1 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
5 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
7 (rs6000_supported_cpu_names): New static variable.
8 (linux_cpu_translation_table): Likewise.
9 (elf_platform) <cpu>: Define new static variable and use it.
10 Translate kernel AT_PLATFORM name to canonical name if needed.
11 Error if platform name is unknown.
13 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
16 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
18 2018-02-01 Jeff Law <law@redhat.com>
21 * config/i386/i386.c (release_scratch_register_on_entry): Add new
22 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
23 the scratch if RELEASE_VIA_POP is false.
24 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
25 If we have to save a temporary register, decrement SIZE appropriately.
26 Pass new arguments to release_scratch_register_on_entry.
27 (ix86_adjust_stack_and_probe): Likewise.
28 (ix86_emit_probe_stack_range): Pass new arguments to
29 release_scratch_register_on_entry.
31 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
33 PR rtl-optimization/84157
34 * combine.c (change_zero_ext): Use REG_P predicate in
35 front of HARD_REGISTER_P predicate.
37 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
39 * config/avr/avr.c (avr_option_override): Move disabling of
40 -fdelete-null-pointer-checks to...
41 * common/config/avr/avr-common.c (avr_option_optimization_table):
44 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
46 PR tree-optimization/81635
47 * tree-data-ref.c (split_constant_offset_1): For types that
48 wrap on overflow, try to use range info to prove that wrapping
51 2018-02-01 Renlin Li <renlin.li@arm.com>
54 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
56 (aarch64_register_move_cost): Likewise.
57 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
59 (REG_CLASS_NAMES): Likewise.
60 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
61 TAILCALL_ADDR_REGS. Remove IP registers.
62 * config/aarch64/aarch64.md (Ucs): Update register constraint.
64 2018-02-01 Richard Biener <rguenther@suse.de>
66 * domwalk.h (dom_walker::dom_walker): Add additional constructor
67 for specifying RPO order and allow NULL for that.
68 * domwalk.c (dom_walker::dom_walker): Likewise.
69 (dom_walker::walk): Handle NULL RPO order.
70 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
72 (rewrite_update_dom_walker): Likewise.
73 (mark_def_dom_walker): Likewise.
75 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
77 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
78 (aarch64_maybe_expand_sve_subreg_move): Declare.
79 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
80 * config/aarch64/predicates.md (aarch64_any_register_operand): New
82 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
83 that are semantically a reverse operation.
84 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
85 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
86 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
88 (aarch64_can_change_mode_class): For big-endian, forbid changes
89 between two SVE modes if they have different element sizes.
91 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
93 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
94 the TImode handling for big-endian targets.
96 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
98 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
99 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
101 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
102 Remove BSWAP handing for big-endian targets and use the form of
103 LD1RQ appropariate for the mode.
105 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
107 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
108 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
111 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
114 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
115 check for operands that need to go through aarch64_sve_reload_be.
117 2018-02-01 Jakub Jelinek <jakub@redhat.com>
119 PR tree-optimization/81661
120 PR tree-optimization/84117
121 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
122 * tree-eh.c: Include gimplify.h.
123 (find_trapping_overflow, replace_trapping_overflow,
124 rewrite_to_non_trapping_overflow): New functions.
125 * tree-vect-loop.c: Include tree-eh.h.
126 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
127 * tree-data-ref.c: Include tree-eh.h.
128 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
130 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
132 PR rtl-optimization/84123
133 * combine.c (change_zero_ext): Check if hard register satisfies
134 can_change_dest_mode before calling gen_lowpart_SUBREG.
136 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
139 * ira.c (ira_init_register_move_cost): Remove assert.
141 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
143 PR rtl-optimization/84071
144 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
145 * doc/tm.texi: Regenerate.
147 2018-01-31 Richard Biener <rguenther@suse.de>
149 PR tree-optimization/84132
150 * tree-data-ref.c (analyze_miv_subscript): Properly
151 check whether evolution_function_is_affine_multivariate_p
152 before calling gcd_of_steps_may_divide_p.
154 2018-01-31 Julia Koval <julia.koval@intel.com>
157 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
158 * config/i386/i386.md (rdpid_rex64) New.
159 (rdpid): Make 32bit only.
161 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
164 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
165 an IDENTIFIER_NODE for FUNCTION_TYPE's.
167 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
170 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
172 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
174 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
176 PR rtl-optimization/84071
177 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
178 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
180 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
182 * config/arc/arc.c (arc_handle_aux_attribute): New function.
183 (arc_attribute_table): Add 'aux' attribute.
184 (arc_in_small_data_p): Consider aux like variables.
185 (arc_is_aux_reg_p): New function.
186 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
187 (arc_get_aux_arg): New function.
188 (prepare_move_operands): Handle aux-register access.
189 (arc_handle_aux_attribute): New function.
190 * doc/extend.texi (ARC Variable attributes): Add subsection.
192 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
194 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
195 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
196 (arc_attribute_table): Add 'uncached' attribute.
197 (arc_print_operand): Print '.di' flag for uncached memory
199 (arc_in_small_data_p): Do not consider for small data the uncached
201 (arc_is_uncached_mem_p): New function.
202 * config/arc/predicates.md (compact_store_memory_operand): Check
203 for uncached memory accesses.
204 (nonvol_nonimm_operand): Likewise.
205 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
207 2018-01-31 Jakub Jelinek <jakub@redhat.com>
210 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
211 falign-loops=): Add Optimization flag.
213 2018-01-30 Jeff Law <law@redhat.com>
216 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
217 INT_REGISTERS_SAVED. Check it prior to calling
218 get_scratch_register_on_entry.
219 (ix86_adjust_stack_and_probe): Similarly.
220 (ix86_emit_probe_stack_range): Similarly.
221 (ix86_expand_prologue): Corresponding changes.
223 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
226 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
227 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
229 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
232 * lra-constraints.c (curr_insn_transform): Process AND in the
235 2018-01-30 Jakub Jelinek <jakub@redhat.com>
237 PR rtl-optimization/83986
238 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
239 dependence against last_pending_memory_flush in addition to
242 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
244 PR tree-optimization/81611
245 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
248 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
250 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
253 2018-01-30 Richard Biener <rguenther@suse.de>
254 Jakub Jelinek <jakub@redhat.com>
256 PR tree-optimization/84111
257 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
258 inner loops added during recursion, as they don't have up-to-date
261 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
264 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
265 (can_inline_edge_by_limits_p): ... here.
266 (can_early_inline_edge_p, check_callers,
267 update_caller_keys, update_callee_keys, recursive_inlining,
268 add_new_edges_to_heap, speculation_useful_p,
269 inline_small_functions,
270 inline_small_functions, flatten_function,
271 inline_to_all_callers_1): Update.
273 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
275 * profile-count.c (profile_count::combine_with_ipa_count): Handle
278 2018-01-30 Richard Biener <rguenther@suse.de>
280 PR tree-optimization/83008
281 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
282 invariant and constant vector uses in stmts when they need
285 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
288 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
289 * configure: Regenerate.
291 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
293 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
295 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
296 Use gen_rtx_REG rather than gen_lowpart.
298 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
300 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
301 rather than 0 when creating partial subregs.
303 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
305 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
308 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
311 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
312 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
313 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
314 flags. This restores the settings used before the 2017-07-24.
315 Turning off pre increment/decrement/modify allows IVOPTS to
316 optimize DF/SF loops where the index is an int.
318 2018-01-29 Richard Biener <rguenther@suse.de>
319 Kelvin Nilsen <kelvin@gcc.gnu.org>
322 * tree-vect-stmts.c (vectorizable_call): Don't call
323 targetm.vectorize_builtin_md_vectorized_function if callee is
326 2018-01-22 Carl Love <cel@us.ibm.com>
328 * doc/extend.tex: Fix typo in second arg in
329 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
331 2018-01-29 Richard Biener <rguenther@suse.de>
333 PR tree-optimization/84086
334 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
335 (flush_ssaname_freelist): When SSA names were released reset
338 2018-01-29 Richard Biener <rguenther@suse.de>
340 PR tree-optimization/84057
341 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
342 removed paths when removing edges.
344 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
346 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
347 -mfunction-return=@var{choice}.
349 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
352 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
354 (layout::print_source_line): Likewise.
355 (test_get_line_width_without_trailing_whitespace): Add test cases.
357 2018-01-27 Jakub Jelinek <jakub@redhat.com>
360 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
363 2018-01-26 Jim Wilson <jimw@sifive.com>
365 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
367 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
370 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
372 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
373 and CMP + SUB-immediate -> SUBS.
375 2018-01-26 Martin Sebor <msebor@redhat.com>
377 PR tree-optimization/83896
378 * tree-ssa-strlen.c (get_string_len): Rename...
379 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
380 Avoid assuming length is constant.
381 (handle_char_store): Use HOST_WIDE_INT for string length.
383 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
386 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
387 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
389 2018-01-26 Richard Biener <rguenther@suse.de>
391 PR rtl-optimization/84003
392 * dse.c (record_store): Only record redundant stores when
393 the earlier store aliases at least all accesses the later one does.
395 2018-01-26 Jakub Jelinek <jakub@redhat.com>
397 PR rtl-optimization/83985
398 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
399 REG_CFA_RESTORE insns.
400 (delete_unmarked_insns): Don't ignore separate shrink wrapping
401 REG_CFA_RESTORE insns here.
404 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
405 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
407 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
409 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
410 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
411 (arc_init): Likewise.
412 (arc_override_options): Likewise.
413 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
415 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
417 * config/arc/arc.h (TARGET_DBNZ): Define.
418 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
419 properly set the tune attribute.
420 (dbnz): Use TARGET_DBNZ guard.
421 * config/arc/arc.opt (mtune): Add core3 option.
423 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
425 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
426 recognize new pic like addresses.
427 (arc_delegitimize_address): Clean up.
429 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
431 * config/arc/arc-arches.def: Option mrf16 valid for all
433 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
434 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
435 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
436 * config/arc/arc-tables.opt: Regenerate.
437 * config/arc/arc.c (arc_conditional_register_usage): Handle
438 reduced register file case.
439 (arc_file_start): Set must have build attributes.
440 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
442 * config/arc/arc.opt (mrf16): Add new option.
443 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
444 * config/arc/genmultilib.awk: Handle new mrf16 option.
445 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
446 * config/arc/t-multilib: Regenerate.
447 * doc/invoke.texi (ARC Options): Document mrf16 option.
449 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
451 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
452 * config/arc/arc.c (arc_handle_secure_attribute): New function.
453 (arc_attribute_table): Add 'secure_call' attribute.
454 (arc_print_operand): Print secure call operand.
455 (arc_function_ok_for_sibcall): Don't optimize tail calls when
457 (arc_is_secure_call_p): New function. * config/arc/arc.md
458 (call_i): Add support for sjli instruction.
459 (call_value_i): Likewise.
460 * config/arc/constraints.md (Csc): New constraint.
462 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
463 John Eric Martin <John.Martin@emmicro-us.com>
465 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
466 * config/arc/arc.c (_arc_jli_section): New struct.
467 (arc_jli_section): New type.
468 (rc_jli_sections): New static variable.
469 (arc_handle_jli_attribute): New function.
470 (arc_attribute_table): Add jli_always and jli_fixed attribute.
471 (arc_file_end): New function.
472 (TARGET_ASM_FILE_END): Define.
473 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
474 (arc_add_jli_section): New function.
475 (jli_call_scan): Likewise.
476 (arc_reorg): Call jli_call_scan.
477 (arc_output_addsi): Remove 'S' from printing asm operand.
478 (arc_is_jli_call_p): New function.
479 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
481 (movhi_insn): Likewise.
482 (movsi_insn): Likewise.
483 (movsi_set_cc_insn): Likewise.
484 (loadqi_update): Likewise.
485 (load_zeroextendqisi_update): Likewise.
486 (load_signextendqisi_update): Likewise.
487 (loadhi_update): Likewise.
488 (load_zeroextendhisi_update): Likewise.
489 (load_signextendhisi_update): Likewise.
490 (loadsi_update): Likewise.
491 (loadsf_update): Likewise.
492 (movsicc_insn): Likewise.
493 (bset_insn): Likewise.
494 (bxor_insn): Likewise.
495 (bclr_insn): Likewise.
496 (bmsk_insn): Likewise.
497 (bicsi3_insn): Likewise.
498 (cmpsi_cc_c_insn): Likewise.
499 (movsi_ne): Likewise.
500 (movsi_cond_exec): Likewise.
501 (clrsbsi2): Likewise.
512 (call_i): Remove 'S' asm letter, add jli instruction.
513 (call_value_i): Likewise.
514 * config/arc/arc.op (mjli-always): New option.
515 * config/arc/constraints.md (Cji): New constraint.
516 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
518 (subsf3_fpx): Likewise.
519 (mulsf3_fpx): Likewise.
520 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
522 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
524 * doc/invoke.texi (ARC): Document mjli-always option.
526 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
528 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
529 and use incw and decw where possible
530 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
532 2018-01-26 Richard Biener <rguenther@suse.de>
534 PR tree-optimization/81082
535 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
536 association if it requires casting to unsigned.
537 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
538 from fold_plusminus_mult_expr to catch important cases late when
539 range info is available.
541 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
543 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
544 * configure.ac (hidden_linkonce): New test.
545 * configure: Regenerate.
546 * config.in: Regenerate.
548 2018-01-26 Julia Koval <julia.koval@intel.com>
550 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
551 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
552 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
553 _mm_mask_bitshuffle_epi64_mask): Fix type.
554 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
555 USI_FTYPE_V4DI_V4DI_USI): Remove.
556 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
557 __builtin_ia32_vpshufbitqmb256_mask,
558 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
559 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
560 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
562 2018-01-26 Alan Modra <amodra@gmail.com>
565 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
566 UNSPEC_VBPERMQ. Sort other unspecs.
568 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
570 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
572 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
575 * predict.c (drop_profile): Do not push/pop cfun; update also
577 (handle_missing_profiles): Fix logic looking for zero profiles.
579 2018-01-25 Jakub Jelinek <jakub@redhat.com>
582 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
583 on functions with #pragma omp declare simd or functions with simd
585 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
586 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
587 Remove trailing \n from warning_at calls.
589 2018-01-25 Tom de Vries <tom@codesourcery.com>
592 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
593 for neutered workers.
595 2018-01-24 Joseph Myers <joseph@codesourcery.com>
598 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
599 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
601 2018-01-24 Jeff Law <law@redhat.com>
604 * i386.c (get_probe_interval): Move to earlier point.
605 (ix86_compute_frame_layout): If -fstack-clash-protection and
606 the frame is larger than the probe interval, then use pushes
607 to save registers rather than reg->mem moves.
608 (ix86_expand_prologue): Remove conditional for int_registers_saved
611 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
614 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
615 min/max for never referenced object.
617 2018-01-24 Jakub Jelinek <jakub@redhat.com>
620 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
622 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
623 attributes from DECL_ATTRIBUTES (decl) without affecting
624 DECL_ATTRIBUTES (current_function_decl).
625 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
626 functions with non-NULL DECL_ABSTRACT_ORIGIN.
628 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
630 PR tree-optimization/83979
631 * fold-const.c (fold_comparison): Use constant_boolean_node
632 instead of boolean_{true,false}_node.
634 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
636 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
639 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
641 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
642 Simplify the clause that sets the length attribute.
643 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
644 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
645 clause that sets the length attribute.
646 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
648 2018-01-24 Tom de Vries <tom@codesourcery.com>
651 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
652 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
653 Add strict parameter.
654 (prevent_branch_around_nothing): Insert dummy insn between branch to
655 label and label with no ptx insn inbetween.
656 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
658 2018-01-24 Tom de Vries <tom@codesourcery.com>
661 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
662 for neutered threads in warp.
663 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
665 2018-01-24 Richard Biener <rguenther@suse.de>
667 PR tree-optimization/83176
668 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
671 2018-01-24 Richard Biener <rguenther@suse.de>
673 PR tree-optimization/82819
674 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
675 code generating pluses that are no-ops in the target precision.
677 2018-01-24 Richard Biener <rguenther@suse.de>
680 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
682 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
684 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
685 to merge probabilities.
686 * predict.c (probably_never_executed): Also mark as cold functions
687 with global 0 profile and guessed local profile.
688 * profile-count.c (profile_probability::combine_with_count): New
690 * profile-count.h (profile_probability::operator*,
691 profile_probability::operator*=, profile_probability::operator/,
692 profile_probability::operator/=): Reduce precision to adjusted
693 and set value to guessed on contradictory divisions.
694 (profile_probability::combine_with_freq): Remove.
695 (profile_probability::combine_wiht_count): Declare.
696 (profile_count::force_nonzero):: Set to adjusted.
697 (profile_count::probability_in):: Set quality to adjusted.
698 * tree-ssa-tail-merge.c (replace_block_by): Use
701 2018-01-23 Andrew Waterman <andrew@sifive.com>
702 Jim Wilson <jimw@sifive.com>
704 * config/riscv/riscv.c (riscv_stack_boundary): New.
705 (riscv_option_override): Set riscv_stack_boundary. Handle
706 riscv_preferred_stack_boundary_arg.
707 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
708 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
709 (STACK_BOUNDARY): Set to riscv_stack_boundary.
710 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
711 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
712 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
714 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
717 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
718 of struct ix86_frame.
719 (ix86_expand_epilogue): Likewise. Add a local variable for
720 the reg_save_offset field in struct ix86_frame.
722 2018-01-23 Bin Cheng <bin.cheng@arm.com>
724 PR tree-optimization/82604
725 * tree-loop-distribution.c (enum partition_kind): New enum item
726 PKIND_PARTIAL_MEMSET.
727 (partition_builtin_p): Support above new enum item.
728 (generate_code_for_partition): Ditto.
729 (compute_access_range): Differentiate cases that equality can be
730 proven at all loops, the innermost loops or no loops.
731 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
732 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
733 (finalize_partitions, distribute_loop): Don't fuse partition of
734 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
735 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
738 2018-01-23 Martin Liska <mliska@suse.cz>
740 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
741 order to ignore the predictor.
742 (PRED_POLYMORPHIC_CALL): Likewise.
743 (PRED_RECURSIVE_CALL): Likewise.
745 2018-01-23 Martin Liska <mliska@suse.cz>
747 * tree-profile.c (tree_profiling): Print function header to
748 aware reader which function we are working on.
749 * value-prof.c (gimple_find_values_to_profile): Do not print
750 not interesting value histograms.
752 2018-01-23 Martin Liska <mliska@suse.cz>
754 * profile-count.h (enum profile_quality): Add
755 profile_uninitialized as the first value. Do not number values
756 as they are zero based.
757 (profile_count::verify): Update sanity check.
758 (profile_probability::verify): Likewise.
760 2018-01-23 Nathan Sidwell <nathan@acm.org>
762 * doc/invoke.texi (ffor-scope): Deprecate.
764 2018-01-23 David Malcolm <dmalcolm@redhat.com>
766 PR tree-optimization/83510
767 * domwalk.c (set_all_edges_as_executable): New function.
768 (dom_walker::dom_walker): Convert bool param
769 "skip_unreachable_blocks" to enum reachability. Move setup of
770 edge flags to set_all_edges_as_executable and only do it when
771 reachability is REACHABLE_BLOCKS.
772 * domwalk.h (enum dom_walker::reachability): New enum.
773 (dom_walker::dom_walker): Convert bool param
774 "skip_unreachable_blocks" to enum reachability.
775 (set_all_edges_as_executable): New decl.
776 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
777 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
779 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
780 but converting true to REACHABLE_BLOCKS.
781 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
783 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
784 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
785 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
787 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
788 if check_all_array_refs will be called.
790 2018-01-23 David Malcolm <dmalcolm@redhat.com>
792 * tree.c (selftest::test_location_wrappers): Add more test
795 2018-01-23 David Malcolm <dmalcolm@redhat.com>
797 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
798 (selftest::test_bit_in_range): Likewise.
800 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
803 * doc/sourcebuild.texi (vect_float): Say that the selector
804 only describes the situation when -funsafe-math-optimizations is on.
805 (vect_float_strict): Document.
807 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
809 PR tree-optimization/83965
810 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
811 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
812 instead of checking only for a reduction.
813 (vect_recog_widen_sum_pattern): Likewise.
815 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
817 * predict.c (probably_never_executed): Only use precise profile info.
818 (compute_function_frequency): Skip after inlining hack since we now
819 have quality checking.
821 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
823 * profile-count.h (profile_probability::very_unlikely,
824 profile_probability::unlikely, profile_probability::even): Set
825 precision to guessed.
827 2018-01-23 Richard Biener <rguenther@suse.de>
829 PR tree-optimization/83963
830 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
831 Properly terminate dominator walk when crossing the exit edge not
832 when visiting its source block.
834 2018-01-23 Jakub Jelinek <jakub@redhat.com>
837 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
838 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
840 2018-01-22 Jakub Jelinek <jakub@redhat.com>
842 PR tree-optimization/83957
843 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
844 semicolon after for body surrounded by braces.
846 PR tree-optimization/83081
847 * profile-count.h (profile_probability::split): New method.
848 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
849 Use profile_probability::split.
850 (do_compare_rtx_and_jump): Fix adjustment of probabilities
851 when splitting a single conditional jump into 2.
