texi: Regenerate.
31 2018-01-31 Richard Biener <rguenther@suse.de>
33 PR tree-optimization/84132
34 * tree-data-ref.c (analyze_miv_subscript): Properly
35 check whether evolution_function_is_affine_multivariate_p
36 before calling gcd_of_steps_may_divide_p.
38 2018-01-31 Julia Koval <julia.koval@intel.com>
41 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
42 * config/i386/i386.md (rdpid_rex64) New.
43 (rdpid): Make 32bit only.
45 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
48 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
49 an IDENTIFIER_NODE for FUNCTION_TYPE's.
51 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
54 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
56 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
58 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
60 PR rtl-optimization/84071
61 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
62 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
64 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
66 * config/arc/arc.c (arc_handle_aux_attribute): New function.
67 (arc_attribute_table): Add 'aux' attribute.
68 (arc_in_small_data_p): Consider aux like variables.
69 (arc_is_aux_reg_p): New function.
70 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
71 (arc_get_aux_arg): New function.
72 (prepare_move_operands): Handle aux-register access.
73 (arc_handle_aux_attribute): New function.
74 * doc/extend.texi (ARC Variable attributes): Add subsection.
76 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
78 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
79 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
80 (arc_attribute_table): Add 'uncached' attribute.
81 (arc_print_operand): Print '.di' flag for uncached memory
83 (arc_in_small_data_p): Do not consider for small data the uncached
85 (arc_is_uncached_mem_p): New function.
86 * config/arc/predicates.md (compact_store_memory_operand): Check
87 for uncached memory accesses.
88 (nonvol_nonimm_operand): Likewise.
89 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
91 2018-01-31 Jakub Jelinek <jakub@redhat.com>
94 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
95 falign-loops=): Add Optimization flag.
97 2017-01-30 Jeff Law <law@redhat.com>
100 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
101 INT_REGISTERS_SAVED. Check it prior to calling
102 get_scratch_register_on_entry.
103 (ix86_adjust_stack_and_probe): Similarly.
104 (ix86_emit_probe_stack_range): Similarly.
105 (ix86_expand_prologue): Corresponding changes.
107 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
110 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
111 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
113 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
116 * lra-constraints.c (curr_insn_transform): Process AND in the
119 2018-01-30 Jakub Jelinek <jakub@redhat.com>
121 PR rtl-optimization/83986
122 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
123 dependence against last_pending_memory_flush in addition to
126 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
128 PR tree-optimization/81611
129 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
132 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
134 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
137 2018-01-30 Richard Biener <rguenther@suse.de>
138 Jakub Jelinek <jakub@redhat.com>
140 PR tree-optimization/84111
141 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
142 inner loops added during recursion, as they don't have up-to-date
145 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
148 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
149 (can_inline_edge_by_limits_p): ... here.
150 (can_early_inline_edge_p, check_callers,
151 update_caller_keys, update_callee_keys, recursive_inlining,
152 add_new_edges_to_heap, speculation_useful_p,
153 inline_small_functions,
154 inline_small_functions, flatten_function,
155 inline_to_all_callers_1): Update.
157 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
159 * profile-count.c (profile_count::combine_with_ipa_count): Handle
162 2018-01-30 Richard Biener <rguenther@suse.de>
164 PR tree-optimization/83008
165 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
166 invariant and constant vector uses in stmts when they need
169 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
172 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
173 * configure: Regenerate.
175 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
177 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
179 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
180 Use gen_rtx_REG rather than gen_lowpart.
182 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
184 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
185 rather than 0 when creating partial subregs.
187 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
189 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
192 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
195 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
196 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
197 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
198 flags. This restores the settings used before the 2017-07-24.
199 Turning off pre increment/decrement/modify allows IVOPTS to
200 optimize DF/SF loops where the index is an int.
202 2018-01-29 Richard Biener <rguenther@suse.de>
203 Kelvin Nilsen <kelvin@gcc.gnu.org>
206 * tree-vect-stmts.c (vectorizable_call): Don't call
207 targetm.vectorize_builtin_md_vectorized_function if callee is
210 2018-01-22 Carl Love <cel@us.ibm.com>
212 * doc/extend.tex: Fix typo in second arg in
213 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
215 2018-01-29 Richard Biener <rguenther@suse.de>
217 PR tree-optimization/84086
218 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
219 (flush_ssaname_freelist): When SSA names were released reset
222 2018-01-29 Richard Biener <rguenther@suse.de>
224 PR tree-optimization/84057
225 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
226 removed paths when removing edges.
228 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
230 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
231 -mfunction-return=@var{choice}.
233 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
236 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
238 (layout::print_source_line): Likewise.
239 (test_get_line_width_without_trailing_whitespace): Add test cases.
241 2018-01-27 Jakub Jelinek <jakub@redhat.com>
244 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
247 2018-01-26 Jim Wilson <jimw@sifive.com>
249 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
251 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
254 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
256 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
257 and CMP + SUB-immediate -> SUBS.
259 2018-01-26 Martin Sebor <msebor@redhat.com>
261 PR tree-optimization/83896
262 * tree-ssa-strlen.c (get_string_len): Rename...
263 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
264 Avoid assuming length is constant.
265 (handle_char_store): Use HOST_WIDE_INT for string length.
267 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
270 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
271 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
273 2018-01-26 Richard Biener <rguenther@suse.de>
275 PR rtl-optimization/84003
276 * dse.c (record_store): Only record redundant stores when
277 the earlier store aliases at least all accesses the later one does.
279 2018-01-26 Jakub Jelinek <jakub@redhat.com>
281 PR rtl-optimization/83985
282 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
283 REG_CFA_RESTORE insns.
284 (delete_unmarked_insns): Don't ignore separate shrink wrapping
285 REG_CFA_RESTORE insns here.
288 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
289 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
291 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
293 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
294 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
295 (arc_init): Likewise.
296 (arc_override_options): Likewise.
297 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
299 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
301 * config/arc/arc.h (TARGET_DBNZ): Define.
302 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
303 properly set the tune attribute.
304 (dbnz): Use TARGET_DBNZ guard.
305 * config/arc/arc.opt (mtune): Add core3 option.
307 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
309 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
310 recognize new pic like addresses.
311 (arc_delegitimize_address): Clean up.
313 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
315 * config/arc/arc-arches.def: Option mrf16 valid for all
317 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
318 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
319 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
320 * config/arc/arc-tables.opt: Regenerate.
321 * config/arc/arc.c (arc_conditional_register_usage): Handle
322 reduced register file case.
323 (arc_file_start): Set must have build attributes.
324 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
326 * config/arc/arc.opt (mrf16): Add new option.
327 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
328 * config/arc/genmultilib.awk: Handle new mrf16 option.
329 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
330 * config/arc/t-multilib: Regenerate.
331 * doc/invoke.texi (ARC Options): Document mrf16 option.
333 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
335 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
336 * config/arc/arc.c (arc_handle_secure_attribute): New function.
337 (arc_attribute_table): Add 'secure_call' attribute.
338 (arc_print_operand): Print secure call operand.
339 (arc_function_ok_for_sibcall): Don't optimize tail calls when
341 (arc_is_secure_call_p): New function. * config/arc/arc.md
342 (call_i): Add support for sjli instruction.
343 (call_value_i): Likewise.
344 * config/arc/constraints.md (Csc): New constraint.
346 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
347 John Eric Martin <John.Martin@emmicro-us.com>
349 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
350 * config/arc/arc.c (_arc_jli_section): New struct.
351 (arc_jli_section): New type.
352 (rc_jli_sections): New static variable.
353 (arc_handle_jli_attribute): New function.
354 (arc_attribute_table): Add jli_always and jli_fixed attribute.
355 (arc_file_end): New function.
356 (TARGET_ASM_FILE_END): Define.
357 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
358 (arc_add_jli_section): New function.
359 (jli_call_scan): Likewise.
360 (arc_reorg): Call jli_call_scan.
361 (arc_output_addsi): Remove 'S' from printing asm operand.
362 (arc_is_jli_call_p): New function.
363 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
365 (movhi_insn): Likewise.
366 (movsi_insn): Likewise.
367 (movsi_set_cc_insn): Likewise.
368 (loadqi_update): Likewise.
369 (load_zeroextendqisi_update): Likewise.
370 (load_signextendqisi_update): Likewise.
371 (loadhi_update): Likewise.
372 (load_zeroextendhisi_update): Likewise.
373 (load_signextendhisi_update): Likewise.
374 (loadsi_update): Likewise.
375 (loadsf_update): Likewise.
376 (movsicc_insn): Likewise.
377 (bset_insn): Likewise.
378 (bxor_insn): Likewise.
379 (bclr_insn): Likewise.
380 (bmsk_insn): Likewise.
381 (bicsi3_insn): Likewise.
382 (cmpsi_cc_c_insn): Likewise.
383 (movsi_ne): Likewise.
384 (movsi_cond_exec): Likewise.
385 (clrsbsi2): Likewise.
396 (call_i): Remove 'S' asm letter, add jli instruction.
397 (call_value_i): Likewise.
398 * config/arc/arc.op (mjli-always): New option.
399 * config/arc/constraints.md (Cji): New constraint.
400 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
402 (subsf3_fpx): Likewise.
403 (mulsf3_fpx): Likewise.
404 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
406 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
408 * doc/invoke.texi (ARC): Document mjli-always option.
410 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
412 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
413 and use incw and decw where possible
414 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
416 2018-01-26 Richard Biener <rguenther@suse.de>
418 PR tree-optimization/81082
419 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
420 association if it requires casting to unsigned.
421 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
422 from fold_plusminus_mult_expr to catch important cases late when
423 range info is available.
425 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
427 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
428 * configure.ac (hidden_linkonce): New test.
429 * configure: Regenerate.
430 * config.in: Regenerate.
432 2018-01-26 Julia Koval <julia.koval@intel.com>
434 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
435 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
436 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
437 _mm_mask_bitshuffle_epi64_mask): Fix type.
438 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
439 USI_FTYPE_V4DI_V4DI_USI): Remove.
440 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
441 __builtin_ia32_vpshufbitqmb256_mask,
442 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
443 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
444 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
446 2018-01-26 Alan Modra <amodra@gmail.com>
449 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
450 UNSPEC_VBPERMQ. Sort other unspecs.
452 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
454 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
456 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
459 * predict.c (drop_profile): Do not push/pop cfun; update also
461 (handle_missing_profiles): Fix logic looking for zero profiles.
463 2018-01-25 Jakub Jelinek <jakub@redhat.com>
466 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
467 on functions with #pragma omp declare simd or functions with simd
469 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
470 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
471 Remove trailing \n from warning_at calls.
473 2018-01-25 Tom de Vries <tom@codesourcery.com>
476 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
477 for neutered workers.
479 2018-01-24 Joseph Myers <joseph@codesourcery.com>
482 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
483 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
485 2017-01-24 Jeff Law <law@redhat.com>
488 * i386.c (get_probe_interval): Move to earlier point.
489 (ix86_compute_frame_layout): If -fstack-clash-protection and
490 the frame is larger than the probe interval, then use pushes
491 to save registers rather than reg->mem moves.
492 (ix86_expand_prologue): Remove conditional for int_registers_saved
495 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
498 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
499 min/max for never referenced object.
501 2018-01-24 Jakub Jelinek <jakub@redhat.com>
504 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
506 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
507 attributes from DECL_ATTRIBUTES (decl) without affecting
508 DECL_ATTRIBUTES (current_function_decl).
509 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
510 functions with non-NULL DECL_ABSTRACT_ORIGIN.
512 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
514 PR tree-optimization/83979
515 * fold-const.c (fold_comparison): Use constant_boolean_node
516 instead of boolean_{true,false}_node.
518 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
520 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
523 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
525 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
526 Simplify the clause that sets the length attribute.
527 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
528 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
529 clause that sets the length attribute.
530 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
532 2018-01-24 Tom de Vries <tom@codesourcery.com>
535 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
536 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
537 Add strict parameter.
538 (prevent_branch_around_nothing): Insert dummy insn between branch to
539 label and label with no ptx insn inbetween.
540 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
542 2018-01-24 Tom de Vries <tom@codesourcery.com>
545 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
546 for neutered threads in warp.
547 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
549 2018-01-24 Richard Biener <rguenther@suse.de>
551 PR tree-optimization/83176
552 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
555 2018-01-24 Richard Biener <rguenther@suse.de>
557 PR tree-optimization/82819
558 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
559 code generating pluses that are no-ops in the target precision.
561 2018-01-24 Richard Biener <rguenther@suse.de>
564 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
566 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
568 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
569 to merge probabilities.
570 * predict.c (probably_never_executed): Also mark as cold functions
571 with global 0 profile and guessed local profile.
572 * profile-count.c (profile_probability::combine_with_count): New
574 * profile-count.h (profile_probability::operator*,
575 profile_probability::operator*=, profile_probability::operator/,
576 profile_probability::operator/=): Reduce precision to adjusted
577 and set value to guessed on contradictory divisions.
578 (profile_probability::combine_with_freq): Remove.
579 (profile_probability::combine_wiht_count): Declare.
580 (profile_count::force_nonzero):: Set to adjusted.
581 (profile_count::probability_in):: Set quality to adjusted.
582 * tree-ssa-tail-merge.c (replace_block_by): Use
585 2018-01-23 Andrew Waterman <andrew@sifive.com>
586 Jim Wilson <jimw@sifive.com>
588 * config/riscv/riscv.c (riscv_stack_boundary): New.
589 (riscv_option_override): Set riscv_stack_boundary. Handle
590 riscv_preferred_stack_boundary_arg.
591 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
592 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
593 (STACK_BOUNDARY): Set to riscv_stack_boundary.
594 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
595 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
596 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
598 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
601 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
602 of struct ix86_frame.
603 (ix86_expand_epilogue): Likewise. Add a local variable for
604 the reg_save_offset field in struct ix86_frame.
606 2018-01-23 Bin Cheng <bin.cheng@arm.com>
608 PR tree-optimization/82604
609 * tree-loop-distribution.c (enum partition_kind): New enum item
610 PKIND_PARTIAL_MEMSET.
611 (partition_builtin_p): Support above new enum item.
612 (generate_code_for_partition): Ditto.
613 (compute_access_range): Differentiate cases that equality can be
614 proven at all loops, the innermost loops or no loops.
615 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
616 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
617 (finalize_partitions, distribute_loop): Don't fuse partition of
618 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
619 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
622 2018-01-23 Martin Liska <mliska@suse.cz>
624 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
625 order to ignore the predictor.
626 (PRED_POLYMORPHIC_CALL): Likewise.
627 (PRED_RECURSIVE_CALL): Likewise.
629 2018-01-23 Martin Liska <mliska@suse.cz>
631 * tree-profile.c (tree_profiling): Print function header to
632 aware reader which function we are working on.
633 * value-prof.c (gimple_find_values_to_profile): Do not print
634 not interesting value histograms.
636 2018-01-23 Martin Liska <mliska@suse.cz>
638 * profile-count.h (enum profile_quality): Add
639 profile_uninitialized as the first value. Do not number values
640 as they are zero based.
641 (profile_count::verify): Update sanity check.
642 (profile_probability::verify): Likewise.
644 2018-01-23 Nathan Sidwell <nathan@acm.org>
646 * doc/invoke.texi (ffor-scope): Deprecate.
648 2018-01-23 David Malcolm <dmalcolm@redhat.com>
650 PR tree-optimization/83510
651 * domwalk.c (set_all_edges_as_executable): New function.
652 (dom_walker::dom_walker): Convert bool param
653 "skip_unreachable_blocks" to enum reachability. Move setup of
654 edge flags to set_all_edges_as_executable and only do it when
655 reachability is REACHABLE_BLOCKS.
656 * domwalk.h (enum dom_walker::reachability): New enum.
657 (dom_walker::dom_walker): Convert bool param
658 "skip_unreachable_blocks" to enum reachability.
659 (set_all_edges_as_executable): New decl.
660 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
661 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
663 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
664 but converting true to REACHABLE_BLOCKS.
665 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
667 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
668 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
669 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
671 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
672 if check_all_array_refs will be called.
674 2018-01-23 David Malcolm <dmalcolm@redhat.com>
676 * tree.c (selftest::test_location_wrappers): Add more test
679 2018-01-23 David Malcolm <dmalcolm@redhat.com>
681 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
682 (selftest::test_bit_in_range): Likewise.
684 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
687 * doc/sourcebuild.texi (vect_float): Say that the selector
688 only describes the situation when -funsafe-math-optimizations is on.
689 (vect_float_strict): Document.
691 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
693 PR tree-optimization/83965
694 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
695 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
696 instead of checking only for a reduction.
697 (vect_recog_widen_sum_pattern): Likewise.
699 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
701 * predict.c (probably_never_executed): Only use precise profile info.
702 (compute_function_frequency): Skip after inlining hack since we now
703 have quality checking.
705 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
707 * profile-count.h (profile_probability::very_unlikely,
708 profile_probability::unlikely, profile_probability::even): Set
709 precision to guessed.
711 2018-01-23 Richard Biener <rguenther@suse.de>
713 PR tree-optimization/83963
714 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
715 Properly terminate dominator walk when crossing the exit edge not
716 when visiting its source block.
718 2018-01-23 Jakub Jelinek <jakub@redhat.com>
721 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
722 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
724 2018-01-22 Jakub Jelinek <jakub@redhat.com>
726 PR tree-optimization/83957
727 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
728 semicolon after for body surrounded by braces.
730 PR tree-optimization/83081
731 * profile-count.h (profile_probability::split): New method.
732 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
733 Use profile_probability::split.
734 (do_compare_rtx_and_jump): Fix adjustment of probabilities
735 when splitting a single conditional jump into 2.
737 2018-01-22 David Malcolm <dmalcolm@redhat.com>
739 PR tree-optimization/69452
740 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
743 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
745 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
746 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
747 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
749 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
751 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
752 * config/rl78/rl78.md: New define_expand "movdi"
753 * config/rl78/rl78.c: New function definition rl78_split_movdi
755 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
758 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
760 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
761 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
762 128-bit to produce an UNSPEC move to get the double word with the
763 signbit and then a shift directly to do signbit.
764 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
765 implementation with a new version that just does either a direct
766 move or a regular move. Move memory interface to separate insns.
767 Move insns so they are next to the expander.
768 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
769 with signbit move. Split big and little endian case.
770 (signbit<mode>2_dm_mem_le): Likewise.
771 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
772 (signbit<mode>2_dm2): Likewise.
774 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
776 * config/rl78/rl78.md: New define_expand "anddi3".
778 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
780 * config/rl78/rl78.md: New define_expand "umindi3".
782 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
784 * config/rl78/rl78.md: New define_expand "smindi3".
786 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
788 * config/rl78/rl78.md: New define_expand "smaxdi3".
790 2018-01-22 Carl Love <cel@us.ibm.com>
792 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
793 LVX_V1TI): Add macro expansion.
794 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
795 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
796 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
797 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
798 Change check to determine if the instruction is a byte reversing
799 entry. Fix typo in comment.
800 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
801 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
802 Add def_builtin calls for new builtins.
803 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
804 Add define_insn expansion.
806 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
808 * config/rl78/rl78.md: New define_expand "umaxdi3".
810 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
812 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
813 for non-QImode registers
815 2018-01-22 Richard Biener <rguenther@suse.de>
817 PR tree-optimization/83963
818 * graphite-scop-detection.c (scop_detection::get_sese): Delay
819 including the loop exit block.
820 (scop_detection::merge_sese): Likewise.
821 (scop_detection::add_scop): Do it here instead.
823 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
825 * doc/sourcebuild.texi (arm_softfloat): Document.
827 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
830 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
831 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
832 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
834 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
835 David Edelsohn <dje.gcc@gmail.com>
838 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
839 Change "crset eq" to "crset 2".
840 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
841 (*call_indirect_aix<mode>_nospec): Likewise.
842 (*call_value_indirect_aix<mode>_nospec): Likewise.
843 (*call_indirect_elfv2<mode>_nospec): Likewise.
844 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
845 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
846 change assembly output from . to $.
847 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
848 (indirect_jump<mode>_nospec): Change assembly output from . to $.
849 (*tablejump<mode>_internal1_nospec): Likewise.
851 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
854 * config/sh/sh_optimize_sett_clrt.cc:
855 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
857 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
859 PR tree-optimization/83940
860 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
861 offset_dt to vect_constant_def rather than vect_unknown_def_type.
862 (vect_check_load_store_mask): Add a mask_dt_out parameter and
863 use it to pass back the definition type.
864 (vect_check_store_rhs): Likewise rhs_dt_out.
865 (vect_build_gather_load_calls): Add a mask_dt argument and use
866 it instead of a call to vect_is_simple_use.
867 (vectorizable_store): Update calls to vect_check_load_store_mask
868 and vect_check_store_rhs. Use the dt returned by the latter instead
869 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
870 instead of calls to vect_is_simple_use. Pass the scalar rather
871 than the vector operand to vect_is_simple_use when handling
872 second and subsequent copies of an rhs value.
