1 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
3 * configure: Regenerate.
5 2018-02-05 Martin Sebor <msebor@redhat.com>
7 PR tree-optimization/83369
8 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
11 2018-02-05 Martin Liska <mliska@suse.cz>
13 * doc/invoke.texi: Cherry-pick upstream r323995.
15 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
17 * ira.c (ira_init_register_move_cost): Adjust comment.
19 2018-02-05 Martin Liska <mliska@suse.cz>
22 * doc/gcov.texi: Fix typo in documentation.
24 2018-02-05 Martin Liska <mliska@suse.cz>
27 * doc/gcov.texi: Document necessity of --dynamic-list-data when
28 using dlopen functionality.
30 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
32 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
33 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
34 _mm_maskz_range_ss, _mm_mask_range_round_ss,
35 _mm_maskz_range_round_ss): New intrinsics.
36 (__builtin_ia32_rangesd128_round)
37 (__builtin_ia32_rangess128_round): Remove.
38 (__builtin_ia32_rangesd128_mask_round,
39 __builtin_ia32_rangess128_mask_round): New builtins.
40 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
41 __builtin_ia32_rangess128_round): Remove.
42 (__builtin_ia32_rangesd128_mask_round,
43 __builtin_ia32_rangess128_mask_round): New builtins.
44 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
45 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
46 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
47 "<round_saeonly_constraint>")): Changed to ...
48 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
49 "<round_saeonly_scalar_constraint>")): ... this.
50 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
51 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
52 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
53 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
54 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
56 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
58 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
60 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
61 Remove all values except native, 8540 and 8548.
63 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
65 * config/i386/i386.c (ix86_output_function_return): Pass
66 INVALID_REGNUM, instead of -1, as invalid register number to
67 indirect_thunk_name and output_indirect_thunk.
69 2018-02-02 Julia Koval <julia.koval@intel.com>
71 * config.gcc: Add -march=icelake.
72 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
73 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
74 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
75 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
76 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
77 (processor_target_table): Add icelake.
78 (ix86_option_override_internal): Handle new PTAs.
79 (get_builtin_code_for_version): Handle icelake.
80 (M_INTEL_COREI7_ICELAKE): New.
81 (fold_builtin_cpu): Handle icelake.
82 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
83 * doc/invoke.texi: Add -march=icelake.
85 2018-02-02 Julia Koval <julia.koval@intel.com>
87 * config/i386/i386.c (ix86_option_override_internal): Change flags type
89 * wide-int-bitmask.h: New.
91 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
94 * config/i386/i386.md: Replace Pmode with word_mode in
95 builtin_setjmp_setup and builtin_longjmp to support x32.
97 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
101 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
103 (rs6000_supported_cpu_names): New static variable.
104 (linux_cpu_translation_table): Likewise.
105 (elf_platform) <cpu>: Define new static variable and use it.
106 Translate kernel AT_PLATFORM name to canonical name if needed.
107 Error if platform name is unknown.
109 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
112 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
114 2018-02-01 Jeff Law <law@redhat.com>
117 * config/i386/i386.c (release_scratch_register_on_entry): Add new
118 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
119 the scratch if RELEASE_VIA_POP is false.
120 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
121 If we have to save a temporary register, decrement SIZE appropriately.
122 Pass new arguments to release_scratch_register_on_entry.
123 (ix86_adjust_stack_and_probe): Likewise.
124 (ix86_emit_probe_stack_range): Pass new arguments to
125 release_scratch_register_on_entry.
127 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
129 PR rtl-optimization/84157
130 * combine.c (change_zero_ext): Use REG_P predicate in
131 front of HARD_REGISTER_P predicate.
133 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
135 * config/avr/avr.c (avr_option_override): Move disabling of
136 -fdelete-null-pointer-checks to...
137 * common/config/avr/avr-common.c (avr_option_optimization_table):
140 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
142 PR tree-optimization/81635
143 * tree-data-ref.c (split_constant_offset_1): For types that
144 wrap on overflow, try to use range info to prove that wrapping
147 2018-02-01 Renlin Li <renlin.li@arm.com>
150 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
152 (aarch64_register_move_cost): Likewise.
153 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
155 (REG_CLASS_NAMES): Likewise.
156 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
157 TAILCALL_ADDR_REGS. Remove IP registers.
158 * config/aarch64/aarch64.md (Ucs): Update register constraint.
160 2018-02-01 Richard Biener <rguenther@suse.de>
162 * domwalk.h (dom_walker::dom_walker): Add additional constructor
163 for specifying RPO order and allow NULL for that.
164 * domwalk.c (dom_walker::dom_walker): Likewise.
165 (dom_walker::walk): Handle NULL RPO order.
166 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
168 (rewrite_update_dom_walker): Likewise.
169 (mark_def_dom_walker): Likewise.
171 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
173 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
174 (aarch64_maybe_expand_sve_subreg_move): Declare.
175 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
176 * config/aarch64/predicates.md (aarch64_any_register_operand): New
178 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
179 that are semantically a reverse operation.
180 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
181 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
182 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
184 (aarch64_can_change_mode_class): For big-endian, forbid changes
185 between two SVE modes if they have different element sizes.
187 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
189 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
190 the TImode handling for big-endian targets.
192 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
194 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
195 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
197 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
198 Remove BSWAP handing for big-endian targets and use the form of
199 LD1RQ appropariate for the mode.
201 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
203 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
204 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
207 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
210 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
211 check for operands that need to go through aarch64_sve_reload_be.
213 2018-02-01 Jakub Jelinek <jakub@redhat.com>
215 PR tree-optimization/81661
216 PR tree-optimization/84117
217 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
218 * tree-eh.c: Include gimplify.h.
219 (find_trapping_overflow, replace_trapping_overflow,
220 rewrite_to_non_trapping_overflow): New functions.
221 * tree-vect-loop.c: Include tree-eh.h.
222 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
223 * tree-data-ref.c: Include tree-eh.h.
224 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
226 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
228 PR rtl-optimization/84123
229 * combine.c (change_zero_ext): Check if hard register satisfies
230 can_change_dest_mode before calling gen_lowpart_SUBREG.
232 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
235 * ira.c (ira_init_register_move_cost): Remove assert.
237 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
239 PR rtl-optimization/84071
240 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
241 * doc/tm.texi: Regenerate.
243 2018-01-31 Richard Biener <rguenther@suse.de>
245 PR tree-optimization/84132
246 * tree-data-ref.c (analyze_miv_subscript): Properly
247 check whether evolution_function_is_affine_multivariate_p
248 before calling gcd_of_steps_may_divide_p.
250 2018-01-31 Julia Koval <julia.koval@intel.com>
253 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
254 * config/i386/i386.md (rdpid_rex64) New.
255 (rdpid): Make 32bit only.
257 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
260 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
261 an IDENTIFIER_NODE for FUNCTION_TYPE's.
263 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
266 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
268 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
270 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
272 PR rtl-optimization/84071
273 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
274 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
276 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
278 * config/arc/arc.c (arc_handle_aux_attribute): New function.
279 (arc_attribute_table): Add 'aux' attribute.
280 (arc_in_small_data_p): Consider aux like variables.
281 (arc_is_aux_reg_p): New function.
282 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
283 (arc_get_aux_arg): New function.
284 (prepare_move_operands): Handle aux-register access.
285 (arc_handle_aux_attribute): New function.
286 * doc/extend.texi (ARC Variable attributes): Add subsection.
288 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
290 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
291 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
292 (arc_attribute_table): Add 'uncached' attribute.
293 (arc_print_operand): Print '.di' flag for uncached memory
295 (arc_in_small_data_p): Do not consider for small data the uncached
297 (arc_is_uncached_mem_p): New function.
298 * config/arc/predicates.md (compact_store_memory_operand): Check
299 for uncached memory accesses.
300 (nonvol_nonimm_operand): Likewise.
301 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
303 2018-01-31 Jakub Jelinek <jakub@redhat.com>
306 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
307 falign-loops=): Add Optimization flag.
309 2018-01-30 Jeff Law <law@redhat.com>
312 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
313 INT_REGISTERS_SAVED. Check it prior to calling
314 get_scratch_register_on_entry.
315 (ix86_adjust_stack_and_probe): Similarly.
316 (ix86_emit_probe_stack_range): Similarly.
317 (ix86_expand_prologue): Corresponding changes.