853 2018-01-22 David Malcolm <dmalcolm@redhat.com>
855 PR tree-optimization/69452
856 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
859 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
861 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
862 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
863 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
865 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
867 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
868 * config/rl78/rl78.md: New define_expand "movdi"
869 * config/rl78/rl78.c: New function definition rl78_split_movdi
871 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
874 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
876 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
877 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
878 128-bit to produce an UNSPEC move to get the double word with the
879 signbit and then a shift directly to do signbit.
880 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
881 implementation with a new version that just does either a direct
882 move or a regular move. Move memory interface to separate insns.
883 Move insns so they are next to the expander.
884 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
885 with signbit move. Split big and little endian case.
886 (signbit<mode>2_dm_mem_le): Likewise.
887 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
888 (signbit<mode>2_dm2): Likewise.
890 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
892 * config/rl78/rl78.md: New define_expand "anddi3".
894 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
896 * config/rl78/rl78.md: New define_expand "umindi3".
898 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
900 * config/rl78/rl78.md: New define_expand "smindi3".
902 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
904 * config/rl78/rl78.md: New define_expand "smaxdi3".
906 2018-01-22 Carl Love <cel@us.ibm.com>
908 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
909 LVX_V1TI): Add macro expansion.
910 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
911 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
912 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
913 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
914 Change check to determine if the instruction is a byte reversing
915 entry. Fix typo in comment.
916 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
917 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
918 Add def_builtin calls for new builtins.
919 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
920 Add define_insn expansion.
922 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
924 * config/rl78/rl78.md: New define_expand "umaxdi3".
926 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
928 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
929 for non-QImode registers
931 2018-01-22 Richard Biener <rguenther@suse.de>
933 PR tree-optimization/83963
934 * graphite-scop-detection.c (scop_detection::get_sese): Delay
935 including the loop exit block.
936 (scop_detection::merge_sese): Likewise.
937 (scop_detection::add_scop): Do it here instead.
939 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
941 * doc/sourcebuild.texi (arm_softfloat): Document.
943 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
946 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
947 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
948 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
950 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
951 David Edelsohn <dje.gcc@gmail.com>
954 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
955 Change "crset eq" to "crset 2".
956 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
957 (*call_indirect_aix<mode>_nospec): Likewise.
958 (*call_value_indirect_aix<mode>_nospec): Likewise.
959 (*call_indirect_elfv2<mode>_nospec): Likewise.
960 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
961 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
962 change assembly output from . to $.
963 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
964 (indirect_jump<mode>_nospec): Change assembly output from . to $.
965 (*tablejump<mode>_internal1_nospec): Likewise.
967 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
970 * config/sh/sh_optimize_sett_clrt.cc:
971 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
973 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
975 PR tree-optimization/83940
976 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
977 offset_dt to vect_constant_def rather than vect_unknown_def_type.
978 (vect_check_load_store_mask): Add a mask_dt_out parameter and
979 use it to pass back the definition type.
980 (vect_check_store_rhs): Likewise rhs_dt_out.
981 (vect_build_gather_load_calls): Add a mask_dt argument and use
982 it instead of a call to vect_is_simple_use.
983 (vectorizable_store): Update calls to vect_check_load_store_mask
984 and vect_check_store_rhs. Use the dt returned by the latter instead
985 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
986 instead of calls to vect_is_simple_use. Pass the scalar rather
987 than the vector operand to vect_is_simple_use when handling
988 second and subsequent copies of an rhs value.
989 (vectorizable_load): Update calls to vect_check_load_store_mask
990 and vect_build_gather_load_calls. Use the cached mask_dt and
991 gs_info.offset_dt instead of calls to vect_is_simple_use.
993 2018-01-20 Jakub Jelinek <jakub@redhat.com>
996 * tree-emutls.c: Include gimplify.h.
997 (lower_emutls_2): New function.
998 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
999 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1000 it before further processing.
1003 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1004 UINTVAL (trueop1) instead of INTVAL (op1).
1006 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1010 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1011 INCOMING_FRAME_SP_OFFSET if not defined.
1012 (scan_trace): Add ENTRY argument. If true and
1013 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1014 emit a note to adjust the CFA offset.
1015 (create_cfi_notes): Adjust scan_trace callers.
1016 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1017 INCOMING_FRAME_SP_OFFSET in the CIE.
1018 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1019 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1021 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1022 * doc/tm.texi: Regenerated.
1024 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1026 PR rtl-optimization/83147
1027 * lra-constraints.c (remove_inheritance_pseudos): Use
1028 lra_substitute_pseudo_within_insn.
1030 2018-01-19 Tom de Vries <tom@codesourcery.com>
1031 Cesar Philippidis <cesar@codesourcery.com>
1034 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1036 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1039 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1040 spaces for function labels.
1042 2018-01-19 Martin Liska <mliska@suse.cz>
1044 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1045 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1046 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1047 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1048 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1049 (PRED_CONST_RETURN): Change from 69 to 65.
1050 (PRED_NULL_RETURN): Change from 91 to 71.
1051 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1052 (PRED_LOOP_GUARD): Change from 66 to 73.
1054 2018-01-19 Martin Liska <mliska@suse.cz>
1056 * predict.c (predict_insn_def): Add new assert.
1057 (struct branch_predictor): Change type to signed integer.
1058 (test_prediction_value_range): Amend test to cover
1060 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1061 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1062 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1063 (PRED_LOOP_IV_COMPARE): Likewise.
1064 * predict.h (PROB_UNINITIALIZED): Define new constant.
1066 2018-01-19 Martin Liska <mliska@suse.cz>
1068 * predict.c (dump_prediction): Add new format for
1069 analyze_brprob.py script which is enabled with -details
1071 * profile-count.h (precise_p): New function.
1073 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1075 PR tree-optimization/83922
1076 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1077 there are no statements that need masking.
1078 (vect_active_double_reduction_p): New function.
1079 (vect_analyze_loop_operations): Use it when handling phis that
1080 are not in the loop header.
1082 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1084 PR tree-optimization/83914
1085 * tree-vect-loop.c (vectorizable_induction): Don't convert
1086 init_expr or apply the peeling adjustment for inductions
1087 that are nested within the vectorized loop.
1089 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1091 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1094 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1098 * function.h (gimplify_parameters): Add gimple_seq * argument.
1099 * function.c: Include gimple.h and options.h.
1100 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1101 for the added local temporaries if needed.
1102 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1103 if there are any parameter cleanups, wrap whole body into a
1104 try/finally with the cleanups.
1106 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1109 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1110 Use GET_MODE_CLASS for scalar floating point.
1112 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1116 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1117 Fix call of call_cgraph_insertion_hooks.
1119 2018-01-18 Martin Sebor <msebor@redhat.com>
1121 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1123 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1126 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1129 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1132 * common.opt: (-ffile-prefix-map): New option.
1133 * opts.c (common_handle_option): Defer it.
1134 * opts-global.c (handle_common_deferred_options): Handle it.
1135 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1136 * file-prefix-map.h: New file.
1137 (remap_debug_filename, add_debug_prefix_map): ...here.
1138 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1139 * final.c (debug_prefix_map, add_debug_prefix_map
1140 remap_debug_filename): Move to...
1141 * file-prefix-map.c: New file.
1142 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1143 generalize, get rid of alloca(), use strrchr() instead of strchr().
1144 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1145 Implement in terms of add_prefix_map().
1146 (remap_macro_filename, remap_debug_filename): Implement in term of
1148 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1149 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1150 * dbxout.c: Include file-prefix-map.h.
1151 * varasm.c: Likewise.
1152 * vmsdbgout.c: Likewise.
1153 * xcoffout.c: Likewise.
1154 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1155 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1156 * doc/invoke.texi (-ffile-prefix-map): Document.
1157 (-fdebug-prefix-map): Update description.
1159 2018-01-18 Martin Liska <mliska@suse.cz>
1161 * config/i386/i386.c (indirect_thunk_name): Document that also
1163 (output_indirect_thunk): Document why both instructions
1164 (pause and lfence) are generated.
1166 2018-01-18 Richard Biener <rguenther@suse.de>
1168 PR tree-optimization/83887
1169 * graphite-scop-detection.c
1170 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1171 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1172 (scop_detection::merge_sese): Re-implement with a flood-fill
1173 algorithm that properly finds a SESE region if it exists.
1175 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1178 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1179 pointer_diff optimizations use view_convert instead of convert.
1181 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1183 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1184 Generate different code for -mno-speculate-indirect-jumps.
1185 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1186 (*call_indirect_aix<mode>): Disable for
1187 -mno-speculate-indirect-jumps.
1188 (*call_indirect_aix<mode>_nospec): New define_insn.
1189 (*call_value_indirect_aix<mode>): Disable for
1190 -mno-speculate-indirect-jumps.
1191 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1192 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1193 -mno-speculate-indirect-jumps.
1194 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1196 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1198 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1199 long double type, set the flags for noting the default long double
1200 type, even if we don't pass or return a long double type.
1202 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1205 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1208 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1210 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1211 support for merge[hl].
1212 (fold_mergehl_helper): New helper function.
1213 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1214 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1215 (altivec_vmrglw_direct): Add xxmrglw insn.
1217 2018-01-17 Andrew Waterman <andrew@sifive.com>
1219 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1220 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1222 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1225 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1226 call the lto_location_cache before reading the
1227 DECL_SOURCE_LOCATION of the types.
1229 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1230 Richard Sandiford <richard.sandiford@linaro.org>
1232 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1233 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1234 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1235 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1236 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1238 * config/aarch64/constraints.md (aarch64_movti_operand):
1240 * config/aarch64/predicates.md (Uti): Add new constraint.
1242 2018-01-17 Carl Love <cel@us.ibm.com>
1243 * config/rs6000/vsx.md (define_expand xl_len_r,
1244 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1245 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1247 (define_expand, define_insn): Move the shift left from the
1248 define_insn to the define_expand for lxvl and stxvl instructions.
1249 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1250 and XL_LEN_R definitions to PURE.
1252 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1254 * config/i386/i386.c (indirect_thunk_name): Declare regno
1255 as unsigned int. Compare regno with INVALID_REGNUM.
1256 (output_indirect_thunk): Ditto.
1257 (output_indirect_thunk_function): Ditto.
1258 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1259 in the call to output_indirect_thunk_function.
1261 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1264 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1265 rather than the size of inner_type to determine the stack slot size
1266 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1268 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1271 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1274 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1276 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1277 endian Linux systems to optionally enable multilibs for selecting
1278 the long double type if the user configured an explicit type.
1279 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1280 have no long double multilibs if not defined.
1281 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1282 warn if the user used -mabi={ieee,ibm}longdouble and we built
1283 multilibs for long double.
1284 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1285 appropriate multilib option.
1286 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1288 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1289 for building long double multilibs.
1290 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1292 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1294 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1297 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1299 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1302 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1305 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1308 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1310 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1311 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1313 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1315 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1316 different rtl trees depending on TARGET_64BIT.
1317 (rs6000_gen_lvx): Likewise.
1319 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1321 * config/visium/visium.md (nop): Tweak comment.
1322 (hazard_nop): Likewise.
1324 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1326 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1327 -mspeculate-indirect-jumps.
1328 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1329 for -mno-speculate-indirect-jumps.
1330 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1331 (*call_value_indirect_elfv2<mode>): Disable for
1332 -mno-speculate-indirect-jumps.
1333 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1334 (indirect_jump): Emit different RTL for
1335 -mno-speculate-indirect-jumps.
1336 (*indirect_jump<mode>): Disable for
1337 -mno-speculate-indirect-jumps.
1338 (*indirect_jump<mode>_nospec): New define_insn.
1339 (tablejump): Emit different RTL for
1340 -mno-speculate-indirect-jumps.
1341 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1342 (tablejumpsi_nospec): New define_expand.
1343 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1344 (tablejumpdi_nospec): New define_expand.
1345 (*tablejump<mode>_internal1): Disable for
1346 -mno-speculate-indirect-jumps.
1347 (*tablejump<mode>_internal1_nospec): New define_insn.
1348 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1351 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1353 * caller-save.c (insert_save): Drop unnecessary parameter. All
1356 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1357 Richard Biener <rguenth@suse.de>
1360 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1361 return early, inline manually is_gimple_sizepos. Make sure if we
1362 call gimplify_expr we don't end up with a gimple constant.
1363 * tree.c (variably_modified_type_p): Don't return true for
1364 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1365 * gimplify.h (is_gimple_sizepos): Remove.
1367 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1369 PR tree-optimization/83857
1370 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1371 vectorizable_live_operation for pure SLP statements.
1372 (vectorizable_live_operation): Handle PHIs.
1374 2018-01-16 Richard Biener <rguenther@suse.de>
1376 PR tree-optimization/83867
1377 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1378 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1380 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1383 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1384 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1385 If off is not INTEGER_CST, issue a may not be aligned warning
1386 rather than isn't aligned. Use isn%'t rather than isn't.
1387 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1389 <case MULT_EXPR>: Improve the case when bottom and one of the
1390 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1391 operand, in that case check if the other operand is multiple of
1392 bottom divided by the INTEGER_CST operand.
1394 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1397 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1398 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1399 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1400 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1401 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1402 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1403 (pa_function_arg_size): New function.
1405 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1407 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1408 in a separate statement.
1410 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1412 PR tree-optimization/83847
1413 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1414 group gathers and scatters.
1416 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1418 PR rtl-optimization/86620
1419 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1421 PR rtl-optimization/83213
1422 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1423 to last if both are JUMP_INSNs.
1425 PR tree-optimization/83843
1426 * gimple-ssa-store-merging.c
1427 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1428 store_immediate_info for bswap/nop orig_stores.
1430 2018-01-15 Andrew Waterman <andrew@sifive.com>
1432 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1434 <UDIV>: Increase cost if !TARGET_DIV.
1436 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1438 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1439 (define_attr "cr_logical_3op"): New.
1440 (cceq_ior_compare): Adjust.
1441 (cceq_ior_compare_complement): Adjust.
1442 (*cceq_rev_compare): Adjust.
1443 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1444 (is_cracked_insn): Adjust.
1445 (insn_must_be_first_in_group): Adjust.
1446 * config/rs6000/40x.md: Adjust.
1447 * config/rs6000/440.md: Adjust.
1448 * config/rs6000/476.md: Adjust.
1449 * config/rs6000/601.md: Adjust.
1450 * config/rs6000/603.md: Adjust.
1451 * config/rs6000/6xx.md: Adjust.
1452 * config/rs6000/7450.md: Adjust.
1453 * config/rs6000/7xx.md: Adjust.
1454 * config/rs6000/8540.md: Adjust.
1455 * config/rs6000/cell.md: Adjust.
1456 * config/rs6000/e300c2c3.md: Adjust.
1457 * config/rs6000/e500mc.md: Adjust.
1458 * config/rs6000/e500mc64.md: Adjust.
1459 * config/rs6000/e5500.md: Adjust.
1460 * config/rs6000/e6500.md: Adjust.
1461 * config/rs6000/mpc.md: Adjust.
1462 * config/rs6000/power4.md: Adjust.
1463 * config/rs6000/power5.md: Adjust.
1464 * config/rs6000/power6.md: Adjust.
1465 * config/rs6000/power7.md: Adjust.
1466 * config/rs6000/power8.md: Adjust.
1467 * config/rs6000/power9.md: Adjust.
1468 * config/rs6000/rs64.md: Adjust.
1469 * config/rs6000/titan.md: Adjust.
1471 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1473 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1474 ix86_indirect_branch_register logic.
1476 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1478 * config/i386/constraints.md (Bs): Update
1479 ix86_indirect_branch_register check. Don't check
1480 ix86_indirect_branch_register with GOT_memory_operand.
1482 * config/i386/predicates.md (GOT_memory_operand): Don't check
1483 ix86_indirect_branch_register here.
1484 (GOT32_symbol_operand): Likewise.
1486 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1488 * config/i386/predicates.md (constant_call_address_operand):
1489 Rewrite ix86_indirect_branch_register logic.
1490 (sibcall_insn_operand): Likewise.
1492 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1494 * config/i386/constraints.md (Bs): Replace
1495 ix86_indirect_branch_thunk_register with
1496 ix86_indirect_branch_register.
1498 * config/i386/i386.md (indirect_jump): Likewise.
1499 (tablejump): Likewise.
1500 (*sibcall_memory): Likewise.
1501 (*sibcall_value_memory): Likewise.
1502 Peepholes of indirect call and jump via memory: Likewise.
1503 * config/i386/i386.opt: Likewise.
1504 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1505 (GOT_memory_operand): Likewise.
1506 (call_insn_operand): Likewise.
1507 (sibcall_insn_operand): Likewise.
1508 (GOT32_symbol_operand): Likewise.
1510 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1513 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1514 type rather than type addr's type points to.
1515 (expand_omp_atomic_mutex): Likewise.
1516 (expand_omp_atomic): Likewise.
1518 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1521 * config/i386/i386.c (output_indirect_thunk_function): Use
1522 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1523 for __x86_return_thunk.
1525 2018-01-15 Richard Biener <rguenther@suse.de>
1528 * expmed.c (extract_bit_field_1): Fix typo.
1530 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1533 * config/arm/iterators.md (VF): New mode iterator.
1534 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1535 Remove integer-related logic from pattern.
1536 (neon_vabd<mode>_3): Likewise.
1538 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1541 * common.opt (fstrict-overflow): No longer an alias.
1542 (fwrapv-pointer): New option.
1543 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1544 also for pointer types based on flag_wrapv_pointer.
1545 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1546 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1547 opts->x_flag_wrapv got set.
1548 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1549 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1550 POINTER_TYPE_OVERFLOW_UNDEFINED.
1551 * match.pd: Likewise in address comparison pattern.
1552 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1554 2018-01-15 Richard Biener <rguenther@suse.de>
1557 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1558 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1559 Reset type names to their identifier if their TYPE_DECL doesn't
1560 have linkage (and thus is used for ODR and devirt).
1561 (save_debug_info_for_decl): Remove.
1562 (save_debug_info_for_type): Likewise.
1563 (add_tree_to_fld_list): Adjust.
1564 * tree-pretty-print.c (dump_generic_node): Make dumping of
1565 type names more robust.
1567 2018-01-15 Richard Biener <rguenther@suse.de>
1569 * BASE-VER: Bump to 8.0.1.
1571 2018-01-14 Martin Sebor <msebor@redhat.com>
1574 * builtins.c (check_access): Avoid warning when the no-warning bit
1577 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1579 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1580 * ira-color (allocno_hard_regs_compare): Likewise.
1582 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1585 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1586 Use .pushsection/.popsection.
1588 2018-01-14 Martin Sebor <msebor@redhat.com>
1591 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1593 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1595 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1596 entry from extra_headers.
1597 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1598 extra_headers, make the list bitwise identical to the i?86-*-* one.
1600 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1602 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1603 -mcmodel=large with -mindirect-branch=thunk,
1604 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1605 -mfunction-return=thunk-extern.
1606 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1607 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1608 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1610 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1612 * config/i386/i386.c (print_reg): Print the name of the full
1613 integer register without '%'.
1614 (ix86_print_operand): Handle 'V'.
1615 * doc/extend.texi: Document 'V' modifier.
1617 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1619 * config/i386/constraints.md (Bs): Disallow memory operand for
1620 -mindirect-branch-register.
1622 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1623 (GOT_memory_operand): Likewise.
1624 (call_insn_operand): Likewise.
1625 (sibcall_insn_operand): Likewise.
1626 (GOT32_symbol_operand): Likewise.
1627 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1628 for -mindirect-branch-register.
1629 (tablejump): Likewise.
1630 (*sibcall_memory): Likewise.
1631 (*sibcall_value_memory): Likewise.
1632 Disallow peepholes of indirect call and jump via memory for
1633 -mindirect-branch-register.
1634 (*call_pop): Replace m with Bw.
1635 (*call_value_pop): Likewise.
1636 (*sibcall_pop_memory): Replace m with Bs.
1637 * config/i386/i386.opt (mindirect-branch-register): New option.
1638 * doc/invoke.texi: Document -mindirect-branch-register option.
1640 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1642 * config/i386/i386-protos.h (ix86_output_function_return): New.
1643 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1644 set function_return_type.
1645 (indirect_thunk_name): Add ret_p to indicate thunk for function
1647 (output_indirect_thunk_function): Pass false to
1648 indirect_thunk_name.
1649 (ix86_output_indirect_branch_via_reg): Likewise.
1650 (ix86_output_indirect_branch_via_push): Likewise.
1651 (output_indirect_thunk_function): Create alias for function
1652 return thunk if regno < 0.
1653 (ix86_output_function_return): New function.
1654 (ix86_handle_fndecl_attribute): Handle function_return.
1655 (ix86_attribute_table): Add function_return.
1656 * config/i386/i386.h (machine_function): Add
1657 function_return_type.
1658 * config/i386/i386.md (simple_return_internal): Use
1659 ix86_output_function_return.
1660 (simple_return_internal_long): Likewise.
1661 * config/i386/i386.opt (mfunction-return=): New option.
1662 (indirect_branch): Mention -mfunction-return=.
1663 * doc/extend.texi: Document function_return function attribute.
1664 * doc/invoke.texi: Document -mfunction-return= option.
1666 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1668 * config/i386/i386-opts.h (indirect_branch): New.