873 (vectorizable_load): Update calls to vect_check_load_store_mask
874 and vect_build_gather_load_calls. Use the cached mask_dt and
875 gs_info.offset_dt instead of calls to vect_is_simple_use.
877 2018-01-20 Jakub Jelinek <jakub@redhat.com>
880 * tree-emutls.c: Include gimplify.h.
881 (lower_emutls_2): New function.
882 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
883 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
884 it before further processing.
887 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
888 UINTVAL (trueop1) instead of INTVAL (op1).
890 2018-01-19 Jakub Jelinek <jakub@redhat.com>
894 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
895 INCOMING_FRAME_SP_OFFSET if not defined.
896 (scan_trace): Add ENTRY argument. If true and
897 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
898 emit a note to adjust the CFA offset.
899 (create_cfi_notes): Adjust scan_trace callers.
900 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
901 INCOMING_FRAME_SP_OFFSET in the CIE.
902 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
903 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
905 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
906 * doc/tm.texi: Regenerated.
908 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
910 PR rtl-optimization/83147
911 * lra-constraints.c (remove_inheritance_pseudos): Use
912 lra_substitute_pseudo_within_insn.
914 2018-01-19 Tom de Vries <tom@codesourcery.com>
915 Cesar Philippidis <cesar@codesourcery.com>
918 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
920 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
923 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
924 spaces for function labels.
926 2018-01-19 Martin Liska <mliska@suse.cz>
928 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
929 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
930 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
931 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
932 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
933 (PRED_CONST_RETURN): Change from 69 to 65.
934 (PRED_NULL_RETURN): Change from 91 to 71.
935 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
936 (PRED_LOOP_GUARD): Change from 66 to 73.
938 2018-01-19 Martin Liska <mliska@suse.cz>
940 * predict.c (predict_insn_def): Add new assert.
941 (struct branch_predictor): Change type to signed integer.
942 (test_prediction_value_range): Amend test to cover
944 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
945 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
946 (PRED_LOOP_ITERATIONS_MAX): Likewise.
947 (PRED_LOOP_IV_COMPARE): Likewise.
948 * predict.h (PROB_UNINITIALIZED): Define new constant.
950 2018-01-19 Martin Liska <mliska@suse.cz>
952 * predict.c (dump_prediction): Add new format for
953 analyze_brprob.py script which is enabled with -details
955 * profile-count.h (precise_p): New function.
957 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
959 PR tree-optimization/83922
960 * tree-vect-loop.c (vect_verify_full_masking): Return false if
961 there are no statements that need masking.
962 (vect_active_double_reduction_p): New function.
963 (vect_analyze_loop_operations): Use it when handling phis that
964 are not in the loop header.
966 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
968 PR tree-optimization/83914
969 * tree-vect-loop.c (vectorizable_induction): Don't convert
970 init_expr or apply the peeling adjustment for inductions
971 that are nested within the vectorized loop.
973 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
975 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
978 2018-01-18 Jakub Jelinek <jakub@redhat.com>
982 * function.h (gimplify_parameters): Add gimple_seq * argument.
983 * function.c: Include gimple.h and options.h.
984 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
985 for the added local temporaries if needed.
986 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
987 if there are any parameter cleanups, wrap whole body into a
988 try/finally with the cleanups.
990 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
993 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
994 Use GET_MODE_CLASS for scalar floating point.
996 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1000 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1001 Fix call of call_cgraph_insertion_hooks.
1003 2018-01-18 Martin Sebor <msebor@redhat.com>
1005 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1007 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1010 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1013 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1016 * common.opt: (-ffile-prefix-map): New option.
1017 * opts.c (common_handle_option): Defer it.
1018 * opts-global.c (handle_common_deferred_options): Handle it.
1019 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1020 * file-prefix-map.h: New file.
1021 (remap_debug_filename, add_debug_prefix_map): ...here.
1022 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1023 * final.c (debug_prefix_map, add_debug_prefix_map
1024 remap_debug_filename): Move to...
1025 * file-prefix-map.c: New file.
1026 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1027 generalize, get rid of alloca(), use strrchr() instead of strchr().
1028 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1029 Implement in terms of add_prefix_map().
1030 (remap_macro_filename, remap_debug_filename): Implement in term of
1032 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1033 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1034 * dbxout.c: Include file-prefix-map.h.
1035 * varasm.c: Likewise.
1036 * vmsdbgout.c: Likewise.
1037 * xcoffout.c: Likewise.
1038 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1039 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1040 * doc/invoke.texi (-ffile-prefix-map): Document.
1041 (-fdebug-prefix-map): Update description.
1043 2018-01-18 Martin Liska <mliska@suse.cz>
1045 * config/i386/i386.c (indirect_thunk_name): Document that also
1047 (output_indirect_thunk): Document why both instructions
1048 (pause and lfence) are generated.
1050 2018-01-18 Richard Biener <rguenther@suse.de>
1052 PR tree-optimization/83887
1053 * graphite-scop-detection.c
1054 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1055 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1056 (scop_detection::merge_sese): Re-implement with a flood-fill
1057 algorithm that properly finds a SESE region if it exists.
1059 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1062 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1063 pointer_diff optimizations use view_convert instead of convert.
1065 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1067 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1068 Generate different code for -mno-speculate-indirect-jumps.
1069 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1070 (*call_indirect_aix<mode>): Disable for
1071 -mno-speculate-indirect-jumps.
1072 (*call_indirect_aix<mode>_nospec): New define_insn.
1073 (*call_value_indirect_aix<mode>): Disable for
1074 -mno-speculate-indirect-jumps.
1075 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1076 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1077 -mno-speculate-indirect-jumps.
1078 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1080 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1082 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1083 long double type, set the flags for noting the default long double
1084 type, even if we don't pass or return a long double type.
1086 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1089 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1092 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1094 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1095 support for merge[hl].
1096 (fold_mergehl_helper): New helper function.
1097 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1098 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1099 (altivec_vmrglw_direct): Add xxmrglw insn.
1101 2018-01-17 Andrew Waterman <andrew@sifive.com>
1103 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1104 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1106 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1109 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1110 call the lto_location_cache before reading the
1111 DECL_SOURCE_LOCATION of the types.
1113 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1114 Richard Sandiford <richard.sandiford@linaro.org>
1116 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1117 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1118 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1119 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1120 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1122 * config/aarch64/constraints.md (aarch64_movti_operand):
1124 * config/aarch64/predicates.md (Uti): Add new constraint.
1126 2018-01-17 Carl Love <cel@us.ibm.com>
1127 * config/rs6000/vsx.md (define_expand xl_len_r,
1128 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1129 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1131 (define_expand, define_insn): Move the shift left from the
1132 define_insn to the define_expand for lxvl and stxvl instructions.
1133 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1134 and XL_LEN_R definitions to PURE.
1136 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1138 * config/i386/i386.c (indirect_thunk_name): Declare regno
1139 as unsigned int. Compare regno with INVALID_REGNUM.
1140 (output_indirect_thunk): Ditto.
1141 (output_indirect_thunk_function): Ditto.
1142 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1143 in the call to output_indirect_thunk_function.
1145 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1148 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1149 rather than the size of inner_type to determine the stack slot size
1150 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1152 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1155 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1158 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1160 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1161 endian Linux systems to optionally enable multilibs for selecting
1162 the long double type if the user configured an explicit type.
1163 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1164 have no long double multilibs if not defined.
1165 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1166 warn if the user used -mabi={ieee,ibm}longdouble and we built
1167 multilibs for long double.
1168 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1169 appropriate multilib option.
1170 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1172 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1173 for building long double multilibs.
1174 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1176 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1178 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1181 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1183 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1186 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1189 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1192 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1194 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1195 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1197 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1199 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1200 different rtl trees depending on TARGET_64BIT.
1201 (rs6000_gen_lvx): Likewise.
1203 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1205 * config/visium/visium.md (nop): Tweak comment.
1206 (hazard_nop): Likewise.
1208 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1210 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1211 -mspeculate-indirect-jumps.
1212 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1213 for -mno-speculate-indirect-jumps.
1214 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1215 (*call_value_indirect_elfv2<mode>): Disable for
1216 -mno-speculate-indirect-jumps.
1217 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1218 (indirect_jump): Emit different RTL for
1219 -mno-speculate-indirect-jumps.
1220 (*indirect_jump<mode>): Disable for
1221 -mno-speculate-indirect-jumps.
1222 (*indirect_jump<mode>_nospec): New define_insn.
1223 (tablejump): Emit different RTL for
1224 -mno-speculate-indirect-jumps.
1225 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1226 (tablejumpsi_nospec): New define_expand.
1227 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1228 (tablejumpdi_nospec): New define_expand.
1229 (*tablejump<mode>_internal1): Disable for
1230 -mno-speculate-indirect-jumps.
1231 (*tablejump<mode>_internal1_nospec): New define_insn.
1232 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1235 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1237 * caller-save.c (insert_save): Drop unnecessary parameter. All
1240 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1241 Richard Biener <rguenth@suse.de>
1244 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1245 return early, inline manually is_gimple_sizepos. Make sure if we
1246 call gimplify_expr we don't end up with a gimple constant.
1247 * tree.c (variably_modified_type_p): Don't return true for
1248 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1249 * gimplify.h (is_gimple_sizepos): Remove.
1251 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1253 PR tree-optimization/83857
1254 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1255 vectorizable_live_operation for pure SLP statements.
1256 (vectorizable_live_operation): Handle PHIs.
1258 2018-01-16 Richard Biener <rguenther@suse.de>
1260 PR tree-optimization/83867
1261 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1262 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1264 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1267 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1268 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1269 If off is not INTEGER_CST, issue a may not be aligned warning
1270 rather than isn't aligned. Use isn%'t rather than isn't.
1271 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1273 <case MULT_EXPR>: Improve the case when bottom and one of the
1274 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1275 operand, in that case check if the other operand is multiple of
1276 bottom divided by the INTEGER_CST operand.
1278 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1281 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1282 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1283 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1284 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1285 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1286 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1287 (pa_function_arg_size): New function.
1289 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1291 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1292 in a separate statement.
1294 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1296 PR tree-optimization/83847
1297 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1298 group gathers and scatters.
1300 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1302 PR rtl-optimization/86620
1303 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1305 PR rtl-optimization/83213
1306 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1307 to last if both are JUMP_INSNs.
1309 PR tree-optimization/83843
1310 * gimple-ssa-store-merging.c
1311 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1312 store_immediate_info for bswap/nop orig_stores.
1314 2018-01-15 Andrew Waterman <andrew@sifive.com>
1316 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1318 <UDIV>: Increase cost if !TARGET_DIV.
1320 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1322 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1323 (define_attr "cr_logical_3op"): New.
1324 (cceq_ior_compare): Adjust.
1325 (cceq_ior_compare_complement): Adjust.
1326 (*cceq_rev_compare): Adjust.
1327 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1328 (is_cracked_insn): Adjust.
1329 (insn_must_be_first_in_group): Adjust.
1330 * config/rs6000/40x.md: Adjust.
1331 * config/rs6000/440.md: Adjust.
1332 * config/rs6000/476.md: Adjust.
1333 * config/rs6000/601.md: Adjust.
1334 * config/rs6000/603.md: Adjust.
1335 * config/rs6000/6xx.md: Adjust.
1336 * config/rs6000/7450.md: Adjust.
1337 * config/rs6000/7xx.md: Adjust.
1338 * config/rs6000/8540.md: Adjust.
1339 * config/rs6000/cell.md: Adjust.
1340 * config/rs6000/e300c2c3.md: Adjust.
1341 * config/rs6000/e500mc.md: Adjust.
1342 * config/rs6000/e500mc64.md: Adjust.
1343 * config/rs6000/e5500.md: Adjust.
1344 * config/rs6000/e6500.md: Adjust.
1345 * config/rs6000/mpc.md: Adjust.
1346 * config/rs6000/power4.md: Adjust.
1347 * config/rs6000/power5.md: Adjust.
1348 * config/rs6000/power6.md: Adjust.
1349 * config/rs6000/power7.md: Adjust.
1350 * config/rs6000/power8.md: Adjust.
1351 * config/rs6000/power9.md: Adjust.
1352 * config/rs6000/rs64.md: Adjust.
1353 * config/rs6000/titan.md: Adjust.
1355 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1357 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1358 ix86_indirect_branch_register logic.
1360 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1362 * config/i386/constraints.md (Bs): Update
1363 ix86_indirect_branch_register check. Don't check
1364 ix86_indirect_branch_register with GOT_memory_operand.
1366 * config/i386/predicates.md (GOT_memory_operand): Don't check
1367 ix86_indirect_branch_register here.
1368 (GOT32_symbol_operand): Likewise.
1370 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1372 * config/i386/predicates.md (constant_call_address_operand):
1373 Rewrite ix86_indirect_branch_register logic.
1374 (sibcall_insn_operand): Likewise.
1376 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1378 * config/i386/constraints.md (Bs): Replace
1379 ix86_indirect_branch_thunk_register with
1380 ix86_indirect_branch_register.
1382 * config/i386/i386.md (indirect_jump): Likewise.
1383 (tablejump): Likewise.
1384 (*sibcall_memory): Likewise.
1385 (*sibcall_value_memory): Likewise.
1386 Peepholes of indirect call and jump via memory: Likewise.
1387 * config/i386/i386.opt: Likewise.
1388 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1389 (GOT_memory_operand): Likewise.
1390 (call_insn_operand): Likewise.
1391 (sibcall_insn_operand): Likewise.
1392 (GOT32_symbol_operand): Likewise.
1394 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1397 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1398 type rather than type addr's type points to.
1399 (expand_omp_atomic_mutex): Likewise.
1400 (expand_omp_atomic): Likewise.
1402 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1405 * config/i386/i386.c (output_indirect_thunk_function): Use
1406 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1407 for __x86_return_thunk.
1409 2018-01-15 Richard Biener <rguenther@suse.de>
1412 * expmed.c (extract_bit_field_1): Fix typo.
1414 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1417 * config/arm/iterators.md (VF): New mode iterator.
1418 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1419 Remove integer-related logic from pattern.
1420 (neon_vabd<mode>_3): Likewise.
1422 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1425 * common.opt (fstrict-overflow): No longer an alias.
1426 (fwrapv-pointer): New option.
1427 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1428 also for pointer types based on flag_wrapv_pointer.
1429 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1430 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1431 opts->x_flag_wrapv got set.
1432 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1433 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1434 POINTER_TYPE_OVERFLOW_UNDEFINED.
1435 * match.pd: Likewise in address comparison pattern.
1436 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1438 2018-01-15 Richard Biener <rguenther@suse.de>
1441 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1442 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1443 Reset type names to their identifier if their TYPE_DECL doesn't
1444 have linkage (and thus is used for ODR and devirt).
1445 (save_debug_info_for_decl): Remove.
1446 (save_debug_info_for_type): Likewise.
1447 (add_tree_to_fld_list): Adjust.
1448 * tree-pretty-print.c (dump_generic_node): Make dumping of
1449 type names more robust.
1451 2018-01-15 Richard Biener <rguenther@suse.de>
1453 * BASE-VER: Bump to 8.0.1.
1455 2018-01-14 Martin Sebor <msebor@redhat.com>
1458 * builtins.c (check_access): Avoid warning when the no-warning bit
1461 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1463 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1464 * ira-color (allocno_hard_regs_compare): Likewise.
1466 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1469 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1470 Use .pushsection/.popsection.
1472 2018-01-14 Martin Sebor <msebor@redhat.com>
1475 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1477 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1479 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1480 entry from extra_headers.
1481 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1482 extra_headers, make the list bitwise identical to the i?86-*-* one.
1484 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1486 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1487 -mcmodel=large with -mindirect-branch=thunk,
1488 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1489 -mfunction-return=thunk-extern.
1490 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1491 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1492 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1494 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1496 * config/i386/i386.c (print_reg): Print the name of the full
1497 integer register without '%'.
1498 (ix86_print_operand): Handle 'V'.
1499 * doc/extend.texi: Document 'V' modifier.
1501 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1503 * config/i386/constraints.md (Bs): Disallow memory operand for
1504 -mindirect-branch-register.
1506 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1507 (GOT_memory_operand): Likewise.
1508 (call_insn_operand): Likewise.
1509 (sibcall_insn_operand): Likewise.
1510 (GOT32_symbol_operand): Likewise.
1511 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1512 for -mindirect-branch-register.
1513 (tablejump): Likewise.
1514 (*sibcall_memory): Likewise.
1515 (*sibcall_value_memory): Likewise.
1516 Disallow peepholes of indirect call and jump via memory for
1517 -mindirect-branch-register.
1518 (*call_pop): Replace m with Bw.
1519 (*call_value_pop): Likewise.
1520 (*sibcall_pop_memory): Replace m with Bs.
1521 * config/i386/i386.opt (mindirect-branch-register): New option.
1522 * doc/invoke.texi: Document -mindirect-branch-register option.
1524 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1526 * config/i386/i386-protos.h (ix86_output_function_return): New.
1527 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1528 set function_return_type.
1529 (indirect_thunk_name): Add ret_p to indicate thunk for function
1531 (output_indirect_thunk_function): Pass false to
1532 indirect_thunk_name.
1533 (ix86_output_indirect_branch_via_reg): Likewise.
1534 (ix86_output_indirect_branch_via_push): Likewise.
1535 (output_indirect_thunk_function): Create alias for function
1536 return thunk if regno < 0.
1537 (ix86_output_function_return): New function.
1538 (ix86_handle_fndecl_attribute): Handle function_return.
1539 (ix86_attribute_table): Add function_return.
1540 * config/i386/i386.h (machine_function): Add
1541 function_return_type.
1542 * config/i386/i386.md (simple_return_internal): Use
1543 ix86_output_function_return.
1544 (simple_return_internal_long): Likewise.
1545 * config/i386/i386.opt (mfunction-return=): New option.
1546 (indirect_branch): Mention -mfunction-return=.
1547 * doc/extend.texi: Document function_return function attribute.
1548 * doc/invoke.texi: Document -mfunction-return= option.
1550 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1552 * config/i386/i386-opts.h (indirect_branch): New.
1553 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1554 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1555 with local indirect jump when converting indirect call and jump.
1556 (ix86_set_indirect_branch_type): New.
1557 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1558 (indirectlabelno): New.
1559 (indirect_thunk_needed): Likewise.
1560 (indirect_thunk_bnd_needed): Likewise.
1561 (indirect_thunks_used): Likewise.
1562 (indirect_thunks_bnd_used): Likewise.
1563 (INDIRECT_LABEL): Likewise.
1564 (indirect_thunk_name): Likewise.
1565 (output_indirect_thunk): Likewise.
1566 (output_indirect_thunk_function): Likewise.
1567 (ix86_output_indirect_branch_via_reg): Likewise.
1568 (ix86_output_indirect_branch_via_push): Likewise.
1569 (ix86_output_indirect_branch): Likewise.
1570 (ix86_output_indirect_jmp): Likewise.
1571 (ix86_code_end): Call output_indirect_thunk_function if needed.
1572 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1574 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1575 (ix86_attribute_table): Add indirect_branch.
1576 * config/i386/i386.h (machine_function): Add indirect_branch_type
1577 and has_local_indirect_jump.
1578 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1580 (tablejump): Likewise.
1581 (*indirect_jump): Use ix86_output_indirect_jmp.
1582 (*tablejump_1): Likewise.
1583 (simple_return_indirect_internal): Likewise.
1584 * config/i386/i386.opt (mindirect-branch=): New option.
1585 (indirect_branch): New.
1588 (thunk-inline): Likewise.
1589 (thunk-extern): Likewise.
1590 * doc/extend.texi: Document indirect_branch function attribute.
1591 * doc/invoke.texi: Document -mindirect-branch= option.
1593 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1596 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1598 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1600 * ipa-inline.c (want_inline_small_function_p): Return false if
1601 inlining has already failed with CIF_FINAL_ERROR.
1602 (update_caller_keys): Call want_inline_small_function_p before
1604 (update_callee_keys): Likewise.
1606 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1608 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1610 (rs6000_quadword_masked_address_p): Likewise.
1611 (quad_aligned_load_p): Likewise.
1612 (quad_aligned_store_p): Likewise.
1613 (const_load_sequence_p): Add comment to describe the outer-most loop.
1614 (mimic_memory_attributes_and_flags): New function.
1615 (rs6000_gen_stvx): Likewise.
1616 (replace_swapped_aligned_store): Likewise.
1617 (rs6000_gen_lvx): Likewise.
1618 (replace_swapped_aligned_load): Likewise.
1619 (replace_swapped_load_constant): Capitalize argument name in
1620 comment describing this function.
1621 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1622 and stores that access quad-word aligned addresses and replace
1623 with stvx or lvx instructions when appropriate.
1624 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1625 New function prototype.
1626 (rs6000_quadword_masked_address_p): Likewise.
1627 (rs6000_gen_lvx): Likewise.