319 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
322 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
323 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
325 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
328 * lra-constraints.c (curr_insn_transform): Process AND in the
331 2018-01-30 Jakub Jelinek <jakub@redhat.com>
333 PR rtl-optimization/83986
334 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
335 dependence against last_pending_memory_flush in addition to
338 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
340 PR tree-optimization/81611
341 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
344 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
346 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
349 2018-01-30 Richard Biener <rguenther@suse.de>
350 Jakub Jelinek <jakub@redhat.com>
352 PR tree-optimization/84111
353 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
354 inner loops added during recursion, as they don't have up-to-date
357 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
360 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
361 (can_inline_edge_by_limits_p): ... here.
362 (can_early_inline_edge_p, check_callers,
363 update_caller_keys, update_callee_keys, recursive_inlining,
364 add_new_edges_to_heap, speculation_useful_p,
365 inline_small_functions,
366 inline_small_functions, flatten_function,
367 inline_to_all_callers_1): Update.
369 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
371 * profile-count.c (profile_count::combine_with_ipa_count): Handle
374 2018-01-30 Richard Biener <rguenther@suse.de>
376 PR tree-optimization/83008
377 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
378 invariant and constant vector uses in stmts when they need
381 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
384 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
385 * configure: Regenerate.
387 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
389 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
391 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
392 Use gen_rtx_REG rather than gen_lowpart.
394 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
396 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
397 rather than 0 when creating partial subregs.
399 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
401 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
404 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
407 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
408 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
409 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
410 flags. This restores the settings used before the 2017-07-24.
411 Turning off pre increment/decrement/modify allows IVOPTS to
412 optimize DF/SF loops where the index is an int.
414 2018-01-29 Richard Biener <rguenther@suse.de>
415 Kelvin Nilsen <kelvin@gcc.gnu.org>
418 * tree-vect-stmts.c (vectorizable_call): Don't call
419 targetm.vectorize_builtin_md_vectorized_function if callee is
422 2018-01-22 Carl Love <cel@us.ibm.com>
424 * doc/extend.tex: Fix typo in second arg in
425 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
427 2018-01-29 Richard Biener <rguenther@suse.de>
429 PR tree-optimization/84086
430 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
431 (flush_ssaname_freelist): When SSA names were released reset
434 2018-01-29 Richard Biener <rguenther@suse.de>
436 PR tree-optimization/84057
437 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
438 removed paths when removing edges.
440 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
442 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
443 -mfunction-return=@var{choice}.
445 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
448 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
450 (layout::print_source_line): Likewise.
451 (test_get_line_width_without_trailing_whitespace): Add test cases.
453 2018-01-27 Jakub Jelinek <jakub@redhat.com>
456 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
459 2018-01-26 Jim Wilson <jimw@sifive.com>
461 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
463 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
466 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
468 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
469 and CMP + SUB-immediate -> SUBS.
471 2018-01-26 Martin Sebor <msebor@redhat.com>
473 PR tree-optimization/83896
474 * tree-ssa-strlen.c (get_string_len): Rename...
475 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
476 Avoid assuming length is constant.
477 (handle_char_store): Use HOST_WIDE_INT for string length.
479 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
482 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
483 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
485 2018-01-26 Richard Biener <rguenther@suse.de>
487 PR rtl-optimization/84003
488 * dse.c (record_store): Only record redundant stores when
489 the earlier store aliases at least all accesses the later one does.
491 2018-01-26 Jakub Jelinek <jakub@redhat.com>
493 PR rtl-optimization/83985
494 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
495 REG_CFA_RESTORE insns.
496 (delete_unmarked_insns): Don't ignore separate shrink wrapping
497 REG_CFA_RESTORE insns here.
500 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
501 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
503 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
505 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
506 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
507 (arc_init): Likewise.
508 (arc_override_options): Likewise.
509 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
511 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
513 * config/arc/arc.h (TARGET_DBNZ): Define.
514 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
515 properly set the tune attribute.
516 (dbnz): Use TARGET_DBNZ guard.
517 * config/arc/arc.opt (mtune): Add core3 option.
519 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
521 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
522 recognize new pic like addresses.
523 (arc_delegitimize_address): Clean up.
525 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
527 * config/arc/arc-arches.def: Option mrf16 valid for all
529 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
530 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
531 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
532 * config/arc/arc-tables.opt: Regenerate.
533 * config/arc/arc.c (arc_conditional_register_usage): Handle
534 reduced register file case.
535 (arc_file_start): Set must have build attributes.
536 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
538 * config/arc/arc.opt (mrf16): Add new option.
539 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
540 * config/arc/genmultilib.awk: Handle new mrf16 option.
541 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
542 * config/arc/t-multilib: Regenerate.
543 * doc/invoke.texi (ARC Options): Document mrf16 option.
545 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
547 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
548 * config/arc/arc.c (arc_handle_secure_attribute): New function.
549 (arc_attribute_table): Add 'secure_call' attribute.
550 (arc_print_operand): Print secure call operand.
551 (arc_function_ok_for_sibcall): Don't optimize tail calls when
553 (arc_is_secure_call_p): New function. * config/arc/arc.md
554 (call_i): Add support for sjli instruction.
555 (call_value_i): Likewise.
556 * config/arc/constraints.md (Csc): New constraint.
558 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
559 John Eric Martin <John.Martin@emmicro-us.com>
561 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
562 * config/arc/arc.c (_arc_jli_section): New struct.
563 (arc_jli_section): New type.
564 (rc_jli_sections): New static variable.
565 (arc_handle_jli_attribute): New function.
566 (arc_attribute_table): Add jli_always and jli_fixed attribute.
567 (arc_file_end): New function.
568 (TARGET_ASM_FILE_END): Define.
569 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
570 (arc_add_jli_section): New function.
571 (jli_call_scan): Likewise.
572 (arc_reorg): Call jli_call_scan.
573 (arc_output_addsi): Remove 'S' from printing asm operand.
574 (arc_is_jli_call_p): New function.
575 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
577 (movhi_insn): Likewise.
578 (movsi_insn): Likewise.
579 (movsi_set_cc_insn): Likewise.
580 (loadqi_update): Likewise.
581 (load_zeroextendqisi_update): Likewise.
582 (load_signextendqisi_update): Likewise.
583 (loadhi_update): Likewise.
584 (load_zeroextendhisi_update): Likewise.
585 (load_signextendhisi_update): Likewise.
586 (loadsi_update): Likewise.
587 (loadsf_update): Likewise.
588 (movsicc_insn): Likewise.
589 (bset_insn): Likewise.
590 (bxor_insn): Likewise.
591 (bclr_insn): Likewise.
592 (bmsk_insn): Likewise.
593 (bicsi3_insn): Likewise.
594 (cmpsi_cc_c_insn): Likewise.
595 (movsi_ne): Likewise.
596 (movsi_cond_exec): Likewise.
597 (clrsbsi2): Likewise.
608 (call_i): Remove 'S' asm letter, add jli instruction.
609 (call_value_i): Likewise.
610 * config/arc/arc.op (mjli-always): New option.
611 * config/arc/constraints.md (Cji): New constraint.
612 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
614 (subsf3_fpx): Likewise.
615 (mulsf3_fpx): Likewise.
616 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
618 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
620 * doc/invoke.texi (ARC): Document mjli-always option.
622 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
624 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
625 and use incw and decw where possible
626 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
628 2018-01-26 Richard Biener <rguenther@suse.de>
630 PR tree-optimization/81082
631 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
632 association if it requires casting to unsigned.
633 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
634 from fold_plusminus_mult_expr to catch important cases late when
635 range info is available.
637 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
639 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
640 * configure.ac (hidden_linkonce): New test.
641 * configure: Regenerate.
642 * config.in: Regenerate.
644 2018-01-26 Julia Koval <julia.koval@intel.com>
646 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
647 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
648 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
649 _mm_mask_bitshuffle_epi64_mask): Fix type.
650 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
651 USI_FTYPE_V4DI_V4DI_USI): Remove.
652 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
653 __builtin_ia32_vpshufbitqmb256_mask,
654 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
655 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
656 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
658 2018-01-26 Alan Modra <amodra@gmail.com>
661 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
662 UNSPEC_VBPERMQ. Sort other unspecs.
664 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
666 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
668 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
671 * predict.c (drop_profile): Do not push/pop cfun; update also
673 (handle_missing_profiles): Fix logic looking for zero profiles.
675 2018-01-25 Jakub Jelinek <jakub@redhat.com>
678 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
679 on functions with #pragma omp declare simd or functions with simd
681 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
682 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
683 Remove trailing \n from warning_at calls.
685 2018-01-25 Tom de Vries <tom@codesourcery.com>
688 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
689 for neutered workers.
691 2018-01-24 Joseph Myers <joseph@codesourcery.com>
694 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
695 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
697 2018-01-24 Jeff Law <law@redhat.com>
700 * i386.c (get_probe_interval): Move to earlier point.