1669 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1670 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1671 with local indirect jump when converting indirect call and jump.
1672 (ix86_set_indirect_branch_type): New.
1673 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1674 (indirectlabelno): New.
1675 (indirect_thunk_needed): Likewise.
1676 (indirect_thunk_bnd_needed): Likewise.
1677 (indirect_thunks_used): Likewise.
1678 (indirect_thunks_bnd_used): Likewise.
1679 (INDIRECT_LABEL): Likewise.
1680 (indirect_thunk_name): Likewise.
1681 (output_indirect_thunk): Likewise.
1682 (output_indirect_thunk_function): Likewise.
1683 (ix86_output_indirect_branch_via_reg): Likewise.
1684 (ix86_output_indirect_branch_via_push): Likewise.
1685 (ix86_output_indirect_branch): Likewise.
1686 (ix86_output_indirect_jmp): Likewise.
1687 (ix86_code_end): Call output_indirect_thunk_function if needed.
1688 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1690 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1691 (ix86_attribute_table): Add indirect_branch.
1692 * config/i386/i386.h (machine_function): Add indirect_branch_type
1693 and has_local_indirect_jump.
1694 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1696 (tablejump): Likewise.
1697 (*indirect_jump): Use ix86_output_indirect_jmp.
1698 (*tablejump_1): Likewise.
1699 (simple_return_indirect_internal): Likewise.
1700 * config/i386/i386.opt (mindirect-branch=): New option.
1701 (indirect_branch): New.
1704 (thunk-inline): Likewise.
1705 (thunk-extern): Likewise.
1706 * doc/extend.texi: Document indirect_branch function attribute.
1707 * doc/invoke.texi: Document -mindirect-branch= option.
1709 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1712 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1714 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1716 * ipa-inline.c (want_inline_small_function_p): Return false if
1717 inlining has already failed with CIF_FINAL_ERROR.
1718 (update_caller_keys): Call want_inline_small_function_p before
1720 (update_callee_keys): Likewise.
1722 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1724 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1726 (rs6000_quadword_masked_address_p): Likewise.
1727 (quad_aligned_load_p): Likewise.
1728 (quad_aligned_store_p): Likewise.
1729 (const_load_sequence_p): Add comment to describe the outer-most loop.
1730 (mimic_memory_attributes_and_flags): New function.
1731 (rs6000_gen_stvx): Likewise.
1732 (replace_swapped_aligned_store): Likewise.
1733 (rs6000_gen_lvx): Likewise.
1734 (replace_swapped_aligned_load): Likewise.
1735 (replace_swapped_load_constant): Capitalize argument name in
1736 comment describing this function.
1737 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1738 and stores that access quad-word aligned addresses and replace
1739 with stvx or lvx instructions when appropriate.
1740 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1741 New function prototype.
1742 (rs6000_quadword_masked_address_p): Likewise.
1743 (rs6000_gen_lvx): Likewise.
1744 (rs6000_gen_stvx): Likewise.
1745 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1746 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1747 when memory address is aligned.
1748 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1749 this split to select lvx instruction when memory address is aligned.
1750 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1751 instruction when memory address is aligned.
1752 (*vsx_le_perm_load_v16qi): Likewise.
1753 (four unnamed splitters): Modify to select the stvx instruction
1754 when memory is aligned.
1756 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1758 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1759 which appears in the queue multiple times.
1761 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1762 Alan Hayward <alan.hayward@arm.com>
1763 David Sherwood <david.sherwood@arm.com>
1765 * tree-vectorizer.h (vec_lower_bound): New structure.
1766 (_loop_vec_info): Add check_nonzero and lower_bounds.
1767 (LOOP_VINFO_CHECK_NONZERO): New macro.
1768 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1769 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1770 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1771 fields. Make seg_len the distance travelled, not including the
1773 (dr_direction_indicator): Declare.
1774 (dr_zero_step_indicator): Likewise.
1775 (dr_known_forward_stride_p): Likewise.
1776 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1778 (runtime_alias_check_p): Allow runtime alias checks with
1780 (operator ==): Compare access_size and align.
1781 (prune_runtime_alias_test_list): Rework for new distinction between
1782 the access_size and seg_len.
1783 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1785 (get_segment_min_max): New function.
1786 (create_intersect_range_checks): Use it.
1787 (dr_step_indicator): New function.
1788 (dr_direction_indicator): Likewise.
1789 (dr_zero_step_indicator): Likewise.
1790 (dr_known_forward_stride_p): Likewise.
1791 * tree-loop-distribution.c (data_ref_segment_size): Return
1792 DR_STEP * (niters - 1).
1793 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1795 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1796 (vect_preserves_scalar_order_p): New function, split out from...
1797 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1798 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1799 (vect_vfa_access_size): New function.
1800 (vect_vfa_align): Likewise.
1801 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1802 (dump_lower_bound): New function.
1803 (vect_check_lower_bound): Likewise.
1804 (vect_small_gap_p): Likewise.
1805 (vectorizable_with_step_bound_p): Likewise.
1806 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1807 depencies if the vectorization factor is 1. Convert the checks
1808 for nonzero steps into checks on the bounds of DR_STEP. Try using
1809 a bunds check for variable steps if the minimum required step is
1810 relatively small. Update calls to the dr_with_seg_len
1811 constructor and to vect_compile_time_alias.
1812 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1814 (vect_loop_versioning): Call it.
1815 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1817 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1819 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1820 Alan Hayward <alan.hayward@arm.com>
1821 David Sherwood <david.sherwood@arm.com>
1823 * doc/sourcebuild.texi (vect_scatter_store): Document.
1824 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1826 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1828 * genopinit.c (main): Add supports_vec_scatter_store and
1829 supports_vec_scatter_store_cached to target_optabs.
1830 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1831 IFN_MASK_SCATTER_STORE.
1832 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1834 * internal-fn.h (internal_store_fn_p): Declare.
1835 (internal_fn_stored_value_index): Likewise.
1836 * internal-fn.c (scatter_store_direct): New macro.
1837 (expand_scatter_store_optab_fn): New function.
1838 (direct_scatter_store_optab_supported_p): New macro.
1839 (internal_store_fn_p): New function.
1840 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1841 IFN_MASK_SCATTER_STORE.
1842 (internal_fn_mask_index): Likewise.
1843 (internal_fn_stored_value_index): New function.
1844 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1846 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1847 * optabs-query.c (supports_vec_scatter_store_p): New function.
1848 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1849 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1850 true for scatter stores.
1851 (vect_gather_scatter_fn_p): Handle scatter stores too.
1852 (vect_check_gather_scatter): Consider using scatter stores if
1853 supports_vec_scatter_store_p.
1854 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1856 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1857 internal_fn_stored_value_index.
1858 (check_load_store_masking): Handle scatter stores too.
1859 (vect_get_store_rhs): Make public.
1860 (vectorizable_call): Use internal_store_fn_p.
1861 (vectorizable_store): Handle scatter store internal functions.
1862 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1863 when deciding whether the end of the group has been reached.
1864 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1865 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1866 (mask_scatter_store<mode>): New insns.
1868 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1869 Alan Hayward <alan.hayward@arm.com>
1870 David Sherwood <david.sherwood@arm.com>
1872 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1873 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1874 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1876 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1877 Use vect_truncate_gather_scatter_offset if we can't treat the
1878 operation as a normal gather load or scatter store.
1879 (get_group_load_store_type): Take the gather_scatter_info
1880 as argument. Try using a gather load or scatter store for
1881 single-element groups.
1882 (get_load_store_type): Update calls to get_group_load_store_type
1883 and vect_use_strided_gather_scatters_p.
1885 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1886 Alan Hayward <alan.hayward@arm.com>
1887 David Sherwood <david.sherwood@arm.com>
1889 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1890 optional tree argument.
1891 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1893 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1894 but continue to use the current value as a fallback.
1895 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1896 to compare the updates.
1897 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1898 (get_load_store_type): Use it when handling a strided access.
1899 (vect_get_strided_load_store_ops): New function.
1900 (vect_get_data_ptr_increment): Likewise.
1901 (vectorizable_load): Handle strided gather loads. Always pass
1902 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1904 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1905 Alan Hayward <alan.hayward@arm.com>
1906 David Sherwood <david.sherwood@arm.com>
1908 * doc/md.texi (gather_load@var{m}): Document.
1909 (mask_gather_load@var{m}): Likewise.
1910 * genopinit.c (main): Add supports_vec_gather_load and
1911 supports_vec_gather_load_cached to target_optabs.
1912 * optabs-tree.c (init_tree_optimization_optabs): Use
1913 ggc_cleared_alloc to allocate target_optabs.
1914 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1915 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1917 * internal-fn.h (internal_load_fn_p): Declare.
1918 (internal_gather_scatter_fn_p): Likewise.
1919 (internal_fn_mask_index): Likewise.
1920 (internal_gather_scatter_fn_supported_p): Likewise.
1921 * internal-fn.c (gather_load_direct): New macro.
1922 (expand_gather_load_optab_fn): New function.
1923 (direct_gather_load_optab_supported_p): New macro.
1924 (direct_internal_fn_optab): New function.
1925 (internal_load_fn_p): Likewise.
1926 (internal_gather_scatter_fn_p): Likewise.
1927 (internal_fn_mask_index): Likewise.
1928 (internal_gather_scatter_fn_supported_p): Likewise.
1929 * optabs-query.c (supports_at_least_one_mode_p): New function.
1930 (supports_vec_gather_load_p): Likewise.
1931 * optabs-query.h (supports_vec_gather_load_p): Declare.
1932 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1933 and memory_type field.
1934 (NUM_PATTERNS): Bump to 15.
1935 * tree-vect-data-refs.c: Include internal-fn.h.
1936 (vect_gather_scatter_fn_p): New function.
1937 (vect_describe_gather_scatter_call): Likewise.
1938 (vect_check_gather_scatter): Try using internal functions for
1939 gather loads. Recognize existing calls to a gather load function.
1940 (vect_analyze_data_refs): Consider using gather loads if
1941 supports_vec_gather_load_p.
1942 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1943 (vect_get_gather_scatter_offset_type): Likewise.
1944 (vect_convert_mask_for_vectype): Likewise.
1945 (vect_add_conversion_to_patterm): Likewise.
1946 (vect_try_gather_scatter_pattern): Likewise.
1947 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1948 (vect_vect_recog_func_ptrs): Add it.
1949 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1950 internal_fn_mask_index and internal_gather_scatter_fn_p.
1951 (check_load_store_masking): Take the gather_scatter_info as an
1952 argument and handle gather loads.
1953 (vect_get_gather_scatter_ops): New function.
1954 (vectorizable_call): Check internal_load_fn_p.
1955 (vectorizable_load): Likewise. Handle gather load internal
1957 (vectorizable_store): Update call to check_load_store_masking.
1958 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1959 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1960 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1961 (aarch64_gather_scale_operand_d): New predicates.
1962 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1963 (mask_gather_load<mode>): New insns.
1965 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1966 Alan Hayward <alan.hayward@arm.com>
1967 David Sherwood <david.sherwood@arm.com>
1969 * optabs.def (fold_left_plus_optab): New optab.
1970 * doc/md.texi (fold_left_plus_@var{m}): Document.
1971 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1972 * internal-fn.c (fold_left_direct): Define.
1973 (expand_fold_left_optab_fn): Likewise.
1974 (direct_fold_left_optab_supported_p): Likewise.
1975 * fold-const-call.c (fold_const_fold_left): New function.
1976 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1977 * tree-parloops.c (valid_reduction_p): New function.
1978 (gather_scalar_reductions): Use it.
1979 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1980 (vect_finish_replace_stmt): Declare.
1981 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1982 (needs_fold_left_reduction_p): New function, split out from...
1983 (vect_is_simple_reduction): ...here. Accept reductions that
1984 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1985 (vect_force_simple_reduction): Also store the reduction type in
1986 the assignment's STMT_VINFO_REDUC_TYPE.
1987 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1988 (merge_with_identity): New function.
1989 (vect_expand_fold_left): Likewise.
1990 (vectorize_fold_left_reduction): Likewise.
1991 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1992 scalar phi in place for it. Check for target support and reject
1993 cases that would reassociate the operation. Defer the transform
1994 phase to vectorize_fold_left_reduction.
1995 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1996 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1997 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1999 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2001 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2002 call to ifc_temp_var.
2004 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2005 Alan Hayward <alan.hayward@arm.com>
2006 David Sherwood <david.sherwood@arm.com>
2008 * target.def (legitimize_address_displacement): Take the original
2009 offset as a poly_int.
2010 * targhooks.h (default_legitimize_address_displacement): Update
2012 * targhooks.c (default_legitimize_address_displacement): Likewise.
2013 * doc/tm.texi: Regenerate.
2014 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2015 as an argument, moving assert of ad->disp == ad->disp_term to...
2016 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2017 Try calling targetm.legitimize_address_displacement before expanding
2018 the address rather than afterwards, and adjust for the new interface.
2019 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2020 Match the new hook interface. Handle SVE addresses.
2021 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2024 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2026 * Makefile.in (OBJS): Add early-remat.o.
2027 * target.def (select_early_remat_modes): New hook.
2028 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2029 * doc/tm.texi: Regenerate.
2030 * targhooks.h (default_select_early_remat_modes): Declare.
2031 * targhooks.c (default_select_early_remat_modes): New function.
2032 * timevar.def (TV_EARLY_REMAT): New timevar.
2033 * passes.def (pass_early_remat): New pass.
2034 * tree-pass.h (make_pass_early_remat): Declare.
2035 * early-remat.c: New file.
2036 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2038 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2041 Alan Hayward <alan.hayward@arm.com>
2042 David Sherwood <david.sherwood@arm.com>
2044 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2045 vfm1 with a bound_epilog parameter.
2046 (vect_do_peeling): Update calls accordingly, and move the prologue
2047 call earlier in the function. Treat the base bound_epilog as 0 for
2048 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2049 this base when peeling for gaps.
2050 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2051 with fully-masked loops.
2052 (vect_estimate_min_profitable_iters): Handle the single peeled
2053 iteration in that case.
2055 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2056 Alan Hayward <alan.hayward@arm.com>
2057 David Sherwood <david.sherwood@arm.com>
2059 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2060 single-element interleaving even if the size is not a power of 2.
2061 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2062 accesses for single-element interleaving if the group size is
2065 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2066 Alan Hayward <alan.hayward@arm.com>
2067 David Sherwood <david.sherwood@arm.com>
2069 * doc/md.texi (fold_extract_last_@var{m}): Document.
2070 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2071 * optabs.def (fold_extract_last_optab): New optab.
2072 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2073 * internal-fn.c (fold_extract_direct): New macro.
2074 (expand_fold_extract_optab_fn): Likewise.
2075 (direct_fold_extract_optab_supported_p): Likewise.
2076 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2077 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2078 EXTRACT_LAST_REDUCTION.
2079 (get_initial_def_for_reduction): Do not create an initial vector
2080 for EXTRACT_LAST_REDUCTION reductions.
2081 (vectorizable_reduction): Leave the scalar phi in place for
2082 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2083 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2084 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2085 transform phase to vectorizable_condition.
2086 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2088 (vect_finish_stmt_generation): ...here.
2089 (vect_finish_replace_stmt): New function.
2090 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2091 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2093 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2095 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2096 Alan Hayward <alan.hayward@arm.com>
2097 David Sherwood <david.sherwood@arm.com>
2099 * doc/md.texi (extract_last_@var{m}): Document.
2100 * optabs.def (extract_last_optab): New optab.
2101 * internal-fn.def (EXTRACT_LAST): New internal function.
2102 * internal-fn.c (cond_unary_direct): New macro.
2103 (expand_cond_unary_optab_fn): Likewise.
2104 (direct_cond_unary_optab_supported_p): Likewise.
2105 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2106 loops using EXTRACT_LAST.
2107 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2108 (extract_last_<mode>): ...this optab.
2109 (vec_extract<mode><Vel>): Update accordingly.
2111 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2112 Alan Hayward <alan.hayward@arm.com>
2113 David Sherwood <david.sherwood@arm.com>
2115 * target.def (empty_mask_is_expensive): New hook.
2116 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2117 * doc/tm.texi: Regenerate.
2118 * targhooks.h (default_empty_mask_is_expensive): Declare.
2119 * targhooks.c (default_empty_mask_is_expensive): New function.
2120 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2121 if the target says that empty masks are expensive.
2122 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2124 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2126 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2127 Alan Hayward <alan.hayward@arm.com>
2128 David Sherwood <david.sherwood@arm.com>
2130 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2131 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2132 (vect_use_loop_mask_for_alignment_p): New function.
2133 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2134 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2135 niters_skip argument. Make sure that the first niters_skip elements
2136 of the first iteration are inactive.
2137 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2138 Update call to vect_set_loop_masks_directly.
2139 (get_misalign_in_elems): New function, split out from...
2140 (vect_gen_prolog_loop_niters): ...here.
2141 (vect_update_init_of_dr): Take a code argument that specifies whether
2142 the adjustment should be added or subtracted.
2143 (vect_update_init_of_drs): Likewise.
2144 (vect_prepare_for_masked_peels): New function.
2145 (vect_do_peeling): Skip prologue peeling if we're using a mask
2146 instead. Update call to vect_update_inits_of_drs.
2147 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2149 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2150 alignment. Do not include the number of peeled iterations in
2151 the minimum threshold in that case.
2152 (vectorizable_induction): Adjust the start value down by
2153 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2154 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2155 Take the number of skipped iterations into account when calculating
2157 * tree-vect-stmts.c (vect_gen_while_not): New function.
2159 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2160 Alan Hayward <alan.hayward@arm.com>
2161 David Sherwood <david.sherwood@arm.com>
2163 * doc/sourcebuild.texi (vect_fully_masked): Document.
2164 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2166 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2168 (vect_analyze_loop_2): ...here. Don't check the vectorization
2169 factor against the number of loop iterations if the loop is
2172 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2173 Alan Hayward <alan.hayward@arm.com>
2174 David Sherwood <david.sherwood@arm.com>
2176 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2177 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2178 (dump_groups): Update accordingly.
2179 (iv_use::mem_type): New member variable.
2180 (address_p): New function.
2181 (record_use): Add a mem_type argument and initialize the new
2183 (record_group_use): Add a mem_type argument. Use address_p.
2184 Remove obsolete null checks of base_object. Update call to record_use.
2185 (find_interesting_uses_op): Update call to record_group_use.
2186 (find_interesting_uses_cond): Likewise.
2187 (find_interesting_uses_address): Likewise.
2188 (get_mem_type_for_internal_fn): New function.
2189 (find_address_like_use): Likewise.
2190 (find_interesting_uses_stmt): Try find_address_like_use before
2191 calling find_interesting_uses_op.
2192 (addr_offset_valid_p): Use the iv mem_type field as the type
2193 of the addressed memory.
2194 (add_autoinc_candidates): Likewise.
2195 (get_address_cost): Likewise.
2196 (split_small_address_groups_p): Use address_p.
2197 (split_address_groups): Likewise.
2198 (add_iv_candidate_for_use): Likewise.
2199 (autoinc_possible_for_pair): Likewise.
2200 (rewrite_groups): Likewise.
2201 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2202 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2203 (get_alias_ptr_type_for_ptr_address): New function.
2204 (rewrite_use_address): Rewrite address uses in calls that were
2205 identified by find_address_like_use.
2207 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2208 Alan Hayward <alan.hayward@arm.com>
2209 David Sherwood <david.sherwood@arm.com>
2211 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2213 * gimple-expr.h (is_gimple_addressable: Likewise.
2214 * gimple-expr.c (is_gimple_address): Likewise.
2215 * internal-fn.c (expand_call_mem_ref): New function.
2216 (expand_mask_load_optab_fn): Use it.
2217 (expand_mask_store_optab_fn): Likewise.
2219 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2220 Alan Hayward <alan.hayward@arm.com>
2221 David Sherwood <david.sherwood@arm.com>
2223 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2224 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2225 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2226 (cond_umax@var{mode}): Document.
2227 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2228 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2229 (cond_umin_optab, cond_umax_optab): New optabs.
2230 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2231 (COND_IOR, COND_XOR): New internal functions.
2232 * internal-fn.h (get_conditional_internal_fn): Declare.
2233 * internal-fn.c (cond_binary_direct): New macro.
2234 (expand_cond_binary_optab_fn): Likewise.
2235 (direct_cond_binary_optab_supported_p): Likewise.
2236 (get_conditional_internal_fn): New function.
2237 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2238 Cope with reduction statements that are vectorized as calls rather
2240 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2241 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2242 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2243 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2244 (UNSPEC_COND_EOR): New unspecs.
2245 (optab): Add mappings for them.
2246 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2247 (sve_int_op, sve_fp_op): New int attributes.
2249 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2250 Alan Hayward <alan.hayward@arm.com>
2251 David Sherwood <david.sherwood@arm.com>
2253 * optabs.def (while_ult_optab): New optab.
2254 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2255 * internal-fn.def (WHILE_ULT): New internal function.
2256 * internal-fn.h (direct_internal_fn_supported_p): New override
2257 that takes two types as argument.
2258 * internal-fn.c (while_direct): New macro.
2259 (expand_while_optab_fn): New function.
2260 (convert_optab_supported_p): Likewise.
2261 (direct_while_optab_supported_p): New macro.