1628 (rs6000_gen_stvx): Likewise.
1629 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1630 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1631 when memory address is aligned.
1632 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1633 this split to select lvx instruction when memory address is aligned.
1634 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1635 instruction when memory address is aligned.
1636 (*vsx_le_perm_load_v16qi): Likewise.
1637 (four unnamed splitters): Modify to select the stvx instruction
1638 when memory is aligned.
1640 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1642 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1643 which appears in the queue multiple times.
1645 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1646 Alan Hayward <alan.hayward@arm.com>
1647 David Sherwood <david.sherwood@arm.com>
1649 * tree-vectorizer.h (vec_lower_bound): New structure.
1650 (_loop_vec_info): Add check_nonzero and lower_bounds.
1651 (LOOP_VINFO_CHECK_NONZERO): New macro.
1652 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1653 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1654 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1655 fields. Make seg_len the distance travelled, not including the
1657 (dr_direction_indicator): Declare.
1658 (dr_zero_step_indicator): Likewise.
1659 (dr_known_forward_stride_p): Likewise.
1660 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1662 (runtime_alias_check_p): Allow runtime alias checks with
1664 (operator ==): Compare access_size and align.
1665 (prune_runtime_alias_test_list): Rework for new distinction between
1666 the access_size and seg_len.
1667 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1669 (get_segment_min_max): New function.
1670 (create_intersect_range_checks): Use it.
1671 (dr_step_indicator): New function.
1672 (dr_direction_indicator): Likewise.
1673 (dr_zero_step_indicator): Likewise.
1674 (dr_known_forward_stride_p): Likewise.
1675 * tree-loop-distribution.c (data_ref_segment_size): Return
1676 DR_STEP * (niters - 1).
1677 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1679 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1680 (vect_preserves_scalar_order_p): New function, split out from...
1681 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1682 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1683 (vect_vfa_access_size): New function.
1684 (vect_vfa_align): Likewise.
1685 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1686 (dump_lower_bound): New function.
1687 (vect_check_lower_bound): Likewise.
1688 (vect_small_gap_p): Likewise.
1689 (vectorizable_with_step_bound_p): Likewise.
1690 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1691 depencies if the vectorization factor is 1. Convert the checks
1692 for nonzero steps into checks on the bounds of DR_STEP. Try using
1693 a bunds check for variable steps if the minimum required step is
1694 relatively small. Update calls to the dr_with_seg_len
1695 constructor and to vect_compile_time_alias.
1696 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1698 (vect_loop_versioning): Call it.
1699 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1701 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1704 Alan Hayward <alan.hayward@arm.com>
1705 David Sherwood <david.sherwood@arm.com>
1707 * doc/sourcebuild.texi (vect_scatter_store): Document.
1708 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1710 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1712 * genopinit.c (main): Add supports_vec_scatter_store and
1713 supports_vec_scatter_store_cached to target_optabs.
1714 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1715 IFN_MASK_SCATTER_STORE.
1716 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1718 * internal-fn.h (internal_store_fn_p): Declare.
1719 (internal_fn_stored_value_index): Likewise.
1720 * internal-fn.c (scatter_store_direct): New macro.
1721 (expand_scatter_store_optab_fn): New function.
1722 (direct_scatter_store_optab_supported_p): New macro.
1723 (internal_store_fn_p): New function.
1724 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1725 IFN_MASK_SCATTER_STORE.
1726 (internal_fn_mask_index): Likewise.
1727 (internal_fn_stored_value_index): New function.
1728 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1730 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1731 * optabs-query.c (supports_vec_scatter_store_p): New function.
1732 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1733 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1734 true for scatter stores.
1735 (vect_gather_scatter_fn_p): Handle scatter stores too.
1736 (vect_check_gather_scatter): Consider using scatter stores if
1737 supports_vec_scatter_store_p.
1738 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1740 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1741 internal_fn_stored_value_index.
1742 (check_load_store_masking): Handle scatter stores too.
1743 (vect_get_store_rhs): Make public.
1744 (vectorizable_call): Use internal_store_fn_p.
1745 (vectorizable_store): Handle scatter store internal functions.
1746 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1747 when deciding whether the end of the group has been reached.
1748 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1749 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1750 (mask_scatter_store<mode>): New insns.
1752 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1753 Alan Hayward <alan.hayward@arm.com>
1754 David Sherwood <david.sherwood@arm.com>
1756 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1757 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1758 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1760 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1761 Use vect_truncate_gather_scatter_offset if we can't treat the
1762 operation as a normal gather load or scatter store.
1763 (get_group_load_store_type): Take the gather_scatter_info
1764 as argument. Try using a gather load or scatter store for
1765 single-element groups.
1766 (get_load_store_type): Update calls to get_group_load_store_type
1767 and vect_use_strided_gather_scatters_p.
1769 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1770 Alan Hayward <alan.hayward@arm.com>
1771 David Sherwood <david.sherwood@arm.com>
1773 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1774 optional tree argument.
1775 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1777 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1778 but continue to use the current value as a fallback.
1779 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1780 to compare the updates.
1781 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1782 (get_load_store_type): Use it when handling a strided access.
1783 (vect_get_strided_load_store_ops): New function.
1784 (vect_get_data_ptr_increment): Likewise.
1785 (vectorizable_load): Handle strided gather loads. Always pass
1786 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1788 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1789 Alan Hayward <alan.hayward@arm.com>
1790 David Sherwood <david.sherwood@arm.com>
1792 * doc/md.texi (gather_load@var{m}): Document.
1793 (mask_gather_load@var{m}): Likewise.
1794 * genopinit.c (main): Add supports_vec_gather_load and
1795 supports_vec_gather_load_cached to target_optabs.
1796 * optabs-tree.c (init_tree_optimization_optabs): Use
1797 ggc_cleared_alloc to allocate target_optabs.
1798 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1799 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1801 * internal-fn.h (internal_load_fn_p): Declare.
1802 (internal_gather_scatter_fn_p): Likewise.
1803 (internal_fn_mask_index): Likewise.
1804 (internal_gather_scatter_fn_supported_p): Likewise.
1805 * internal-fn.c (gather_load_direct): New macro.
1806 (expand_gather_load_optab_fn): New function.
1807 (direct_gather_load_optab_supported_p): New macro.
1808 (direct_internal_fn_optab): New function.
1809 (internal_load_fn_p): Likewise.
1810 (internal_gather_scatter_fn_p): Likewise.
1811 (internal_fn_mask_index): Likewise.
1812 (internal_gather_scatter_fn_supported_p): Likewise.
1813 * optabs-query.c (supports_at_least_one_mode_p): New function.
1814 (supports_vec_gather_load_p): Likewise.
1815 * optabs-query.h (supports_vec_gather_load_p): Declare.
1816 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1817 and memory_type field.
1818 (NUM_PATTERNS): Bump to 15.
1819 * tree-vect-data-refs.c: Include internal-fn.h.
1820 (vect_gather_scatter_fn_p): New function.
1821 (vect_describe_gather_scatter_call): Likewise.
1822 (vect_check_gather_scatter): Try using internal functions for
1823 gather loads. Recognize existing calls to a gather load function.
1824 (vect_analyze_data_refs): Consider using gather loads if
1825 supports_vec_gather_load_p.
1826 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1827 (vect_get_gather_scatter_offset_type): Likewise.
1828 (vect_convert_mask_for_vectype): Likewise.
1829 (vect_add_conversion_to_patterm): Likewise.
1830 (vect_try_gather_scatter_pattern): Likewise.
1831 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1832 (vect_vect_recog_func_ptrs): Add it.
1833 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1834 internal_fn_mask_index and internal_gather_scatter_fn_p.
1835 (check_load_store_masking): Take the gather_scatter_info as an
1836 argument and handle gather loads.
1837 (vect_get_gather_scatter_ops): New function.
1838 (vectorizable_call): Check internal_load_fn_p.
1839 (vectorizable_load): Likewise. Handle gather load internal
1841 (vectorizable_store): Update call to check_load_store_masking.
1842 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1843 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1844 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1845 (aarch64_gather_scale_operand_d): New predicates.
1846 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1847 (mask_gather_load<mode>): New insns.
1849 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1850 Alan Hayward <alan.hayward@arm.com>
1851 David Sherwood <david.sherwood@arm.com>
1853 * optabs.def (fold_left_plus_optab): New optab.
1854 * doc/md.texi (fold_left_plus_@var{m}): Document.
1855 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1856 * internal-fn.c (fold_left_direct): Define.
1857 (expand_fold_left_optab_fn): Likewise.
1858 (direct_fold_left_optab_supported_p): Likewise.
1859 * fold-const-call.c (fold_const_fold_left): New function.
1860 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1861 * tree-parloops.c (valid_reduction_p): New function.
1862 (gather_scalar_reductions): Use it.
1863 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1864 (vect_finish_replace_stmt): Declare.
1865 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1866 (needs_fold_left_reduction_p): New function, split out from...
1867 (vect_is_simple_reduction): ...here. Accept reductions that
1868 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1869 (vect_force_simple_reduction): Also store the reduction type in
1870 the assignment's STMT_VINFO_REDUC_TYPE.
1871 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1872 (merge_with_identity): New function.
1873 (vect_expand_fold_left): Likewise.
1874 (vectorize_fold_left_reduction): Likewise.
1875 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1876 scalar phi in place for it. Check for target support and reject
1877 cases that would reassociate the operation. Defer the transform
1878 phase to vectorize_fold_left_reduction.
1879 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1880 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1881 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1883 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1885 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1886 call to ifc_temp_var.
1888 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1889 Alan Hayward <alan.hayward@arm.com>
1890 David Sherwood <david.sherwood@arm.com>
1892 * target.def (legitimize_address_displacement): Take the original
1893 offset as a poly_int.
1894 * targhooks.h (default_legitimize_address_displacement): Update
1896 * targhooks.c (default_legitimize_address_displacement): Likewise.
1897 * doc/tm.texi: Regenerate.
1898 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1899 as an argument, moving assert of ad->disp == ad->disp_term to...
1900 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1901 Try calling targetm.legitimize_address_displacement before expanding
1902 the address rather than afterwards, and adjust for the new interface.
1903 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1904 Match the new hook interface. Handle SVE addresses.
1905 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1908 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1910 * Makefile.in (OBJS): Add early-remat.o.
1911 * target.def (select_early_remat_modes): New hook.
1912 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1913 * doc/tm.texi: Regenerate.
1914 * targhooks.h (default_select_early_remat_modes): Declare.
1915 * targhooks.c (default_select_early_remat_modes): New function.
1916 * timevar.def (TV_EARLY_REMAT): New timevar.
1917 * passes.def (pass_early_remat): New pass.
1918 * tree-pass.h (make_pass_early_remat): Declare.
1919 * early-remat.c: New file.
1920 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1922 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1924 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1925 Alan Hayward <alan.hayward@arm.com>
1926 David Sherwood <david.sherwood@arm.com>
1928 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1929 vfm1 with a bound_epilog parameter.
1930 (vect_do_peeling): Update calls accordingly, and move the prologue
1931 call earlier in the function. Treat the base bound_epilog as 0 for
1932 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1933 this base when peeling for gaps.
1934 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1935 with fully-masked loops.
1936 (vect_estimate_min_profitable_iters): Handle the single peeled
1937 iteration in that case.
1939 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1940 Alan Hayward <alan.hayward@arm.com>
1941 David Sherwood <david.sherwood@arm.com>
1943 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1944 single-element interleaving even if the size is not a power of 2.
1945 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1946 accesses for single-element interleaving if the group size is
1949 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1950 Alan Hayward <alan.hayward@arm.com>
1951 David Sherwood <david.sherwood@arm.com>
1953 * doc/md.texi (fold_extract_last_@var{m}): Document.
1954 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1955 * optabs.def (fold_extract_last_optab): New optab.
1956 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1957 * internal-fn.c (fold_extract_direct): New macro.
1958 (expand_fold_extract_optab_fn): Likewise.
1959 (direct_fold_extract_optab_supported_p): Likewise.
1960 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1961 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1962 EXTRACT_LAST_REDUCTION.
1963 (get_initial_def_for_reduction): Do not create an initial vector
1964 for EXTRACT_LAST_REDUCTION reductions.
1965 (vectorizable_reduction): Leave the scalar phi in place for
1966 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1967 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1968 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1969 transform phase to vectorizable_condition.
1970 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1972 (vect_finish_stmt_generation): ...here.
1973 (vect_finish_replace_stmt): New function.
1974 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1975 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1977 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1979 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1980 Alan Hayward <alan.hayward@arm.com>
1981 David Sherwood <david.sherwood@arm.com>
1983 * doc/md.texi (extract_last_@var{m}): Document.
1984 * optabs.def (extract_last_optab): New optab.
1985 * internal-fn.def (EXTRACT_LAST): New internal function.
1986 * internal-fn.c (cond_unary_direct): New macro.
1987 (expand_cond_unary_optab_fn): Likewise.
1988 (direct_cond_unary_optab_supported_p): Likewise.
1989 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1990 loops using EXTRACT_LAST.
1991 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1992 (extract_last_<mode>): ...this optab.
1993 (vec_extract<mode><Vel>): Update accordingly.
1995 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1996 Alan Hayward <alan.hayward@arm.com>
1997 David Sherwood <david.sherwood@arm.com>
1999 * target.def (empty_mask_is_expensive): New hook.
2000 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2001 * doc/tm.texi: Regenerate.
2002 * targhooks.h (default_empty_mask_is_expensive): Declare.
2003 * targhooks.c (default_empty_mask_is_expensive): New function.
2004 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2005 if the target says that empty masks are expensive.
2006 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2008 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2010 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2011 Alan Hayward <alan.hayward@arm.com>
2012 David Sherwood <david.sherwood@arm.com>
2014 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2015 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2016 (vect_use_loop_mask_for_alignment_p): New function.
2017 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2018 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2019 niters_skip argument. Make sure that the first niters_skip elements
2020 of the first iteration are inactive.
2021 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2022 Update call to vect_set_loop_masks_directly.
2023 (get_misalign_in_elems): New function, split out from...
2024 (vect_gen_prolog_loop_niters): ...here.
2025 (vect_update_init_of_dr): Take a code argument that specifies whether
2026 the adjustment should be added or subtracted.
2027 (vect_update_init_of_drs): Likewise.
2028 (vect_prepare_for_masked_peels): New function.
2029 (vect_do_peeling): Skip prologue peeling if we're using a mask
2030 instead. Update call to vect_update_inits_of_drs.
2031 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2033 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2034 alignment. Do not include the number of peeled iterations in
2035 the minimum threshold in that case.
2036 (vectorizable_induction): Adjust the start value down by
2037 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2038 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2039 Take the number of skipped iterations into account when calculating
2041 * tree-vect-stmts.c (vect_gen_while_not): New function.
2043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2044 Alan Hayward <alan.hayward@arm.com>
2045 David Sherwood <david.sherwood@arm.com>
2047 * doc/sourcebuild.texi (vect_fully_masked): Document.
2048 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2050 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2052 (vect_analyze_loop_2): ...here. Don't check the vectorization
2053 factor against the number of loop iterations if the loop is
2056 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2057 Alan Hayward <alan.hayward@arm.com>
2058 David Sherwood <david.sherwood@arm.com>
2060 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2061 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2062 (dump_groups): Update accordingly.
2063 (iv_use::mem_type): New member variable.
2064 (address_p): New function.
2065 (record_use): Add a mem_type argument and initialize the new
2067 (record_group_use): Add a mem_type argument. Use address_p.
2068 Remove obsolete null checks of base_object. Update call to record_use.
2069 (find_interesting_uses_op): Update call to record_group_use.
2070 (find_interesting_uses_cond): Likewise.
2071 (find_interesting_uses_address): Likewise.
2072 (get_mem_type_for_internal_fn): New function.
2073 (find_address_like_use): Likewise.
2074 (find_interesting_uses_stmt): Try find_address_like_use before
2075 calling find_interesting_uses_op.
2076 (addr_offset_valid_p): Use the iv mem_type field as the type
2077 of the addressed memory.
2078 (add_autoinc_candidates): Likewise.
2079 (get_address_cost): Likewise.
2080 (split_small_address_groups_p): Use address_p.
2081 (split_address_groups): Likewise.
2082 (add_iv_candidate_for_use): Likewise.
2083 (autoinc_possible_for_pair): Likewise.
2084 (rewrite_groups): Likewise.
2085 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2086 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2087 (get_alias_ptr_type_for_ptr_address): New function.
2088 (rewrite_use_address): Rewrite address uses in calls that were
2089 identified by find_address_like_use.
2091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2092 Alan Hayward <alan.hayward@arm.com>
2093 David Sherwood <david.sherwood@arm.com>
2095 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2097 * gimple-expr.h (is_gimple_addressable: Likewise.
2098 * gimple-expr.c (is_gimple_address): Likewise.
2099 * internal-fn.c (expand_call_mem_ref): New function.
2100 (expand_mask_load_optab_fn): Use it.
2101 (expand_mask_store_optab_fn): Likewise.
2103 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2104 Alan Hayward <alan.hayward@arm.com>
2105 David Sherwood <david.sherwood@arm.com>
2107 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2108 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2109 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2110 (cond_umax@var{mode}): Document.
2111 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2112 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2113 (cond_umin_optab, cond_umax_optab): New optabs.
2114 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2115 (COND_IOR, COND_XOR): New internal functions.
2116 * internal-fn.h (get_conditional_internal_fn): Declare.
2117 * internal-fn.c (cond_binary_direct): New macro.
2118 (expand_cond_binary_optab_fn): Likewise.
2119 (direct_cond_binary_optab_supported_p): Likewise.
2120 (get_conditional_internal_fn): New function.
2121 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2122 Cope with reduction statements that are vectorized as calls rather
2124 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2125 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2126 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2127 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2128 (UNSPEC_COND_EOR): New unspecs.
2129 (optab): Add mappings for them.
2130 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2131 (sve_int_op, sve_fp_op): New int attributes.
2133 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2134 Alan Hayward <alan.hayward@arm.com>
2135 David Sherwood <david.sherwood@arm.com>
2137 * optabs.def (while_ult_optab): New optab.
2138 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2139 * internal-fn.def (WHILE_ULT): New internal function.
2140 * internal-fn.h (direct_internal_fn_supported_p): New override
2141 that takes two types as argument.
2142 * internal-fn.c (while_direct): New macro.
2143 (expand_while_optab_fn): New function.
2144 (convert_optab_supported_p): Likewise.
2145 (direct_while_optab_supported_p): New macro.
2146 * wide-int.h (wi::udiv_ceil): New function.
2147 * tree-vectorizer.h (rgroup_masks): New structure.
2148 (vec_loop_masks): New typedef.
2149 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2151 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2152 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2153 (vect_max_vf): New function.
2154 (slpeel_make_loop_iterate_ntimes): Delete.
2155 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2156 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2157 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2158 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2159 internal-fn.h, stor-layout.h and optabs-query.h.
2160 (vect_set_loop_mask): New function.
2161 (add_preheader_seq): Likewise.
2162 (add_header_seq): Likewise.
2163 (interleave_supported_p): Likewise.
2164 (vect_maybe_permute_loop_masks): Likewise.
2165 (vect_set_loop_masks_directly): Likewise.
2166 (vect_set_loop_condition_masked): Likewise.
2167 (vect_set_loop_condition_unmasked): New function, split out from
2168 slpeel_make_loop_iterate_ntimes.
2169 (slpeel_make_loop_iterate_ntimes): Rename to..
2170 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2171 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2172 (vect_do_peeling): Update call accordingly.
2173 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2175 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2176 mask_compare_type, can_fully_mask_p and fully_masked_p.
2177 (release_vec_loop_masks): New function.
2178 (_loop_vec_info): Use it to free the loop masks.
2179 (can_produce_all_loop_masks_p): New function.
2180 (vect_get_max_nscalars_per_iter): Likewise.
2181 (vect_verify_full_masking): Likewise.
2182 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2183 retries, and free the mask rgroups before retrying. Check loop-wide
2184 reasons for disallowing fully-masked loops. Make the final decision
2185 about whether use a fully-masked loop or not.
2186 (vect_estimate_min_profitable_iters): Do not assume that peeling
2187 for the number of iterations will be needed for fully-masked loops.
2188 (vectorizable_reduction): Disable fully-masked loops.
2189 (vectorizable_live_operation): Likewise.
2190 (vect_halve_mask_nunits): New function.
2191 (vect_double_mask_nunits): Likewise.
2192 (vect_record_loop_mask): Likewise.
2193 (vect_get_loop_mask): Likewise.
2194 (vect_transform_loop): Handle the case in which the final loop
2195 iteration might handle a partial vector. Call vect_set_loop_condition
2196 instead of slpeel_make_loop_iterate_ntimes.
2197 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2198 (check_load_store_masking): New function.
2199 (prepare_load_store_mask): Likewise.
2200 (vectorizable_store): Handle fully-masked loops.
2201 (vectorizable_load): Likewise.