701 (ix86_compute_frame_layout): If -fstack-clash-protection and
702 the frame is larger than the probe interval, then use pushes
703 to save registers rather than reg->mem moves.
704 (ix86_expand_prologue): Remove conditional for int_registers_saved
707 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
710 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
711 min/max for never referenced object.
713 2018-01-24 Jakub Jelinek <jakub@redhat.com>
716 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
718 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
719 attributes from DECL_ATTRIBUTES (decl) without affecting
720 DECL_ATTRIBUTES (current_function_decl).
721 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
722 functions with non-NULL DECL_ABSTRACT_ORIGIN.
724 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
726 PR tree-optimization/83979
727 * fold-const.c (fold_comparison): Use constant_boolean_node
728 instead of boolean_{true,false}_node.
730 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
732 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
735 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
737 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
738 Simplify the clause that sets the length attribute.
739 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
740 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
741 clause that sets the length attribute.
742 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
744 2018-01-24 Tom de Vries <tom@codesourcery.com>
747 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
748 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
749 Add strict parameter.
750 (prevent_branch_around_nothing): Insert dummy insn between branch to
751 label and label with no ptx insn inbetween.
752 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
754 2018-01-24 Tom de Vries <tom@codesourcery.com>
757 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
758 for neutered threads in warp.
759 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
761 2018-01-24 Richard Biener <rguenther@suse.de>
763 PR tree-optimization/83176
764 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
767 2018-01-24 Richard Biener <rguenther@suse.de>
769 PR tree-optimization/82819
770 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
771 code generating pluses that are no-ops in the target precision.
773 2018-01-24 Richard Biener <rguenther@suse.de>
776 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
778 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
780 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
781 to merge probabilities.
782 * predict.c (probably_never_executed): Also mark as cold functions
783 with global 0 profile and guessed local profile.
784 * profile-count.c (profile_probability::combine_with_count): New
786 * profile-count.h (profile_probability::operator*,
787 profile_probability::operator*=, profile_probability::operator/,
788 profile_probability::operator/=): Reduce precision to adjusted
789 and set value to guessed on contradictory divisions.
790 (profile_probability::combine_with_freq): Remove.
791 (profile_probability::combine_wiht_count): Declare.
792 (profile_count::force_nonzero):: Set to adjusted.
793 (profile_count::probability_in):: Set quality to adjusted.
794 * tree-ssa-tail-merge.c (replace_block_by): Use
797 2018-01-23 Andrew Waterman <andrew@sifive.com>
798 Jim Wilson <jimw@sifive.com>
800 * config/riscv/riscv.c (riscv_stack_boundary): New.
801 (riscv_option_override): Set riscv_stack_boundary. Handle
802 riscv_preferred_stack_boundary_arg.
803 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
804 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
805 (STACK_BOUNDARY): Set to riscv_stack_boundary.
806 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
807 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
808 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
810 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
813 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
814 of struct ix86_frame.
815 (ix86_expand_epilogue): Likewise. Add a local variable for
816 the reg_save_offset field in struct ix86_frame.
818 2018-01-23 Bin Cheng <bin.cheng@arm.com>
820 PR tree-optimization/82604
821 * tree-loop-distribution.c (enum partition_kind): New enum item
822 PKIND_PARTIAL_MEMSET.
823 (partition_builtin_p): Support above new enum item.
824 (generate_code_for_partition): Ditto.
825 (compute_access_range): Differentiate cases that equality can be
826 proven at all loops, the innermost loops or no loops.
827 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
828 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
829 (finalize_partitions, distribute_loop): Don't fuse partition of
830 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
831 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
834 2018-01-23 Martin Liska <mliska@suse.cz>
836 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
837 order to ignore the predictor.
838 (PRED_POLYMORPHIC_CALL): Likewise.
839 (PRED_RECURSIVE_CALL): Likewise.
841 2018-01-23 Martin Liska <mliska@suse.cz>
843 * tree-profile.c (tree_profiling): Print function header to
844 aware reader which function we are working on.
845 * value-prof.c (gimple_find_values_to_profile): Do not print
846 not interesting value histograms.
848 2018-01-23 Martin Liska <mliska@suse.cz>
850 * profile-count.h (enum profile_quality): Add
851 profile_uninitialized as the first value. Do not number values
852 as they are zero based.
853 (profile_count::verify): Update sanity check.
854 (profile_probability::verify): Likewise.
856 2018-01-23 Nathan Sidwell <nathan@acm.org>
858 * doc/invoke.texi (ffor-scope): Deprecate.
860 2018-01-23 David Malcolm <dmalcolm@redhat.com>
862 PR tree-optimization/83510
863 * domwalk.c (set_all_edges_as_executable): New function.
864 (dom_walker::dom_walker): Convert bool param
865 "skip_unreachable_blocks" to enum reachability. Move setup of
866 edge flags to set_all_edges_as_executable and only do it when
867 reachability is REACHABLE_BLOCKS.
868 * domwalk.h (enum dom_walker::reachability): New enum.
869 (dom_walker::dom_walker): Convert bool param
870 "skip_unreachable_blocks" to enum reachability.
871 (set_all_edges_as_executable): New decl.
872 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
873 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
875 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
876 but converting true to REACHABLE_BLOCKS.
877 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
879 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
880 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
881 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
883 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
884 if check_all_array_refs will be called.
886 2018-01-23 David Malcolm <dmalcolm@redhat.com>
888 * tree.c (selftest::test_location_wrappers): Add more test
891 2018-01-23 David Malcolm <dmalcolm@redhat.com>
893 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
894 (selftest::test_bit_in_range): Likewise.
896 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
899 * doc/sourcebuild.texi (vect_float): Say that the selector
900 only describes the situation when -funsafe-math-optimizations is on.
901 (vect_float_strict): Document.
903 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
905 PR tree-optimization/83965
906 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
907 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
908 instead of checking only for a reduction.
909 (vect_recog_widen_sum_pattern): Likewise.
911 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
913 * predict.c (probably_never_executed): Only use precise profile info.
914 (compute_function_frequency): Skip after inlining hack since we now
915 have quality checking.
917 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
919 * profile-count.h (profile_probability::very_unlikely,
920 profile_probability::unlikely, profile_probability::even): Set
921 precision to guessed.
923 2018-01-23 Richard Biener <rguenther@suse.de>
925 PR tree-optimization/83963
926 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
927 Properly terminate dominator walk when crossing the exit edge not
928 when visiting its source block.
930 2018-01-23 Jakub Jelinek <jakub@redhat.com>
933 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
934 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
936 2018-01-22 Jakub Jelinek <jakub@redhat.com>
938 PR tree-optimization/83957
939 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
940 semicolon after for body surrounded by braces.
942 PR tree-optimization/83081
943 * profile-count.h (profile_probability::split): New method.
944 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
945 Use profile_probability::split.
946 (do_compare_rtx_and_jump): Fix adjustment of probabilities
947 when splitting a single conditional jump into 2.
949 2018-01-22 David Malcolm <dmalcolm@redhat.com>
951 PR tree-optimization/69452
952 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
955 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
957 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
958 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
959 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
961 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
963 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
964 * config/rl78/rl78.md: New define_expand "movdi"
965 * config/rl78/rl78.c: New function definition rl78_split_movdi
967 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
970 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
972 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
973 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
974 128-bit to produce an UNSPEC move to get the double word with the
975 signbit and then a shift directly to do signbit.
976 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
977 implementation with a new version that just does either a direct
978 move or a regular move. Move memory interface to separate insns.
979 Move insns so they are next to the expander.
980 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
981 with signbit move. Split big and little endian case.
982 (signbit<mode>2_dm_mem_le): Likewise.
983 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
984 (signbit<mode>2_dm2): Likewise.
986 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
988 * config/rl78/rl78.md: New define_expand "anddi3".
990 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
992 * config/rl78/rl78.md: New define_expand "umindi3".
994 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
996 * config/rl78/rl78.md: New define_expand "smindi3".
998 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1000 * config/rl78/rl78.md: New define_expand "smaxdi3".
1002 2018-01-22 Carl Love <cel@us.ibm.com>
1004 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1005 LVX_V1TI): Add macro expansion.
1006 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1007 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1008 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1009 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1010 Change check to determine if the instruction is a byte reversing
1011 entry. Fix typo in comment.
1012 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1013 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1014 Add def_builtin calls for new builtins.
1015 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1016 Add define_insn expansion.
1018 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1020 * config/rl78/rl78.md: New define_expand "umaxdi3".
1022 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1024 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1025 for non-QImode registers
1027 2018-01-22 Richard Biener <rguenther@suse.de>
1029 PR tree-optimization/83963
1030 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1031 including the loop exit block.