2262 * wide-int.h (wi::udiv_ceil): New function.
2263 * tree-vectorizer.h (rgroup_masks): New structure.
2264 (vec_loop_masks): New typedef.
2265 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2267 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2268 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2269 (vect_max_vf): New function.
2270 (slpeel_make_loop_iterate_ntimes): Delete.
2271 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2272 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2273 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2274 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2275 internal-fn.h, stor-layout.h and optabs-query.h.
2276 (vect_set_loop_mask): New function.
2277 (add_preheader_seq): Likewise.
2278 (add_header_seq): Likewise.
2279 (interleave_supported_p): Likewise.
2280 (vect_maybe_permute_loop_masks): Likewise.
2281 (vect_set_loop_masks_directly): Likewise.
2282 (vect_set_loop_condition_masked): Likewise.
2283 (vect_set_loop_condition_unmasked): New function, split out from
2284 slpeel_make_loop_iterate_ntimes.
2285 (slpeel_make_loop_iterate_ntimes): Rename to..
2286 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2287 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2288 (vect_do_peeling): Update call accordingly.
2289 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2291 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2292 mask_compare_type, can_fully_mask_p and fully_masked_p.
2293 (release_vec_loop_masks): New function.
2294 (_loop_vec_info): Use it to free the loop masks.
2295 (can_produce_all_loop_masks_p): New function.
2296 (vect_get_max_nscalars_per_iter): Likewise.
2297 (vect_verify_full_masking): Likewise.
2298 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2299 retries, and free the mask rgroups before retrying. Check loop-wide
2300 reasons for disallowing fully-masked loops. Make the final decision
2301 about whether use a fully-masked loop or not.
2302 (vect_estimate_min_profitable_iters): Do not assume that peeling
2303 for the number of iterations will be needed for fully-masked loops.
2304 (vectorizable_reduction): Disable fully-masked loops.
2305 (vectorizable_live_operation): Likewise.
2306 (vect_halve_mask_nunits): New function.
2307 (vect_double_mask_nunits): Likewise.
2308 (vect_record_loop_mask): Likewise.
2309 (vect_get_loop_mask): Likewise.
2310 (vect_transform_loop): Handle the case in which the final loop
2311 iteration might handle a partial vector. Call vect_set_loop_condition
2312 instead of slpeel_make_loop_iterate_ntimes.
2313 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2314 (check_load_store_masking): New function.
2315 (prepare_load_store_mask): Likewise.
2316 (vectorizable_store): Handle fully-masked loops.
2317 (vectorizable_load): Likewise.
2318 (supportable_widening_operation): Use vect_halve_mask_nunits for
2320 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2321 (vect_gen_while): New function.
2322 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2323 (aarch64_uqdec<mode>): New insn.
2325 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2326 Alan Hayward <alan.hayward@arm.com>
2327 David Sherwood <david.sherwood@arm.com>
2329 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2330 (reduc_xor_scal_optab): New optabs.
2331 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2332 (reduc_xor_scal_@var{m}): Document.
2333 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2334 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2336 * fold-const-call.c (fold_const_call): Handle them.
2337 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2338 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2339 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2340 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2341 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2342 (UNSPEC_XORV): New unspecs.
2343 (optab): Add entries for them.
2344 (BITWISEV): New int iterator.
2345 (bit_reduc_op): New int attributes.
2347 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2348 Alan Hayward <alan.hayward@arm.com>
2349 David Sherwood <david.sherwood@arm.com>
2351 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2352 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2353 * optabs.def (vec_shl_insert_optab): New optab.
2354 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2355 (duplicate_and_interleave): Likewise.
2356 * tree-vect-loop.c: Include internal-fn.h.
2357 (neutral_op_for_slp_reduction): New function, split out from
2358 get_initial_defs_for_reduction.
2359 (get_initial_def_for_reduction): Handle option 2 for variable-length
2360 vectors by loading the neutral value into a vector and then shifting
2361 the initial value into element 0.
2362 (get_initial_defs_for_reduction): Replace the code argument with
2363 the neutral value calculated by neutral_op_for_slp_reduction.
2364 Use gimple_build_vector for constant-length vectors.
2365 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2366 but the first group_size elements have a neutral value.
2367 Use duplicate_and_interleave otherwise.
2368 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2369 Update call to get_initial_defs_for_reduction. Handle SLP
2370 reductions for variable-length vectors by creating one vector
2371 result for each scalar result, with the elements associated
2372 with other scalar results stubbed out with the neutral value.
2373 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2374 Require IFN_VEC_SHL_INSERT for double reductions on
2375 variable-length vectors, or SLP reductions that have
2376 a neutral value. Require can_duplicate_and_interleave_p
2377 support for variable-length unchained SLP reductions if there
2378 is no neutral value, such as for MIN/MAX reductions. Also require
2379 the number of vector elements to be a multiple of the number of
2380 SLP statements when doing variable-length unchained SLP reductions.
2381 Update call to vect_create_epilog_for_reduction.
2382 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2383 and remove initial values.
2384 (duplicate_and_interleave): Make public.
2385 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2386 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2388 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2389 Alan Hayward <alan.hayward@arm.com>
2390 David Sherwood <david.sherwood@arm.com>
2392 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2393 (can_duplicate_and_interleave_p): New function.
2394 (vect_get_and_check_slp_defs): Take the vector of statements
2395 rather than just the current one. Remove excess parentheses.
2396 Restriction rejectinon of vect_constant_def and vect_external_def
2397 for variable-length vectors to boolean types, or types for which
2398 can_duplicate_and_interleave_p is false.
2399 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2400 (duplicate_and_interleave): New function.
2401 (vect_get_constant_vectors): Use gimple_build_vector for
2402 constant-length vectors and suitable variable-length constant
2403 vectors. Use duplicate_and_interleave for other variable-length
2404 vectors. Don't defer the update when inserting new statements.
2406 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2407 Alan Hayward <alan.hayward@arm.com>
2408 David Sherwood <david.sherwood@arm.com>
2410 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2411 min_profitable_iters doesn't go negative.
2413 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2414 Alan Hayward <alan.hayward@arm.com>
2415 David Sherwood <david.sherwood@arm.com>
2417 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2418 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2419 * optabs.def (vec_mask_load_lanes_optab): New optab.
2420 (vec_mask_store_lanes_optab): Likewise.
2421 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2422 (MASK_STORE_LANES): Likewise.
2423 * internal-fn.c (mask_load_lanes_direct): New macro.
2424 (mask_store_lanes_direct): Likewise.
2425 (expand_mask_load_optab_fn): Handle masked operations.
2426 (expand_mask_load_lanes_optab_fn): New macro.
2427 (expand_mask_store_optab_fn): Handle masked operations.
2428 (expand_mask_store_lanes_optab_fn): New macro.
2429 (direct_mask_load_lanes_optab_supported_p): Likewise.
2430 (direct_mask_store_lanes_optab_supported_p): Likewise.
2431 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2433 (vect_load_lanes_supported): Likewise.
2434 * tree-vect-data-refs.c (strip_conversion): New function.
2435 (can_group_stmts_p): Likewise.
2436 (vect_analyze_data_ref_accesses): Use it instead of checking
2437 for a pair of assignments.
2438 (vect_store_lanes_supported): Take a masked_p parameter.
2439 (vect_load_lanes_supported): Likewise.
2440 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2441 vect_store_lanes_supported and vect_load_lanes_supported.
2442 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2443 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2444 parameter. Don't allow gaps for masked accesses.
2445 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2446 and vect_load_lanes_supported.
2447 (get_load_store_type): Take a masked_p parameter and update
2448 call to get_group_load_store_type.
2449 (vectorizable_store): Update call to get_load_store_type.
2450 Handle IFN_MASK_STORE_LANES.
2451 (vectorizable_load): Update call to get_load_store_type.
2452 Handle IFN_MASK_LOAD_LANES.
2454 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2455 Alan Hayward <alan.hayward@arm.com>
2456 David Sherwood <david.sherwood@arm.com>
2458 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2460 * config/aarch64/aarch64-protos.h
2461 (aarch64_sve_struct_memory_operand_p): Declare.
2462 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2463 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2464 (VPRED, vpred): Handle SVE structure modes.
2465 * config/aarch64/constraints.md (Utx): New constraint.
2466 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2467 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2468 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2469 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2470 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2471 structure modes. Split into pieces after RA.
2472 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2473 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2475 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2476 SVE structure modes.
2477 (aarch64_classify_address): Likewise.
2478 (sizetochar): Move earlier in file.
2479 (aarch64_print_operand): Handle SVE register lists.
2480 (aarch64_array_mode): New function.
2481 (aarch64_sve_struct_memory_operand_p): Likewise.
2482 (TARGET_ARRAY_MODE): Redefine.
2484 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2485 Alan Hayward <alan.hayward@arm.com>
2486 David Sherwood <david.sherwood@arm.com>
2488 * target.def (array_mode): New target hook.
2489 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2490 * doc/tm.texi: Regenerate.
2491 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2492 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2493 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2495 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2498 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2499 Alan Hayward <alan.hayward@arm.com>
2500 David Sherwood <david.sherwood@arm.com>
2502 * fold-const.c (fold_binary_loc): Check the argument types
2503 rather than the result type when testing for a vector operation.
2505 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2507 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2508 * doc/tm.texi: Regenerate.
2510 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2511 Alan Hayward <alan.hayward@arm.com>
2512 David Sherwood <david.sherwood@arm.com>
2514 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2515 (sve): Document new AArch64 extension.
2516 * doc/md.texi (w): Extend the description of the AArch64
2517 constraint to include SVE vectors.
2518 (Upl, Upa): Document new AArch64 predicate constraints.
2519 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2521 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2522 (msve-vector-bits=): New option.
2523 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2524 SVE when these are disabled.
2525 (sve): New extension.
2526 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2527 modes. Adjust their number of units based on aarch64_sve_vg.
2528 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2529 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2530 aarch64_addr_query_type.
2531 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2532 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2533 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2534 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2535 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2536 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2537 (aarch64_simd_imm_zero_p): Delete.
2538 (aarch64_check_zero_based_sve_index_immediate): Declare.
2539 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2540 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2541 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2542 (aarch64_sve_float_mul_immediate_p): Likewise.
2543 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2545 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2546 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2547 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2548 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2549 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2550 (aarch64_regmode_natural_size): Likewise.
2551 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2552 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2554 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2555 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2556 for VG and the SVE predicate registers.
2557 (V_ALIASES): Add a "z"-prefixed alias.
2558 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2559 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2560 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2561 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2562 (REG_CLASS_NAMES): Add entries for them.
2563 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2564 and the predicate registers.
2565 (aarch64_sve_vg): Declare.
2566 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2567 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2568 (REGMODE_NATURAL_SIZE): Define.
2569 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2571 * config/aarch64/aarch64.c: Include cfgrtl.h.
2572 (simd_immediate_info): Add a constructor for series vectors,
2573 and an associated step field.
2574 (aarch64_sve_vg): New variable.
2575 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2576 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2577 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2578 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2579 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2580 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2581 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2582 (aarch64_get_mask_mode): New functions.
2583 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2584 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2585 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2586 predicate modes and predicate registers. Explicitly restrict
2587 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2588 to store a vector mode if it is recognized by
2589 aarch64_classify_vector_mode.
2590 (aarch64_regmode_natural_size): New function.
2591 (aarch64_hard_regno_caller_save_mode): Return the original mode
2593 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2594 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2595 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2596 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2598 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2599 does not overlap dest if the function is frame-related. Handle
2601 (aarch64_split_add_offset): New function.
2602 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2603 them aarch64_add_offset.
2604 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2605 and update call to aarch64_sub_sp.
2606 (aarch64_add_cfa_expression): New function.
2607 (aarch64_expand_prologue): Pass extra temporary registers to the
2608 functions above. Handle the case in which we need to emit new
2609 DW_CFA_expressions for registers that were originally saved
2610 relative to the stack pointer, but now have to be expressed
2611 relative to the frame pointer.
2612 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2614 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2615 IP0 and IP1 values for SVE frames.
2616 (aarch64_expand_vec_series): New function.
2617 (aarch64_expand_sve_widened_duplicate): Likewise.
2618 (aarch64_expand_sve_const_vector): Likewise.
2619 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2620 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2621 into the register, rather than emitting a SET directly.
2622 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2623 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2624 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2625 (offset_9bit_signed_scaled_p): New functions.
2626 (aarch64_replicate_bitmask_imm): New function.
2627 (aarch64_bitmask_imm): Use it.
2628 (aarch64_cannot_force_const_mem): Reject expressions involving
2629 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2630 (aarch64_classify_index): Handle SVE indices, by requiring
2631 a plain register index with a scale that matches the element size.
2632 (aarch64_classify_address): Handle SVE addresses. Assert that
2633 the mode of the address is VOIDmode or an integer mode.
2634 Update call to aarch64_classify_symbol.
2635 (aarch64_classify_symbolic_expression): Update call to
2636 aarch64_classify_symbol.
2637 (aarch64_const_vec_all_in_range_p): New function.
2638 (aarch64_print_vector_float_operand): Likewise.
2639 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2640 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2641 and the FP immediates 1.0 and 0.5.
2642 (aarch64_print_address_internal): Handle SVE addresses.
2643 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2644 (aarch64_regno_regclass): Handle predicate registers.
2645 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2647 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2648 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2649 (aarch64_convert_sve_vector_bits): New function.
2650 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2651 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2653 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2654 Handle SVE vector and predicate modes. Accept VL-based constants
2655 that need only one temporary register, and VL offsets that require
2656 no temporary registers.
2657 (aarch64_conditional_register_usage): Mark the predicate registers
2658 as fixed if SVE isn't available.
2659 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2660 Return true for SVE vector and predicate modes.
2661 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2662 rather than an unsigned int. Handle SVE modes.
2663 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2665 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2667 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2668 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2669 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2670 (aarch64_sve_float_mul_immediate_p): New functions.
2671 (aarch64_sve_valid_immediate): New function.
2672 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2673 Explicitly reject structure modes. Check for INDEX constants.
2674 Handle PTRUE and PFALSE constants.
2675 (aarch64_check_zero_based_sve_index_immediate): New function.
2676 (aarch64_simd_imm_zero_p): Delete.
2677 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2678 vector modes. Accept constants in the range of CNT[BHWD].
2679 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2680 ask for an Advanced SIMD mode.
2681 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2682 (aarch64_simd_vector_alignment): Handle SVE predicates.
2683 (aarch64_vectorize_preferred_vector_alignment): New function.
2684 (aarch64_simd_vector_alignment_reachable): Use it instead of
2686 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2687 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2689 (MAX_VECT_LEN): Delete.
2690 (expand_vec_perm_d): Add a vec_flags field.
2691 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2692 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2693 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2695 (aarch64_evpc_rev): Rename to...
2696 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2697 (aarch64_evpc_rev_global): New function.
2698 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2699 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2701 (aarch64_evpc_sve_tbl): New function.
2702 (aarch64_expand_vec_perm_const_1): Update after rename of
2703 aarch64_evpc_rev. Handle SVE permutes too, trying
2704 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2705 than aarch64_evpc_tbl.
2706 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2707 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2708 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2709 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2710 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2711 (aarch64_expand_sve_vcond): New functions.
2712 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2713 of aarch64_vector_mode_p.
2714 (aarch64_dwarf_poly_indeterminate_value): New function.
2715 (aarch64_compute_pressure_classes): Likewise.
2716 (aarch64_can_change_mode_class): Likewise.
2717 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2718 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2719 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2720 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2721 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2722 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2723 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2724 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2726 (Dn, Dl, Dr): Accept const as well as const_vector.
2727 (Dz): Likewise. Compare against CONST0_RTX.
2728 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2729 of "vector" where appropriate.
2730 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2731 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2732 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2733 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2734 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2735 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2736 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2737 (v_int_equiv): Extend to SVE modes.
2738 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2740 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2741 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2742 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2743 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2744 (SVE_COND_FP_CMP): New int iterators.
2745 (perm_hilo): Handle the new unpack unspecs.
2746 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2748 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2749 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2750 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2751 (aarch64_equality_operator, aarch64_constant_vector_operand)
2752 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2753 (aarch64_sve_nonimmediate_operand): Likewise.
2754 (aarch64_sve_general_operand): Likewise.
2755 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2756 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2757 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2758 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2759 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2760 (aarch64_sve_float_arith_immediate): Likewise.
2761 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2762 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2763 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2764 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2765 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2766 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2767 (aarch64_sve_float_arith_operand): Likewise.
2768 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2769 (aarch64_sve_float_mul_operand): Likewise.
2770 (aarch64_sve_vec_perm_operand): Likewise.
2771 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2772 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2773 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2774 as well as const_vector.
2775 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2776 in file. Use CONST0_RTX and CONSTM1_RTX.
2777 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2778 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2779 Use aarch64_simd_imm_zero.
2780 * config/aarch64/aarch64-sve.md: New file.
2781 * config/aarch64/aarch64.md: Include it.
2782 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2783 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2784 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2785 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2786 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2787 (sve): New attribute.
2788 (enabled): Disable instructions with the sve attribute unless
2790 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2791 aarch64_expand_mov_immediate.
2792 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2793 CNT[BHSD] immediates.
2794 (movti): Split CONST_POLY_INT moves into two halves.
2795 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2796 Split additions that need a temporary here if the destination
2797 is the stack pointer.
2798 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2799 (*add<mode>3_poly_1): New instruction.
2800 (set_clobber_cc): New expander.
2802 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2804 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2805 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2806 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2807 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2808 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2809 Change innermode from fixed_mode_size to machine_mode.
2810 (simplify_subreg): Update call accordingly. Handle a constant-sized
2811 subreg of a variable-length CONST_VECTOR.
2813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2814 Alan Hayward <alan.hayward@arm.com>
2815 David Sherwood <david.sherwood@arm.com>
2817 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2818 (add_offset_to_base): New function, split out from...
2819 (create_mem_ref): ...here. When handling a scale other than 1,
2820 check first whether the address is valid without the offset.
2821 Add it into the base if so, leaving the index and scale as-is.
2823 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2826 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2827 fold_for_warn before checking if arg2 is INTEGER_CST.
2829 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2831 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2832 (store_multiple_operation): Delete.
2833 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2834 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2835 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2836 guarded by TARGET_STRING.
2837 (rs6000_output_load_multiple): Delete.
2838 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2839 OPTION_MASK_STRING / TARGET_STRING handling.
2840 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2841 (const rs6000_opt_masks) <"string">: Change mask to 0.
2842 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2843 (MASK_STRING): Delete.
2844 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2846 (load_multiple): Delete.
2853 (store_multiple): Delete.
2860 (movmemsi_8reg): Delete.
2861 (corresponding unnamed define_insn): Delete.
2862 (movmemsi_6reg): Delete.
2863 (corresponding unnamed define_insn): Delete.
2864 (movmemsi_4reg): Delete.
2865 (corresponding unnamed define_insn): Delete.
2866 (movmemsi_2reg): Delete.
2867 (corresponding unnamed define_insn): Delete.
2868 (movmemsi_1reg): Delete.
2869 (corresponding unnamed define_insn): Delete.
2870 * config/rs6000/rs6000.opt (mno-string): New.
2871 (mstring): Replace by deprecation warning stub.
2872 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2874 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2876 * regrename.c (regrename_do_replace): If replacing the same
2877 reg multiple times, try to reuse last created gen_raw_REG.
2880 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2881 main to workaround a bug in GDB.
2883 2018-01-12 Tom de Vries <tom@codesourcery.com>
2886 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2888 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2890 PR rtl-optimization/80481
2891 * ira-color.c (get_cap_member): New function.
2892 (allocnos_conflict_by_live_ranges_p): Use it.
2893 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2894 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2896 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2899 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2900 (*saddl_se_1): Ditto.
2902 (*ssubl_se_1): Ditto.
2904 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2906 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2907 rather than wi::to_widest for DR_INITs.
2908 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2909 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2910 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2912 (vect_analyze_group_access_1): Note that here.
2914 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2916 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2917 polynomial type sizes.
2919 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2921 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2922 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2923 (gimple_add_tmp_var): Likewise.
2925 2018-01-12 Martin Liska <mliska@suse.cz>
2927 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2928 (gimple_alloc_sizes): Likewise.
2929 (dump_gimple_statistics): Use PRIu64 in printf format.
2930 * gimple.h: Change uint64_t to int.
2932 2018-01-12 Martin Liska <mliska@suse.cz>
2934 * tree-core.h: Use uint64_t instead of int.
2935 * tree.c (tree_node_counts): Likewise.
2936 (tree_node_sizes): Likewise.
2937 (dump_tree_statistics): Use PRIu64 in printf format.
2939 2018-01-12 Martin Liska <mliska@suse.cz>
2941 * Makefile.in: As qsort_chk is implemented in vec.c, add
2942 vec.o to linkage of gencfn-macros.
2943 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2944 passing the info to record_node_allocation_statistics.
2945 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2947 * ggc-common.c (struct ggc_usage): Add operator== and use
2948 it in operator< and compare function.
2949 * mem-stats.h (struct mem_usage): Likewise.
2950 * vec.c (struct vec_usage): Remove operator< and compare
2951 function. Can be simply inherited.