2202 (supportable_widening_operation): Use vect_halve_mask_nunits for
2204 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2205 (vect_gen_while): New function.
2206 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2207 (aarch64_uqdec<mode>): New insn.
2209 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2210 Alan Hayward <alan.hayward@arm.com>
2211 David Sherwood <david.sherwood@arm.com>
2213 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2214 (reduc_xor_scal_optab): New optabs.
2215 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2216 (reduc_xor_scal_@var{m}): Document.
2217 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2218 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2220 * fold-const-call.c (fold_const_call): Handle them.
2221 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2222 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2223 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2224 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2225 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2226 (UNSPEC_XORV): New unspecs.
2227 (optab): Add entries for them.
2228 (BITWISEV): New int iterator.
2229 (bit_reduc_op): New int attributes.
2231 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2232 Alan Hayward <alan.hayward@arm.com>
2233 David Sherwood <david.sherwood@arm.com>
2235 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2236 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2237 * optabs.def (vec_shl_insert_optab): New optab.
2238 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2239 (duplicate_and_interleave): Likewise.
2240 * tree-vect-loop.c: Include internal-fn.h.
2241 (neutral_op_for_slp_reduction): New function, split out from
2242 get_initial_defs_for_reduction.
2243 (get_initial_def_for_reduction): Handle option 2 for variable-length
2244 vectors by loading the neutral value into a vector and then shifting
2245 the initial value into element 0.
2246 (get_initial_defs_for_reduction): Replace the code argument with
2247 the neutral value calculated by neutral_op_for_slp_reduction.
2248 Use gimple_build_vector for constant-length vectors.
2249 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2250 but the first group_size elements have a neutral value.
2251 Use duplicate_and_interleave otherwise.
2252 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2253 Update call to get_initial_defs_for_reduction. Handle SLP
2254 reductions for variable-length vectors by creating one vector
2255 result for each scalar result, with the elements associated
2256 with other scalar results stubbed out with the neutral value.
2257 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2258 Require IFN_VEC_SHL_INSERT for double reductions on
2259 variable-length vectors, or SLP reductions that have
2260 a neutral value. Require can_duplicate_and_interleave_p
2261 support for variable-length unchained SLP reductions if there
2262 is no neutral value, such as for MIN/MAX reductions. Also require
2263 the number of vector elements to be a multiple of the number of
2264 SLP statements when doing variable-length unchained SLP reductions.
2265 Update call to vect_create_epilog_for_reduction.
2266 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2267 and remove initial values.
2268 (duplicate_and_interleave): Make public.
2269 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2270 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2272 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2273 Alan Hayward <alan.hayward@arm.com>
2274 David Sherwood <david.sherwood@arm.com>
2276 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2277 (can_duplicate_and_interleave_p): New function.
2278 (vect_get_and_check_slp_defs): Take the vector of statements
2279 rather than just the current one. Remove excess parentheses.
2280 Restriction rejectinon of vect_constant_def and vect_external_def
2281 for variable-length vectors to boolean types, or types for which
2282 can_duplicate_and_interleave_p is false.
2283 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2284 (duplicate_and_interleave): New function.
2285 (vect_get_constant_vectors): Use gimple_build_vector for
2286 constant-length vectors and suitable variable-length constant
2287 vectors. Use duplicate_and_interleave for other variable-length
2288 vectors. Don't defer the update when inserting new statements.
2290 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2291 Alan Hayward <alan.hayward@arm.com>
2292 David Sherwood <david.sherwood@arm.com>
2294 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2295 min_profitable_iters doesn't go negative.
2297 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2298 Alan Hayward <alan.hayward@arm.com>
2299 David Sherwood <david.sherwood@arm.com>
2301 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2302 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2303 * optabs.def (vec_mask_load_lanes_optab): New optab.
2304 (vec_mask_store_lanes_optab): Likewise.
2305 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2306 (MASK_STORE_LANES): Likewise.
2307 * internal-fn.c (mask_load_lanes_direct): New macro.
2308 (mask_store_lanes_direct): Likewise.
2309 (expand_mask_load_optab_fn): Handle masked operations.
2310 (expand_mask_load_lanes_optab_fn): New macro.
2311 (expand_mask_store_optab_fn): Handle masked operations.
2312 (expand_mask_store_lanes_optab_fn): New macro.
2313 (direct_mask_load_lanes_optab_supported_p): Likewise.
2314 (direct_mask_store_lanes_optab_supported_p): Likewise.
2315 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2317 (vect_load_lanes_supported): Likewise.
2318 * tree-vect-data-refs.c (strip_conversion): New function.
2319 (can_group_stmts_p): Likewise.
2320 (vect_analyze_data_ref_accesses): Use it instead of checking
2321 for a pair of assignments.
2322 (vect_store_lanes_supported): Take a masked_p parameter.
2323 (vect_load_lanes_supported): Likewise.
2324 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2325 vect_store_lanes_supported and vect_load_lanes_supported.
2326 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2327 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2328 parameter. Don't allow gaps for masked accesses.
2329 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2330 and vect_load_lanes_supported.
2331 (get_load_store_type): Take a masked_p parameter and update
2332 call to get_group_load_store_type.
2333 (vectorizable_store): Update call to get_load_store_type.
2334 Handle IFN_MASK_STORE_LANES.
2335 (vectorizable_load): Update call to get_load_store_type.
2336 Handle IFN_MASK_LOAD_LANES.
2338 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2339 Alan Hayward <alan.hayward@arm.com>
2340 David Sherwood <david.sherwood@arm.com>
2342 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2344 * config/aarch64/aarch64-protos.h
2345 (aarch64_sve_struct_memory_operand_p): Declare.
2346 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2347 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2348 (VPRED, vpred): Handle SVE structure modes.
2349 * config/aarch64/constraints.md (Utx): New constraint.
2350 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2351 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2352 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2353 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2354 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2355 structure modes. Split into pieces after RA.
2356 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2357 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2359 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2360 SVE structure modes.
2361 (aarch64_classify_address): Likewise.
2362 (sizetochar): Move earlier in file.
2363 (aarch64_print_operand): Handle SVE register lists.
2364 (aarch64_array_mode): New function.
2365 (aarch64_sve_struct_memory_operand_p): Likewise.
2366 (TARGET_ARRAY_MODE): Redefine.
2368 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2369 Alan Hayward <alan.hayward@arm.com>
2370 David Sherwood <david.sherwood@arm.com>
2372 * target.def (array_mode): New target hook.
2373 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2374 * doc/tm.texi: Regenerate.
2375 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2376 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2377 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2379 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2382 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2383 Alan Hayward <alan.hayward@arm.com>
2384 David Sherwood <david.sherwood@arm.com>
2386 * fold-const.c (fold_binary_loc): Check the argument types
2387 rather than the result type when testing for a vector operation.
2389 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2391 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2392 * doc/tm.texi: Regenerate.
2394 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2395 Alan Hayward <alan.hayward@arm.com>
2396 David Sherwood <david.sherwood@arm.com>
2398 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2399 (sve): Document new AArch64 extension.
2400 * doc/md.texi (w): Extend the description of the AArch64
2401 constraint to include SVE vectors.
2402 (Upl, Upa): Document new AArch64 predicate constraints.
2403 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2405 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2406 (msve-vector-bits=): New option.
2407 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2408 SVE when these are disabled.
2409 (sve): New extension.
2410 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2411 modes. Adjust their number of units based on aarch64_sve_vg.
2412 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2413 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2414 aarch64_addr_query_type.
2415 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2416 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2417 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2418 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2419 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2420 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2421 (aarch64_simd_imm_zero_p): Delete.
2422 (aarch64_check_zero_based_sve_index_immediate): Declare.
2423 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2424 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2425 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2426 (aarch64_sve_float_mul_immediate_p): Likewise.
2427 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2429 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2430 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2431 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2432 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2433 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2434 (aarch64_regmode_natural_size): Likewise.
2435 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2436 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2438 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2439 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2440 for VG and the SVE predicate registers.
2441 (V_ALIASES): Add a "z"-prefixed alias.
2442 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2443 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2444 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2445 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2446 (REG_CLASS_NAMES): Add entries for them.
2447 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2448 and the predicate registers.
2449 (aarch64_sve_vg): Declare.
2450 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2451 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2452 (REGMODE_NATURAL_SIZE): Define.
2453 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2455 * config/aarch64/aarch64.c: Include cfgrtl.h.
2456 (simd_immediate_info): Add a constructor for series vectors,
2457 and an associated step field.
2458 (aarch64_sve_vg): New variable.
2459 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2460 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2461 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2462 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2463 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2464 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2465 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2466 (aarch64_get_mask_mode): New functions.
2467 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2468 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2469 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2470 predicate modes and predicate registers. Explicitly restrict
2471 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2472 to store a vector mode if it is recognized by
2473 aarch64_classify_vector_mode.
2474 (aarch64_regmode_natural_size): New function.
2475 (aarch64_hard_regno_caller_save_mode): Return the original mode
2477 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2478 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2479 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2480 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2482 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2483 does not overlap dest if the function is frame-related. Handle
2485 (aarch64_split_add_offset): New function.
2486 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2487 them aarch64_add_offset.
2488 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2489 and update call to aarch64_sub_sp.
2490 (aarch64_add_cfa_expression): New function.
2491 (aarch64_expand_prologue): Pass extra temporary registers to the
2492 functions above. Handle the case in which we need to emit new
2493 DW_CFA_expressions for registers that were originally saved
2494 relative to the stack pointer, but now have to be expressed
2495 relative to the frame pointer.
2496 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2498 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2499 IP0 and IP1 values for SVE frames.
2500 (aarch64_expand_vec_series): New function.
2501 (aarch64_expand_sve_widened_duplicate): Likewise.
2502 (aarch64_expand_sve_const_vector): Likewise.
2503 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2504 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2505 into the register, rather than emitting a SET directly.
2506 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2507 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2508 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2509 (offset_9bit_signed_scaled_p): New functions.
2510 (aarch64_replicate_bitmask_imm): New function.
2511 (aarch64_bitmask_imm): Use it.
2512 (aarch64_cannot_force_const_mem): Reject expressions involving
2513 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2514 (aarch64_classify_index): Handle SVE indices, by requiring
2515 a plain register index with a scale that matches the element size.
2516 (aarch64_classify_address): Handle SVE addresses. Assert that
2517 the mode of the address is VOIDmode or an integer mode.
2518 Update call to aarch64_classify_symbol.
2519 (aarch64_classify_symbolic_expression): Update call to
2520 aarch64_classify_symbol.
2521 (aarch64_const_vec_all_in_range_p): New function.
2522 (aarch64_print_vector_float_operand): Likewise.
2523 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2524 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2525 and the FP immediates 1.0 and 0.5.
2526 (aarch64_print_address_internal): Handle SVE addresses.
2527 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2528 (aarch64_regno_regclass): Handle predicate registers.
2529 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2531 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2532 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2533 (aarch64_convert_sve_vector_bits): New function.
2534 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2535 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2537 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2538 Handle SVE vector and predicate modes. Accept VL-based constants
2539 that need only one temporary register, and VL offsets that require
2540 no temporary registers.
2541 (aarch64_conditional_register_usage): Mark the predicate registers
2542 as fixed if SVE isn't available.
2543 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2544 Return true for SVE vector and predicate modes.
2545 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2546 rather than an unsigned int. Handle SVE modes.
2547 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2549 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2551 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2552 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2553 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2554 (aarch64_sve_float_mul_immediate_p): New functions.
2555 (aarch64_sve_valid_immediate): New function.
2556 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2557 Explicitly reject structure modes. Check for INDEX constants.
2558 Handle PTRUE and PFALSE constants.
2559 (aarch64_check_zero_based_sve_index_immediate): New function.
2560 (aarch64_simd_imm_zero_p): Delete.
2561 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2562 vector modes. Accept constants in the range of CNT[BHWD].
2563 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2564 ask for an Advanced SIMD mode.
2565 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2566 (aarch64_simd_vector_alignment): Handle SVE predicates.
2567 (aarch64_vectorize_preferred_vector_alignment): New function.
2568 (aarch64_simd_vector_alignment_reachable): Use it instead of
2570 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2571 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2573 (MAX_VECT_LEN): Delete.
2574 (expand_vec_perm_d): Add a vec_flags field.
2575 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2576 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2577 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2579 (aarch64_evpc_rev): Rename to...
2580 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2581 (aarch64_evpc_rev_global): New function.
2582 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2583 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2585 (aarch64_evpc_sve_tbl): New function.
2586 (aarch64_expand_vec_perm_const_1): Update after rename of
2587 aarch64_evpc_rev. Handle SVE permutes too, trying
2588 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2589 than aarch64_evpc_tbl.
2590 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2591 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2592 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2593 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2594 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2595 (aarch64_expand_sve_vcond): New functions.
2596 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2597 of aarch64_vector_mode_p.
2598 (aarch64_dwarf_poly_indeterminate_value): New function.
2599 (aarch64_compute_pressure_classes): Likewise.
2600 (aarch64_can_change_mode_class): Likewise.
2601 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2602 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2603 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2604 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2605 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2606 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2607 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2608 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2610 (Dn, Dl, Dr): Accept const as well as const_vector.
2611 (Dz): Likewise. Compare against CONST0_RTX.
2612 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2613 of "vector" where appropriate.
2614 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2615 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2616 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2617 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2618 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2619 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2620 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2621 (v_int_equiv): Extend to SVE modes.
2622 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2624 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2625 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2626 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2627 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2628 (SVE_COND_FP_CMP): New int iterators.
2629 (perm_hilo): Handle the new unpack unspecs.
2630 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2632 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2633 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2634 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2635 (aarch64_equality_operator, aarch64_constant_vector_operand)
2636 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2637 (aarch64_sve_nonimmediate_operand): Likewise.
2638 (aarch64_sve_general_operand): Likewise.
2639 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2640 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2641 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2642 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2643 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2644 (aarch64_sve_float_arith_immediate): Likewise.
2645 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2646 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2647 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2648 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2649 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2650 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2651 (aarch64_sve_float_arith_operand): Likewise.
2652 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2653 (aarch64_sve_float_mul_operand): Likewise.
2654 (aarch64_sve_vec_perm_operand): Likewise.
2655 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2656 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2657 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2658 as well as const_vector.
2659 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2660 in file. Use CONST0_RTX and CONSTM1_RTX.
2661 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2662 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2663 Use aarch64_simd_imm_zero.
2664 * config/aarch64/aarch64-sve.md: New file.
2665 * config/aarch64/aarch64.md: Include it.
2666 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2667 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2668 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2669 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2670 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2671 (sve): New attribute.
2672 (enabled): Disable instructions with the sve attribute unless
2674 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2675 aarch64_expand_mov_immediate.
2676 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2677 CNT[BHSD] immediates.
2678 (movti): Split CONST_POLY_INT moves into two halves.
2679 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2680 Split additions that need a temporary here if the destination
2681 is the stack pointer.
2682 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2683 (*add<mode>3_poly_1): New instruction.
2684 (set_clobber_cc): New expander.
2686 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2688 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2689 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2690 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2691 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2692 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2693 Change innermode from fixed_mode_size to machine_mode.
2694 (simplify_subreg): Update call accordingly. Handle a constant-sized
2695 subreg of a variable-length CONST_VECTOR.
2697 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2698 Alan Hayward <alan.hayward@arm.com>
2699 David Sherwood <david.sherwood@arm.com>
2701 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2702 (add_offset_to_base): New function, split out from...
2703 (create_mem_ref): ...here. When handling a scale other than 1,
2704 check first whether the address is valid without the offset.
2705 Add it into the base if so, leaving the index and scale as-is.
2707 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2710 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2711 fold_for_warn before checking if arg2 is INTEGER_CST.
2713 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2715 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2716 (store_multiple_operation): Delete.
2717 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2718 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2719 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2720 guarded by TARGET_STRING.
2721 (rs6000_output_load_multiple): Delete.
2722 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2723 OPTION_MASK_STRING / TARGET_STRING handling.
2724 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2725 (const rs6000_opt_masks) <"string">: Change mask to 0.
2726 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2727 (MASK_STRING): Delete.
2728 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2730 (load_multiple): Delete.
2737 (store_multiple): Delete.
2744 (movmemsi_8reg): Delete.
2745 (corresponding unnamed define_insn): Delete.
2746 (movmemsi_6reg): Delete.
2747 (corresponding unnamed define_insn): Delete.
2748 (movmemsi_4reg): Delete.
2749 (corresponding unnamed define_insn): Delete.
2750 (movmemsi_2reg): Delete.
2751 (corresponding unnamed define_insn): Delete.
2752 (movmemsi_1reg): Delete.
2753 (corresponding unnamed define_insn): Delete.
2754 * config/rs6000/rs6000.opt (mno-string): New.
2755 (mstring): Replace by deprecation warning stub.
2756 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2758 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2760 * regrename.c (regrename_do_replace): If replacing the same
2761 reg multiple times, try to reuse last created gen_raw_REG.
2764 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2765 main to workaround a bug in GDB.
2767 2018-01-12 Tom de Vries <tom@codesourcery.com>
2770 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2772 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2774 PR rtl-optimization/80481
2775 * ira-color.c (get_cap_member): New function.
2776 (allocnos_conflict_by_live_ranges_p): Use it.
2777 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2778 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2780 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2783 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2784 (*saddl_se_1): Ditto.
2786 (*ssubl_se_1): Ditto.
2788 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2790 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2791 rather than wi::to_widest for DR_INITs.
2792 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2793 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2794 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2796 (vect_analyze_group_access_1): Note that here.
2798 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2800 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2801 polynomial type sizes.
2803 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2805 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2806 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2807 (gimple_add_tmp_var): Likewise.
2809 2018-01-12 Martin Liska <mliska@suse.cz>
2811 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2812 (gimple_alloc_sizes): Likewise.
2813 (dump_gimple_statistics): Use PRIu64 in printf format.
2814 * gimple.h: Change uint64_t to int.
2816 2018-01-12 Martin Liska <mliska@suse.cz>
2818 * tree-core.h: Use uint64_t instead of int.
2819 * tree.c (tree_node_counts): Likewise.
2820 (tree_node_sizes): Likewise.
2821 (dump_tree_statistics): Use PRIu64 in printf format.
2823 2018-01-12 Martin Liska <mliska@suse.cz>
2825 * Makefile.in: As qsort_chk is implemented in vec.c, add
2826 vec.o to linkage of gencfn-macros.
2827 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2828 passing the info to record_node_allocation_statistics.
2829 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2831 * ggc-common.c (struct ggc_usage): Add operator== and use
2832 it in operator< and compare function.
2833 * mem-stats.h (struct mem_usage): Likewise.
2834 * vec.c (struct vec_usage): Remove operator< and compare
2835 function. Can be simply inherited.
2837 2018-01-12 Martin Jambor <mjambor@suse.cz>
2840 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2841 * tree-ssa-math-opts.c: Include domwalk.h.
2842 (convert_mult_to_fma_1): New function.
2843 (fma_transformation_info): New type.
2844 (fma_deferring_state): Likewise.
2845 (cancel_fma_deferring): New function.
2846 (result_of_phi): Likewise.
2847 (last_fma_candidate_feeds_initial_phi): Likewise.
2848 (convert_mult_to_fma): Added deferring logic, split actual
2849 transformation to convert_mult_to_fma_1.
2850 (math_opts_dom_walker): New type.
2851 (math_opts_dom_walker::after_dom_children): New method, body moved
2852 here from pass_optimize_widening_mul::execute, added deferring logic
2854 (pass_optimize_widening_mul::execute): Moved most of code to
2855 math_opts_dom_walker::after_dom_children.
2856 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2857 * config/i386/i386.c (ix86_option_override_internal): Added
2858 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2860 2018-01-12 Richard Biener <rguenther@suse.de>
2863 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2864 inline instance vars.
2866 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2869 * config/rx/rx.c (rx_is_restricted_memory_address):
2872 2018-01-12 Richard Biener <rguenther@suse.de>
2874 PR tree-optimization/80846
2875 * target.def (split_reduction): New target hook.
2876 * targhooks.c (default_split_reduction): New function.
2877 * targhooks.h (default_split_reduction): Declare.
2878 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2879 target requests first reduce vectors by combining low and high
2881 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2882 (get_vectype_for_scalar_type_and_size): Export.
2883 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2884 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2885 * doc/tm.texi: Regenerate.
2886 * config/i386/i386.c (ix86_split_reduction): Implement
2887 TARGET_VECTORIZE_SPLIT_REDUCTION.
2889 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2892 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2893 in PIC mode except for TARGET_VXWORKS_RTP.
2894 * config/sparc/sparc.c: Include cfgrtl.h.
2895 (TARGET_INIT_PIC_REG): Define.
2896 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2897 (sparc_pic_register_p): New predicate.
2898 (sparc_legitimate_address_p): Use it.
2899 (sparc_legitimize_pic_address): Likewise.
2900 (sparc_delegitimize_address): Likewise.