1032 (scop_detection::merge_sese): Likewise.
1033 (scop_detection::add_scop): Do it here instead.
1035 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1037 * doc/sourcebuild.texi (arm_softfloat): Document.
1039 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1042 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1043 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1044 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1046 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1047 David Edelsohn <dje.gcc@gmail.com>
1050 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1051 Change "crset eq" to "crset 2".
1052 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1053 (*call_indirect_aix<mode>_nospec): Likewise.
1054 (*call_value_indirect_aix<mode>_nospec): Likewise.
1055 (*call_indirect_elfv2<mode>_nospec): Likewise.
1056 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1057 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1058 change assembly output from . to $.
1059 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1060 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1061 (*tablejump<mode>_internal1_nospec): Likewise.
1063 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1066 * config/sh/sh_optimize_sett_clrt.cc:
1067 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1069 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1071 PR tree-optimization/83940
1072 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1073 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1074 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1075 use it to pass back the definition type.
1076 (vect_check_store_rhs): Likewise rhs_dt_out.
1077 (vect_build_gather_load_calls): Add a mask_dt argument and use
1078 it instead of a call to vect_is_simple_use.
1079 (vectorizable_store): Update calls to vect_check_load_store_mask
1080 and vect_check_store_rhs. Use the dt returned by the latter instead
1081 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1082 instead of calls to vect_is_simple_use. Pass the scalar rather
1083 than the vector operand to vect_is_simple_use when handling
1084 second and subsequent copies of an rhs value.
1085 (vectorizable_load): Update calls to vect_check_load_store_mask
1086 and vect_build_gather_load_calls. Use the cached mask_dt and
1087 gs_info.offset_dt instead of calls to vect_is_simple_use.
1089 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1092 * tree-emutls.c: Include gimplify.h.
1093 (lower_emutls_2): New function.
1094 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1095 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1096 it before further processing.
1099 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1100 UINTVAL (trueop1) instead of INTVAL (op1).
1102 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1106 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1107 INCOMING_FRAME_SP_OFFSET if not defined.
1108 (scan_trace): Add ENTRY argument. If true and
1109 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1110 emit a note to adjust the CFA offset.
1111 (create_cfi_notes): Adjust scan_trace callers.
1112 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1113 INCOMING_FRAME_SP_OFFSET in the CIE.
1114 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1115 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1117 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1118 * doc/tm.texi: Regenerated.
1120 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1122 PR rtl-optimization/83147
1123 * lra-constraints.c (remove_inheritance_pseudos): Use
1124 lra_substitute_pseudo_within_insn.
1126 2018-01-19 Tom de Vries <tom@codesourcery.com>
1127 Cesar Philippidis <cesar@codesourcery.com>
1130 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1132 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1135 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1136 spaces for function labels.
1138 2018-01-19 Martin Liska <mliska@suse.cz>
1140 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1141 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1142 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1143 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1144 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1145 (PRED_CONST_RETURN): Change from 69 to 65.
1146 (PRED_NULL_RETURN): Change from 91 to 71.
1147 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1148 (PRED_LOOP_GUARD): Change from 66 to 73.
1150 2018-01-19 Martin Liska <mliska@suse.cz>
1152 * predict.c (predict_insn_def): Add new assert.
1153 (struct branch_predictor): Change type to signed integer.
1154 (test_prediction_value_range): Amend test to cover
1156 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1157 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1158 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1159 (PRED_LOOP_IV_COMPARE): Likewise.
1160 * predict.h (PROB_UNINITIALIZED): Define new constant.
1162 2018-01-19 Martin Liska <mliska@suse.cz>
1164 * predict.c (dump_prediction): Add new format for
1165 analyze_brprob.py script which is enabled with -details
1167 * profile-count.h (precise_p): New function.
1169 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1171 PR tree-optimization/83922
1172 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1173 there are no statements that need masking.
1174 (vect_active_double_reduction_p): New function.
1175 (vect_analyze_loop_operations): Use it when handling phis that
1176 are not in the loop header.
1178 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1180 PR tree-optimization/83914
1181 * tree-vect-loop.c (vectorizable_induction): Don't convert
1182 init_expr or apply the peeling adjustment for inductions
1183 that are nested within the vectorized loop.
1185 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1187 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1190 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1194 * function.h (gimplify_parameters): Add gimple_seq * argument.
1195 * function.c: Include gimple.h and options.h.
1196 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1197 for the added local temporaries if needed.
1198 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1199 if there are any parameter cleanups, wrap whole body into a
1200 try/finally with the cleanups.
1202 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1205 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1206 Use GET_MODE_CLASS for scalar floating point.
1208 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1212 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1213 Fix call of call_cgraph_insertion_hooks.
1215 2018-01-18 Martin Sebor <msebor@redhat.com>
1217 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1219 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1222 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1225 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1228 * common.opt: (-ffile-prefix-map): New option.
1229 * opts.c (common_handle_option): Defer it.
1230 * opts-global.c (handle_common_deferred_options): Handle it.
1231 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1232 * file-prefix-map.h: New file.
1233 (remap_debug_filename, add_debug_prefix_map): ...here.
1234 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1235 * final.c (debug_prefix_map, add_debug_prefix_map
1236 remap_debug_filename): Move to...
1237 * file-prefix-map.c: New file.
1238 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1239 generalize, get rid of alloca(), use strrchr() instead of strchr().
1240 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1241 Implement in terms of add_prefix_map().
1242 (remap_macro_filename, remap_debug_filename): Implement in term of
1244 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1245 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1246 * dbxout.c: Include file-prefix-map.h.
1247 * varasm.c: Likewise.
1248 * vmsdbgout.c: Likewise.
1249 * xcoffout.c: Likewise.
1250 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1251 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1252 * doc/invoke.texi (-ffile-prefix-map): Document.
1253 (-fdebug-prefix-map): Update description.
1255 2018-01-18 Martin Liska <mliska@suse.cz>
1257 * config/i386/i386.c (indirect_thunk_name): Document that also
1259 (output_indirect_thunk): Document why both instructions
1260 (pause and lfence) are generated.
1262 2018-01-18 Richard Biener <rguenther@suse.de>
1264 PR tree-optimization/83887
1265 * graphite-scop-detection.c
1266 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1267 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1268 (scop_detection::merge_sese): Re-implement with a flood-fill
1269 algorithm that properly finds a SESE region if it exists.
1271 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1274 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1275 pointer_diff optimizations use view_convert instead of convert.
1277 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1279 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1280 Generate different code for -mno-speculate-indirect-jumps.
1281 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1282 (*call_indirect_aix<mode>): Disable for
1283 -mno-speculate-indirect-jumps.
1284 (*call_indirect_aix<mode>_nospec): New define_insn.
1285 (*call_value_indirect_aix<mode>): Disable for
1286 -mno-speculate-indirect-jumps.
1287 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1288 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1289 -mno-speculate-indirect-jumps.
1290 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1292 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1294 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1295 long double type, set the flags for noting the default long double
1296 type, even if we don't pass or return a long double type.
1298 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1301 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1304 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1306 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1307 support for merge[hl].
1308 (fold_mergehl_helper): New helper function.
1309 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1310 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1311 (altivec_vmrglw_direct): Add xxmrglw insn.
1313 2018-01-17 Andrew Waterman <andrew@sifive.com>
1315 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1316 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1318 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1321 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1322 call the lto_location_cache before reading the
1323 DECL_SOURCE_LOCATION of the types.
1325 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1326 Richard Sandiford <richard.sandiford@linaro.org>
1328 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1329 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1330 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1331 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1332 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1334 * config/aarch64/constraints.md (aarch64_movti_operand):
1336 * config/aarch64/predicates.md (Uti): Add new constraint.
1338 2018-01-17 Carl Love <cel@us.ibm.com>
1339 * config/rs6000/vsx.md (define_expand xl_len_r,
1340 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1341 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1343 (define_expand, define_insn): Move the shift left from the
1344 define_insn to the define_expand for lxvl and stxvl instructions.
1345 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1346 and XL_LEN_R definitions to PURE.
1348 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1350 * config/i386/i386.c (indirect_thunk_name): Declare regno
1351 as unsigned int. Compare regno with INVALID_REGNUM.
1352 (output_indirect_thunk): Ditto.
1353 (output_indirect_thunk_function): Ditto.
1354 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1355 in the call to output_indirect_thunk_function.
1357 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1360 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1361 rather than the size of inner_type to determine the stack slot size
1362 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1364 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1367 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1370 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1372 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1373 endian Linux systems to optionally enable multilibs for selecting
1374 the long double type if the user configured an explicit type.
1375 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1376 have no long double multilibs if not defined.