2953 2018-01-12 Martin Jambor <mjambor@suse.cz>
2956 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2957 * tree-ssa-math-opts.c: Include domwalk.h.
2958 (convert_mult_to_fma_1): New function.
2959 (fma_transformation_info): New type.
2960 (fma_deferring_state): Likewise.
2961 (cancel_fma_deferring): New function.
2962 (result_of_phi): Likewise.
2963 (last_fma_candidate_feeds_initial_phi): Likewise.
2964 (convert_mult_to_fma): Added deferring logic, split actual
2965 transformation to convert_mult_to_fma_1.
2966 (math_opts_dom_walker): New type.
2967 (math_opts_dom_walker::after_dom_children): New method, body moved
2968 here from pass_optimize_widening_mul::execute, added deferring logic
2970 (pass_optimize_widening_mul::execute): Moved most of code to
2971 math_opts_dom_walker::after_dom_children.
2972 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2973 * config/i386/i386.c (ix86_option_override_internal): Added
2974 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2976 2018-01-12 Richard Biener <rguenther@suse.de>
2979 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2980 inline instance vars.
2982 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2985 * config/rx/rx.c (rx_is_restricted_memory_address):
2988 2018-01-12 Richard Biener <rguenther@suse.de>
2990 PR tree-optimization/80846
2991 * target.def (split_reduction): New target hook.
2992 * targhooks.c (default_split_reduction): New function.
2993 * targhooks.h (default_split_reduction): Declare.
2994 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2995 target requests first reduce vectors by combining low and high
2997 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2998 (get_vectype_for_scalar_type_and_size): Export.
2999 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3000 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3001 * doc/tm.texi: Regenerate.
3002 * config/i386/i386.c (ix86_split_reduction): Implement
3003 TARGET_VECTORIZE_SPLIT_REDUCTION.
3005 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3008 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3009 in PIC mode except for TARGET_VXWORKS_RTP.
3010 * config/sparc/sparc.c: Include cfgrtl.h.
3011 (TARGET_INIT_PIC_REG): Define.
3012 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3013 (sparc_pic_register_p): New predicate.
3014 (sparc_legitimate_address_p): Use it.
3015 (sparc_legitimize_pic_address): Likewise.
3016 (sparc_delegitimize_address): Likewise.
3017 (sparc_mode_dependent_address_p): Likewise.
3018 (gen_load_pcrel_sym): Remove 4th parameter.
3019 (load_got_register): Adjust call to above. Remove obsolete stuff.
3020 (sparc_expand_prologue): Do not call load_got_register here.
3021 (sparc_flat_expand_prologue): Likewise.
3022 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3023 (sparc_use_pseudo_pic_reg): New function.
3024 (sparc_init_pic_reg): Likewise.
3025 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3026 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3028 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3030 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3031 Add item for branch_cost.
3033 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3035 PR rtl-optimization/83565
3036 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3037 not extend the result to a larger mode for rotate operations.
3038 (num_sign_bit_copies1): Likewise.
3040 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3043 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3045 Use values-Xc.o for -pedantic.
3046 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3048 2018-01-12 Martin Liska <mliska@suse.cz>
3051 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3053 (possible_polymorphic_call_targets): Use it.
3054 (ipa_devirt): Likewise.
3056 2018-01-12 Martin Liska <mliska@suse.cz>
3058 * profile-count.h (enum profile_quality): Use 0 as invalid
3059 enum value of profile_quality.
3061 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3063 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3064 -mext-string options.
3066 2018-01-12 Richard Biener <rguenther@suse.de>
3068 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3069 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3070 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3072 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3074 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3076 * configure.ac (--with-long-double-format): Add support for the
3077 configuration option to change the default long double format on
3079 * config.gcc (powerpc*-linux*-*): Likewise.
3080 * configure: Regenerate.
3081 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3082 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3083 used without modification.
3085 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3087 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3088 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3089 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3090 MISC_BUILTIN_SPEC_BARRIER.
3091 (rs6000_init_builtins): Likewise.
3092 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3094 (speculation_barrier): New define_insn.
3095 * doc/extend.texi: Document __builtin_speculation_barrier.
3097 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3100 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3101 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3102 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3104 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3105 integral modes instead of "ss" and "sd".
3106 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3107 vectors with 32-bit and 64-bit elements.
3108 (vecdupssescalarmodesuffix): New mode attribute.
3109 (vec_dup<mode>): Use it.
3111 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3114 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3115 frame if argument is passed on stack.
3117 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3120 * ree.c (combine_reaching_defs): Optimize also
3121 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3122 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3124 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3127 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3129 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3132 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3133 after they are computed.
3135 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3137 PR tree-optimization/83695
3138 * gimple-loop-linterchange.cc
3139 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3140 reset cached scev information after interchange.
3141 (pass_linterchange::execute): Remove call to scev_reset_htab.
3143 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3145 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3146 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3147 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3148 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3149 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3150 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3151 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3152 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3153 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3154 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3155 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3156 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3157 (V_lane_reg): Likewise.
3158 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3160 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3161 (vfmal_lane_low<mode>_intrinsic,
3162 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3163 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3164 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3165 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3166 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3167 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3169 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3171 * config/arm/arm-cpus.in (fp16fml): New feature.
3172 (ALL_SIMD): Add fp16fml.
3173 (armv8.2-a): Add fp16fml as an option.
3174 (armv8.3-a): Likewise.
3175 (armv8.4-a): Add fp16fml as part of fp16.
3176 * config/arm/arm.h (TARGET_FP16FML): Define.
3177 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3179 * config/arm/arm-modes.def (V2HF): Define.
3180 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3181 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3182 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3183 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3184 vfmsl_low, vfmsl_high): New set of builtins.
3185 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3186 (vfml_op): New code attribute.
3187 (VFMLHALVES): New int iterator.
3188 (VFML, VFMLSEL): New mode attributes.
3189 (V_reg): Define mapping for V2HF.
3190 (V_hi, V_lo): New mode attributes.
3191 (VF_constraint): Likewise.
3192 (vfml_half, vfml_half_selector): New int attributes.
3193 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3195 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3196 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3198 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3199 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3200 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3201 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3203 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3204 Document new effective target and option set.
3206 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3208 * config/arm/arm-cpus.in (armv8_4): New feature.
3209 (ARMv8_4a): New fgroup.
3210 (armv8.4-a): New arch.
3211 * config/arm/arm-tables.opt: Regenerate.
3212 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3213 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3214 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3215 Add matching rules for -march=armv8.4-a and extensions.
3216 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3218 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3221 * config/rx/rx.md (BW): New mode attribute.
3222 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3224 2018-01-11 Richard Biener <rguenther@suse.de>
3226 PR tree-optimization/83435
3227 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3228 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3229 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3231 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3232 Alan Hayward <alan.hayward@arm.com>
3233 David Sherwood <david.sherwood@arm.com>
3235 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3237 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3238 (aarch64_print_address_internal): Use it to check for a zero offset.
3240 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3241 Alan Hayward <alan.hayward@arm.com>
3242 David Sherwood <david.sherwood@arm.com>
3244 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3245 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3246 Return a poly_int64 rather than a HOST_WIDE_INT.
3247 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3248 rather than a HOST_WIDE_INT.
3249 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3250 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3251 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3252 final_offset from HOST_WIDE_INT to poly_int64.
3253 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3254 to_constant when getting the number of units in an Advanced SIMD
3256 (aarch64_builtin_vectorized_function): Check for a constant number
3258 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3260 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3261 attribute instead of GET_MODE_NUNITS.
3262 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3263 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3264 GET_MODE_SIZE for fixed-size registers.
3265 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3266 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3267 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3268 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3269 (aarch64_print_operand, aarch64_print_address_internal)
3270 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3271 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3272 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3273 Handle polynomial GET_MODE_SIZE.
3274 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3275 wider than SImode without modification.
3276 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3277 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3278 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3279 passing and returning SVE modes.
3280 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3281 rather than GEN_INT.
3282 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3283 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3284 (aarch64_allocate_and_probe_stack_space): Likewise.
3285 (aarch64_layout_frame): Cope with polynomial offsets.
3286 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3287 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3289 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3290 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3291 poly_int64 rather than a HOST_WIDE_INT.
3292 (aarch64_get_separate_components, aarch64_process_components)
3293 (aarch64_expand_prologue, aarch64_expand_epilogue)
3294 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3295 (aarch64_anchor_offset): New function, split out from...
3296 (aarch64_legitimize_address): ...here.
3297 (aarch64_builtin_vectorization_cost): Handle polynomial
3298 TYPE_VECTOR_SUBPARTS.
3299 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3301 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3302 number of elements from the PARALLEL rather than the mode.
3303 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3304 rather than GET_MODE_BITSIZE.
3305 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3306 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3307 (aarch64_expand_vec_perm_const_1): Handle polynomial
3308 d->perm.length () and d->perm elements.
3309 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3310 Apply to_constant to d->perm elements.
3311 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3312 polynomial CONST_VECTOR_NUNITS.
3313 (aarch64_move_pointer): Take amount as a poly_int64 rather
3315 (aarch64_progress_pointer): Avoid temporary variable.
3316 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3317 the mode attribute instead of GET_MODE.
3319 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3320 Alan Hayward <alan.hayward@arm.com>
3321 David Sherwood <david.sherwood@arm.com>
3323 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3324 x exists before using it.
3325 (aarch64_add_constant_internal): Rename to...
3326 (aarch64_add_offset_1): ...this. Replace regnum with separate
3327 src and dest rtxes. Handle the case in which they're different,
3328 including when the offset is zero. Replace scratchreg with an rtx.
3329 Use 2 additions if there is no spare register into which we can
3330 move a 16-bit constant.
3331 (aarch64_add_constant): Delete.
3332 (aarch64_add_offset): Replace reg with separate src and dest
3333 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3334 Use aarch64_add_offset_1.
3335 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3336 an rtx rather than an int. Take the delta as a poly_int64
3337 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3338 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3339 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3340 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3341 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3343 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3344 aarch64_add_constant.
3346 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3348 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3349 Use scalar_float_mode.
3351 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3353 * config/aarch64/aarch64-simd.md
3354 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3355 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3356 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3357 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3358 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3359 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3360 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3361 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3362 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3363 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3365 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3368 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3369 targ_options->x_arm_arch_string is non NULL.
3371 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3373 * config/aarch64/aarch64.h
3374 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3376 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3379 * expmed.c (emit_store_flag_force): Swap if const op0
3380 and change VOIDmode to mode of op0.
3382 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3384 PR rtl-optimization/83761
3385 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3386 than bytes to mode_for_size.
3388 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3391 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3392 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3395 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3398 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3399 when in layout mode.
3400 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3401 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3404 2018-01-10 Michael Collison <michael.collison@arm.com>
3406 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3407 * config/aarch64/aarch64-option-extension.def: Add
3408 AARCH64_OPT_EXTENSION of 'fp16fml'.
3409 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3410 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3411 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3412 * config/aarch64/constraints.md (Ui7): New constraint.
3413 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3414 (VFMLA_SEL_W): Ditto.
3417 (VFMLA16_LOW): New int iterator.
3418 (VFMLA16_HIGH): Ditto.
3419 (UNSPEC_FMLAL): New unspec.
3420 (UNSPEC_FMLSL): Ditto.
3421 (UNSPEC_FMLAL2): Ditto.
3422 (UNSPEC_FMLSL2): Ditto.
3423 (f16mac): New code attribute.
3424 * config/aarch64/aarch64-simd-builtins.def
3425 (aarch64_fmlal_lowv2sf): Ditto.
3426 (aarch64_fmlsl_lowv2sf): Ditto.
3427 (aarch64_fmlalq_lowv4sf): Ditto.
3428 (aarch64_fmlslq_lowv4sf): Ditto.
3429 (aarch64_fmlal_highv2sf): Ditto.
3430 (aarch64_fmlsl_highv2sf): Ditto.
3431 (aarch64_fmlalq_highv4sf): Ditto.
3432 (aarch64_fmlslq_highv4sf): Ditto.
3433 (aarch64_fmlal_lane_lowv2sf): Ditto.
3434 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3435 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3436 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3437 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3438 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3439 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3440 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3441 (aarch64_fmlal_lane_highv2sf): Ditto.
3442 (aarch64_fmlsl_lane_highv2sf): Ditto.
3443 (aarch64_fmlal_laneq_highv2sf): Ditto.
3444 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3445 (aarch64_fmlalq_lane_highv4sf): Ditto.
3446 (aarch64_fmlsl_lane_highv4sf): Ditto.
3447 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3448 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3449 * config/aarch64/aarch64-simd.md:
3450 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3451 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3452 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3453 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3454 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3455 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3456 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3457 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3458 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3459 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3460 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3461 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3462 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3463 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3464 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3465 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3466 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3467 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3468 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3469 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3470 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3471 (vfmlsl_low_u32): Ditto.
3472 (vfmlalq_low_u32): Ditto.
3473 (vfmlslq_low_u32): Ditto.
3474 (vfmlal_high_u32): Ditto.
3475 (vfmlsl_high_u32): Ditto.
3476 (vfmlalq_high_u32): Ditto.
3477 (vfmlslq_high_u32): Ditto.
3478 (vfmlal_lane_low_u32): Ditto.
3479 (vfmlsl_lane_low_u32): Ditto.
3480 (vfmlal_laneq_low_u32): Ditto.
3481 (vfmlsl_laneq_low_u32): Ditto.
3482 (vfmlalq_lane_low_u32): Ditto.
3483 (vfmlslq_lane_low_u32): Ditto.
3484 (vfmlalq_laneq_low_u32): Ditto.
3485 (vfmlslq_laneq_low_u32): Ditto.
3486 (vfmlal_lane_high_u32): Ditto.
3487 (vfmlsl_lane_high_u32): Ditto.
3488 (vfmlal_laneq_high_u32): Ditto.
3489 (vfmlsl_laneq_high_u32): Ditto.
3490 (vfmlalq_lane_high_u32): Ditto.
3491 (vfmlslq_lane_high_u32): Ditto.
3492 (vfmlalq_laneq_high_u32): Ditto.
3493 (vfmlslq_laneq_high_u32): Ditto.
3494 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3495 (AARCH64_FL_FOR_ARCH8_4): New.
3496 (AARCH64_ISA_F16FML): New ISA flag.
3497 (TARGET_F16FML): New feature flag for fp16fml.
3498 (doc/invoke.texi): Document new fp16fml option.
3500 2018-01-10 Michael Collison <michael.collison@arm.com>
3502 * config/aarch64/aarch64-builtins.c:
3503 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3504 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3505 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3506 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3507 (AARCH64_ISA_SHA3): New ISA flag.
3508 (TARGET_SHA3): New feature flag for sha3.
3509 * config/aarch64/iterators.md (sha512_op): New int attribute.
3510 (CRYPTO_SHA512): New int iterator.
3511 (UNSPEC_SHA512H): New unspec.
3512 (UNSPEC_SHA512H2): Ditto.
3513 (UNSPEC_SHA512SU0): Ditto.
3514 (UNSPEC_SHA512SU1): Ditto.
3515 * config/aarch64/aarch64-simd-builtins.def
3516 (aarch64_crypto_sha512hqv2di): New builtin.
3517 (aarch64_crypto_sha512h2qv2di): Ditto.
3518 (aarch64_crypto_sha512su0qv2di): Ditto.
3519 (aarch64_crypto_sha512su1qv2di): Ditto.
3520 (aarch64_eor3qv8hi): Ditto.
3521 (aarch64_rax1qv2di): Ditto.
3522 (aarch64_xarqv2di): Ditto.
3523 (aarch64_bcaxqv8hi): Ditto.
3524 * config/aarch64/aarch64-simd.md:
3525 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3526 (aarch64_crypto_sha512su0qv2di): Ditto.
3527 (aarch64_crypto_sha512su1qv2di): Ditto.
3528 (aarch64_eor3qv8hi): Ditto.
3529 (aarch64_rax1qv2di): Ditto.
3530 (aarch64_xarqv2di): Ditto.
3531 (aarch64_bcaxqv8hi): Ditto.
3532 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3533 (vsha512h2q_u64): Ditto.
3534 (vsha512su0q_u64): Ditto.
3535 (vsha512su1q_u64): Ditto.
3536 (veor3q_u16): Ditto.
3537 (vrax1q_u64): Ditto.
3539 (vbcaxq_u16): Ditto.
3540 * config/arm/types.md (crypto_sha512): New type attribute.
3541 (crypto_sha3): Ditto.
3542 (doc/invoke.texi): Document new sha3 option.
3544 2018-01-10 Michael Collison <michael.collison@arm.com>
3546 * config/aarch64/aarch64-builtins.c:
3547 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3548 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3549 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3550 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3551 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3552 (AARCH64_ISA_SM4): New ISA flag.
3553 (TARGET_SM4): New feature flag for sm4.
3554 * config/aarch64/aarch64-simd-builtins.def
3555 (aarch64_sm3ss1qv4si): Ditto.
3556 (aarch64_sm3tt1aq4si): Ditto.
3557 (aarch64_sm3tt1bq4si): Ditto.
3558 (aarch64_sm3tt2aq4si): Ditto.
3559 (aarch64_sm3tt2bq4si): Ditto.
3560 (aarch64_sm3partw1qv4si): Ditto.
3561 (aarch64_sm3partw2qv4si): Ditto.
3562 (aarch64_sm4eqv4si): Ditto.
3563 (aarch64_sm4ekeyqv4si): Ditto.
3564 * config/aarch64/aarch64-simd.md:
3565 (aarch64_sm3ss1qv4si): Ditto.
3566 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3567 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3568 (aarch64_sm4eqv4si): Ditto.
3569 (aarch64_sm4ekeyqv4si): Ditto.
3570 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3571 (sm3part_op): Ditto.
3572 (CRYPTO_SM3TT): Ditto.
3573 (CRYPTO_SM3PART): Ditto.
3574 (UNSPEC_SM3SS1): New unspec.
3575 (UNSPEC_SM3TT1A): Ditto.
3576 (UNSPEC_SM3TT1B): Ditto.
3577 (UNSPEC_SM3TT2A): Ditto.
3578 (UNSPEC_SM3TT2B): Ditto.
3579 (UNSPEC_SM3PARTW1): Ditto.
3580 (UNSPEC_SM3PARTW2): Ditto.
3581 (UNSPEC_SM4E): Ditto.
3582 (UNSPEC_SM4EKEY): Ditto.
3583 * config/aarch64/constraints.md (Ui2): New constraint.
3584 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3585 * config/arm/types.md (crypto_sm3): New type attribute.
3586 (crypto_sm4): Ditto.
3587 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3588 (vsm3tt1aq_u32): Ditto.
3589 (vsm3tt1bq_u32): Ditto.
3590 (vsm3tt2aq_u32): Ditto.
3591 (vsm3tt2bq_u32): Ditto.
3592 (vsm3partw1q_u32): Ditto.
3593 (vsm3partw2q_u32): Ditto.
3594 (vsm4eq_u32): Ditto.
3595 (vsm4ekeyq_u32): Ditto.
3596 (doc/invoke.texi): Document new sm4 option.
3598 2018-01-10 Michael Collison <michael.collison@arm.com>
3600 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3601 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3602 (AARCH64_FL_FOR_ARCH8_4): New.
3603 (AARCH64_FL_V8_4): New flag.
3604 (doc/invoke.texi): Document new armv8.4-a option.
3606 2018-01-10 Michael Collison <michael.collison@arm.com>
3608 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3609 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3610 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3611 * config/aarch64/aarch64-option-extension.def: Add
3612 AARCH64_OPT_EXTENSION of 'sha2'.
3613 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3614 (crypto): Disable sha2 and aes if crypto disabled.
3615 (crypto): Enable aes and sha2 if enabled.
3616 (simd): Disable sha2 and aes if simd disabled.
3617 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3619 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3620 (TARGET_SHA2): New feature flag for sha2.
3621 (TARGET_AES): New feature flag for aes.
3622 * config/aarch64/aarch64-simd.md:
3623 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3624 conditional on TARGET_AES.
3625 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3626 (aarch64_crypto_sha1hsi): Make pattern conditional
3628 (aarch64_crypto_sha1hv4si): Ditto.
3629 (aarch64_be_crypto_sha1hv4si): Ditto.
3630 (aarch64_crypto_sha1su1v4si): Ditto.
3631 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3632 (aarch64_crypto_sha1su0v4si): Ditto.
3633 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3634 (aarch64_crypto_sha256su0v4si): Ditto.
3635 (aarch64_crypto_sha256su1v4si): Ditto.
3636 (doc/invoke.texi): Document new aes and sha2 options.
3638 2018-01-10 Martin Sebor <msebor@redhat.com>
3640 PR tree-optimization/83781
3641 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3644 2018-01-11 Martin Sebor <msebor@gmail.com>
3645 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3647 PR tree-optimization/83501
3648 PR tree-optimization/81703
3650 * tree-ssa-strlen.c (get_string_cst): Rename...
3651 (get_string_len): ...to this. Handle global constants.
3652 (handle_char_store): Adjust.
3654 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3655 Jim Wilson <jimw@sifive.com>
3657 * config/riscv/riscv-protos.h (riscv_output_return): New.
3658 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3659 (riscv_attribute_table, riscv_output_return),
3660 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3661 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3662 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3664 (riscv_expand_prologue): Add early return for naked function.