2901 (sparc_mode_dependent_address_p): Likewise.
2902 (gen_load_pcrel_sym): Remove 4th parameter.
2903 (load_got_register): Adjust call to above. Remove obsolete stuff.
2904 (sparc_expand_prologue): Do not call load_got_register here.
2905 (sparc_flat_expand_prologue): Likewise.
2906 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2907 (sparc_use_pseudo_pic_reg): New function.
2908 (sparc_init_pic_reg): Likewise.
2909 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2910 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2912 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2914 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2915 Add item for branch_cost.
2917 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2919 PR rtl-optimization/83565
2920 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2921 not extend the result to a larger mode for rotate operations.
2922 (num_sign_bit_copies1): Likewise.
2924 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2927 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2929 Use values-Xc.o for -pedantic.
2930 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2932 2018-01-12 Martin Liska <mliska@suse.cz>
2935 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2937 (possible_polymorphic_call_targets): Use it.
2938 (ipa_devirt): Likewise.
2940 2018-01-12 Martin Liska <mliska@suse.cz>
2942 * profile-count.h (enum profile_quality): Use 0 as invalid
2943 enum value of profile_quality.
2945 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2947 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2948 -mext-string options.
2950 2018-01-12 Richard Biener <rguenther@suse.de>
2952 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2953 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2954 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2956 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2958 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2960 * configure.ac (--with-long-double-format): Add support for the
2961 configuration option to change the default long double format on
2963 * config.gcc (powerpc*-linux*-*): Likewise.
2964 * configure: Regenerate.
2965 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2966 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2967 used without modification.
2969 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2971 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2972 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2973 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2974 MISC_BUILTIN_SPEC_BARRIER.
2975 (rs6000_init_builtins): Likewise.
2976 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2978 (speculation_barrier): New define_insn.
2979 * doc/extend.texi: Document __builtin_speculation_barrier.
2981 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2984 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2985 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2986 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2988 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2989 integral modes instead of "ss" and "sd".
2990 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2991 vectors with 32-bit and 64-bit elements.
2992 (vecdupssescalarmodesuffix): New mode attribute.
2993 (vec_dup<mode>): Use it.
2995 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2998 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2999 frame if argument is passed on stack.
3001 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3004 * ree.c (combine_reaching_defs): Optimize also
3005 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3006 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3008 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3011 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3013 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3016 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3017 after they are computed.
3019 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3021 PR tree-optimization/83695
3022 * gimple-loop-linterchange.cc
3023 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3024 reset cached scev information after interchange.
3025 (pass_linterchange::execute): Remove call to scev_reset_htab.
3027 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3029 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3030 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3031 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3032 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3033 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3034 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3035 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3036 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3037 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3038 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3039 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3040 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3041 (V_lane_reg): Likewise.
3042 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3044 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3045 (vfmal_lane_low<mode>_intrinsic,
3046 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3047 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3048 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3049 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3050 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3051 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3053 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3055 * config/arm/arm-cpus.in (fp16fml): New feature.
3056 (ALL_SIMD): Add fp16fml.
3057 (armv8.2-a): Add fp16fml as an option.
3058 (armv8.3-a): Likewise.
3059 (armv8.4-a): Add fp16fml as part of fp16.
3060 * config/arm/arm.h (TARGET_FP16FML): Define.
3061 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3063 * config/arm/arm-modes.def (V2HF): Define.
3064 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3065 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3066 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3067 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3068 vfmsl_low, vfmsl_high): New set of builtins.
3069 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3070 (vfml_op): New code attribute.
3071 (VFMLHALVES): New int iterator.
3072 (VFML, VFMLSEL): New mode attributes.
3073 (V_reg): Define mapping for V2HF.
3074 (V_hi, V_lo): New mode attributes.
3075 (VF_constraint): Likewise.
3076 (vfml_half, vfml_half_selector): New int attributes.
3077 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3079 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3080 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3082 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3083 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3084 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3085 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3087 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3088 Document new effective target and option set.
3090 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3092 * config/arm/arm-cpus.in (armv8_4): New feature.
3093 (ARMv8_4a): New fgroup.
3094 (armv8.4-a): New arch.
3095 * config/arm/arm-tables.opt: Regenerate.
3096 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3097 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3098 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3099 Add matching rules for -march=armv8.4-a and extensions.
3100 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3102 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3105 * config/rx/rx.md (BW): New mode attribute.
3106 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3108 2018-01-11 Richard Biener <rguenther@suse.de>
3110 PR tree-optimization/83435
3111 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3112 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3113 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3115 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3116 Alan Hayward <alan.hayward@arm.com>
3117 David Sherwood <david.sherwood@arm.com>
3119 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3121 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3122 (aarch64_print_address_internal): Use it to check for a zero offset.
3124 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3125 Alan Hayward <alan.hayward@arm.com>
3126 David Sherwood <david.sherwood@arm.com>
3128 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3129 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3130 Return a poly_int64 rather than a HOST_WIDE_INT.
3131 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3132 rather than a HOST_WIDE_INT.
3133 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3134 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3135 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3136 final_offset from HOST_WIDE_INT to poly_int64.
3137 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3138 to_constant when getting the number of units in an Advanced SIMD
3140 (aarch64_builtin_vectorized_function): Check for a constant number
3142 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3144 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3145 attribute instead of GET_MODE_NUNITS.
3146 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3147 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3148 GET_MODE_SIZE for fixed-size registers.
3149 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3150 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3151 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3152 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3153 (aarch64_print_operand, aarch64_print_address_internal)
3154 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3155 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3156 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3157 Handle polynomial GET_MODE_SIZE.
3158 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3159 wider than SImode without modification.
3160 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3161 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3162 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3163 passing and returning SVE modes.
3164 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3165 rather than GEN_INT.
3166 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3167 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3168 (aarch64_allocate_and_probe_stack_space): Likewise.
3169 (aarch64_layout_frame): Cope with polynomial offsets.
3170 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3171 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3173 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3174 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3175 poly_int64 rather than a HOST_WIDE_INT.
3176 (aarch64_get_separate_components, aarch64_process_components)
3177 (aarch64_expand_prologue, aarch64_expand_epilogue)
3178 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3179 (aarch64_anchor_offset): New function, split out from...
3180 (aarch64_legitimize_address): ...here.
3181 (aarch64_builtin_vectorization_cost): Handle polynomial
3182 TYPE_VECTOR_SUBPARTS.
3183 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3185 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3186 number of elements from the PARALLEL rather than the mode.
3187 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3188 rather than GET_MODE_BITSIZE.
3189 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3190 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3191 (aarch64_expand_vec_perm_const_1): Handle polynomial
3192 d->perm.length () and d->perm elements.
3193 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3194 Apply to_constant to d->perm elements.
3195 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3196 polynomial CONST_VECTOR_NUNITS.
3197 (aarch64_move_pointer): Take amount as a poly_int64 rather
3199 (aarch64_progress_pointer): Avoid temporary variable.
3200 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3201 the mode attribute instead of GET_MODE.
3203 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3204 Alan Hayward <alan.hayward@arm.com>
3205 David Sherwood <david.sherwood@arm.com>
3207 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3208 x exists before using it.
3209 (aarch64_add_constant_internal): Rename to...
3210 (aarch64_add_offset_1): ...this. Replace regnum with separate
3211 src and dest rtxes. Handle the case in which they're different,
3212 including when the offset is zero. Replace scratchreg with an rtx.
3213 Use 2 additions if there is no spare register into which we can
3214 move a 16-bit constant.
3215 (aarch64_add_constant): Delete.
3216 (aarch64_add_offset): Replace reg with separate src and dest
3217 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3218 Use aarch64_add_offset_1.
3219 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3220 an rtx rather than an int. Take the delta as a poly_int64
3221 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3222 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3223 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3224 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3225 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3227 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3228 aarch64_add_constant.
3230 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3232 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3233 Use scalar_float_mode.
3235 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3237 * config/aarch64/aarch64-simd.md
3238 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3239 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3240 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3241 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3242 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3243 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3244 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3245 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3246 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3247 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3249 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3252 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3253 targ_options->x_arm_arch_string is non NULL.
3255 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3257 * config/aarch64/aarch64.h
3258 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3260 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3263 * expmed.c (emit_store_flag_force): Swap if const op0
3264 and change VOIDmode to mode of op0.
3266 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3268 PR rtl-optimization/83761
3269 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3270 than bytes to mode_for_size.
3272 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3275 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3276 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3279 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3282 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3283 when in layout mode.
3284 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3285 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3288 2018-01-10 Michael Collison <michael.collison@arm.com>
3290 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3291 * config/aarch64/aarch64-option-extension.def: Add
3292 AARCH64_OPT_EXTENSION of 'fp16fml'.
3293 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3294 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3295 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3296 * config/aarch64/constraints.md (Ui7): New constraint.
3297 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3298 (VFMLA_SEL_W): Ditto.
3301 (VFMLA16_LOW): New int iterator.
3302 (VFMLA16_HIGH): Ditto.
3303 (UNSPEC_FMLAL): New unspec.
3304 (UNSPEC_FMLSL): Ditto.
3305 (UNSPEC_FMLAL2): Ditto.
3306 (UNSPEC_FMLSL2): Ditto.
3307 (f16mac): New code attribute.
3308 * config/aarch64/aarch64-simd-builtins.def
3309 (aarch64_fmlal_lowv2sf): Ditto.
3310 (aarch64_fmlsl_lowv2sf): Ditto.
3311 (aarch64_fmlalq_lowv4sf): Ditto.
3312 (aarch64_fmlslq_lowv4sf): Ditto.
3313 (aarch64_fmlal_highv2sf): Ditto.
3314 (aarch64_fmlsl_highv2sf): Ditto.
3315 (aarch64_fmlalq_highv4sf): Ditto.
3316 (aarch64_fmlslq_highv4sf): Ditto.
3317 (aarch64_fmlal_lane_lowv2sf): Ditto.
3318 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3319 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3320 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3321 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3322 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3323 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3324 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3325 (aarch64_fmlal_lane_highv2sf): Ditto.
3326 (aarch64_fmlsl_lane_highv2sf): Ditto.
3327 (aarch64_fmlal_laneq_highv2sf): Ditto.
3328 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3329 (aarch64_fmlalq_lane_highv4sf): Ditto.
3330 (aarch64_fmlsl_lane_highv4sf): Ditto.
3331 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3332 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3333 * config/aarch64/aarch64-simd.md:
3334 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3335 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3336 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3337 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3338 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3339 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3340 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3341 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3342 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3343 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3344 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3345 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3346 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3347 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3348 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3349 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3350 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3351 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3352 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3353 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3354 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3355 (vfmlsl_low_u32): Ditto.
3356 (vfmlalq_low_u32): Ditto.
3357 (vfmlslq_low_u32): Ditto.
3358 (vfmlal_high_u32): Ditto.
3359 (vfmlsl_high_u32): Ditto.
3360 (vfmlalq_high_u32): Ditto.
3361 (vfmlslq_high_u32): Ditto.
3362 (vfmlal_lane_low_u32): Ditto.
3363 (vfmlsl_lane_low_u32): Ditto.
3364 (vfmlal_laneq_low_u32): Ditto.
3365 (vfmlsl_laneq_low_u32): Ditto.
3366 (vfmlalq_lane_low_u32): Ditto.
3367 (vfmlslq_lane_low_u32): Ditto.
3368 (vfmlalq_laneq_low_u32): Ditto.
3369 (vfmlslq_laneq_low_u32): Ditto.
3370 (vfmlal_lane_high_u32): Ditto.
3371 (vfmlsl_lane_high_u32): Ditto.
3372 (vfmlal_laneq_high_u32): Ditto.
3373 (vfmlsl_laneq_high_u32): Ditto.
3374 (vfmlalq_lane_high_u32): Ditto.
3375 (vfmlslq_lane_high_u32): Ditto.
3376 (vfmlalq_laneq_high_u32): Ditto.
3377 (vfmlslq_laneq_high_u32): Ditto.
3378 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3379 (AARCH64_FL_FOR_ARCH8_4): New.
3380 (AARCH64_ISA_F16FML): New ISA flag.
3381 (TARGET_F16FML): New feature flag for fp16fml.
3382 (doc/invoke.texi): Document new fp16fml option.
3384 2018-01-10 Michael Collison <michael.collison@arm.com>
3386 * config/aarch64/aarch64-builtins.c:
3387 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3388 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3389 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3390 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3391 (AARCH64_ISA_SHA3): New ISA flag.
3392 (TARGET_SHA3): New feature flag for sha3.
3393 * config/aarch64/iterators.md (sha512_op): New int attribute.
3394 (CRYPTO_SHA512): New int iterator.
3395 (UNSPEC_SHA512H): New unspec.
3396 (UNSPEC_SHA512H2): Ditto.
3397 (UNSPEC_SHA512SU0): Ditto.
3398 (UNSPEC_SHA512SU1): Ditto.
3399 * config/aarch64/aarch64-simd-builtins.def
3400 (aarch64_crypto_sha512hqv2di): New builtin.
3401 (aarch64_crypto_sha512h2qv2di): Ditto.
3402 (aarch64_crypto_sha512su0qv2di): Ditto.
3403 (aarch64_crypto_sha512su1qv2di): Ditto.
3404 (aarch64_eor3qv8hi): Ditto.
3405 (aarch64_rax1qv2di): Ditto.
3406 (aarch64_xarqv2di): Ditto.
3407 (aarch64_bcaxqv8hi): Ditto.
3408 * config/aarch64/aarch64-simd.md:
3409 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3410 (aarch64_crypto_sha512su0qv2di): Ditto.
3411 (aarch64_crypto_sha512su1qv2di): Ditto.
3412 (aarch64_eor3qv8hi): Ditto.
3413 (aarch64_rax1qv2di): Ditto.
3414 (aarch64_xarqv2di): Ditto.
3415 (aarch64_bcaxqv8hi): Ditto.
3416 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3417 (vsha512h2q_u64): Ditto.
3418 (vsha512su0q_u64): Ditto.
3419 (vsha512su1q_u64): Ditto.
3420 (veor3q_u16): Ditto.
3421 (vrax1q_u64): Ditto.
3423 (vbcaxq_u16): Ditto.
3424 * config/arm/types.md (crypto_sha512): New type attribute.
3425 (crypto_sha3): Ditto.
3426 (doc/invoke.texi): Document new sha3 option.
3428 2018-01-10 Michael Collison <michael.collison@arm.com>
3430 * config/aarch64/aarch64-builtins.c:
3431 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3432 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3433 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3434 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3435 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3436 (AARCH64_ISA_SM4): New ISA flag.
3437 (TARGET_SM4): New feature flag for sm4.
3438 * config/aarch64/aarch64-simd-builtins.def
3439 (aarch64_sm3ss1qv4si): Ditto.
3440 (aarch64_sm3tt1aq4si): Ditto.
3441 (aarch64_sm3tt1bq4si): Ditto.
3442 (aarch64_sm3tt2aq4si): Ditto.
3443 (aarch64_sm3tt2bq4si): Ditto.
3444 (aarch64_sm3partw1qv4si): Ditto.
3445 (aarch64_sm3partw2qv4si): Ditto.
3446 (aarch64_sm4eqv4si): Ditto.
3447 (aarch64_sm4ekeyqv4si): Ditto.
3448 * config/aarch64/aarch64-simd.md:
3449 (aarch64_sm3ss1qv4si): Ditto.
3450 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3451 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3452 (aarch64_sm4eqv4si): Ditto.
3453 (aarch64_sm4ekeyqv4si): Ditto.
3454 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3455 (sm3part_op): Ditto.
3456 (CRYPTO_SM3TT): Ditto.
3457 (CRYPTO_SM3PART): Ditto.
3458 (UNSPEC_SM3SS1): New unspec.
3459 (UNSPEC_SM3TT1A): Ditto.
3460 (UNSPEC_SM3TT1B): Ditto.
3461 (UNSPEC_SM3TT2A): Ditto.
3462 (UNSPEC_SM3TT2B): Ditto.
3463 (UNSPEC_SM3PARTW1): Ditto.
3464 (UNSPEC_SM3PARTW2): Ditto.
3465 (UNSPEC_SM4E): Ditto.
3466 (UNSPEC_SM4EKEY): Ditto.
3467 * config/aarch64/constraints.md (Ui2): New constraint.
3468 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3469 * config/arm/types.md (crypto_sm3): New type attribute.
3470 (crypto_sm4): Ditto.
3471 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3472 (vsm3tt1aq_u32): Ditto.
3473 (vsm3tt1bq_u32): Ditto.
3474 (vsm3tt2aq_u32): Ditto.
3475 (vsm3tt2bq_u32): Ditto.
3476 (vsm3partw1q_u32): Ditto.
3477 (vsm3partw2q_u32): Ditto.
3478 (vsm4eq_u32): Ditto.
3479 (vsm4ekeyq_u32): Ditto.
3480 (doc/invoke.texi): Document new sm4 option.
3482 2018-01-10 Michael Collison <michael.collison@arm.com>
3484 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3485 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3486 (AARCH64_FL_FOR_ARCH8_4): New.
3487 (AARCH64_FL_V8_4): New flag.
3488 (doc/invoke.texi): Document new armv8.4-a option.
3490 2018-01-10 Michael Collison <michael.collison@arm.com>
3492 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3493 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3494 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3495 * config/aarch64/aarch64-option-extension.def: Add
3496 AARCH64_OPT_EXTENSION of 'sha2'.
3497 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3498 (crypto): Disable sha2 and aes if crypto disabled.
3499 (crypto): Enable aes and sha2 if enabled.
3500 (simd): Disable sha2 and aes if simd disabled.
3501 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3503 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3504 (TARGET_SHA2): New feature flag for sha2.
3505 (TARGET_AES): New feature flag for aes.
3506 * config/aarch64/aarch64-simd.md:
3507 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3508 conditional on TARGET_AES.
3509 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3510 (aarch64_crypto_sha1hsi): Make pattern conditional
3512 (aarch64_crypto_sha1hv4si): Ditto.
3513 (aarch64_be_crypto_sha1hv4si): Ditto.
3514 (aarch64_crypto_sha1su1v4si): Ditto.
3515 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3516 (aarch64_crypto_sha1su0v4si): Ditto.
3517 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3518 (aarch64_crypto_sha256su0v4si): Ditto.
3519 (aarch64_crypto_sha256su1v4si): Ditto.
3520 (doc/invoke.texi): Document new aes and sha2 options.
3522 2018-01-10 Martin Sebor <msebor@redhat.com>
3524 PR tree-optimization/83781
3525 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3528 2018-01-11 Martin Sebor <msebor@gmail.com>
3529 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3531 PR tree-optimization/83501
3532 PR tree-optimization/81703
3534 * tree-ssa-strlen.c (get_string_cst): Rename...
3535 (get_string_len): ...to this. Handle global constants.
3536 (handle_char_store): Adjust.
3538 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3539 Jim Wilson <jimw@sifive.com>
3541 * config/riscv/riscv-protos.h (riscv_output_return): New.
3542 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3543 (riscv_attribute_table, riscv_output_return),
3544 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3545 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3546 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3548 (riscv_expand_prologue): Add early return for naked function.
3549 (riscv_expand_epilogue): Likewise.
3550 (riscv_function_ok_for_sibcall): Return false for naked function.
3551 (riscv_set_current_function): New.
3552 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3553 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3554 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3555 * doc/extend.texi (RISC-V Function Attributes): New.
3557 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3559 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3560 check for 128-bit long double before checking TCmode.
3561 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3562 128-bit long doubles before checking TFmode or TCmode.
3563 (FLOAT128_IBM_P): Likewise.
3565 2018-01-10 Martin Sebor <msebor@redhat.com>
3567 PR tree-optimization/83671
3568 * builtins.c (c_strlen): Unconditionally return zero for the empty
3570 Use -Warray-bounds for warnings.
3571 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3572 for non-constant array indices with COMPONENT_REF, arrays of
3573 arrays, and pointers to arrays.
3574 (gimple_fold_builtin_strlen): Determine and set length range for
3575 non-constant character arrays.
3577 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3580 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3583 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3585 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3587 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3590 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3591 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3592 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3593 indexed_or_indirect_operand predicate.
3594 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3595 (*vsx_le_perm_load_v8hi): Likewise.
3596 (*vsx_le_perm_load_v16qi): Likewise.
3597 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3598 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3599 (*vsx_le_perm_store_v8hi): Likewise.
3600 (*vsx_le_perm_store_v16qi): Likewise.
3601 (eight unnamed splitters): Likewise.
3603 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3605 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3606 * config/rs6000/emmintrin.h: Likewise.
3607 * config/rs6000/mmintrin.h: Likewise.
3608 * config/rs6000/xmmintrin.h: Likewise.
3610 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3613 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3615 * tree.c (tree_nop_conversion): Return true for location wrapper
3617 (maybe_wrap_with_location): New function.
3618 (selftest::check_strip_nops): New function.
3619 (selftest::test_location_wrappers): New function.