1377 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1378 warn if the user used -mabi={ieee,ibm}longdouble and we built
1379 multilibs for long double.
1380 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1381 appropriate multilib option.
1382 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1384 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1385 for building long double multilibs.
1386 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1388 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1390 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1393 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1395 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1398 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1401 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1404 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1406 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1407 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1409 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1411 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1412 different rtl trees depending on TARGET_64BIT.
1413 (rs6000_gen_lvx): Likewise.
1415 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1417 * config/visium/visium.md (nop): Tweak comment.
1418 (hazard_nop): Likewise.
1420 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1422 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1423 -mspeculate-indirect-jumps.
1424 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1425 for -mno-speculate-indirect-jumps.
1426 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1427 (*call_value_indirect_elfv2<mode>): Disable for
1428 -mno-speculate-indirect-jumps.
1429 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1430 (indirect_jump): Emit different RTL for
1431 -mno-speculate-indirect-jumps.
1432 (*indirect_jump<mode>): Disable for
1433 -mno-speculate-indirect-jumps.
1434 (*indirect_jump<mode>_nospec): New define_insn.
1435 (tablejump): Emit different RTL for
1436 -mno-speculate-indirect-jumps.
1437 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1438 (tablejumpsi_nospec): New define_expand.
1439 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1440 (tablejumpdi_nospec): New define_expand.
1441 (*tablejump<mode>_internal1): Disable for
1442 -mno-speculate-indirect-jumps.
1443 (*tablejump<mode>_internal1_nospec): New define_insn.
1444 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1447 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1449 * caller-save.c (insert_save): Drop unnecessary parameter. All
1452 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1453 Richard Biener <rguenth@suse.de>
1456 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1457 return early, inline manually is_gimple_sizepos. Make sure if we
1458 call gimplify_expr we don't end up with a gimple constant.
1459 * tree.c (variably_modified_type_p): Don't return true for
1460 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1461 * gimplify.h (is_gimple_sizepos): Remove.
1463 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1465 PR tree-optimization/83857
1466 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1467 vectorizable_live_operation for pure SLP statements.
1468 (vectorizable_live_operation): Handle PHIs.
1470 2018-01-16 Richard Biener <rguenther@suse.de>
1472 PR tree-optimization/83867
1473 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1474 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1476 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1479 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1480 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1481 If off is not INTEGER_CST, issue a may not be aligned warning
1482 rather than isn't aligned. Use isn%'t rather than isn't.
1483 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1485 <case MULT_EXPR>: Improve the case when bottom and one of the
1486 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1487 operand, in that case check if the other operand is multiple of
1488 bottom divided by the INTEGER_CST operand.
1490 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1493 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1494 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1495 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1496 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1497 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1498 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1499 (pa_function_arg_size): New function.
1501 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1503 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1504 in a separate statement.
1506 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1508 PR tree-optimization/83847
1509 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1510 group gathers and scatters.
1512 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1514 PR rtl-optimization/86620
1515 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1517 PR rtl-optimization/83213
1518 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1519 to last if both are JUMP_INSNs.
1521 PR tree-optimization/83843
1522 * gimple-ssa-store-merging.c
1523 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1524 store_immediate_info for bswap/nop orig_stores.
1526 2018-01-15 Andrew Waterman <andrew@sifive.com>
1528 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1530 <UDIV>: Increase cost if !TARGET_DIV.
1532 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1534 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1535 (define_attr "cr_logical_3op"): New.
1536 (cceq_ior_compare): Adjust.
1537 (cceq_ior_compare_complement): Adjust.
1538 (*cceq_rev_compare): Adjust.
1539 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1540 (is_cracked_insn): Adjust.
1541 (insn_must_be_first_in_group): Adjust.
1542 * config/rs6000/40x.md: Adjust.
1543 * config/rs6000/440.md: Adjust.
1544 * config/rs6000/476.md: Adjust.
1545 * config/rs6000/601.md: Adjust.
1546 * config/rs6000/603.md: Adjust.
1547 * config/rs6000/6xx.md: Adjust.
1548 * config/rs6000/7450.md: Adjust.
1549 * config/rs6000/7xx.md: Adjust.
1550 * config/rs6000/8540.md: Adjust.
1551 * config/rs6000/cell.md: Adjust.
1552 * config/rs6000/e300c2c3.md: Adjust.
1553 * config/rs6000/e500mc.md: Adjust.
1554 * config/rs6000/e500mc64.md: Adjust.
1555 * config/rs6000/e5500.md: Adjust.
1556 * config/rs6000/e6500.md: Adjust.
1557 * config/rs6000/mpc.md: Adjust.
1558 * config/rs6000/power4.md: Adjust.
1559 * config/rs6000/power5.md: Adjust.
1560 * config/rs6000/power6.md: Adjust.
1561 * config/rs6000/power7.md: Adjust.
1562 * config/rs6000/power8.md: Adjust.
1563 * config/rs6000/power9.md: Adjust.
1564 * config/rs6000/rs64.md: Adjust.
1565 * config/rs6000/titan.md: Adjust.
1567 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1569 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1570 ix86_indirect_branch_register logic.
1572 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1574 * config/i386/constraints.md (Bs): Update
1575 ix86_indirect_branch_register check. Don't check
1576 ix86_indirect_branch_register with GOT_memory_operand.
1578 * config/i386/predicates.md (GOT_memory_operand): Don't check
1579 ix86_indirect_branch_register here.
1580 (GOT32_symbol_operand): Likewise.
1582 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1584 * config/i386/predicates.md (constant_call_address_operand):
1585 Rewrite ix86_indirect_branch_register logic.
1586 (sibcall_insn_operand): Likewise.
1588 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1590 * config/i386/constraints.md (Bs): Replace
1591 ix86_indirect_branch_thunk_register with
1592 ix86_indirect_branch_register.
1594 * config/i386/i386.md (indirect_jump): Likewise.
1595 (tablejump): Likewise.
1596 (*sibcall_memory): Likewise.
1597 (*sibcall_value_memory): Likewise.
1598 Peepholes of indirect call and jump via memory: Likewise.
1599 * config/i386/i386.opt: Likewise.
1600 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1601 (GOT_memory_operand): Likewise.
1602 (call_insn_operand): Likewise.
1603 (sibcall_insn_operand): Likewise.
1604 (GOT32_symbol_operand): Likewise.
1606 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1609 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1610 type rather than type addr's type points to.
1611 (expand_omp_atomic_mutex): Likewise.
1612 (expand_omp_atomic): Likewise.
1614 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1617 * config/i386/i386.c (output_indirect_thunk_function): Use
1618 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1619 for __x86_return_thunk.
1621 2018-01-15 Richard Biener <rguenther@suse.de>
1624 * expmed.c (extract_bit_field_1): Fix typo.
1626 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1629 * config/arm/iterators.md (VF): New mode iterator.
1630 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1631 Remove integer-related logic from pattern.
1632 (neon_vabd<mode>_3): Likewise.
1634 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1637 * common.opt (fstrict-overflow): No longer an alias.
1638 (fwrapv-pointer): New option.
1639 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1640 also for pointer types based on flag_wrapv_pointer.
1641 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1642 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1643 opts->x_flag_wrapv got set.
1644 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1645 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1646 POINTER_TYPE_OVERFLOW_UNDEFINED.
1647 * match.pd: Likewise in address comparison pattern.
1648 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1650 2018-01-15 Richard Biener <rguenther@suse.de>
1653 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1654 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1655 Reset type names to their identifier if their TYPE_DECL doesn't
1656 have linkage (and thus is used for ODR and devirt).
1657 (save_debug_info_for_decl): Remove.
1658 (save_debug_info_for_type): Likewise.
1659 (add_tree_to_fld_list): Adjust.
1660 * tree-pretty-print.c (dump_generic_node): Make dumping of
1661 type names more robust.
1663 2018-01-15 Richard Biener <rguenther@suse.de>
1665 * BASE-VER: Bump to 8.0.1.
1667 2018-01-14 Martin Sebor <msebor@redhat.com>
1670 * builtins.c (check_access): Avoid warning when the no-warning bit
1673 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1675 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1676 * ira-color (allocno_hard_regs_compare): Likewise.
1678 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1681 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1682 Use .pushsection/.popsection.
1684 2018-01-14 Martin Sebor <msebor@redhat.com>
1687 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1689 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1691 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1692 entry from extra_headers.
1693 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1694 extra_headers, make the list bitwise identical to the i?86-*-* one.
1696 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1698 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1699 -mcmodel=large with -mindirect-branch=thunk,
1700 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1701 -mfunction-return=thunk-extern.