3665 (riscv_expand_epilogue): Likewise.
3666 (riscv_function_ok_for_sibcall): Return false for naked function.
3667 (riscv_set_current_function): New.
3668 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3669 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3670 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3671 * doc/extend.texi (RISC-V Function Attributes): New.
3673 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3675 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3676 check for 128-bit long double before checking TCmode.
3677 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3678 128-bit long doubles before checking TFmode or TCmode.
3679 (FLOAT128_IBM_P): Likewise.
3681 2018-01-10 Martin Sebor <msebor@redhat.com>
3683 PR tree-optimization/83671
3684 * builtins.c (c_strlen): Unconditionally return zero for the empty
3686 Use -Warray-bounds for warnings.
3687 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3688 for non-constant array indices with COMPONENT_REF, arrays of
3689 arrays, and pointers to arrays.
3690 (gimple_fold_builtin_strlen): Determine and set length range for
3691 non-constant character arrays.
3693 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3696 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3699 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3701 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3703 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3706 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3707 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3708 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3709 indexed_or_indirect_operand predicate.
3710 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3711 (*vsx_le_perm_load_v8hi): Likewise.
3712 (*vsx_le_perm_load_v16qi): Likewise.
3713 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3714 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3715 (*vsx_le_perm_store_v8hi): Likewise.
3716 (*vsx_le_perm_store_v16qi): Likewise.
3717 (eight unnamed splitters): Likewise.
3719 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3721 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3722 * config/rs6000/emmintrin.h: Likewise.
3723 * config/rs6000/mmintrin.h: Likewise.
3724 * config/rs6000/xmmintrin.h: Likewise.
3726 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3729 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3731 * tree.c (tree_nop_conversion): Return true for location wrapper
3733 (maybe_wrap_with_location): New function.
3734 (selftest::check_strip_nops): New function.
3735 (selftest::test_location_wrappers): New function.
3736 (selftest::tree_c_tests): Call it.
3737 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3738 (maybe_wrap_with_location): New decl.
3739 (EXPR_LOCATION_WRAPPER_P): New macro.
3740 (location_wrapper_p): New inline function.
3741 (tree_strip_any_location_wrapper): New inline function.
3743 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3746 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3747 stack_realign_offset for the largest alignment of stack slot
3749 (ix86_find_max_used_stack_alignment): New function.
3750 (ix86_finalize_stack_frame_flags): Use it. Set
3751 max_used_stack_alignment if we don't realign stack.
3752 * config/i386/i386.h (machine_function): Add
3753 max_used_stack_alignment.
3755 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3757 * config/arm/arm.opt (-mbranch-cost): New option.
3758 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3761 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3764 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3765 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3767 2018-01-10 Richard Biener <rguenther@suse.de>
3770 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3771 early out so it also covers the case where we have a non-NULL
3774 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3776 PR tree-optimization/83753
3777 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3778 for non-strided grouped accesses if the number of elements is 1.
3780 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3783 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3784 * i386.h (TARGET_USE_GATHER): Define.
3785 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3787 2018-01-10 Martin Liska <mliska@suse.cz>
3790 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3791 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3793 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3794 CLEANUP_NO_PARTITIONING is not set.
3796 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3798 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3799 for vectors, as a partial revert of r254296.
3800 * rtl.h (const_vec_p): Delete.
3801 (const_vec_duplicate_p): Don't test for vector CONSTs.
3802 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3803 * expmed.c (make_tree): Likewise.
3806 * common.md (E, F): Use CONSTANT_P instead of checking for
3808 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3809 checking for CONST_VECTOR.
3811 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3814 * predict.c (force_edge_cold): Handle in more sane way edges
3817 2018-01-09 Carl Love <cel@us.ibm.com>
3819 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3821 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3822 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3823 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3824 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3825 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3826 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3827 * config/rs6000/rs6000-protos.h: Add extern defition for
3828 rs6000_generate_float2_double_code.
3829 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3831 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3832 (float2_v2df): Add define_expand.
3834 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3837 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3838 op_mode in the force_to_mode call.
3840 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3842 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3843 instead of checking each element individually.
3844 (aarch64_evpc_uzp): Likewise.
3845 (aarch64_evpc_zip): Likewise.
3846 (aarch64_evpc_ext): Likewise.
3847 (aarch64_evpc_rev): Likewise.
3848 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3849 instead of checking each element individually. Return true without
3851 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3852 whether all selected elements come from the same input, instead of
3853 checking each element individually. Remove calls to gen_rtx_REG,
3854 start_sequence and end_sequence and instead assert that no rtl is
3857 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3859 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3860 order of HIGH and CONST checks.
3862 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3864 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3865 if the destination isn't an SSA_NAME.
3867 2018-01-09 Richard Biener <rguenther@suse.de>
3869 PR tree-optimization/83668
3870 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3872 (canonicalize_loop_form): ... here, renamed from ...
3873 (canonicalize_loop_closed_ssa_form): ... this and amended to
3874 swap successor edges for loop exit blocks to make us use
3875 the RPO order we need for initial schedule generation.
3877 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3879 PR tree-optimization/64811
3880 * match.pd: When optimizing comparisons with Inf, avoid
3881 introducing or losing exceptions from comparisons with NaN.
3883 2018-01-09 Martin Liska <mliska@suse.cz>
3886 * asan.c (shadow_mem_size): Add gcc_assert.
3888 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3890 Don't save registers in main().
3893 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3894 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3895 * config/avr/avr.c (avr_set_current_function): Don't error if
3896 naked, OS_task or OS_main are specified at the same time.
3897 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3899 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3901 * common/config/avr/avr-common.c (avr_option_optimization_table):
3902 Switch on -mmain-is-OS_task for optimizing compilations.
3904 2018-01-09 Richard Biener <rguenther@suse.de>
3906 PR tree-optimization/83572
3907 * graphite.c: Include cfganal.h.
3908 (graphite_transform_loops): Connect infinite loops to exit
3909 and remove fake edges at the end.
3911 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3913 * ipa-inline.c (edge_badness): Revert accidental checkin.
3915 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3918 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3919 symbols; not inline clones.
3921 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3924 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3925 hard registers. Formatting fixes.
3927 PR preprocessor/83722
3928 * gcc.c (try_generate_repro): Pass
3929 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3930 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3933 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3934 Kito Cheng <kito.cheng@gmail.com>
3936 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3937 (riscv_leaf_function_p): Delete.
3938 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3940 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3942 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3944 (do_ifelse): New function.
3945 (do_isel): New function.
3946 (do_sub3): New function.
3947 (do_add3): New function.
3948 (do_load_mask_compare): New function.
3949 (do_overlap_load_compare): New function.
3950 (expand_compare_loop): New function.
3951 (expand_block_compare): Call expand_compare_loop() when appropriate.
3952 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3954 (-mblock-compare-inline-loop-limit): New option.
3956 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3959 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3960 Reverse order of second and third operands in first alternative.
3961 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3962 of first and second elements in UNSPEC_VPERMR vector.
3963 (altivec_expand_vec_perm_le): Likewise.
3965 2018-01-08 Jeff Law <law@redhat.com>
3967 PR rtl-optimizatin/81308
3968 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3969 (process_switch): If group_case_labels makes a change, then set
3971 (pass_convert_switch::execute): If a switch is converted, then
3972 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3974 PR rtl-optimization/81308
3975 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3978 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3980 PR target/83663 - Revert r255946
3981 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3982 generation for cases where splatting a value is not useful.
3983 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3984 across a vec_duplicate and a paradoxical subreg forming a vector
3985 mode to a vec_concat.
3987 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3989 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3990 -march=armv8.3-a variants.
3991 * config/arm/t-multilib: Likewise.
3992 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3994 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3996 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3998 (cceq_ior_compare_complement): Give it a name so I can use it, and
3999 change boolean_or_operator predicate to boolean_operator so it can
4000 be used to generate a crand.
4001 (eqne): New code iterator.
4002 (bd/bd_neg): New code_attrs.
4003 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4004 a single define_insn.
4005 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4006 decrement (bdnzt/bdnzf/bdzt/bdzf).
4007 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4008 with the new names of the branch decrement patterns, and added the
4009 names of the branch decrement conditional patterns.
4011 2018-01-08 Richard Biener <rguenther@suse.de>
4013 PR tree-optimization/83563
4014 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4017 2018-01-08 Richard Biener <rguenther@suse.de>
4020 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4022 2018-01-08 Richard Biener <rguenther@suse.de>
4024 PR tree-optimization/83685
4025 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4026 references to abnormals.
4028 2018-01-08 Richard Biener <rguenther@suse.de>
4031 * dwarf2out.c (output_indirect_strings): Handle empty
4032 skeleton_debug_str_hash.
4033 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4035 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4037 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4038 (emit_store_direct): Likewise.
4039 (arc_trampoline_adjust_address): Likewise.
4040 (arc_asm_trampoline_template): New function.
4041 (arc_initialize_trampoline): Use asm_trampoline_template.
4042 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4043 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4044 * config/arc/arc.md (flush_icache): Delete pattern.
4046 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4048 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4049 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4052 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4055 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4056 by not USED_FOR_TARGET.
4057 (make_pass_resolve_sw_modes): Likewise.
4059 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4061 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4064 2018-01-08 Richard Biener <rguenther@suse.de>
4067 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4069 2018-01-08 Richard Biener <rguenther@suse.de>
4072 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4074 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4077 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4078 basic blocks with a small number of successors.
4079 (convert_control_dep_chain_into_preds): Improve handling of
4081 (dump_predicates): Split apart into...
4082 (dump_pred_chain): ...here...
4083 (dump_pred_info): ...and here.
4084 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4085 (can_chain_union_be_invalidated_p): Improve check for invalidation
4087 (uninit_uses_cannot_happen): Avoid unnecessary if
4088 convert_control_dep_chain_into_preds yielded nothing.
4090 2018-01-06 Martin Sebor <msebor@redhat.com>
4092 PR tree-optimization/83640
4093 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4094 subtracting negative offset from size.
4095 (builtin_access::overlap): Adjust offset bounds of the access to fall
4096 within the size of the object if possible.
4098 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4100 PR rtl-optimization/83699
4101 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4102 extract_bit_field_as_subreg to cases in which the extracted
4103 value is also a vector.
4105 * lra-constraints.c (process_alt_operands): Test for the equivalence
4106 substitutions when detecting a possible reload cycle.
4108 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4111 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4112 by default if flag_selective_schedling{,2}. Formatting fixes.
4114 PR rtl-optimization/83682
4115 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4116 if it has non-VECTOR_MODE element mode.
4117 (vec_duplicate_p): Likewise.
4120 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4121 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4123 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4126 * config/i386/i386-builtin.def
4127 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4128 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4129 Require also OPTION_MASK_ISA_AVX512F in addition to
4130 OPTION_MASK_ISA_GFNI.
4131 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4132 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4133 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4134 to OPTION_MASK_ISA_GFNI.
4135 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4136 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4137 OPTION_MASK_ISA_AVX512BW.
4138 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4139 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4140 addition to OPTION_MASK_ISA_GFNI.
4141 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4142 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4143 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4144 to OPTION_MASK_ISA_GFNI.
4145 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4146 a requirement for all ISAs rather than any of them with a few
4148 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4150 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4151 bitmasks to be enabled with 3 exceptions, instead of requiring any
4152 enabled ISA with lots of exceptions.
4153 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4154 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4155 Change avx512bw in isa attribute to avx512f.
4156 * config/i386/sgxintrin.h: Add license boilerplate.
4157 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4158 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4159 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4160 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4162 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4163 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4164 temporarily sse2 rather than sse if not enabled already.
4167 * config/i386/sse.md (VI248_VLBW): Rename to ...
4168 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4169 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4170 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4171 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4172 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4173 mode iterator instead of VI248_VLBW.
4175 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4177 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4178 (record_modified): Skip clobbers; add debug output.
4179 (param_change_prob): Use sreal frequencies.
4181 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4183 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4184 punt for user-aligned variables.
4186 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4188 * tree-chrec.c (chrec_contains_symbols): Return true for
4191 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4194 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4195 of (x|y) == x for BICS pattern.
4197 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4199 PR tree-optimization/83605
4200 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4201 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4204 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4206 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4207 * config/epiphany/rtems.h: New file.
4209 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4210 Uros Bizjak <ubizjak@gmail.com>
4213 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4214 QIreg_operand instead of register_operand predicate.
4215 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4216 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4217 comments instead of -fmitigate[-_]rop.
4219 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4222 * cgraphunit.c (symbol_table::compile): Switch to text_section
4223 before calling assembly_start debug hook.
4224 * run-rtl-passes.c (run_rtl_passes): Likewise.
4227 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4229 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4230 range_int_cst_p rather than !symbolic_range_p before calling
4231 extract_range_from_multiplicative_op_1.
4233 2018-01-04 Jeff Law <law@redhat.com>
4235 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4236 redundant test in assertion.
4238 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4240 * doc/rtl.texi: Document machine_mode wrapper classes.
4242 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4244 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4247 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4249 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4250 the VEC_PERM_EXPR fold to fail.
4252 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4255 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4256 to switched_sections.
4258 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4261 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4264 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4267 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4268 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4270 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4273 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4274 is BLKmode and bitpos not zero or mode change is needed.
4276 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4279 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4282 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4285 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4286 instead of MULT rtx. Update all corresponding splitters.
4288 (*ssub<modesuffix>): Ditto.
4290 (*cmp_sadd_di): Update split patterns.
4291 (*cmp_sadd_si): Ditto.
4292 (*cmp_sadd_sidi): Ditto.
4293 (*cmp_ssub_di): Ditto.
4294 (*cmp_ssub_si): Ditto.
4295 (*cmp_ssub_sidi): Ditto.
4296 * config/alpha/predicates.md (const23_operand): New predicate.
4297 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4298 Look for ASHIFT, not MULT inner operand.
4299 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4301 2018-01-04 Martin Liska <mliska@suse.cz>
4303 PR gcov-profile/83669
4304 * gcov.c (output_intermediate_file): Add version to intermediate
4306 * doc/gcov.texi: Document new field 'version' in intermediate
4307 file format. Fix location of '-k' option of gcov command.
4309 2018-01-04 Martin Liska <mliska@suse.cz>
4312 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4314 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4316 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4318 2018-01-03 Martin Sebor <msebor@redhat.com>
4320 PR tree-optimization/83655
4321 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4322 checking calls with invalid arguments.
4324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4326 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4327 (vectorizable_mask_load_store): Delete.
4328 (vectorizable_call): Return false for masked loads and stores.
4329 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4330 instead of gimple_assign_rhs1.
4331 (vectorizable_load): Handle IFN_MASK_LOAD.
4332 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4336 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4338 (vectorizable_mask_load_store): ...here.
4339 (vectorizable_load): ...and here.
4341 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4343 * tree-vect-stmts.c (vect_build_all_ones_mask)
4344 (vect_build_zero_merge_argument): New functions, split out from...
4345 (vectorizable_load): ...here.
4347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4349 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4351 (vectorizable_mask_load_store): ...here.
4352 (vectorizable_store): ...and here.
4354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4356 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4358 (vectorizable_mask_load_store): ...here.
4360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4362 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4363 (vect_model_store_cost): Take a vec_load_store_type instead of a
4365 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4366 (vect_model_store_cost): Take a vec_load_store_type instead of a
4368 (vectorizable_mask_load_store): Update accordingly.
4369 (vectorizable_store): Likewise.
4370 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4374 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4375 IFN_MASK_LOAD calls here rather than...
4376 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4378 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4379 Alan Hayward <alan.hayward@arm.com>
4380 David Sherwood <david.sherwood@arm.com>
4382 * expmed.c (extract_bit_field_1): For vector extracts,
4383 fall back to extract_bit_field_as_subreg if vec_extract
4386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4387 Alan Hayward <alan.hayward@arm.com>
4388 David Sherwood <david.sherwood@arm.com>
4390 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4391 they are variable or constant sized.
4392 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4393 slots for constant-sized data.
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 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4400 handling COND_EXPRs with boolean comparisons, try to find a better
4401 basis for the mask type than the boolean itself.
4403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4405 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4406 is calculated and how it can be overridden.
4407 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4408 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4410 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4413 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4414 Alan Hayward <alan.hayward@arm.com>
4415 David Sherwood <david.sherwood@arm.com>
4417 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4418 Remove the mode argument.
4419 (aarch64_simd_valid_immediate): Remove the mode and inverse
4421 * config/aarch64/iterators.md (bitsize): New iterator.
4422 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4423 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4424 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4425 aarch64_simd_valid_immediate.
4426 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4427 (aarch64_reg_or_bic_imm): Likewise.
4428 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4429 with an insn_type enum and msl with a modifier_type enum.
4430 Replace element_width with a scalar_mode. Change the shift
4431 to unsigned int. Add constructors for scalar_float_mode and
4432 scalar_int_mode elements.
4433 (aarch64_vect_float_const_representable_p): Delete.
4434 (aarch64_can_const_movi_rtx_p)
4435 (aarch64_simd_scalar_immediate_valid_for_move)
4436 (aarch64_simd_make_constant): Update call to
4437 aarch64_simd_valid_immediate.
4438 (aarch64_advsimd_valid_immediate_hs): New function.
4439 (aarch64_advsimd_valid_immediate): Likewise.
4440 (aarch64_simd_valid_immediate): Remove mode and inverse
4441 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4442 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4443 and aarch64_float_const_representable_p on the result.
4444 (aarch64_output_simd_mov_immediate): Remove mode argument.
4445 Update call to aarch64_simd_valid_immediate and use of
4446 simd_immediate_info.
4447 (aarch64_output_scalar_simd_mov_immediate): Update call
4450 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4451 Alan Hayward <alan.hayward@arm.com>
4452 David Sherwood <david.sherwood@arm.com>
4454 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4455 (mode_nunits): Likewise CONST_MODE_NUNITS.
4456 * machmode.def (ADJUST_NUNITS): Document.
4457 * genmodes.c (mode_data::need_nunits_adj): New field.
4458 (blank_mode): Update accordingly.
4459 (adj_nunits): New variable.
4460 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4462 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4463 listed in adj_nunits.
4464 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4465 listed in adj_nunits. Don't emit case statements for such modes.
4466 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4467 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4468 nothing if adj_nunits is nonnull.
4469 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4470 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4471 (emit_mode_fbit): Update use of print_maybe_const_decl.
4472 (emit_move_size): Likewise. Treat the array as non-const
4474 (emit_mode_adjustments): Handle adj_nunits.
4476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4478 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4479 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4480 (VECTOR_MODES): Use it.
4481 (make_vector_modes): Take the prefix as an argument.
4483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4484 Alan Hayward <alan.hayward@arm.com>
4485 David Sherwood <david.sherwood@arm.com>
4487 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4488 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4489 for MODE_VECTOR_BOOL.
4490 * machmode.def (VECTOR_BOOL_MODE): Document.
4491 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4492 (make_vector_bool_mode): New function.
4493 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4495 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4496 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4498 * stor-layout.c (int_mode_for_mode): Likewise.
4499 * tree.c (build_vector_type_for_mode): Likewise.
4500 * varasm.c (output_constant_pool_2): Likewise.
4501 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4502 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4503 for MODE_VECTOR_BOOL.
4504 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4505 of mode class checks.
4506 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4507 instead of a list of mode class checks.
4508 (expand_vector_scalar_condition): Likewise.
4509 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4512 Alan Hayward <alan.hayward@arm.com>
4513 David Sherwood <david.sherwood@arm.com>
4515 * machmode.h (mode_size): Change from unsigned short to
4517 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4518 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4519 or if measurement_type is not polynomial.
4520 (fixed_size_mode::includes_p): Check for constant-sized modes.
4521 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4522 return a poly_uint16 rather than an unsigned short.
4523 (emit_mode_size): Change the type of mode_size from unsigned short
4524 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4525 (emit_mode_adjustments): Cope with polynomial vector sizes.
4526 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4528 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4530 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4531 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4532 * caller-save.c (setup_save_areas): Likewise.
4533 (replace_reg_with_saved_mem): Likewise.
4534 * calls.c (emit_library_call_value_1): Likewise.
4535 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4536 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4537 (gen_lowpart_for_combine): Likewise.
4538 * convert.c (convert_to_integer_1): Likewise.
4539 * cse.c (equiv_constant, cse_insn): Likewise.
4540 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4541 (cselib_subst_to_values): Likewise.
4542 * dce.c (word_dce_process_block): Likewise.
4543 * df-problems.c (df_word_lr_mark_ref): Likewise.
4544 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4545 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4546 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4547 (rtl_for_decl_location): Likewise.
4548 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4549 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4550 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4551 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4552 (expand_expr_real_1): Likewise.
4553 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4554 (pad_below): Likewise.
4555 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4556 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4557 * ira.c (get_subreg_tracking_sizes): Likewise.
4558 * ira-build.c (ira_create_allocno_objects): Likewise.
4559 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4560 (ira_sort_regnos_for_alter_reg): Likewise.
4561 * ira-costs.c (record_operand_costs): Likewise.
4562 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4563 (resolve_simple_move): Likewise.
4564 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4565 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4566 (lra_constraints): Likewise.
4567 (CONST_POOL_OK_P): Reject variable-sized modes.
4568 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4569 (add_pseudo_to_slot, lra_spill): Likewise.