3620 (selftest::tree_c_tests): Call it.
3621 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3622 (maybe_wrap_with_location): New decl.
3623 (EXPR_LOCATION_WRAPPER_P): New macro.
3624 (location_wrapper_p): New inline function.
3625 (tree_strip_any_location_wrapper): New inline function.
3627 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3630 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3631 stack_realign_offset for the largest alignment of stack slot
3633 (ix86_find_max_used_stack_alignment): New function.
3634 (ix86_finalize_stack_frame_flags): Use it. Set
3635 max_used_stack_alignment if we don't realign stack.
3636 * config/i386/i386.h (machine_function): Add
3637 max_used_stack_alignment.
3639 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3641 * config/arm/arm.opt (-mbranch-cost): New option.
3642 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3645 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3648 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3649 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3651 2018-01-10 Richard Biener <rguenther@suse.de>
3654 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3655 early out so it also covers the case where we have a non-NULL
3658 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3660 PR tree-optimization/83753
3661 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3662 for non-strided grouped accesses if the number of elements is 1.
3664 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3667 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3668 * i386.h (TARGET_USE_GATHER): Define.
3669 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3671 2018-01-10 Martin Liska <mliska@suse.cz>
3674 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3675 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3677 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3678 CLEANUP_NO_PARTITIONING is not set.
3680 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3682 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3683 for vectors, as a partial revert of r254296.
3684 * rtl.h (const_vec_p): Delete.
3685 (const_vec_duplicate_p): Don't test for vector CONSTs.
3686 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3687 * expmed.c (make_tree): Likewise.
3690 * common.md (E, F): Use CONSTANT_P instead of checking for
3692 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3693 checking for CONST_VECTOR.
3695 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3698 * predict.c (force_edge_cold): Handle in more sane way edges
3701 2018-01-09 Carl Love <cel@us.ibm.com>
3703 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3705 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3706 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3707 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3708 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3709 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3710 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3711 * config/rs6000/rs6000-protos.h: Add extern defition for
3712 rs6000_generate_float2_double_code.
3713 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3715 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3716 (float2_v2df): Add define_expand.
3718 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3721 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3722 op_mode in the force_to_mode call.
3724 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3726 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3727 instead of checking each element individually.
3728 (aarch64_evpc_uzp): Likewise.
3729 (aarch64_evpc_zip): Likewise.
3730 (aarch64_evpc_ext): Likewise.
3731 (aarch64_evpc_rev): Likewise.
3732 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3733 instead of checking each element individually. Return true without
3735 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3736 whether all selected elements come from the same input, instead of
3737 checking each element individually. Remove calls to gen_rtx_REG,
3738 start_sequence and end_sequence and instead assert that no rtl is
3741 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3743 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3744 order of HIGH and CONST checks.
3746 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3748 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3749 if the destination isn't an SSA_NAME.
3751 2018-01-09 Richard Biener <rguenther@suse.de>
3753 PR tree-optimization/83668
3754 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3756 (canonicalize_loop_form): ... here, renamed from ...
3757 (canonicalize_loop_closed_ssa_form): ... this and amended to
3758 swap successor edges for loop exit blocks to make us use
3759 the RPO order we need for initial schedule generation.
3761 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3763 PR tree-optimization/64811
3764 * match.pd: When optimizing comparisons with Inf, avoid
3765 introducing or losing exceptions from comparisons with NaN.
3767 2018-01-09 Martin Liska <mliska@suse.cz>
3770 * asan.c (shadow_mem_size): Add gcc_assert.
3772 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3774 Don't save registers in main().
3777 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3778 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3779 * config/avr/avr.c (avr_set_current_function): Don't error if
3780 naked, OS_task or OS_main are specified at the same time.
3781 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3783 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3785 * common/config/avr/avr-common.c (avr_option_optimization_table):
3786 Switch on -mmain-is-OS_task for optimizing compilations.
3788 2018-01-09 Richard Biener <rguenther@suse.de>
3790 PR tree-optimization/83572
3791 * graphite.c: Include cfganal.h.
3792 (graphite_transform_loops): Connect infinite loops to exit
3793 and remove fake edges at the end.
3795 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3797 * ipa-inline.c (edge_badness): Revert accidental checkin.
3799 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3802 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3803 symbols; not inline clones.
3805 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3808 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3809 hard registers. Formatting fixes.
3811 PR preprocessor/83722
3812 * gcc.c (try_generate_repro): Pass
3813 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3814 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3817 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3818 Kito Cheng <kito.cheng@gmail.com>
3820 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3821 (riscv_leaf_function_p): Delete.
3822 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3824 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3826 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3828 (do_ifelse): New function.
3829 (do_isel): New function.
3830 (do_sub3): New function.
3831 (do_add3): New function.
3832 (do_load_mask_compare): New function.
3833 (do_overlap_load_compare): New function.
3834 (expand_compare_loop): New function.
3835 (expand_block_compare): Call expand_compare_loop() when appropriate.
3836 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3838 (-mblock-compare-inline-loop-limit): New option.
3840 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3843 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3844 Reverse order of second and third operands in first alternative.
3845 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3846 of first and second elements in UNSPEC_VPERMR vector.
3847 (altivec_expand_vec_perm_le): Likewise.
3849 2017-01-08 Jeff Law <law@redhat.com>
3851 PR rtl-optimizatin/81308
3852 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3853 (process_switch): If group_case_labels makes a change, then set
3855 (pass_convert_switch::execute): If a switch is converted, then
3856 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3858 PR rtl-optimization/81308
3859 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3862 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3864 PR target/83663 - Revert r255946
3865 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3866 generation for cases where splatting a value is not useful.
3867 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3868 across a vec_duplicate and a paradoxical subreg forming a vector
3869 mode to a vec_concat.
3871 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3873 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3874 -march=armv8.3-a variants.
3875 * config/arm/t-multilib: Likewise.
3876 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3878 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3880 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3882 (cceq_ior_compare_complement): Give it a name so I can use it, and
3883 change boolean_or_operator predicate to boolean_operator so it can
3884 be used to generate a crand.
3885 (eqne): New code iterator.
3886 (bd/bd_neg): New code_attrs.
3887 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3888 a single define_insn.
3889 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3890 decrement (bdnzt/bdnzf/bdzt/bdzf).
3891 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3892 with the new names of the branch decrement patterns, and added the
3893 names of the branch decrement conditional patterns.
3895 2018-01-08 Richard Biener <rguenther@suse.de>
3897 PR tree-optimization/83563
3898 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3901 2018-01-08 Richard Biener <rguenther@suse.de>
3904 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3906 2018-01-08 Richard Biener <rguenther@suse.de>
3908 PR tree-optimization/83685
3909 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3910 references to abnormals.
3912 2018-01-08 Richard Biener <rguenther@suse.de>
3915 * dwarf2out.c (output_indirect_strings): Handle empty
3916 skeleton_debug_str_hash.
3917 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3919 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3921 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3922 (emit_store_direct): Likewise.
3923 (arc_trampoline_adjust_address): Likewise.
3924 (arc_asm_trampoline_template): New function.
3925 (arc_initialize_trampoline): Use asm_trampoline_template.
3926 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3927 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3928 * config/arc/arc.md (flush_icache): Delete pattern.
3930 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3932 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3933 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3936 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3939 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3940 by not USED_FOR_TARGET.
3941 (make_pass_resolve_sw_modes): Likewise.
3943 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3945 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3948 2018-01-08 Richard Biener <rguenther@suse.de>
3951 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3953 2018-01-08 Richard Biener <rguenther@suse.de>
3956 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3958 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3961 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3962 basic blocks with a small number of successors.
3963 (convert_control_dep_chain_into_preds): Improve handling of
3965 (dump_predicates): Split apart into...
3966 (dump_pred_chain): ...here...
3967 (dump_pred_info): ...and here.
3968 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3969 (can_chain_union_be_invalidated_p): Improve check for invalidation
3971 (uninit_uses_cannot_happen): Avoid unnecessary if
3972 convert_control_dep_chain_into_preds yielded nothing.
3974 2018-01-06 Martin Sebor <msebor@redhat.com>
3976 PR tree-optimization/83640
3977 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3978 subtracting negative offset from size.
3979 (builtin_access::overlap): Adjust offset bounds of the access to fall
3980 within the size of the object if possible.
3982 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3984 PR rtl-optimization/83699
3985 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3986 extract_bit_field_as_subreg to cases in which the extracted
3987 value is also a vector.
3989 * lra-constraints.c (process_alt_operands): Test for the equivalence
3990 substitutions when detecting a possible reload cycle.
3992 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3995 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3996 by default if flag_selective_schedling{,2}. Formatting fixes.
3998 PR rtl-optimization/83682
3999 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4000 if it has non-VECTOR_MODE element mode.
4001 (vec_duplicate_p): Likewise.
4004 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4005 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4007 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4010 * config/i386/i386-builtin.def
4011 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4012 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4013 Require also OPTION_MASK_ISA_AVX512F in addition to
4014 OPTION_MASK_ISA_GFNI.
4015 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4016 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4017 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4018 to OPTION_MASK_ISA_GFNI.
4019 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4020 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4021 OPTION_MASK_ISA_AVX512BW.
4022 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4023 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4024 addition to OPTION_MASK_ISA_GFNI.
4025 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4026 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4027 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4028 to OPTION_MASK_ISA_GFNI.
4029 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4030 a requirement for all ISAs rather than any of them with a few
4032 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4034 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4035 bitmasks to be enabled with 3 exceptions, instead of requiring any
4036 enabled ISA with lots of exceptions.
4037 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4038 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4039 Change avx512bw in isa attribute to avx512f.
4040 * config/i386/sgxintrin.h: Add license boilerplate.
4041 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4042 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4043 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4044 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4046 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4047 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4048 temporarily sse2 rather than sse if not enabled already.
4051 * config/i386/sse.md (VI248_VLBW): Rename to ...
4052 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4053 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4054 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4055 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4056 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4057 mode iterator instead of VI248_VLBW.
4059 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4061 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4062 (record_modified): Skip clobbers; add debug output.
4063 (param_change_prob): Use sreal frequencies.
4065 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4067 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4068 punt for user-aligned variables.
4070 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4072 * tree-chrec.c (chrec_contains_symbols): Return true for
4075 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4078 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4079 of (x|y) == x for BICS pattern.
4081 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4083 PR tree-optimization/83605
4084 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4085 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4088 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4090 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4091 * config/epiphany/rtems.h: New file.
4093 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4094 Uros Bizjak <ubizjak@gmail.com>
4097 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4098 QIreg_operand instead of register_operand predicate.
4099 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4100 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4101 comments instead of -fmitigate[-_]rop.
4103 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4106 * cgraphunit.c (symbol_table::compile): Switch to text_section
4107 before calling assembly_start debug hook.
4108 * run-rtl-passes.c (run_rtl_passes): Likewise.
4111 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4113 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4114 range_int_cst_p rather than !symbolic_range_p before calling
4115 extract_range_from_multiplicative_op_1.
4117 2017-01-04 Jeff Law <law@redhat.com>
4119 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4120 redundant test in assertion.
4122 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4124 * doc/rtl.texi: Document machine_mode wrapper classes.
4126 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4128 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4131 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4133 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4134 the VEC_PERM_EXPR fold to fail.
4136 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4139 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4140 to switched_sections.
4142 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4145 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4148 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4151 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4152 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4154 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4157 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4158 is BLKmode and bitpos not zero or mode change is needed.
4160 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4163 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4166 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4169 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4170 instead of MULT rtx. Update all corresponding splitters.
4172 (*ssub<modesuffix>): Ditto.
4174 (*cmp_sadd_di): Update split patterns.
4175 (*cmp_sadd_si): Ditto.
4176 (*cmp_sadd_sidi): Ditto.
4177 (*cmp_ssub_di): Ditto.
4178 (*cmp_ssub_si): Ditto.
4179 (*cmp_ssub_sidi): Ditto.
4180 * config/alpha/predicates.md (const23_operand): New predicate.
4181 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4182 Look for ASHIFT, not MULT inner operand.
4183 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4185 2018-01-04 Martin Liska <mliska@suse.cz>
4187 PR gcov-profile/83669
4188 * gcov.c (output_intermediate_file): Add version to intermediate
4190 * doc/gcov.texi: Document new field 'version' in intermediate
4191 file format. Fix location of '-k' option of gcov command.
4193 2018-01-04 Martin Liska <mliska@suse.cz>
4196 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4198 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4200 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4202 2018-01-03 Martin Sebor <msebor@redhat.com>
4204 PR tree-optimization/83655
4205 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4206 checking calls with invalid arguments.
4208 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4210 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4211 (vectorizable_mask_load_store): Delete.
4212 (vectorizable_call): Return false for masked loads and stores.
4213 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4214 instead of gimple_assign_rhs1.
4215 (vectorizable_load): Handle IFN_MASK_LOAD.
4216 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4218 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4220 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4222 (vectorizable_mask_load_store): ...here.
4223 (vectorizable_load): ...and here.
4225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4227 * tree-vect-stmts.c (vect_build_all_ones_mask)
4228 (vect_build_zero_merge_argument): New functions, split out from...
4229 (vectorizable_load): ...here.
4231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4233 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4235 (vectorizable_mask_load_store): ...here.
4236 (vectorizable_store): ...and here.
4238 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4240 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4242 (vectorizable_mask_load_store): ...here.
4244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4246 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4247 (vect_model_store_cost): Take a vec_load_store_type instead of a
4249 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4250 (vect_model_store_cost): Take a vec_load_store_type instead of a
4252 (vectorizable_mask_load_store): Update accordingly.
4253 (vectorizable_store): Likewise.
4254 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4256 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4258 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4259 IFN_MASK_LOAD calls here rather than...
4260 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4262 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4263 Alan Hayward <alan.hayward@arm.com>
4264 David Sherwood <david.sherwood@arm.com>
4266 * expmed.c (extract_bit_field_1): For vector extracts,
4267 fall back to extract_bit_field_as_subreg if vec_extract
4270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4271 Alan Hayward <alan.hayward@arm.com>
4272 David Sherwood <david.sherwood@arm.com>
4274 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4275 they are variable or constant sized.
4276 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4277 slots for constant-sized data.
4279 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4280 Alan Hayward <alan.hayward@arm.com>
4281 David Sherwood <david.sherwood@arm.com>
4283 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4284 handling COND_EXPRs with boolean comparisons, try to find a better
4285 basis for the mask type than the boolean itself.
4287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4289 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4290 is calculated and how it can be overridden.
4291 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4292 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4294 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4298 Alan Hayward <alan.hayward@arm.com>
4299 David Sherwood <david.sherwood@arm.com>
4301 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4302 Remove the mode argument.
4303 (aarch64_simd_valid_immediate): Remove the mode and inverse
4305 * config/aarch64/iterators.md (bitsize): New iterator.
4306 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4307 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4308 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4309 aarch64_simd_valid_immediate.
4310 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4311 (aarch64_reg_or_bic_imm): Likewise.
4312 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4313 with an insn_type enum and msl with a modifier_type enum.
4314 Replace element_width with a scalar_mode. Change the shift
4315 to unsigned int. Add constructors for scalar_float_mode and
4316 scalar_int_mode elements.
4317 (aarch64_vect_float_const_representable_p): Delete.
4318 (aarch64_can_const_movi_rtx_p)
4319 (aarch64_simd_scalar_immediate_valid_for_move)
4320 (aarch64_simd_make_constant): Update call to
4321 aarch64_simd_valid_immediate.
4322 (aarch64_advsimd_valid_immediate_hs): New function.
4323 (aarch64_advsimd_valid_immediate): Likewise.
4324 (aarch64_simd_valid_immediate): Remove mode and inverse
4325 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4326 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4327 and aarch64_float_const_representable_p on the result.
4328 (aarch64_output_simd_mov_immediate): Remove mode argument.
4329 Update call to aarch64_simd_valid_immediate and use of
4330 simd_immediate_info.
4331 (aarch64_output_scalar_simd_mov_immediate): Update call
4334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4335 Alan Hayward <alan.hayward@arm.com>
4336 David Sherwood <david.sherwood@arm.com>
4338 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4339 (mode_nunits): Likewise CONST_MODE_NUNITS.
4340 * machmode.def (ADJUST_NUNITS): Document.
4341 * genmodes.c (mode_data::need_nunits_adj): New field.
4342 (blank_mode): Update accordingly.
4343 (adj_nunits): New variable.
4344 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4346 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4347 listed in adj_nunits.
4348 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4349 listed in adj_nunits. Don't emit case statements for such modes.
4350 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4351 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4352 nothing if adj_nunits is nonnull.
4353 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4354 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4355 (emit_mode_fbit): Update use of print_maybe_const_decl.
4356 (emit_move_size): Likewise. Treat the array as non-const
4358 (emit_mode_adjustments): Handle adj_nunits.
4360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4362 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4363 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4364 (VECTOR_MODES): Use it.
4365 (make_vector_modes): Take the prefix as an argument.
4367 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4368 Alan Hayward <alan.hayward@arm.com>
4369 David Sherwood <david.sherwood@arm.com>
4371 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4372 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4373 for MODE_VECTOR_BOOL.
4374 * machmode.def (VECTOR_BOOL_MODE): Document.
4375 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4376 (make_vector_bool_mode): New function.
4377 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4379 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4380 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4382 * stor-layout.c (int_mode_for_mode): Likewise.
4383 * tree.c (build_vector_type_for_mode): Likewise.
4384 * varasm.c (output_constant_pool_2): Likewise.
4385 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4386 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4387 for MODE_VECTOR_BOOL.
4388 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4389 of mode class checks.
4390 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4391 instead of a list of mode class checks.
4392 (expand_vector_scalar_condition): Likewise.
4393 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4396 Alan Hayward <alan.hayward@arm.com>
4397 David Sherwood <david.sherwood@arm.com>
4399 * machmode.h (mode_size): Change from unsigned short to
4401 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4402 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4403 or if measurement_type is not polynomial.
4404 (fixed_size_mode::includes_p): Check for constant-sized modes.
4405 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4406 return a poly_uint16 rather than an unsigned short.
4407 (emit_mode_size): Change the type of mode_size from unsigned short
4408 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4409 (emit_mode_adjustments): Cope with polynomial vector sizes.
4410 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4412 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4414 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4415 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4416 * caller-save.c (setup_save_areas): Likewise.
4417 (replace_reg_with_saved_mem): Likewise.
4418 * calls.c (emit_library_call_value_1): Likewise.
4419 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4420 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4421 (gen_lowpart_for_combine): Likewise.
4422 * convert.c (convert_to_integer_1): Likewise.
4423 * cse.c (equiv_constant, cse_insn): Likewise.
4424 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4425 (cselib_subst_to_values): Likewise.
4426 * dce.c (word_dce_process_block): Likewise.
4427 * df-problems.c (df_word_lr_mark_ref): Likewise.
4428 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4429 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4430 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4431 (rtl_for_decl_location): Likewise.
4432 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4433 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4434 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4435 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4436 (expand_expr_real_1): Likewise.
4437 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4438 (pad_below): Likewise.
4439 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4440 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4441 * ira.c (get_subreg_tracking_sizes): Likewise.
4442 * ira-build.c (ira_create_allocno_objects): Likewise.
4443 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4444 (ira_sort_regnos_for_alter_reg): Likewise.
4445 * ira-costs.c (record_operand_costs): Likewise.
4446 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4447 (resolve_simple_move): Likewise.
4448 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4449 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4450 (lra_constraints): Likewise.
4451 (CONST_POOL_OK_P): Reject variable-sized modes.
4452 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4453 (add_pseudo_to_slot, lra_spill): Likewise.
4454 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4455 * optabs-query.c (get_best_extraction_insn): Likewise.
4456 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4457 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4458 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4459 * recog.c (offsettable_address_addr_space_p): Likewise.
4460 * regcprop.c (maybe_mode_change): Likewise.
4461 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4462 * regrename.c (build_def_use): Likewise.
4463 * regstat.c (dump_reg_info): Likewise.
4464 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4465 (find_reloads, find_reloads_subreg_address): Likewise.
4466 * reload1.c (eliminate_regs_1): Likewise.
4467 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4468 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4469 (simplify_binary_operation_1, simplify_subreg): Likewise.
4470 * targhooks.c (default_function_arg_padding): Likewise.
4471 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4472 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4473 (verify_gimple_assign_ternary): Likewise.
4474 * tree-inline.c (estimate_move_cost): Likewise.
4475 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4476 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4477 (get_address_cost_ainc): Likewise.
4478 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4479 (vect_supportable_dr_alignment): Likewise.
4480 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4481 (vectorizable_reduction): Likewise.
4482 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4483 (vectorizable_operation, vectorizable_load): Likewise.
4484 * tree.c (build_same_sized_truth_vector_type): Likewise.
4485 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4486 * var-tracking.c (emit_note_insn_var_location): Likewise.
4487 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4488 (ADDR_VEC_ALIGN): Likewise.