1702 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1703 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1704 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1706 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1708 * config/i386/i386.c (print_reg): Print the name of the full
1709 integer register without '%'.
1710 (ix86_print_operand): Handle 'V'.
1711 * doc/extend.texi: Document 'V' modifier.
1713 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1715 * config/i386/constraints.md (Bs): Disallow memory operand for
1716 -mindirect-branch-register.
1718 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1719 (GOT_memory_operand): Likewise.
1720 (call_insn_operand): Likewise.
1721 (sibcall_insn_operand): Likewise.
1722 (GOT32_symbol_operand): Likewise.
1723 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1724 for -mindirect-branch-register.
1725 (tablejump): Likewise.
1726 (*sibcall_memory): Likewise.
1727 (*sibcall_value_memory): Likewise.
1728 Disallow peepholes of indirect call and jump via memory for
1729 -mindirect-branch-register.
1730 (*call_pop): Replace m with Bw.
1731 (*call_value_pop): Likewise.
1732 (*sibcall_pop_memory): Replace m with Bs.
1733 * config/i386/i386.opt (mindirect-branch-register): New option.
1734 * doc/invoke.texi: Document -mindirect-branch-register option.
1736 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1738 * config/i386/i386-protos.h (ix86_output_function_return): New.
1739 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1740 set function_return_type.
1741 (indirect_thunk_name): Add ret_p to indicate thunk for function
1743 (output_indirect_thunk_function): Pass false to
1744 indirect_thunk_name.
1745 (ix86_output_indirect_branch_via_reg): Likewise.
1746 (ix86_output_indirect_branch_via_push): Likewise.
1747 (output_indirect_thunk_function): Create alias for function
1748 return thunk if regno < 0.
1749 (ix86_output_function_return): New function.
1750 (ix86_handle_fndecl_attribute): Handle function_return.
1751 (ix86_attribute_table): Add function_return.
1752 * config/i386/i386.h (machine_function): Add
1753 function_return_type.
1754 * config/i386/i386.md (simple_return_internal): Use
1755 ix86_output_function_return.
1756 (simple_return_internal_long): Likewise.
1757 * config/i386/i386.opt (mfunction-return=): New option.
1758 (indirect_branch): Mention -mfunction-return=.
1759 * doc/extend.texi: Document function_return function attribute.
1760 * doc/invoke.texi: Document -mfunction-return= option.
1762 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1764 * config/i386/i386-opts.h (indirect_branch): New.
1765 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1766 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1767 with local indirect jump when converting indirect call and jump.
1768 (ix86_set_indirect_branch_type): New.
1769 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1770 (indirectlabelno): New.
1771 (indirect_thunk_needed): Likewise.
1772 (indirect_thunk_bnd_needed): Likewise.
1773 (indirect_thunks_used): Likewise.
1774 (indirect_thunks_bnd_used): Likewise.
1775 (INDIRECT_LABEL): Likewise.
1776 (indirect_thunk_name): Likewise.
1777 (output_indirect_thunk): Likewise.
1778 (output_indirect_thunk_function): Likewise.
1779 (ix86_output_indirect_branch_via_reg): Likewise.
1780 (ix86_output_indirect_branch_via_push): Likewise.
1781 (ix86_output_indirect_branch): Likewise.
1782 (ix86_output_indirect_jmp): Likewise.
1783 (ix86_code_end): Call output_indirect_thunk_function if needed.
1784 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1786 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1787 (ix86_attribute_table): Add indirect_branch.
1788 * config/i386/i386.h (machine_function): Add indirect_branch_type
1789 and has_local_indirect_jump.
1790 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1792 (tablejump): Likewise.
1793 (*indirect_jump): Use ix86_output_indirect_jmp.
1794 (*tablejump_1): Likewise.
1795 (simple_return_indirect_internal): Likewise.
1796 * config/i386/i386.opt (mindirect-branch=): New option.
1797 (indirect_branch): New.
1800 (thunk-inline): Likewise.
1801 (thunk-extern): Likewise.
1802 * doc/extend.texi: Document indirect_branch function attribute.
1803 * doc/invoke.texi: Document -mindirect-branch= option.
1805 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1808 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1810 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1812 * ipa-inline.c (want_inline_small_function_p): Return false if
1813 inlining has already failed with CIF_FINAL_ERROR.
1814 (update_caller_keys): Call want_inline_small_function_p before
1816 (update_callee_keys): Likewise.
1818 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1820 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1822 (rs6000_quadword_masked_address_p): Likewise.
1823 (quad_aligned_load_p): Likewise.
1824 (quad_aligned_store_p): Likewise.
1825 (const_load_sequence_p): Add comment to describe the outer-most loop.
1826 (mimic_memory_attributes_and_flags): New function.
1827 (rs6000_gen_stvx): Likewise.
1828 (replace_swapped_aligned_store): Likewise.
1829 (rs6000_gen_lvx): Likewise.
1830 (replace_swapped_aligned_load): Likewise.
1831 (replace_swapped_load_constant): Capitalize argument name in
1832 comment describing this function.
1833 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1834 and stores that access quad-word aligned addresses and replace
1835 with stvx or lvx instructions when appropriate.
1836 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1837 New function prototype.
1838 (rs6000_quadword_masked_address_p): Likewise.
1839 (rs6000_gen_lvx): Likewise.
1840 (rs6000_gen_stvx): Likewise.
1841 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1842 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1843 when memory address is aligned.
1844 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1845 this split to select lvx instruction when memory address is aligned.
1846 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1847 instruction when memory address is aligned.
1848 (*vsx_le_perm_load_v16qi): Likewise.
1849 (four unnamed splitters): Modify to select the stvx instruction
1850 when memory is aligned.
1852 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1854 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1855 which appears in the queue multiple times.
1857 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1858 Alan Hayward <alan.hayward@arm.com>
1859 David Sherwood <david.sherwood@arm.com>
1861 * tree-vectorizer.h (vec_lower_bound): New structure.
1862 (_loop_vec_info): Add check_nonzero and lower_bounds.
1863 (LOOP_VINFO_CHECK_NONZERO): New macro.
1864 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1865 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1866 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1867 fields. Make seg_len the distance travelled, not including the
1869 (dr_direction_indicator): Declare.
1870 (dr_zero_step_indicator): Likewise.
1871 (dr_known_forward_stride_p): Likewise.
1872 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1874 (runtime_alias_check_p): Allow runtime alias checks with
1876 (operator ==): Compare access_size and align.
1877 (prune_runtime_alias_test_list): Rework for new distinction between
1878 the access_size and seg_len.
1879 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1881 (get_segment_min_max): New function.
1882 (create_intersect_range_checks): Use it.
1883 (dr_step_indicator): New function.
1884 (dr_direction_indicator): Likewise.
1885 (dr_zero_step_indicator): Likewise.
1886 (dr_known_forward_stride_p): Likewise.
1887 * tree-loop-distribution.c (data_ref_segment_size): Return
1888 DR_STEP * (niters - 1).
1889 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1891 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1892 (vect_preserves_scalar_order_p): New function, split out from...
1893 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1894 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1895 (vect_vfa_access_size): New function.
1896 (vect_vfa_align): Likewise.
1897 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1898 (dump_lower_bound): New function.
1899 (vect_check_lower_bound): Likewise.
1900 (vect_small_gap_p): Likewise.
1901 (vectorizable_with_step_bound_p): Likewise.
1902 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1903 depencies if the vectorization factor is 1. Convert the checks
1904 for nonzero steps into checks on the bounds of DR_STEP. Try using
1905 a bunds check for variable steps if the minimum required step is
1906 relatively small. Update calls to the dr_with_seg_len
1907 constructor and to vect_compile_time_alias.
1908 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1910 (vect_loop_versioning): Call it.
1911 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1913 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1915 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1916 Alan Hayward <alan.hayward@arm.com>
1917 David Sherwood <david.sherwood@arm.com>
1919 * doc/sourcebuild.texi (vect_scatter_store): Document.
1920 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1922 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1924 * genopinit.c (main): Add supports_vec_scatter_store and
1925 supports_vec_scatter_store_cached to target_optabs.
1926 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1927 IFN_MASK_SCATTER_STORE.
1928 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1930 * internal-fn.h (internal_store_fn_p): Declare.
1931 (internal_fn_stored_value_index): Likewise.
1932 * internal-fn.c (scatter_store_direct): New macro.
1933 (expand_scatter_store_optab_fn): New function.
1934 (direct_scatter_store_optab_supported_p): New macro.
1935 (internal_store_fn_p): New function.
1936 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1937 IFN_MASK_SCATTER_STORE.
1938 (internal_fn_mask_index): Likewise.
1939 (internal_fn_stored_value_index): New function.