4570 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4571 * optabs-query.c (get_best_extraction_insn): Likewise.
4572 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4573 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4574 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4575 * recog.c (offsettable_address_addr_space_p): Likewise.
4576 * regcprop.c (maybe_mode_change): Likewise.
4577 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4578 * regrename.c (build_def_use): Likewise.
4579 * regstat.c (dump_reg_info): Likewise.
4580 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4581 (find_reloads, find_reloads_subreg_address): Likewise.
4582 * reload1.c (eliminate_regs_1): Likewise.
4583 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4584 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4585 (simplify_binary_operation_1, simplify_subreg): Likewise.
4586 * targhooks.c (default_function_arg_padding): Likewise.
4587 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4588 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4589 (verify_gimple_assign_ternary): Likewise.
4590 * tree-inline.c (estimate_move_cost): Likewise.
4591 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4592 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4593 (get_address_cost_ainc): Likewise.
4594 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4595 (vect_supportable_dr_alignment): Likewise.
4596 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4597 (vectorizable_reduction): Likewise.
4598 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4599 (vectorizable_operation, vectorizable_load): Likewise.
4600 * tree.c (build_same_sized_truth_vector_type): Likewise.
4601 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4602 * var-tracking.c (emit_note_insn_var_location): Likewise.
4603 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4604 (ADDR_VEC_ALIGN): Likewise.
4606 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4607 Alan Hayward <alan.hayward@arm.com>
4608 David Sherwood <david.sherwood@arm.com>
4610 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4612 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4613 or if measurement_type is polynomial.
4614 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4615 * combine.c (make_extraction): Likewise.
4616 * dse.c (find_shift_sequence): Likewise.
4617 * dwarf2out.c (mem_loc_descriptor): Likewise.
4618 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4619 (extract_bit_field, extract_low_bits): Likewise.
4620 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4621 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4622 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4623 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4624 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4625 * reload.c (find_reloads): Likewise.
4626 * reload1.c (alter_reg): Likewise.
4627 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4628 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4629 * tree-if-conv.c (predicate_mem_writes): Likewise.
4630 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4631 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4632 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4633 * valtrack.c (dead_debug_insert_temp): Likewise.
4634 * varasm.c (mergeable_constant_section): Likewise.
4635 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4638 Alan Hayward <alan.hayward@arm.com>
4639 David Sherwood <david.sherwood@arm.com>
4641 * expr.c (expand_assignment): Cope with polynomial mode sizes
4642 when assigning to a CONCAT.
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 * machmode.h (mode_precision): Change from unsigned short to
4650 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4652 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4653 or if measurement_type is not polynomial.
4654 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4655 in which the mode is already known to be a scalar_int_mode.
4656 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4657 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4659 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4660 for GET_MODE_PRECISION.
4661 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4662 for GET_MODE_PRECISION.
4663 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4665 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4666 (expand_field_assignment, make_extraction): Likewise.
4667 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4668 (get_last_value): Likewise.
4669 * convert.c (convert_to_integer_1): Likewise.
4670 * cse.c (cse_insn): Likewise.
4671 * expr.c (expand_expr_real_1): Likewise.
4672 * lra-constraints.c (simplify_operand_subreg): Likewise.
4673 * optabs-query.c (can_atomic_load_p): Likewise.
4674 * optabs.c (expand_atomic_load): Likewise.
4675 (expand_atomic_store): Likewise.
4676 * ree.c (combine_reaching_defs): Likewise.
4677 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4678 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4679 * tree.h (type_has_mode_precision_p): Likewise.
4680 * ubsan.c (instrument_si_overflow): Likewise.
4682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4683 Alan Hayward <alan.hayward@arm.com>
4684 David Sherwood <david.sherwood@arm.com>
4686 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4687 polynomial numbers of units.
4688 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4689 (valid_vector_subparts_p): New function.
4690 (build_vector_type): Remove temporary shim and take the number
4691 of units as a poly_uint64 rather than an int.
4692 (build_opaque_vector_type): Take the number of units as a
4693 poly_uint64 rather than an int.
4694 * tree.c (build_vector_from_ctor): Handle polynomial
4695 TYPE_VECTOR_SUBPARTS.
4696 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4697 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4698 (build_vector_from_val): If the number of units is variable,
4699 use build_vec_duplicate_cst for constant operands and
4700 VEC_DUPLICATE_EXPR otherwise.
4701 (make_vector_type): Remove temporary is_constant ().
4702 (build_vector_type, build_opaque_vector_type): Take the number of
4703 units as a poly_uint64 rather than an int.
4704 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4706 * cfgexpand.c (expand_debug_expr): Likewise.
4707 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4708 (store_constructor, expand_expr_real_1): Likewise.
4709 (const_scalar_mask_from_tree): Likewise.
4710 * fold-const-call.c (fold_const_reduction): Likewise.
4711 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4712 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4713 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4714 (fold_relational_const): Likewise.
4715 (native_interpret_vector): Likewise. Change the size from an
4716 int to an unsigned int.
4717 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4718 TYPE_VECTOR_SUBPARTS.
4719 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4720 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4721 duplicating a non-constant operand into a variable-length vector.
4722 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4723 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4724 * ipa-icf.c (sem_variable::equals): Likewise.
4725 * match.pd: Likewise.
4726 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4727 * print-tree.c (print_node): Likewise.
4728 * stor-layout.c (layout_type): Likewise.
4729 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4730 * tree-cfg.c (verify_gimple_comparison): Likewise.
4731 (verify_gimple_assign_binary): Likewise.
4732 (verify_gimple_assign_ternary): Likewise.
4733 (verify_gimple_assign_single): Likewise.
4734 * tree-pretty-print.c (dump_generic_node): Likewise.
4735 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4736 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4737 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4738 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4739 (vect_shift_permute_load_chain): Likewise.
4740 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4741 (expand_vector_condition, optimize_vector_constructor): Likewise.
4742 (lower_vec_perm, get_compute_type): Likewise.
4743 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4744 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4745 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4746 (vect_recog_mask_conversion_pattern): Likewise.
4747 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4748 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4749 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4750 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4751 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4752 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4753 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4754 (supportable_widening_operation): Likewise.
4755 (supportable_narrowing_operation): Likewise.
4756 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4758 * varasm.c (output_constant): Likewise.
4760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4761 Alan Hayward <alan.hayward@arm.com>
4762 David Sherwood <david.sherwood@arm.com>
4764 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4765 so that both the length == 3 and length != 3 cases set up their
4766 own permute vectors. Add comments explaining why we know the
4767 number of elements is constant.
4768 (vect_permute_load_chain): Likewise.
4770 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4771 Alan Hayward <alan.hayward@arm.com>
4772 David Sherwood <david.sherwood@arm.com>
4774 * machmode.h (mode_nunits): Change from unsigned char to
4776 (ONLY_FIXED_SIZE_MODES): New macro.
4777 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4778 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4779 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4781 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4782 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4783 or if measurement_type is not polynomial.
4784 * genmodes.c (ZERO_COEFFS): New macro.
4785 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4787 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4788 Use ZERO_COEFFS when emitting initializers.
4789 * data-streamer.h (bp_pack_poly_value): New function.
4790 (bp_unpack_poly_value): Likewise.
4791 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4792 for GET_MODE_NUNITS.
4793 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4794 for GET_MODE_NUNITS.
4795 * tree.c (make_vector_type): Remove temporary shim and make
4796 the real function take the number of units as a poly_uint64
4798 (build_vector_type_for_mode): Handle polynomial nunits.
4799 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4800 * emit-rtl.c (const_vec_series_p_1): Likewise.
4801 (gen_rtx_CONST_VECTOR): Likewise.
4802 * fold-const.c (test_vec_duplicate_folding): Likewise.
4803 * genrecog.c (validate_pattern): Likewise.
4804 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4805 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4806 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4807 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4808 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4809 * rtlanal.c (subreg_get_info): Likewise.
4810 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4811 (vect_grouped_load_supported): Likewise.
4812 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4813 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4814 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4815 (simplify_const_unary_operation, simplify_binary_operation_1)
4816 (simplify_const_binary_operation, simplify_ternary_operation)
4817 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4818 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4819 instead of CONST_VECTOR_NUNITS.
4820 * varasm.c (output_constant_pool_2): Likewise.
4821 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4822 explicit-encoded elements in the XVEC for variable-length vectors.
4824 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4826 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4829 Alan Hayward <alan.hayward@arm.com>
4830 David Sherwood <david.sherwood@arm.com>
4832 * coretypes.h (fixed_size_mode): Declare.
4833 (fixed_size_mode_pod): New typedef.
4834 * builtins.h (target_builtins::x_apply_args_mode)
4835 (target_builtins::x_apply_result_mode): Change type to
4836 fixed_size_mode_pod.
4837 * builtins.c (apply_args_size, apply_result_size, result_vector)
4838 (expand_builtin_apply_args_1, expand_builtin_apply)
4839 (expand_builtin_return): Update accordingly.
4841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4843 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4844 * cselib.c (cselib_hash_rtx): Likewise.
4845 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4846 CONST_VECTOR encoding.
4848 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4849 Jeff Law <law@redhat.com>
4852 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4853 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4854 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4855 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4858 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4859 explicitly probe *sp in a noreturn function if there were any callee
4860 register saves or frame pointer is needed.
4862 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4865 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4866 BLKmode for ternary, binary or unary expressions.
4869 * var-tracking.c (delete_vta_debug_insn): New inline function.
4870 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4871 insns from get_insns () to NULL instead of each bb separately.
4872 Use delete_vta_debug_insn. No longer static.
4873 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4874 delete_vta_debug_insns callers.
4875 * rtl.h (delete_vta_debug_insns): Declare.
4876 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4877 instead of variable_tracking_main.
4879 2018-01-03 Martin Sebor <msebor@redhat.com>
4881 PR tree-optimization/83603
4882 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4883 arguments past the endof the argument list in functions declared
4884 without a prototype.
4885 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4886 Avoid checking when arguments are null.
4888 2018-01-03 Martin Sebor <msebor@redhat.com>
4891 * doc/extend.texi (attribute const): Fix a typo.
4892 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4893 issuing -Wsuggest-attribute for void functions.
4895 2018-01-03 Martin Sebor <msebor@redhat.com>
4897 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4898 offset_int::from instead of wide_int::to_shwi.
4899 (maybe_diag_overlap): Remove assertion.
4900 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4901 * gimple-ssa-sprintf.c (format_directive): Same.
4902 (parse_directive): Same.
4903 (sprintf_dom_walker::compute_format_length): Same.
4904 (try_substitute_return_value): Same.
4906 2018-01-03 Jeff Law <law@redhat.com>
4909 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4910 non-constant residual for zero at runtime and avoid probing in
4911 that case. Reorganize code for trailing problem to mirror handling
4914 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4916 PR tree-optimization/83501
4917 * tree-ssa-strlen.c (get_string_cst): New.
4918 (handle_char_store): Call get_string_cst.
4920 2018-01-03 Martin Liska <mliska@suse.cz>
4922 PR tree-optimization/83593
4923 * tree-ssa-strlen.c: Include tree-cfg.h.
4924 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4925 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4926 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4928 (strlen_dom_walker::before_dom_children): Call
4929 gimple_purge_dead_eh_edges. Dump tranformation with details
4931 (strlen_dom_walker::before_dom_children): Update call by adding
4932 new argument cleanup_eh.
4933 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4935 2018-01-03 Martin Liska <mliska@suse.cz>
4938 * cif-code.def (VARIADIC_THUNK): New enum value.
4939 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4942 2018-01-03 Jan Beulich <jbeulich@suse.com>
4944 * sse.md (mov<mode>_internal): Tighten condition for when to use
4945 vmovdqu<ssescalarsize> for TI and OI modes.
4947 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4949 Update copyright years.
4951 2018-01-03 Martin Liska <mliska@suse.cz>
4954 * ipa-visibility.c (function_and_variable_visibility): Skip
4955 functions with noipa attribure.
4957 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4959 * gcc.c (process_command): Update copyright notice dates.
4960 * gcov-dump.c (print_version): Ditto.
4961 * gcov.c (print_version): Ditto.
4962 * gcov-tool.c (print_version): Ditto.
4963 * gengtype.c (create_file): Ditto.
4964 * doc/cpp.texi: Bump @copying's copyright year.
4965 * doc/cppinternals.texi: Ditto.
4966 * doc/gcc.texi: Ditto.
4967 * doc/gccint.texi: Ditto.
4968 * doc/gcov.texi: Ditto.
4969 * doc/install.texi: Ditto.
4970 * doc/invoke.texi: Ditto.
4972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4974 * vector-builder.h (vector_builder::m_full_nelts): Change from
4975 unsigned int to poly_uint64.
4976 (vector_builder::full_nelts): Update prototype accordingly.
4977 (vector_builder::new_vector): Likewise.
4978 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4979 (vector_builder::operator ==): Likewise.
4980 (vector_builder::finalize): Likewise.
4981 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4982 Take the number of elements as a poly_uint64 rather than an
4984 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4985 from unsigned int to poly_uint64.
4986 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4987 (vec_perm_indices::new_vector): Likewise.
4988 (vec_perm_indices::length): Likewise.
4989 (vec_perm_indices::nelts_per_input): Likewise.
4990 (vec_perm_indices::input_nelts): Likewise.
4991 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4992 number of elements per input as a poly_uint64 rather than an
4993 unsigned int. Use the original encoding for variable-length
4994 vectors, rather than clamping each individual element.
4995 For the second and subsequent elements in each pattern,
4996 clamp the step and base before clamping their sum.
4997 (vec_perm_indices::series_p): Handle polynomial element counts.
4998 (vec_perm_indices::all_in_range_p): Likewise.
4999 (vec_perm_indices_to_tree): Likewise.
5000 (vec_perm_indices_to_rtx): Likewise.
5001 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5002 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5003 (tree_vector_builder::new_binary_operation): Handle polynomial
5004 element counts. Return false if we need to know the number
5005 of elements at compile time.
5006 * fold-const.c (fold_vec_perm): Punt if the number of elements
5007 isn't known at compile time.
5009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5011 * vec-perm-indices.h (vec_perm_builder): Change element type
5012 from HOST_WIDE_INT to poly_int64.
5013 (vec_perm_indices::element_type): Update accordingly.
5014 (vec_perm_indices::clamp): Handle polynomial element_types.
5015 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5016 (vec_perm_indices::all_in_range_p): Likewise.
5017 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5019 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5020 polynomial vec_perm_indices element types.
5021 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5022 * fold-const.c (fold_vec_perm): Likewise.
5023 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5024 * tree-vect-generic.c (lower_vec_perm): Likewise.
5025 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5026 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5027 element type to HOST_WIDE_INT.
5029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5030 Alan Hayward <alan.hayward@arm.com>
5031 David Sherwood <david.sherwood@arm.com>
5033 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5034 rather than an int. Use plus_constant.
5035 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5036 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 Alan Hayward <alan.hayward@arm.com>
5040 David Sherwood <david.sherwood@arm.com>
5042 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5043 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
5050 mode sizes. Require a constant size for BLKmode parameters
5051 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5052 forces a parameter to be padded at the lsb end in order to
5053 fill a complete number of words, require the parameter size
5054 to be ordered wrt UNITS_PER_WORD.
5056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5057 Alan Hayward <alan.hayward@arm.com>
5058 David Sherwood <david.sherwood@arm.com>
5060 * reload1.c (spill_stack_slot_width): Change element type
5061 from unsigned int to poly_uint64_pod.
5062 (alter_reg): Treat mode sizes as polynomial.
5064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5065 Alan Hayward <alan.hayward@arm.com>
5066 David Sherwood <david.sherwood@arm.com>
5068 * reload.c (complex_word_subreg_p): New function.
5069 (reload_inner_reg_of_subreg, push_reload): Use it.
5071 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5072 Alan Hayward <alan.hayward@arm.com>
5073 David Sherwood <david.sherwood@arm.com>
5075 * lra-constraints.c (process_alt_operands): Reject matched
5076 operands whose sizes aren't ordered.
5077 (match_reload): Refer to this check here.
5079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5080 Alan Hayward <alan.hayward@arm.com>
5081 David Sherwood <david.sherwood@arm.com>
5083 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5084 that the mode size is in the set {1, 2, 4, 8, 16}.
5086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5087 Alan Hayward <alan.hayward@arm.com>
5088 David Sherwood <david.sherwood@arm.com>
5090 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5091 Use plus_constant instead of gen_rtx_PLUS.
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 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5098 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5099 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5100 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5101 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5102 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5103 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5104 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5105 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5106 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5108 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5109 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5110 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5111 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5112 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5113 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5114 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5115 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5116 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5117 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5119 * expr.c (emit_move_resolve_push): Treat the input and result
5120 of PUSH_ROUNDING as a poly_int64.
5121 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5122 (emit_push_insn): Likewise.
5123 * lra-eliminations.c (mark_not_eliminable): Likewise.
5124 * recog.c (push_operand): Likewise.
5125 * reload1.c (elimination_effects): Likewise.
5126 * rtlanal.c (nonzero_bits1): Likewise.
5127 * calls.c (store_one_arg): Likewise. Require the padding to be
5128 known at compile time.
5130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5131 Alan Hayward <alan.hayward@arm.com>
5132 David Sherwood <david.sherwood@arm.com>
5134 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5135 Use plus_constant instead of gen_rtx_PLUS.
5137 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5138 Alan Hayward <alan.hayward@arm.com>
5139 David Sherwood <david.sherwood@arm.com>
5141 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5144 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5145 Alan Hayward <alan.hayward@arm.com>
5146 David Sherwood <david.sherwood@arm.com>
5148 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5149 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5150 via stack temporaries. Treat the mode size as polynomial too.
5152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5153 Alan Hayward <alan.hayward@arm.com>
5154 David Sherwood <david.sherwood@arm.com>
5156 * expr.c (expand_expr_real_2): When handling conversions involving
5157 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5158 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5159 as a poly_uint64 too.
5161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5162 Alan Hayward <alan.hayward@arm.com>
5163 David Sherwood <david.sherwood@arm.com>
5165 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5168 Alan Hayward <alan.hayward@arm.com>
5169 David Sherwood <david.sherwood@arm.com>
5171 * combine.c (can_change_dest_mode): Handle polynomial
5172 REGMODE_NATURAL_SIZE.
5173 * expmed.c (store_bit_field_1): Likewise.
5174 * expr.c (store_constructor): Likewise.
5175 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5176 and polynomial REGMODE_NATURAL_SIZE.
5177 (gen_lowpart_common): Likewise.
5178 * reginfo.c (record_subregs_of_mode): Likewise.
5179 * rtlanal.c (read_modify_subreg_p): Likewise.
5181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5182 Alan Hayward <alan.hayward@arm.com>
5183 David Sherwood <david.sherwood@arm.com>
5185 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5186 numbers of elements.
5188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5189 Alan Hayward <alan.hayward@arm.com>
5190 David Sherwood <david.sherwood@arm.com>
5192 * match.pd: Cope with polynomial numbers of vector elements.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5195 Alan Hayward <alan.hayward@arm.com>
5196 David Sherwood <david.sherwood@arm.com>
5198 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5199 in a POINTER_PLUS_EXPR.
5201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5202 Alan Hayward <alan.hayward@arm.com>
5203 David Sherwood <david.sherwood@arm.com>
5205 * omp-simd-clone.c (simd_clone_subparts): New function.
5206 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5207 (ipa_simd_modify_function_body): Likewise.
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 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5214 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5215 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5216 (expand_vector_condition, vector_element): Likewise.
5217 (subparts_gt): New function.
5218 (get_compute_type): Use subparts_gt.
5219 (count_type_subparts): Delete.
5220 (expand_vector_operations_1): Use subparts_gt instead of
5221 count_type_subparts.
5223 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5224 Alan Hayward <alan.hayward@arm.com>
5225 David Sherwood <david.sherwood@arm.com>
5227 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5228 (vect_compile_time_alias): ...this new function. Do the calculation
5229 on poly_ints rather than trees.
5230 (vect_prune_runtime_alias_test_list): Update call accordingly.
5232 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5233 Alan Hayward <alan.hayward@arm.com>
5234 David Sherwood <david.sherwood@arm.com>
5236 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5238 (vect_schedule_slp_instance): Likewise.
5240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5241 Alan Hayward <alan.hayward@arm.com>
5242 David Sherwood <david.sherwood@arm.com>
5244 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5245 constant and extern definitions for variable-length vectors.
5246 (vect_get_constant_vectors): Note that the number of units
5247 is known to be constant.
5249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5250 Alan Hayward <alan.hayward@arm.com>
5251 David Sherwood <david.sherwood@arm.com>
5253 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5254 of units as polynomial. Choose between WIDE and NARROW based
5257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5258 Alan Hayward <alan.hayward@arm.com>
5259 David Sherwood <david.sherwood@arm.com>
5261 * tree-vect-stmts.c (simd_clone_subparts): New function.
5262 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5265 Alan Hayward <alan.hayward@arm.com>
5266 David Sherwood <david.sherwood@arm.com>
5268 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5269 vectors as polynomial. Use build_index_vector for
5272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5273 Alan Hayward <alan.hayward@arm.com>
5274 David Sherwood <david.sherwood@arm.com>
5276 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5277 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5278 for variable-length vectors.