4490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4491 Alan Hayward <alan.hayward@arm.com>
4492 David Sherwood <david.sherwood@arm.com>
4494 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4496 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4497 or if measurement_type is polynomial.
4498 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4499 * combine.c (make_extraction): Likewise.
4500 * dse.c (find_shift_sequence): Likewise.
4501 * dwarf2out.c (mem_loc_descriptor): Likewise.
4502 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4503 (extract_bit_field, extract_low_bits): Likewise.
4504 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4505 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4506 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4507 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4508 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4509 * reload.c (find_reloads): Likewise.
4510 * reload1.c (alter_reg): Likewise.
4511 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4512 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4513 * tree-if-conv.c (predicate_mem_writes): Likewise.
4514 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4515 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4516 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4517 * valtrack.c (dead_debug_insert_temp): Likewise.
4518 * varasm.c (mergeable_constant_section): Likewise.
4519 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4522 Alan Hayward <alan.hayward@arm.com>
4523 David Sherwood <david.sherwood@arm.com>
4525 * expr.c (expand_assignment): Cope with polynomial mode sizes
4526 when assigning to a CONCAT.
4528 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4529 Alan Hayward <alan.hayward@arm.com>
4530 David Sherwood <david.sherwood@arm.com>
4532 * machmode.h (mode_precision): Change from unsigned short to
4534 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4536 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4537 or if measurement_type is not polynomial.
4538 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4539 in which the mode is already known to be a scalar_int_mode.
4540 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4541 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4543 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4544 for GET_MODE_PRECISION.
4545 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4546 for GET_MODE_PRECISION.
4547 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4549 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4550 (expand_field_assignment, make_extraction): Likewise.
4551 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4552 (get_last_value): Likewise.
4553 * convert.c (convert_to_integer_1): Likewise.
4554 * cse.c (cse_insn): Likewise.
4555 * expr.c (expand_expr_real_1): Likewise.
4556 * lra-constraints.c (simplify_operand_subreg): Likewise.
4557 * optabs-query.c (can_atomic_load_p): Likewise.
4558 * optabs.c (expand_atomic_load): Likewise.
4559 (expand_atomic_store): Likewise.
4560 * ree.c (combine_reaching_defs): Likewise.
4561 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4562 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4563 * tree.h (type_has_mode_precision_p): Likewise.
4564 * ubsan.c (instrument_si_overflow): Likewise.
4566 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4567 Alan Hayward <alan.hayward@arm.com>
4568 David Sherwood <david.sherwood@arm.com>
4570 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4571 polynomial numbers of units.
4572 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4573 (valid_vector_subparts_p): New function.
4574 (build_vector_type): Remove temporary shim and take the number
4575 of units as a poly_uint64 rather than an int.
4576 (build_opaque_vector_type): Take the number of units as a
4577 poly_uint64 rather than an int.
4578 * tree.c (build_vector_from_ctor): Handle polynomial
4579 TYPE_VECTOR_SUBPARTS.
4580 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4581 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4582 (build_vector_from_val): If the number of units is variable,
4583 use build_vec_duplicate_cst for constant operands and
4584 VEC_DUPLICATE_EXPR otherwise.
4585 (make_vector_type): Remove temporary is_constant ().
4586 (build_vector_type, build_opaque_vector_type): Take the number of
4587 units as a poly_uint64 rather than an int.
4588 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4590 * cfgexpand.c (expand_debug_expr): Likewise.
4591 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4592 (store_constructor, expand_expr_real_1): Likewise.
4593 (const_scalar_mask_from_tree): Likewise.
4594 * fold-const-call.c (fold_const_reduction): Likewise.
4595 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4596 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4597 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4598 (fold_relational_const): Likewise.
4599 (native_interpret_vector): Likewise. Change the size from an
4600 int to an unsigned int.
4601 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4602 TYPE_VECTOR_SUBPARTS.
4603 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4604 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4605 duplicating a non-constant operand into a variable-length vector.
4606 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4607 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4608 * ipa-icf.c (sem_variable::equals): Likewise.
4609 * match.pd: Likewise.
4610 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4611 * print-tree.c (print_node): Likewise.
4612 * stor-layout.c (layout_type): Likewise.
4613 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4614 * tree-cfg.c (verify_gimple_comparison): Likewise.
4615 (verify_gimple_assign_binary): Likewise.
4616 (verify_gimple_assign_ternary): Likewise.
4617 (verify_gimple_assign_single): Likewise.
4618 * tree-pretty-print.c (dump_generic_node): Likewise.
4619 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4620 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4621 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4622 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4623 (vect_shift_permute_load_chain): Likewise.
4624 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4625 (expand_vector_condition, optimize_vector_constructor): Likewise.
4626 (lower_vec_perm, get_compute_type): Likewise.
4627 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4628 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4629 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4630 (vect_recog_mask_conversion_pattern): Likewise.
4631 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4632 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4633 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4634 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4635 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4636 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4637 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4638 (supportable_widening_operation): Likewise.
4639 (supportable_narrowing_operation): Likewise.
4640 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4642 * varasm.c (output_constant): Likewise.
4644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4645 Alan Hayward <alan.hayward@arm.com>
4646 David Sherwood <david.sherwood@arm.com>
4648 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4649 so that both the length == 3 and length != 3 cases set up their
4650 own permute vectors. Add comments explaining why we know the
4651 number of elements is constant.
4652 (vect_permute_load_chain): Likewise.
4654 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4655 Alan Hayward <alan.hayward@arm.com>
4656 David Sherwood <david.sherwood@arm.com>
4658 * machmode.h (mode_nunits): Change from unsigned char to
4660 (ONLY_FIXED_SIZE_MODES): New macro.
4661 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4662 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4663 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4665 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4666 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4667 or if measurement_type is not polynomial.
4668 * genmodes.c (ZERO_COEFFS): New macro.
4669 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4671 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4672 Use ZERO_COEFFS when emitting initializers.
4673 * data-streamer.h (bp_pack_poly_value): New function.
4674 (bp_unpack_poly_value): Likewise.
4675 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4676 for GET_MODE_NUNITS.
4677 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4678 for GET_MODE_NUNITS.
4679 * tree.c (make_vector_type): Remove temporary shim and make
4680 the real function take the number of units as a poly_uint64
4682 (build_vector_type_for_mode): Handle polynomial nunits.
4683 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4684 * emit-rtl.c (const_vec_series_p_1): Likewise.
4685 (gen_rtx_CONST_VECTOR): Likewise.
4686 * fold-const.c (test_vec_duplicate_folding): Likewise.
4687 * genrecog.c (validate_pattern): Likewise.
4688 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4689 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4690 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4691 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4692 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4693 * rtlanal.c (subreg_get_info): Likewise.
4694 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4695 (vect_grouped_load_supported): Likewise.
4696 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4697 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4698 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4699 (simplify_const_unary_operation, simplify_binary_operation_1)
4700 (simplify_const_binary_operation, simplify_ternary_operation)
4701 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4702 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4703 instead of CONST_VECTOR_NUNITS.
4704 * varasm.c (output_constant_pool_2): Likewise.
4705 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4706 explicit-encoded elements in the XVEC for variable-length vectors.
4708 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4710 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4713 Alan Hayward <alan.hayward@arm.com>
4714 David Sherwood <david.sherwood@arm.com>
4716 * coretypes.h (fixed_size_mode): Declare.
4717 (fixed_size_mode_pod): New typedef.
4718 * builtins.h (target_builtins::x_apply_args_mode)
4719 (target_builtins::x_apply_result_mode): Change type to
4720 fixed_size_mode_pod.
4721 * builtins.c (apply_args_size, apply_result_size, result_vector)
4722 (expand_builtin_apply_args_1, expand_builtin_apply)
4723 (expand_builtin_return): Update accordingly.
4725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4727 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4728 * cselib.c (cselib_hash_rtx): Likewise.
4729 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4730 CONST_VECTOR encoding.
4732 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4733 Jeff Law <law@redhat.com>
4736 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4737 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4738 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4739 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4742 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4743 explicitly probe *sp in a noreturn function if there were any callee
4744 register saves or frame pointer is needed.
4746 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4749 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4750 BLKmode for ternary, binary or unary expressions.
4753 * var-tracking.c (delete_vta_debug_insn): New inline function.
4754 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4755 insns from get_insns () to NULL instead of each bb separately.
4756 Use delete_vta_debug_insn. No longer static.
4757 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4758 delete_vta_debug_insns callers.
4759 * rtl.h (delete_vta_debug_insns): Declare.
4760 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4761 instead of variable_tracking_main.
4763 2018-01-03 Martin Sebor <msebor@redhat.com>
4765 PR tree-optimization/83603
4766 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4767 arguments past the endof the argument list in functions declared
4768 without a prototype.
4769 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4770 Avoid checking when arguments are null.
4772 2018-01-03 Martin Sebor <msebor@redhat.com>
4775 * doc/extend.texi (attribute const): Fix a typo.
4776 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4777 issuing -Wsuggest-attribute for void functions.
4779 2018-01-03 Martin Sebor <msebor@redhat.com>
4781 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4782 offset_int::from instead of wide_int::to_shwi.
4783 (maybe_diag_overlap): Remove assertion.
4784 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4785 * gimple-ssa-sprintf.c (format_directive): Same.
4786 (parse_directive): Same.
4787 (sprintf_dom_walker::compute_format_length): Same.
4788 (try_substitute_return_value): Same.
4790 2017-01-03 Jeff Law <law@redhat.com>
4793 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4794 non-constant residual for zero at runtime and avoid probing in
4795 that case. Reorganize code for trailing problem to mirror handling
4798 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4800 PR tree-optimization/83501
4801 * tree-ssa-strlen.c (get_string_cst): New.
4802 (handle_char_store): Call get_string_cst.
4804 2018-01-03 Martin Liska <mliska@suse.cz>
4806 PR tree-optimization/83593
4807 * tree-ssa-strlen.c: Include tree-cfg.h.
4808 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4809 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4810 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4812 (strlen_dom_walker::before_dom_children): Call
4813 gimple_purge_dead_eh_edges. Dump tranformation with details
4815 (strlen_dom_walker::before_dom_children): Update call by adding
4816 new argument cleanup_eh.
4817 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4819 2018-01-03 Martin Liska <mliska@suse.cz>
4822 * cif-code.def (VARIADIC_THUNK): New enum value.
4823 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4826 2018-01-03 Jan Beulich <jbeulich@suse.com>
4828 * sse.md (mov<mode>_internal): Tighten condition for when to use
4829 vmovdqu<ssescalarsize> for TI and OI modes.
4831 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4833 Update copyright years.
4835 2018-01-03 Martin Liska <mliska@suse.cz>
4838 * ipa-visibility.c (function_and_variable_visibility): Skip
4839 functions with noipa attribure.
4841 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4843 * gcc.c (process_command): Update copyright notice dates.
4844 * gcov-dump.c (print_version): Ditto.
4845 * gcov.c (print_version): Ditto.
4846 * gcov-tool.c (print_version): Ditto.
4847 * gengtype.c (create_file): Ditto.
4848 * doc/cpp.texi: Bump @copying's copyright year.
4849 * doc/cppinternals.texi: Ditto.
4850 * doc/gcc.texi: Ditto.
4851 * doc/gccint.texi: Ditto.
4852 * doc/gcov.texi: Ditto.
4853 * doc/install.texi: Ditto.
4854 * doc/invoke.texi: Ditto.
4856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4858 * vector-builder.h (vector_builder::m_full_nelts): Change from
4859 unsigned int to poly_uint64.
4860 (vector_builder::full_nelts): Update prototype accordingly.
4861 (vector_builder::new_vector): Likewise.
4862 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4863 (vector_builder::operator ==): Likewise.
4864 (vector_builder::finalize): Likewise.
4865 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4866 Take the number of elements as a poly_uint64 rather than an
4868 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4869 from unsigned int to poly_uint64.
4870 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4871 (vec_perm_indices::new_vector): Likewise.
4872 (vec_perm_indices::length): Likewise.
4873 (vec_perm_indices::nelts_per_input): Likewise.
4874 (vec_perm_indices::input_nelts): Likewise.
4875 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4876 number of elements per input as a poly_uint64 rather than an
4877 unsigned int. Use the original encoding for variable-length
4878 vectors, rather than clamping each individual element.
4879 For the second and subsequent elements in each pattern,
4880 clamp the step and base before clamping their sum.
4881 (vec_perm_indices::series_p): Handle polynomial element counts.
4882 (vec_perm_indices::all_in_range_p): Likewise.
4883 (vec_perm_indices_to_tree): Likewise.
4884 (vec_perm_indices_to_rtx): Likewise.
4885 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4886 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4887 (tree_vector_builder::new_binary_operation): Handle polynomial
4888 element counts. Return false if we need to know the number
4889 of elements at compile time.
4890 * fold-const.c (fold_vec_perm): Punt if the number of elements
4891 isn't known at compile time.
4893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4895 * vec-perm-indices.h (vec_perm_builder): Change element type
4896 from HOST_WIDE_INT to poly_int64.
4897 (vec_perm_indices::element_type): Update accordingly.
4898 (vec_perm_indices::clamp): Handle polynomial element_types.
4899 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4900 (vec_perm_indices::all_in_range_p): Likewise.
4901 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4903 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4904 polynomial vec_perm_indices element types.
4905 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4906 * fold-const.c (fold_vec_perm): Likewise.
4907 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4908 * tree-vect-generic.c (lower_vec_perm): Likewise.
4909 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4910 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4911 element type to HOST_WIDE_INT.
4913 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4914 Alan Hayward <alan.hayward@arm.com>
4915 David Sherwood <david.sherwood@arm.com>
4917 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4918 rather than an int. Use plus_constant.
4919 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4920 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4923 Alan Hayward <alan.hayward@arm.com>
4924 David Sherwood <david.sherwood@arm.com>
4926 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4927 a HOST_WIDE_INT to a poly_int64.
4929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4930 Alan Hayward <alan.hayward@arm.com>
4931 David Sherwood <david.sherwood@arm.com>
4933 * calls.c (load_register_parameters): Cope with polynomial
4934 mode sizes. Require a constant size for BLKmode parameters
4935 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4936 forces a parameter to be padded at the lsb end in order to
4937 fill a complete number of words, require the parameter size
4938 to be ordered wrt UNITS_PER_WORD.
4940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4941 Alan Hayward <alan.hayward@arm.com>
4942 David Sherwood <david.sherwood@arm.com>
4944 * reload1.c (spill_stack_slot_width): Change element type
4945 from unsigned int to poly_uint64_pod.
4946 (alter_reg): Treat mode sizes as polynomial.
4948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4949 Alan Hayward <alan.hayward@arm.com>
4950 David Sherwood <david.sherwood@arm.com>
4952 * reload.c (complex_word_subreg_p): New function.
4953 (reload_inner_reg_of_subreg, push_reload): Use it.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4956 Alan Hayward <alan.hayward@arm.com>
4957 David Sherwood <david.sherwood@arm.com>
4959 * lra-constraints.c (process_alt_operands): Reject matched
4960 operands whose sizes aren't ordered.
4961 (match_reload): Refer to this check here.
4963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4964 Alan Hayward <alan.hayward@arm.com>
4965 David Sherwood <david.sherwood@arm.com>
4967 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4968 that the mode size is in the set {1, 2, 4, 8, 16}.
4970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4971 Alan Hayward <alan.hayward@arm.com>
4972 David Sherwood <david.sherwood@arm.com>
4974 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4975 Use plus_constant instead of gen_rtx_PLUS.
4977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4978 Alan Hayward <alan.hayward@arm.com>
4979 David Sherwood <david.sherwood@arm.com>
4981 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4982 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4983 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4984 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4985 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4986 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4987 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4988 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4989 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4990 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4992 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4993 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4994 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4995 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4996 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4997 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4998 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4999 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5000 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5001 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5003 * expr.c (emit_move_resolve_push): Treat the input and result
5004 of PUSH_ROUNDING as a poly_int64.
5005 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5006 (emit_push_insn): Likewise.
5007 * lra-eliminations.c (mark_not_eliminable): Likewise.
5008 * recog.c (push_operand): Likewise.
5009 * reload1.c (elimination_effects): Likewise.
5010 * rtlanal.c (nonzero_bits1): Likewise.
5011 * calls.c (store_one_arg): Likewise. Require the padding to be
5012 known at compile time.
5014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5015 Alan Hayward <alan.hayward@arm.com>
5016 David Sherwood <david.sherwood@arm.com>
5018 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5019 Use plus_constant instead of gen_rtx_PLUS.
5021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5022 Alan Hayward <alan.hayward@arm.com>
5023 David Sherwood <david.sherwood@arm.com>
5025 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5029 Alan Hayward <alan.hayward@arm.com>
5030 David Sherwood <david.sherwood@arm.com>
5032 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5033 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5034 via stack temporaries. Treat the mode size as polynomial too.
5036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5037 Alan Hayward <alan.hayward@arm.com>
5038 David Sherwood <david.sherwood@arm.com>
5040 * expr.c (expand_expr_real_2): When handling conversions involving
5041 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5042 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5043 as a poly_uint64 too.
5045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5046 Alan Hayward <alan.hayward@arm.com>
5047 David Sherwood <david.sherwood@arm.com>
5049 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5052 Alan Hayward <alan.hayward@arm.com>
5053 David Sherwood <david.sherwood@arm.com>
5055 * combine.c (can_change_dest_mode): Handle polynomial
5056 REGMODE_NATURAL_SIZE.
5057 * expmed.c (store_bit_field_1): Likewise.
5058 * expr.c (store_constructor): Likewise.
5059 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5060 and polynomial REGMODE_NATURAL_SIZE.
5061 (gen_lowpart_common): Likewise.
5062 * reginfo.c (record_subregs_of_mode): Likewise.
5063 * rtlanal.c (read_modify_subreg_p): Likewise.
5065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5066 Alan Hayward <alan.hayward@arm.com>
5067 David Sherwood <david.sherwood@arm.com>
5069 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5070 numbers of elements.
5072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5073 Alan Hayward <alan.hayward@arm.com>
5074 David Sherwood <david.sherwood@arm.com>
5076 * match.pd: Cope with polynomial numbers of vector elements.
5078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 Alan Hayward <alan.hayward@arm.com>
5080 David Sherwood <david.sherwood@arm.com>
5082 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5083 in a POINTER_PLUS_EXPR.
5085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5086 Alan Hayward <alan.hayward@arm.com>
5087 David Sherwood <david.sherwood@arm.com>
5089 * omp-simd-clone.c (simd_clone_subparts): New function.
5090 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5091 (ipa_simd_modify_function_body): Likewise.
5093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5094 Alan Hayward <alan.hayward@arm.com>
5095 David Sherwood <david.sherwood@arm.com>
5097 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5098 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5099 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5100 (expand_vector_condition, vector_element): Likewise.
5101 (subparts_gt): New function.
5102 (get_compute_type): Use subparts_gt.
5103 (count_type_subparts): Delete.
5104 (expand_vector_operations_1): Use subparts_gt instead of
5105 count_type_subparts.
5107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5108 Alan Hayward <alan.hayward@arm.com>
5109 David Sherwood <david.sherwood@arm.com>
5111 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5112 (vect_compile_time_alias): ...this new function. Do the calculation
5113 on poly_ints rather than trees.
5114 (vect_prune_runtime_alias_test_list): Update call accordingly.
5116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5117 Alan Hayward <alan.hayward@arm.com>
5118 David Sherwood <david.sherwood@arm.com>
5120 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5122 (vect_schedule_slp_instance): Likewise.
5124 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5125 Alan Hayward <alan.hayward@arm.com>
5126 David Sherwood <david.sherwood@arm.com>
5128 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5129 constant and extern definitions for variable-length vectors.
5130 (vect_get_constant_vectors): Note that the number of units
5131 is known to be constant.
5133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5134 Alan Hayward <alan.hayward@arm.com>
5135 David Sherwood <david.sherwood@arm.com>
5137 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5138 of units as polynomial. Choose between WIDE and NARROW based
5141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5142 Alan Hayward <alan.hayward@arm.com>
5143 David Sherwood <david.sherwood@arm.com>
5145 * tree-vect-stmts.c (simd_clone_subparts): New function.
5146 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5149 Alan Hayward <alan.hayward@arm.com>
5150 David Sherwood <david.sherwood@arm.com>
5152 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5153 vectors as polynomial. Use build_index_vector for
5156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5157 Alan Hayward <alan.hayward@arm.com>
5158 David Sherwood <david.sherwood@arm.com>
5160 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5161 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5162 for variable-length vectors.
5163 (vectorizable_mask_load_store): Treat the number of units as
5164 polynomial, asserting that it is constant if the condition has
5165 already been enforced.
5166 (vectorizable_store, vectorizable_load): Likewise.
5168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5169 Alan Hayward <alan.hayward@arm.com>
5170 David Sherwood <david.sherwood@arm.com>
5172 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5173 of units as polynomial. Punt if we can't tell at compile time
5174 which vector contains the final result.