1940 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1942 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1943 * optabs-query.c (supports_vec_scatter_store_p): New function.
1944 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1945 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1946 true for scatter stores.
1947 (vect_gather_scatter_fn_p): Handle scatter stores too.
1948 (vect_check_gather_scatter): Consider using scatter stores if
1949 supports_vec_scatter_store_p.
1950 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1952 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1953 internal_fn_stored_value_index.
1954 (check_load_store_masking): Handle scatter stores too.
1955 (vect_get_store_rhs): Make public.
1956 (vectorizable_call): Use internal_store_fn_p.
1957 (vectorizable_store): Handle scatter store internal functions.
1958 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1959 when deciding whether the end of the group has been reached.
1960 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1961 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1962 (mask_scatter_store<mode>): New insns.
1964 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1965 Alan Hayward <alan.hayward@arm.com>
1966 David Sherwood <david.sherwood@arm.com>
1968 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1969 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1970 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1972 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1973 Use vect_truncate_gather_scatter_offset if we can't treat the
1974 operation as a normal gather load or scatter store.
1975 (get_group_load_store_type): Take the gather_scatter_info
1976 as argument. Try using a gather load or scatter store for
1977 single-element groups.
1978 (get_load_store_type): Update calls to get_group_load_store_type
1979 and vect_use_strided_gather_scatters_p.
1981 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1982 Alan Hayward <alan.hayward@arm.com>
1983 David Sherwood <david.sherwood@arm.com>
1985 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1986 optional tree argument.
1987 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1989 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1990 but continue to use the current value as a fallback.
1991 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1992 to compare the updates.
1993 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1994 (get_load_store_type): Use it when handling a strided access.
1995 (vect_get_strided_load_store_ops): New function.
1996 (vect_get_data_ptr_increment): Likewise.
1997 (vectorizable_load): Handle strided gather loads. Always pass
1998 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2000 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2001 Alan Hayward <alan.hayward@arm.com>
2002 David Sherwood <david.sherwood@arm.com>
2004 * doc/md.texi (gather_load@var{m}): Document.
2005 (mask_gather_load@var{m}): Likewise.
2006 * genopinit.c (main): Add supports_vec_gather_load and
2007 supports_vec_gather_load_cached to target_optabs.
2008 * optabs-tree.c (init_tree_optimization_optabs): Use
2009 ggc_cleared_alloc to allocate target_optabs.
2010 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2011 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2013 * internal-fn.h (internal_load_fn_p): Declare.
2014 (internal_gather_scatter_fn_p): Likewise.
2015 (internal_fn_mask_index): Likewise.
2016 (internal_gather_scatter_fn_supported_p): Likewise.
2017 * internal-fn.c (gather_load_direct): New macro.
2018 (expand_gather_load_optab_fn): New function.
2019 (direct_gather_load_optab_supported_p): New macro.
2020 (direct_internal_fn_optab): New function.
2021 (internal_load_fn_p): Likewise.
2022 (internal_gather_scatter_fn_p): Likewise.
2023 (internal_fn_mask_index): Likewise.
2024 (internal_gather_scatter_fn_supported_p): Likewise.
2025 * optabs-query.c (supports_at_least_one_mode_p): New function.
2026 (supports_vec_gather_load_p): Likewise.
2027 * optabs-query.h (supports_vec_gather_load_p): Declare.
2028 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2029 and memory_type field.
2030 (NUM_PATTERNS): Bump to 15.
2031 * tree-vect-data-refs.c: Include internal-fn.h.
2032 (vect_gather_scatter_fn_p): New function.
2033 (vect_describe_gather_scatter_call): Likewise.
2034 (vect_check_gather_scatter): Try using internal functions for
2035 gather loads. Recognize existing calls to a gather load function.
2036 (vect_analyze_data_refs): Consider using gather loads if
2037 supports_vec_gather_load_p.
2038 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2039 (vect_get_gather_scatter_offset_type): Likewise.
2040 (vect_convert_mask_for_vectype): Likewise.
2041 (vect_add_conversion_to_patterm): Likewise.
2042 (vect_try_gather_scatter_pattern): Likewise.
2043 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2044 (vect_vect_recog_func_ptrs): Add it.
2045 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2046 internal_fn_mask_index and internal_gather_scatter_fn_p.
2047 (check_load_store_masking): Take the gather_scatter_info as an
2048 argument and handle gather loads.
2049 (vect_get_gather_scatter_ops): New function.
2050 (vectorizable_call): Check internal_load_fn_p.
2051 (vectorizable_load): Likewise. Handle gather load internal
2053 (vectorizable_store): Update call to check_load_store_masking.
2054 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2055 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2056 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2057 (aarch64_gather_scale_operand_d): New predicates.
2058 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2059 (mask_gather_load<mode>): New insns.
2061 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2062 Alan Hayward <alan.hayward@arm.com>
2063 David Sherwood <david.sherwood@arm.com>
2065 * optabs.def (fold_left_plus_optab): New optab.
2066 * doc/md.texi (fold_left_plus_@var{m}): Document.
2067 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2068 * internal-fn.c (fold_left_direct): Define.
2069 (expand_fold_left_optab_fn): Likewise.
2070 (direct_fold_left_optab_supported_p): Likewise.
2071 * fold-const-call.c (fold_const_fold_left): New function.
2072 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2073 * tree-parloops.c (valid_reduction_p): New function.
2074 (gather_scalar_reductions): Use it.
2075 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2076 (vect_finish_replace_stmt): Declare.
2077 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2078 (needs_fold_left_reduction_p): New function, split out from...
2079 (vect_is_simple_reduction): ...here. Accept reductions that
2080 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2081 (vect_force_simple_reduction): Also store the reduction type in
2082 the assignment's STMT_VINFO_REDUC_TYPE.
2083 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2084 (merge_with_identity): New function.
2085 (vect_expand_fold_left): Likewise.
2086 (vectorize_fold_left_reduction): Likewise.
2087 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2088 scalar phi in place for it. Check for target support and reject
2089 cases that would reassociate the operation. Defer the transform
2090 phase to vectorize_fold_left_reduction.
2091 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2092 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2093 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2095 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2097 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2098 call to ifc_temp_var.
2100 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2101 Alan Hayward <alan.hayward@arm.com>
2102 David Sherwood <david.sherwood@arm.com>
2104 * target.def (legitimize_address_displacement): Take the original
2105 offset as a poly_int.
2106 * targhooks.h (default_legitimize_address_displacement): Update
2108 * targhooks.c (default_legitimize_address_displacement): Likewise.
2109 * doc/tm.texi: Regenerate.
2110 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2111 as an argument, moving assert of ad->disp == ad->disp_term to...
2112 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2113 Try calling targetm.legitimize_address_displacement before expanding
2114 the address rather than afterwards, and adjust for the new interface.
2115 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2116 Match the new hook interface. Handle SVE addresses.
2117 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2120 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2122 * Makefile.in (OBJS): Add early-remat.o.
2123 * target.def (select_early_remat_modes): New hook.
2124 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2125 * doc/tm.texi: Regenerate.
2126 * targhooks.h (default_select_early_remat_modes): Declare.
2127 * targhooks.c (default_select_early_remat_modes): New function.
2128 * timevar.def (TV_EARLY_REMAT): New timevar.
2129 * passes.def (pass_early_remat): New pass.
2130 * tree-pass.h (make_pass_early_remat): Declare.
2131 * early-remat.c: New file.
2132 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2134 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2136 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2137 Alan Hayward <alan.hayward@arm.com>
2138 David Sherwood <david.sherwood@arm.com>
2140 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2141 vfm1 with a bound_epilog parameter.
2142 (vect_do_peeling): Update calls accordingly, and move the prologue
2143 call earlier in the function. Treat the base bound_epilog as 0 for
2144 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2145 this base when peeling for gaps.
2146 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2147 with fully-masked loops.
2148 (vect_estimate_min_profitable_iters): Handle the single peeled
2149 iteration in that case.
2151 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2152 Alan Hayward <alan.hayward@arm.com>
2153 David Sherwood <david.sherwood@arm.com>
2155 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2156 single-element interleaving even if the size is not a power of 2.
2157 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2158 accesses for single-element interleaving if the group size is
2161 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2162 Alan Hayward <alan.hayward@arm.com>
2163 David Sherwood <david.sherwood@arm.com>
2165 * doc/md.texi (fold_extract_last_@var{m}): Document.
2166 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2167 * optabs.def (fold_extract_last_optab): New optab.
2168 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2169 * internal-fn.c (fold_extract_direct): New macro.
2170 (expand_fold_extract_optab_fn): Likewise.