5279 (vectorizable_mask_load_store): Treat the number of units as
5280 polynomial, asserting that it is constant if the condition has
5281 already been enforced.
5282 (vectorizable_store, vectorizable_load): Likewise.
5284 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5285 Alan Hayward <alan.hayward@arm.com>
5286 David Sherwood <david.sherwood@arm.com>
5288 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5289 of units as polynomial. Punt if we can't tell at compile time
5290 which vector contains the final result.
5292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5293 Alan Hayward <alan.hayward@arm.com>
5294 David Sherwood <david.sherwood@arm.com>
5296 * tree-vect-loop.c (vectorizable_induction): Treat the number
5297 of units as polynomial. Punt on SLP inductions. Use an integer
5298 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5299 cast of such a series for variable-length floating-point
5302 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5303 Alan Hayward <alan.hayward@arm.com>
5304 David Sherwood <david.sherwood@arm.com>
5306 * tree.h (build_index_vector): Declare.
5307 * tree.c (build_index_vector): New function.
5308 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5309 of units as polynomial, forcibly converting it to a constant if
5310 vectorizable_reduction has already enforced the condition.
5311 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5312 to create a {1,2,3,...} vector.
5313 (vectorizable_reduction): Treat the number of units as polynomial.
5314 Choose vectype_in based on the largest scalar element size rather
5315 than the smallest number of units. Enforce the restrictions
5318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5319 Alan Hayward <alan.hayward@arm.com>
5320 David Sherwood <david.sherwood@arm.com>
5322 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5323 number of units as polynomial.
5325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5326 Alan Hayward <alan.hayward@arm.com>
5327 David Sherwood <david.sherwood@arm.com>
5329 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5330 * target.def (autovectorize_vector_sizes): Return the vector sizes
5331 by pointer, using vector_sizes rather than a bitmask.
5332 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5333 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5334 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5336 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5337 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5338 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5339 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5340 * omp-general.c (omp_max_vf): Likewise.
5341 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5342 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5343 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5344 * tree-vect-slp.c (vect_slp_bb): Likewise.
5345 * doc/tm.texi: Regenerate.
5346 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5348 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5349 the vector size as a poly_uint64 rather than an unsigned int.
5350 (current_vector_size): Change from an unsigned int to a poly_uint64.
5351 (get_vectype_for_scalar_type): Update accordingly.
5352 * tree.h (build_truth_vector_type): Take the size and number of
5353 units as a poly_uint64 rather than an unsigned int.
5354 (build_vector_type): Add a temporary overload that takes
5355 the number of units as a poly_uint64 rather than an unsigned int.
5356 * tree.c (make_vector_type): Likewise.
5357 (build_truth_vector_type): Take the number of units as a poly_uint64
5358 rather than an unsigned int.
5360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5361 Alan Hayward <alan.hayward@arm.com>
5362 David Sherwood <david.sherwood@arm.com>
5364 * target.def (get_mask_mode): Take the number of units and length
5365 as poly_uint64s rather than unsigned ints.
5366 * targhooks.h (default_get_mask_mode): Update accordingly.
5367 * targhooks.c (default_get_mask_mode): Likewise.
5368 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5369 * doc/tm.texi: Regenerate.
5371 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5372 Alan Hayward <alan.hayward@arm.com>
5373 David Sherwood <david.sherwood@arm.com>
5375 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5376 * omp-general.c (omp_max_vf): Likewise.
5377 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5378 (expand_omp_simd): Handle polynomial safelen.
5379 * omp-low.c (omplow_simd_context): Add a default constructor.
5380 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5381 (lower_rec_simd_input_clauses): Update accordingly.
5382 (lower_rec_input_clauses): Likewise.
5384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5385 Alan Hayward <alan.hayward@arm.com>
5386 David Sherwood <david.sherwood@arm.com>
5388 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5389 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5390 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5391 (vect_analyze_slp_cost): Likewise.
5392 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5393 (vect_model_load_cost): Likewise.
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5400 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5401 from an unsigned int * to a poly_uint64_pod *.
5402 (calculate_unrolling_factor): New function.
5403 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5405 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5406 Alan Hayward <alan.hayward@arm.com>
5407 David Sherwood <david.sherwood@arm.com>
5409 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5410 from an unsigned int to a poly_uint64.
5411 (_loop_vec_info::slp_unrolling_factor): Likewise.
5412 (_loop_vec_info::vectorization_factor): Change from an int
5414 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5415 (vect_get_num_vectors): New function.
5416 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5417 (vect_get_num_copies): Use vect_get_num_vectors.
5418 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5419 to an unsigned int *.
5420 (vect_analyze_data_refs): Change min_vf from an int * to a
5422 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5423 than an unsigned HOST_WIDE_INT.
5424 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5425 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5426 to an unsigned int *.
5427 (vect_analyze_data_ref_dependences): Likewise.
5428 (vect_compute_data_ref_alignment): Handle polynomial vf.
5429 (vect_enhance_data_refs_alignment): Likewise.
5430 (vect_prune_runtime_alias_test_list): Likewise.
5431 (vect_shift_permute_load_chain): Likewise.
5432 (vect_supportable_dr_alignment): Likewise.
5433 (dependence_distance_ge_vf): Take the vectorization factor as a
5434 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5435 (vect_analyze_data_refs): Change min_vf from an int * to a
5437 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5438 vfm1 as a poly_uint64 rather than an int. Make the same change
5439 for the returned bound_scalar.
5440 (vect_gen_vector_loop_niters): Handle polynomial vf.
5441 (vect_do_peeling): Likewise. Update call to
5442 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5443 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5445 * tree-vect-loop.c (vect_determine_vectorization_factor)
5446 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5447 (vect_get_known_peeling_cost): Likewise.
5448 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5449 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5450 (vect_transform_loop): Likewise. Use the lowest possible VF when
5451 updating the upper bounds of the loop.
5452 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5454 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5455 polynomial unroll factors.
5456 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5457 (vect_make_slp_decision): Likewise.
5458 (vect_supported_load_permutation_p): Likewise, and polynomial
5460 (vect_analyze_slp_cost): Handle polynomial vf.
5461 (vect_slp_analyze_node_operations): Likewise.
5462 (vect_slp_analyze_bb_1): Likewise.
5463 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5464 than an unsigned HOST_WIDE_INT.
5465 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5466 (vectorizable_load): Handle polynomial vf.
5467 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5469 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5471 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5472 Alan Hayward <alan.hayward@arm.com>
5473 David Sherwood <david.sherwood@arm.com>
5475 * match.pd: Handle bit operations involving three constants
5476 and try to fold one pair.
5478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5480 * tree-vect-loop-manip.c: Include gimple-fold.h.
5481 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5482 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5483 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5484 Add a path that uses a step of VF instead of 1, but disable it
5486 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5487 and niters_no_overflow parameters. Update calls to
5488 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5489 Create a new SSA name if the latter choses to use a ste other
5490 than zero, and return it via niters_vector_mult_vf_var.
5491 * tree-vect-loop.c (vect_transform_loop): Update calls to
5492 vect_do_peeling, vect_gen_vector_loop_niters and
5493 slpeel_make_loop_iterate_ntimes.
5494 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5495 (vect_gen_vector_loop_niters): Update declarations after above changes.
5497 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5499 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5500 128-bit round to integer instructions.
5501 (ceil<mode>2): Likewise.
5502 (btrunc<mode>2): Likewise.
5503 (round<mode>2): Likewise.
5505 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5507 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5508 unaligned VSX load/store on P8/P9.
5509 (expand_block_clear): Allow the use of unaligned VSX
5510 load/store on P8/P9.
5512 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5514 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5516 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5517 swap associated with both a load and a store.
5519 2018-01-02 Andrew Waterman <andrew@sifive.com>
5521 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5522 * config/riscv/riscv.md (clear_cache): Use it.
5524 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5526 * web.c: Remove out-of-date comment.
5528 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5530 * expr.c (fixup_args_size_notes): Check that any existing
5531 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5532 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5533 (emit_single_push_insn): ...here.
5535 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5537 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5538 (const_vector_encoded_nelts): New function.
5539 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5540 (const_vector_int_elt, const_vector_elt): Declare.
5541 * emit-rtl.c (const_vector_int_elt_1): New function.
5542 (const_vector_elt): Likewise.
5543 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5544 of CONST_VECTOR_ELT.
5546 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5548 * expr.c: Include rtx-vector-builder.h.
5549 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5550 directly on the tree encoding.
5551 (const_vector_from_tree): Likewise.
5552 * optabs.c: Include rtx-vector-builder.h.
5553 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5554 sequence of "u" values.
5555 * vec-perm-indices.c: Include rtx-vector-builder.h.
5556 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5557 directly on the vec_perm_indices encoding.
5559 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5561 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5562 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5563 * rtx-vector-builder.h: New file.
5564 * rtx-vector-builder.c: Likewise.
5565 * rtl.h (rtx_def::u2): Add a const_vector field.
5566 (CONST_VECTOR_NPATTERNS): New macro.
5567 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5568 (CONST_VECTOR_DUPLICATE_P): Likewise.
5569 (CONST_VECTOR_STEPPED_P): Likewise.
5570 (CONST_VECTOR_ENCODED_ELT): Likewise.
5571 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5572 (unwrap_const_vec_duplicate): Likewise.
5573 (const_vec_series_p): Check for a non-duplicated vector encoding.
5574 Say that the function only returns true for integer vectors.
5575 * emit-rtl.c: Include rtx-vector-builder.h.
5576 (gen_const_vec_duplicate_1): Delete.
5577 (gen_const_vector): Call gen_const_vec_duplicate instead of
5578 gen_const_vec_duplicate_1.
5579 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5580 (gen_const_vec_duplicate): Use rtx_vector_builder.
5581 (gen_const_vec_series): Likewise.
5582 (gen_rtx_CONST_VECTOR): Likewise.
5583 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5584 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5585 Build a new vector rather than modifying a CONST_VECTOR in-place.
5586 (handle_special_swappables): Update call accordingly.
5587 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5588 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5589 Build a new vector rather than modifying a CONST_VECTOR in-place.
5590 (handle_special_swappables): Update call accordingly.
5592 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5594 * simplify-rtx.c (simplify_const_binary_operation): Use
5595 CONST_VECTOR_ELT instead of XVECEXP.
5597 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5599 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5600 the selector elements to be different from the data elements
5601 if the selector is a VECTOR_CST.
5602 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5603 ssizetype for the selector.
5605 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5607 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5608 before testing each element individually.
5609 * tree-vect-generic.c (lower_vec_perm): Likewise.
5611 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5613 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5614 * selftest-run-tests.c (selftest::run_tests): Call it.
5615 * vector-builder.h (vector_builder::operator ==): New function.
5616 (vector_builder::operator !=): Likewise.
5617 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5618 (vec_perm_indices::all_from_input_p): New function.
5619 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5620 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5621 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5622 instead of reading the VECTOR_CST directly. Detect whether both
5623 vector inputs are the same before constructing the vec_perm_indices,
5624 and update the number of inputs argument accordingly. Use the
5625 utility functions added above. Only construct sel2 if we need to.
5627 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5629 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5630 the broadcast of the low byte.
5631 (expand_mult_highpart): Use an explicit encoding for the permutes.
5632 * optabs-query.c (can_mult_highpart_p): Likewise.
5633 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5634 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5635 (vectorizable_bswap): Likewise.
5636 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5637 explicit encoding for the power-of-2 permutes.
5638 (vect_permute_store_chain): Likewise.
5639 (vect_grouped_load_supported): Likewise.
5640 (vect_permute_load_chain): Likewise.
5642 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5644 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5645 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5646 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5647 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5648 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5649 (vect_gen_perm_mask_any): Likewise.
5651 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5653 * int-vector-builder.h: New file.
5654 * vec-perm-indices.h: Include int-vector-builder.h.
5655 (vec_perm_indices): Redefine as an int_vector_builder.
5656 (auto_vec_perm_indices): Delete.
5657 (vec_perm_builder): Redefine as a stand-alone class.
5658 (vec_perm_indices::vec_perm_indices): New function.
5659 (vec_perm_indices::clamp): Likewise.
5660 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5661 (vec_perm_indices::new_vector): New function.
5662 (vec_perm_indices::new_expanded_vector): Update for new
5663 vec_perm_indices class.
5664 (vec_perm_indices::rotate_inputs): New function.
5665 (vec_perm_indices::all_in_range_p): Operate directly on the
5666 encoded form, without computing elided elements.
5667 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5668 encoding. Update for new vec_perm_indices class.
5669 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5670 the given vec_perm_builder.
5671 (expand_vec_perm_var): Update vec_perm_builder constructor.
5672 (expand_mult_highpart): Use vec_perm_builder instead of
5673 auto_vec_perm_indices.
5674 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5675 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5676 or double series encoding as appropriate.
5677 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5678 vec_perm_indices instead of auto_vec_perm_indices.
5679 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5680 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5681 (vect_permute_store_chain): Likewise.
5682 (vect_grouped_load_supported): Likewise.
5683 (vect_permute_load_chain): Likewise.
5684 (vect_shift_permute_load_chain): Likewise.
5685 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5686 (vect_transform_slp_perm_load): Likewise.
5687 (vect_schedule_slp_instance): Likewise.
5688 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5689 (vectorizable_mask_load_store): Likewise.
5690 (vectorizable_bswap): Likewise.
5691 (vectorizable_store): Likewise.
5692 (vectorizable_load): Likewise.
5693 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5694 vec_perm_indices instead of auto_vec_perm_indices. Use
5695 tree_to_vec_perm_builder to read the vector from a tree.
5696 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5697 vec_perm_builder instead of a vec_perm_indices.
5698 (have_whole_vector_shift): Use vec_perm_builder and
5699 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5700 truncation to calc_vec_perm_mask_for_shift.
5701 (vect_create_epilog_for_reduction): Likewise.
5702 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5703 from auto_vec_perm_indices to vec_perm_indices.
5704 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5705 instead of changing individual elements.
5706 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5707 the vector in d.perm.
5708 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5709 from auto_vec_perm_indices to vec_perm_indices.
5710 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5711 instead of changing individual elements.
5712 (arm_vectorize_vec_perm_const): Use new_vector to install
5713 the vector in d.perm.
5714 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5715 Update vec_perm_builder constructor.
5716 (rs6000_expand_interleave): Likewise.
5717 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5718 (rs6000_expand_interleave): Likewise.
5720 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5722 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5723 to qimode could truncate the indices.
5724 * optabs.c (expand_vec_perm_var): Likewise.
5726 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5728 * Makefile.in (OBJS): Add vec-perm-indices.o.
5729 * vec-perm-indices.h: New file.
5730 * vec-perm-indices.c: Likewise.
5731 * target.h (vec_perm_indices): Replace with a forward class
5733 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5734 * optabs.h: Include vec-perm-indices.h.
5735 (expand_vec_perm): Delete.
5736 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5737 (expand_vec_perm_const): Declare.
5738 * target.def (vec_perm_const_ok): Replace with...
5739 (vec_perm_const): ...this new hook.
5740 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5741 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5742 * doc/tm.texi: Regenerate.
5743 * optabs.def (vec_perm_const): Delete.
5744 * doc/md.texi (vec_perm_const): Likewise.
5745 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5746 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5747 expand_vec_perm for constant permutation vectors. Assert that
5748 the mode of variable permutation vectors is the integer equivalent
5749 of the mode that is being permuted.
5750 * optabs-query.h (selector_fits_mode_p): Declare.
5751 * optabs-query.c: Include vec-perm-indices.h.
5752 (selector_fits_mode_p): New function.
5753 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5754 is defined, instead of checking whether the vec_perm_const_optab
5755 exists. Use targetm.vectorize.vec_perm_const instead of
5756 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5757 fit in the vector mode before using a variable permute.
5758 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5759 vec_perm_indices instead of an rtx.
5760 (expand_vec_perm): Replace with...
5761 (expand_vec_perm_const): ...this new function. Take the selector
5762 as a vec_perm_indices rather than an rtx. Also take the mode of
5763 the selector. Update call to shift_amt_for_vec_perm_mask.
5764 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5765 Use vec_perm_indices::new_expanded_vector to expand the original
5766 selector into bytes. Check whether the indices fit in the vector
5767 mode before using a variable permute.
5768 (expand_vec_perm_var): Make global.
5769 (expand_mult_highpart): Use expand_vec_perm_const.
5770 * fold-const.c: Includes vec-perm-indices.h.
5771 * tree-ssa-forwprop.c: Likewise.
5772 * tree-vect-data-refs.c: Likewise.
5773 * tree-vect-generic.c: Likewise.
5774 * tree-vect-loop.c: Likewise.
5775 * tree-vect-slp.c: Likewise.
5776 * tree-vect-stmts.c: Likewise.
5777 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5779 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5780 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5781 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5782 (aarch64_vectorize_vec_perm_const): ...this new function.
5783 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5784 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5785 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5786 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5787 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5788 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5789 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5791 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5792 check for NEON modes.
5793 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5794 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5795 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5796 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5798 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5799 the old VEC_PERM_CONST conditions.
5800 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5801 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5802 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5803 (ia64_vectorize_vec_perm_const_ok): Merge into...
5804 (ia64_vectorize_vec_perm_const): ...this new function.
5805 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5806 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5807 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5808 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5809 * config/mips/mips.c (mips_expand_vec_perm_const)
5810 (mips_vectorize_vec_perm_const_ok): Merge into...
5811 (mips_vectorize_vec_perm_const): ...this new function.
5812 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5813 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5814 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5815 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5816 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5817 (rs6000_expand_vec_perm_const): Delete.
5818 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5820 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5821 (altivec_expand_vec_perm_const_le): Take each operand individually.
5822 Operate on constant selectors rather than rtxes.
5823 (altivec_expand_vec_perm_const): Likewise. Update call to
5824 altivec_expand_vec_perm_const_le.
5825 (rs6000_expand_vec_perm_const): Delete.
5826 (rs6000_vectorize_vec_perm_const_ok): Delete.
5827 (rs6000_vectorize_vec_perm_const): New function.
5828 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5829 an element count and rtx array.
5830 (rs6000_expand_extract_even): Update call accordingly.
5831 (rs6000_expand_interleave): Likewise.
5832 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5833 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5834 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5835 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5836 (rs6000_expand_vec_perm_const): Delete.
5837 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5838 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5839 (altivec_expand_vec_perm_const_le): Take each operand individually.
5840 Operate on constant selectors rather than rtxes.
5841 (altivec_expand_vec_perm_const): Likewise. Update call to
5842 altivec_expand_vec_perm_const_le.
5843 (rs6000_expand_vec_perm_const): Delete.
5844 (rs6000_vectorize_vec_perm_const_ok): Delete.
5845 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5846 reference to the SPE evmerge intructions.
5847 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5848 an element count and rtx array.
5849 (rs6000_expand_extract_even): Update call accordingly.
5850 (rs6000_expand_interleave): Likewise.
5851 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5852 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5854 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5856 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5858 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5859 vector mode and that that mode matches the mode of the data
5861 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5862 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5863 directly using expand_vec_perm_1 when forcing selectors into
5865 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5867 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5869 * optabs-query.h (can_vec_perm_p): Delete.
5870 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5871 * optabs-query.c (can_vec_perm_p): Split into...
5872 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5873 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5874 particular selector is valid.
5875 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5876 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5877 (vect_grouped_load_supported): Likewise.
5878 (vect_shift_permute_load_chain): Likewise.
5879 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5880 (vect_transform_slp_perm_load): Likewise.
5881 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5882 (vectorizable_bswap): Likewise.
5883 (vect_gen_perm_mask_checked): Likewise.
5884 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5885 implementations of variable permutation vectors into account
5886 when deciding which selector to use.
5887 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5888 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5889 with a false third argument.
5890 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5891 to test whether the constant selector is valid and can_vec_perm_var_p
5892 to test whether a variable selector is valid.
5894 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5896 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5897 * optabs-query.c (can_vec_perm_p): Likewise.
5898 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5899 instead of vec_perm_indices.
5900 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5901 (vect_gen_perm_mask_checked): Likewise,
5902 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5903 (vect_gen_perm_mask_checked): Likewise,
5905 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5907 * optabs-query.h (qimode_for_vec_perm): Declare.
5908 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5909 (qimode_for_vec_perm): ...this new function.
5910 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5912 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5914 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5915 does not have a conditional at the top.
5917 2018-01-02 Richard Biener <rguenther@suse.de>
5919 * ipa-inline.c (big_speedup_p): Fix expression.
5921 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5924 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5927 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5931 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5932 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5933 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5934 cond_taken_branch_cost 3->4.
5936 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5938 PR tree-optimization/83581
5939 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5940 TODO_cleanup_cfg if any changes have been made.
5943 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5944 convert_modes if target mode has the right side, but different mode
5948 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5949 last argument when extracting from CONCAT. If either from_real or
5950 from_imag is NULL, use expansion through memory. If result is not
5951 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5952 the parts directly to inner mode, if even that fails, use expansion
5956 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5957 check for bswap in mode rather than HImode and use that in expand_unop
5960 Copyright (C) 2018 Free Software Foundation, Inc.
5962 Copying and distribution of this file, with or without modification,
5963 are permitted in any medium without royalty provided the copyright
5964 notice and this notice are preserved.