5176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5177 Alan Hayward <alan.hayward@arm.com>
5178 David Sherwood <david.sherwood@arm.com>
5180 * tree-vect-loop.c (vectorizable_induction): Treat the number
5181 of units as polynomial. Punt on SLP inductions. Use an integer
5182 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5183 cast of such a series for variable-length floating-point
5186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5187 Alan Hayward <alan.hayward@arm.com>
5188 David Sherwood <david.sherwood@arm.com>
5190 * tree.h (build_index_vector): Declare.
5191 * tree.c (build_index_vector): New function.
5192 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5193 of units as polynomial, forcibly converting it to a constant if
5194 vectorizable_reduction has already enforced the condition.
5195 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5196 to create a {1,2,3,...} vector.
5197 (vectorizable_reduction): Treat the number of units as polynomial.
5198 Choose vectype_in based on the largest scalar element size rather
5199 than the smallest number of units. Enforce the restrictions
5202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5203 Alan Hayward <alan.hayward@arm.com>
5204 David Sherwood <david.sherwood@arm.com>
5206 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5207 number of units as polynomial.
5209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5210 Alan Hayward <alan.hayward@arm.com>
5211 David Sherwood <david.sherwood@arm.com>
5213 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5214 * target.def (autovectorize_vector_sizes): Return the vector sizes
5215 by pointer, using vector_sizes rather than a bitmask.
5216 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5217 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5218 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5220 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5221 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5222 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5223 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5224 * omp-general.c (omp_max_vf): Likewise.
5225 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5226 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5227 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5228 * tree-vect-slp.c (vect_slp_bb): Likewise.
5229 * doc/tm.texi: Regenerate.
5230 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5232 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5233 the vector size as a poly_uint64 rather than an unsigned int.
5234 (current_vector_size): Change from an unsigned int to a poly_uint64.
5235 (get_vectype_for_scalar_type): Update accordingly.
5236 * tree.h (build_truth_vector_type): Take the size and number of
5237 units as a poly_uint64 rather than an unsigned int.
5238 (build_vector_type): Add a temporary overload that takes
5239 the number of units as a poly_uint64 rather than an unsigned int.
5240 * tree.c (make_vector_type): Likewise.
5241 (build_truth_vector_type): Take the number of units as a poly_uint64
5242 rather than an unsigned int.
5244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5245 Alan Hayward <alan.hayward@arm.com>
5246 David Sherwood <david.sherwood@arm.com>
5248 * target.def (get_mask_mode): Take the number of units and length
5249 as poly_uint64s rather than unsigned ints.
5250 * targhooks.h (default_get_mask_mode): Update accordingly.
5251 * targhooks.c (default_get_mask_mode): Likewise.
5252 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5253 * doc/tm.texi: Regenerate.
5255 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5256 Alan Hayward <alan.hayward@arm.com>
5257 David Sherwood <david.sherwood@arm.com>
5259 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5260 * omp-general.c (omp_max_vf): Likewise.
5261 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5262 (expand_omp_simd): Handle polynomial safelen.
5263 * omp-low.c (omplow_simd_context): Add a default constructor.
5264 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5265 (lower_rec_simd_input_clauses): Update accordingly.
5266 (lower_rec_input_clauses): Likewise.
5268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5269 Alan Hayward <alan.hayward@arm.com>
5270 David Sherwood <david.sherwood@arm.com>
5272 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5273 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5274 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5275 (vect_analyze_slp_cost): Likewise.
5276 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5277 (vect_model_load_cost): Likewise.
5279 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5280 Alan Hayward <alan.hayward@arm.com>
5281 David Sherwood <david.sherwood@arm.com>
5283 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5284 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5285 from an unsigned int * to a poly_uint64_pod *.
5286 (calculate_unrolling_factor): New function.
5287 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5289 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5290 Alan Hayward <alan.hayward@arm.com>
5291 David Sherwood <david.sherwood@arm.com>
5293 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5294 from an unsigned int to a poly_uint64.
5295 (_loop_vec_info::slp_unrolling_factor): Likewise.
5296 (_loop_vec_info::vectorization_factor): Change from an int
5298 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5299 (vect_get_num_vectors): New function.
5300 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5301 (vect_get_num_copies): Use vect_get_num_vectors.
5302 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5303 to an unsigned int *.
5304 (vect_analyze_data_refs): Change min_vf from an int * to a
5306 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5307 than an unsigned HOST_WIDE_INT.
5308 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5309 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5310 to an unsigned int *.
5311 (vect_analyze_data_ref_dependences): Likewise.
5312 (vect_compute_data_ref_alignment): Handle polynomial vf.
5313 (vect_enhance_data_refs_alignment): Likewise.
5314 (vect_prune_runtime_alias_test_list): Likewise.
5315 (vect_shift_permute_load_chain): Likewise.
5316 (vect_supportable_dr_alignment): Likewise.
5317 (dependence_distance_ge_vf): Take the vectorization factor as a
5318 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5319 (vect_analyze_data_refs): Change min_vf from an int * to a
5321 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5322 vfm1 as a poly_uint64 rather than an int. Make the same change
5323 for the returned bound_scalar.
5324 (vect_gen_vector_loop_niters): Handle polynomial vf.
5325 (vect_do_peeling): Likewise. Update call to
5326 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5327 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5329 * tree-vect-loop.c (vect_determine_vectorization_factor)
5330 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5331 (vect_get_known_peeling_cost): Likewise.
5332 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5333 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5334 (vect_transform_loop): Likewise. Use the lowest possible VF when
5335 updating the upper bounds of the loop.
5336 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5338 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5339 polynomial unroll factors.
5340 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5341 (vect_make_slp_decision): Likewise.
5342 (vect_supported_load_permutation_p): Likewise, and polynomial
5344 (vect_analyze_slp_cost): Handle polynomial vf.
5345 (vect_slp_analyze_node_operations): Likewise.
5346 (vect_slp_analyze_bb_1): Likewise.
5347 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5348 than an unsigned HOST_WIDE_INT.
5349 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5350 (vectorizable_load): Handle polynomial vf.
5351 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5353 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5356 Alan Hayward <alan.hayward@arm.com>
5357 David Sherwood <david.sherwood@arm.com>
5359 * match.pd: Handle bit operations involving three constants
5360 and try to fold one pair.
5362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5364 * tree-vect-loop-manip.c: Include gimple-fold.h.
5365 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5366 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5367 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5368 Add a path that uses a step of VF instead of 1, but disable it
5370 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5371 and niters_no_overflow parameters. Update calls to
5372 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5373 Create a new SSA name if the latter choses to use a ste other
5374 than zero, and return it via niters_vector_mult_vf_var.
5375 * tree-vect-loop.c (vect_transform_loop): Update calls to
5376 vect_do_peeling, vect_gen_vector_loop_niters and
5377 slpeel_make_loop_iterate_ntimes.
5378 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5379 (vect_gen_vector_loop_niters): Update declarations after above changes.
5381 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5383 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5384 128-bit round to integer instructions.
5385 (ceil<mode>2): Likewise.
5386 (btrunc<mode>2): Likewise.
5387 (round<mode>2): Likewise.
5389 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5391 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5392 unaligned VSX load/store on P8/P9.
5393 (expand_block_clear): Allow the use of unaligned VSX
5394 load/store on P8/P9.
5396 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5398 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5400 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5401 swap associated with both a load and a store.
5403 2018-01-02 Andrew Waterman <andrew@sifive.com>
5405 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5406 * config/riscv/riscv.md (clear_cache): Use it.
5408 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5410 * web.c: Remove out-of-date comment.
5412 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5414 * expr.c (fixup_args_size_notes): Check that any existing
5415 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5416 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5417 (emit_single_push_insn): ...here.
5419 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5421 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5422 (const_vector_encoded_nelts): New function.
5423 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5424 (const_vector_int_elt, const_vector_elt): Declare.
5425 * emit-rtl.c (const_vector_int_elt_1): New function.
5426 (const_vector_elt): Likewise.
5427 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5428 of CONST_VECTOR_ELT.
5430 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5432 * expr.c: Include rtx-vector-builder.h.
5433 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5434 directly on the tree encoding.
5435 (const_vector_from_tree): Likewise.
5436 * optabs.c: Include rtx-vector-builder.h.
5437 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5438 sequence of "u" values.
5439 * vec-perm-indices.c: Include rtx-vector-builder.h.
5440 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5441 directly on the vec_perm_indices encoding.
5443 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5445 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5446 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5447 * rtx-vector-builder.h: New file.
5448 * rtx-vector-builder.c: Likewise.
5449 * rtl.h (rtx_def::u2): Add a const_vector field.
5450 (CONST_VECTOR_NPATTERNS): New macro.
5451 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5452 (CONST_VECTOR_DUPLICATE_P): Likewise.
5453 (CONST_VECTOR_STEPPED_P): Likewise.
5454 (CONST_VECTOR_ENCODED_ELT): Likewise.
5455 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5456 (unwrap_const_vec_duplicate): Likewise.
5457 (const_vec_series_p): Check for a non-duplicated vector encoding.
5458 Say that the function only returns true for integer vectors.
5459 * emit-rtl.c: Include rtx-vector-builder.h.
5460 (gen_const_vec_duplicate_1): Delete.
5461 (gen_const_vector): Call gen_const_vec_duplicate instead of
5462 gen_const_vec_duplicate_1.
5463 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5464 (gen_const_vec_duplicate): Use rtx_vector_builder.
5465 (gen_const_vec_series): Likewise.
5466 (gen_rtx_CONST_VECTOR): Likewise.
5467 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5468 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5469 Build a new vector rather than modifying a CONST_VECTOR in-place.
5470 (handle_special_swappables): Update call accordingly.
5471 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5472 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5473 Build a new vector rather than modifying a CONST_VECTOR in-place.
5474 (handle_special_swappables): Update call accordingly.
5476 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5478 * simplify-rtx.c (simplify_const_binary_operation): Use
5479 CONST_VECTOR_ELT instead of XVECEXP.
5481 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5483 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5484 the selector elements to be different from the data elements
5485 if the selector is a VECTOR_CST.
5486 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5487 ssizetype for the selector.
5489 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5491 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5492 before testing each element individually.
5493 * tree-vect-generic.c (lower_vec_perm): Likewise.
5495 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5497 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5498 * selftest-run-tests.c (selftest::run_tests): Call it.
5499 * vector-builder.h (vector_builder::operator ==): New function.
5500 (vector_builder::operator !=): Likewise.
5501 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5502 (vec_perm_indices::all_from_input_p): New function.
5503 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5504 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5505 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5506 instead of reading the VECTOR_CST directly. Detect whether both
5507 vector inputs are the same before constructing the vec_perm_indices,
5508 and update the number of inputs argument accordingly. Use the
5509 utility functions added above. Only construct sel2 if we need to.
5511 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5513 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5514 the broadcast of the low byte.
5515 (expand_mult_highpart): Use an explicit encoding for the permutes.
5516 * optabs-query.c (can_mult_highpart_p): Likewise.
5517 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5518 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5519 (vectorizable_bswap): Likewise.
5520 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5521 explicit encoding for the power-of-2 permutes.
5522 (vect_permute_store_chain): Likewise.
5523 (vect_grouped_load_supported): Likewise.
5524 (vect_permute_load_chain): Likewise.
5526 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5528 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5529 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5530 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5531 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5532 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5533 (vect_gen_perm_mask_any): Likewise.
5535 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5537 * int-vector-builder.h: New file.
5538 * vec-perm-indices.h: Include int-vector-builder.h.
5539 (vec_perm_indices): Redefine as an int_vector_builder.
5540 (auto_vec_perm_indices): Delete.
5541 (vec_perm_builder): Redefine as a stand-alone class.
5542 (vec_perm_indices::vec_perm_indices): New function.
5543 (vec_perm_indices::clamp): Likewise.
5544 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5545 (vec_perm_indices::new_vector): New function.
5546 (vec_perm_indices::new_expanded_vector): Update for new
5547 vec_perm_indices class.
5548 (vec_perm_indices::rotate_inputs): New function.
5549 (vec_perm_indices::all_in_range_p): Operate directly on the
5550 encoded form, without computing elided elements.
5551 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5552 encoding. Update for new vec_perm_indices class.
5553 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5554 the given vec_perm_builder.
5555 (expand_vec_perm_var): Update vec_perm_builder constructor.
5556 (expand_mult_highpart): Use vec_perm_builder instead of
5557 auto_vec_perm_indices.
5558 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5559 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5560 or double series encoding as appropriate.
5561 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5562 vec_perm_indices instead of auto_vec_perm_indices.
5563 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5564 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5565 (vect_permute_store_chain): Likewise.
5566 (vect_grouped_load_supported): Likewise.
5567 (vect_permute_load_chain): Likewise.
5568 (vect_shift_permute_load_chain): Likewise.
5569 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5570 (vect_transform_slp_perm_load): Likewise.
5571 (vect_schedule_slp_instance): Likewise.
5572 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5573 (vectorizable_mask_load_store): Likewise.
5574 (vectorizable_bswap): Likewise.
5575 (vectorizable_store): Likewise.
5576 (vectorizable_load): Likewise.
5577 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5578 vec_perm_indices instead of auto_vec_perm_indices. Use
5579 tree_to_vec_perm_builder to read the vector from a tree.
5580 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5581 vec_perm_builder instead of a vec_perm_indices.
5582 (have_whole_vector_shift): Use vec_perm_builder and
5583 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5584 truncation to calc_vec_perm_mask_for_shift.
5585 (vect_create_epilog_for_reduction): Likewise.
5586 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5587 from auto_vec_perm_indices to vec_perm_indices.
5588 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5589 instead of changing individual elements.
5590 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5591 the vector in d.perm.
5592 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5593 from auto_vec_perm_indices to vec_perm_indices.
5594 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5595 instead of changing individual elements.
5596 (arm_vectorize_vec_perm_const): Use new_vector to install
5597 the vector in d.perm.
5598 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5599 Update vec_perm_builder constructor.
5600 (rs6000_expand_interleave): Likewise.
5601 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5602 (rs6000_expand_interleave): Likewise.
5604 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5606 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5607 to qimode could truncate the indices.
5608 * optabs.c (expand_vec_perm_var): Likewise.
5610 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5612 * Makefile.in (OBJS): Add vec-perm-indices.o.
5613 * vec-perm-indices.h: New file.
5614 * vec-perm-indices.c: Likewise.
5615 * target.h (vec_perm_indices): Replace with a forward class
5617 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5618 * optabs.h: Include vec-perm-indices.h.
5619 (expand_vec_perm): Delete.
5620 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5621 (expand_vec_perm_const): Declare.
5622 * target.def (vec_perm_const_ok): Replace with...
5623 (vec_perm_const): ...this new hook.
5624 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5625 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5626 * doc/tm.texi: Regenerate.
5627 * optabs.def (vec_perm_const): Delete.
5628 * doc/md.texi (vec_perm_const): Likewise.
5629 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5630 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5631 expand_vec_perm for constant permutation vectors. Assert that
5632 the mode of variable permutation vectors is the integer equivalent
5633 of the mode that is being permuted.
5634 * optabs-query.h (selector_fits_mode_p): Declare.
5635 * optabs-query.c: Include vec-perm-indices.h.
5636 (selector_fits_mode_p): New function.
5637 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5638 is defined, instead of checking whether the vec_perm_const_optab
5639 exists. Use targetm.vectorize.vec_perm_const instead of
5640 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5641 fit in the vector mode before using a variable permute.
5642 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5643 vec_perm_indices instead of an rtx.
5644 (expand_vec_perm): Replace with...
5645 (expand_vec_perm_const): ...this new function. Take the selector
5646 as a vec_perm_indices rather than an rtx. Also take the mode of
5647 the selector. Update call to shift_amt_for_vec_perm_mask.
5648 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5649 Use vec_perm_indices::new_expanded_vector to expand the original
5650 selector into bytes. Check whether the indices fit in the vector
5651 mode before using a variable permute.
5652 (expand_vec_perm_var): Make global.
5653 (expand_mult_highpart): Use expand_vec_perm_const.
5654 * fold-const.c: Includes vec-perm-indices.h.
5655 * tree-ssa-forwprop.c: Likewise.
5656 * tree-vect-data-refs.c: Likewise.
5657 * tree-vect-generic.c: Likewise.
5658 * tree-vect-loop.c: Likewise.
5659 * tree-vect-slp.c: Likewise.
5660 * tree-vect-stmts.c: Likewise.
5661 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5663 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5664 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5665 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5666 (aarch64_vectorize_vec_perm_const): ...this new function.
5667 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5668 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5669 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5670 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5671 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5672 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5673 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5675 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5676 check for NEON modes.
5677 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5678 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5679 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5680 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5682 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5683 the old VEC_PERM_CONST conditions.
5684 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5685 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5686 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5687 (ia64_vectorize_vec_perm_const_ok): Merge into...
5688 (ia64_vectorize_vec_perm_const): ...this new function.
5689 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5690 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5691 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5692 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5693 * config/mips/mips.c (mips_expand_vec_perm_const)
5694 (mips_vectorize_vec_perm_const_ok): Merge into...
5695 (mips_vectorize_vec_perm_const): ...this new function.
5696 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5697 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5698 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5699 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5700 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5701 (rs6000_expand_vec_perm_const): Delete.
5702 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5704 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5705 (altivec_expand_vec_perm_const_le): Take each operand individually.
5706 Operate on constant selectors rather than rtxes.
5707 (altivec_expand_vec_perm_const): Likewise. Update call to
5708 altivec_expand_vec_perm_const_le.
5709 (rs6000_expand_vec_perm_const): Delete.
5710 (rs6000_vectorize_vec_perm_const_ok): Delete.
5711 (rs6000_vectorize_vec_perm_const): New function.
5712 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5713 an element count and rtx array.
5714 (rs6000_expand_extract_even): Update call accordingly.
5715 (rs6000_expand_interleave): Likewise.
5716 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5717 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5718 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5719 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5720 (rs6000_expand_vec_perm_const): Delete.
5721 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5722 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5723 (altivec_expand_vec_perm_const_le): Take each operand individually.
5724 Operate on constant selectors rather than rtxes.
5725 (altivec_expand_vec_perm_const): Likewise. Update call to
5726 altivec_expand_vec_perm_const_le.
5727 (rs6000_expand_vec_perm_const): Delete.
5728 (rs6000_vectorize_vec_perm_const_ok): Delete.
5729 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5730 reference to the SPE evmerge intructions.
5731 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5732 an element count and rtx array.
5733 (rs6000_expand_extract_even): Update call accordingly.
5734 (rs6000_expand_interleave): Likewise.
5735 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5736 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5738 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5740 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5742 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5743 vector mode and that that mode matches the mode of the data
5745 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5746 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5747 directly using expand_vec_perm_1 when forcing selectors into
5749 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5751 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5753 * optabs-query.h (can_vec_perm_p): Delete.
5754 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5755 * optabs-query.c (can_vec_perm_p): Split into...
5756 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5757 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5758 particular selector is valid.
5759 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5760 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5761 (vect_grouped_load_supported): Likewise.
5762 (vect_shift_permute_load_chain): Likewise.
5763 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5764 (vect_transform_slp_perm_load): Likewise.
5765 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5766 (vectorizable_bswap): Likewise.
5767 (vect_gen_perm_mask_checked): Likewise.
5768 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5769 implementations of variable permutation vectors into account
5770 when deciding which selector to use.
5771 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5772 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5773 with a false third argument.
5774 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5775 to test whether the constant selector is valid and can_vec_perm_var_p
5776 to test whether a variable selector is valid.
5778 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5780 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5781 * optabs-query.c (can_vec_perm_p): Likewise.
5782 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5783 instead of vec_perm_indices.
5784 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5785 (vect_gen_perm_mask_checked): Likewise,
5786 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5787 (vect_gen_perm_mask_checked): Likewise,
5789 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5791 * optabs-query.h (qimode_for_vec_perm): Declare.
5792 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5793 (qimode_for_vec_perm): ...this new function.
5794 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5796 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5798 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5799 does not have a conditional at the top.
5801 2018-01-02 Richard Biener <rguenther@suse.de>
5803 * ipa-inline.c (big_speedup_p): Fix expression.
5805 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5808 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5811 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5815 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5816 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5817 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5818 cond_taken_branch_cost 3->4.
5820 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5822 PR tree-optimization/83581
5823 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5824 TODO_cleanup_cfg if any changes have been made.
5827 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5828 convert_modes if target mode has the right side, but different mode
5832 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5833 last argument when extracting from CONCAT. If either from_real or
5834 from_imag is NULL, use expansion through memory. If result is not
5835 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5836 the parts directly to inner mode, if even that fails, use expansion
5840 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5841 check for bswap in mode rather than HImode and use that in expand_unop
5844 Copyright (C) 2018 Free Software Foundation, Inc.
5846 Copying and distribution of this file, with or without modification,
5847 are permitted in any medium without royalty provided the copyright
5848 notice and this notice are preserved.