2171 (direct_fold_extract_optab_supported_p): Likewise.
2172 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2173 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2174 EXTRACT_LAST_REDUCTION.
2175 (get_initial_def_for_reduction): Do not create an initial vector
2176 for EXTRACT_LAST_REDUCTION reductions.
2177 (vectorizable_reduction): Leave the scalar phi in place for
2178 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2179 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2180 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2181 transform phase to vectorizable_condition.
2182 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2184 (vect_finish_stmt_generation): ...here.
2185 (vect_finish_replace_stmt): New function.
2186 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2187 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2189 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2191 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2192 Alan Hayward <alan.hayward@arm.com>
2193 David Sherwood <david.sherwood@arm.com>
2195 * doc/md.texi (extract_last_@var{m}): Document.
2196 * optabs.def (extract_last_optab): New optab.
2197 * internal-fn.def (EXTRACT_LAST): New internal function.
2198 * internal-fn.c (cond_unary_direct): New macro.
2199 (expand_cond_unary_optab_fn): Likewise.
2200 (direct_cond_unary_optab_supported_p): Likewise.
2201 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2202 loops using EXTRACT_LAST.
2203 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2204 (extract_last_<mode>): ...this optab.
2205 (vec_extract<mode><Vel>): Update accordingly.
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 * target.def (empty_mask_is_expensive): New hook.
2212 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2213 * doc/tm.texi: Regenerate.
2214 * targhooks.h (default_empty_mask_is_expensive): Declare.
2215 * targhooks.c (default_empty_mask_is_expensive): New function.
2216 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2217 if the target says that empty masks are expensive.
2218 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2220 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2222 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2223 Alan Hayward <alan.hayward@arm.com>
2224 David Sherwood <david.sherwood@arm.com>
2226 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2227 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2228 (vect_use_loop_mask_for_alignment_p): New function.
2229 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2230 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2231 niters_skip argument. Make sure that the first niters_skip elements
2232 of the first iteration are inactive.
2233 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2234 Update call to vect_set_loop_masks_directly.
2235 (get_misalign_in_elems): New function, split out from...
2236 (vect_gen_prolog_loop_niters): ...here.
2237 (vect_update_init_of_dr): Take a code argument that specifies whether
2238 the adjustment should be added or subtracted.
2239 (vect_update_init_of_drs): Likewise.
2240 (vect_prepare_for_masked_peels): New function.
2241 (vect_do_peeling): Skip prologue peeling if we're using a mask
2242 instead. Update call to vect_update_inits_of_drs.
2243 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2245 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2246 alignment. Do not include the number of peeled iterations in
2247 the minimum threshold in that case.
2248 (vectorizable_induction): Adjust the start value down by
2249 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2250 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2251 Take the number of skipped iterations into account when calculating
2253 * tree-vect-stmts.c (vect_gen_while_not): New function.
2255 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2256 Alan Hayward <alan.hayward@arm.com>
2257 David Sherwood <david.sherwood@arm.com>
2259 * doc/sourcebuild.texi (vect_fully_masked): Document.
2260 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2262 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2264 (vect_analyze_loop_2): ...here. Don't check the vectorization
2265 factor against the number of loop iterations if the loop is
2268 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2269 Alan Hayward <alan.hayward@arm.com>
2270 David Sherwood <david.sherwood@arm.com>
2272 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2273 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2274 (dump_groups): Update accordingly.
2275 (iv_use::mem_type): New member variable.
2276 (address_p): New function.
2277 (record_use): Add a mem_type argument and initialize the new
2279 (record_group_use): Add a mem_type argument. Use address_p.
2280 Remove obsolete null checks of base_object. Update call to record_use.
2281 (find_interesting_uses_op): Update call to record_group_use.
2282 (find_interesting_uses_cond): Likewise.
2283 (find_interesting_uses_address): Likewise.
2284 (get_mem_type_for_internal_fn): New function.
2285 (find_address_like_use): Likewise.
2286 (find_interesting_uses_stmt): Try find_address_like_use before
2287 calling find_interesting_uses_op.
2288 (addr_offset_valid_p): Use the iv mem_type field as the type
2289 of the addressed memory.
2290 (add_autoinc_candidates): Likewise.
2291 (get_address_cost): Likewise.
2292 (split_small_address_groups_p): Use address_p.
2293 (split_address_groups): Likewise.
2294 (add_iv_candidate_for_use): Likewise.
2295 (autoinc_possible_for_pair): Likewise.
2296 (rewrite_groups): Likewise.
2297 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2298 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2299 (get_alias_ptr_type_for_ptr_address): New function.
2300 (rewrite_use_address): Rewrite address uses in calls that were
2301 identified by find_address_like_use.
2303 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2304 Alan Hayward <alan.hayward@arm.com>
2305 David Sherwood <david.sherwood@arm.com>
2307 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2309 * gimple-expr.h (is_gimple_addressable: Likewise.
2310 * gimple-expr.c (is_gimple_address): Likewise.
2311 * internal-fn.c (expand_call_mem_ref): New function.
2312 (expand_mask_load_optab_fn): Use it.
2313 (expand_mask_store_optab_fn): Likewise.
2315 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2316 Alan Hayward <alan.hayward@arm.com>
2317 David Sherwood <david.sherwood@arm.com>
2319 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2320 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2321 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2322 (cond_umax@var{mode}): Document.
2323 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2324 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2325 (cond_umin_optab, cond_umax_optab): New optabs.
2326 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2327 (COND_IOR, COND_XOR): New internal functions.
2328 * internal-fn.h (get_conditional_internal_fn): Declare.
2329 * internal-fn.c (cond_binary_direct): New macro.
2330 (expand_cond_binary_optab_fn): Likewise.
2331 (direct_cond_binary_optab_supported_p): Likewise.
2332 (get_conditional_internal_fn): New function.
2333 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2334 Cope with reduction statements that are vectorized as calls rather
2336 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2337 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2338 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2339 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2340 (UNSPEC_COND_EOR): New unspecs.
2341 (optab): Add mappings for them.
2342 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2343 (sve_int_op, sve_fp_op): New int attributes.
2345 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2346 Alan Hayward <alan.hayward@arm.com>
2347 David Sherwood <david.sherwood@arm.com>
2349 * optabs.def (while_ult_optab): New optab.
2350 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2351 * internal-fn.def (WHILE_ULT): New internal function.
2352 * internal-fn.h (direct_internal_fn_supported_p): New override
2353 that takes two types as argument.
2354 * internal-fn.c (while_direct): New macro.
2355 (expand_while_optab_fn): New function.
2356 (convert_optab_supported_p): Likewise.
2357 (direct_while_optab_supported_p): New macro.
2358 * wide-int.h (wi::udiv_ceil): New function.
2359 * tree-vectorizer.h (rgroup_masks): New structure.
2360 (vec_loop_masks): New typedef.
2361 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2363 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2364 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2365 (vect_max_vf): New function.
2366 (slpeel_make_loop_iterate_ntimes): Delete.
2367 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2368 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2369 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2370 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2371 internal-fn.h, stor-layout.h and optabs-query.h.
2372 (vect_set_loop_mask): New function.
2373 (add_preheader_seq): Likewise.
2374 (add_header_seq): Likewise.
2375 (interleave_supported_p): Likewise.
2376 (vect_maybe_permute_loop_masks): Likewise.
2377 (vect_set_loop_masks_directly): Likewise.
2378 (vect_set_loop_condition_masked): Likewise.
2379 (vect_set_loop_condition_unmasked): New function, split out from
2380 slpeel_make_loop_iterate_ntimes.
2381 (slpeel_make_loop_iterate_ntimes): Rename to..
2382 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2383 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2384 (vect_do_peeling): Update call accordingly.
2385 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2387 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2388 mask_compare_type, can_fully_mask_p and fully_masked_p.
2389 (release_vec_loop_masks): New function.
2390 (_loop_vec_info): Use it to free the loop masks.
2391 (can_produce_all_loop_masks_p): New function.
2392 (vect_get_max_nscalars_per_iter): Likewise.
2393 (vect_verify_full_masking): Likewise.
2394 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2395 retries, and free the mask rgroups before retrying. Check loop-wide
2396 reasons for disallowing fully-masked loops. Make the final decision
2397 about whether use a fully-masked loop or not.
2398 (vect_estimate_min_profitable_iters): Do not assume that peeling
2399 for the number of iterations will be needed for fully-masked loops.
2400 (vectorizable_reduction): Disable fully-masked loops.
2401 (vectorizable_live_operation): Likewise.
2402 (vect_halve_mask_nunits): New function.
2403 (vect_double_mask_nunits): Likewise.
2404 (vect_record_loop_mask): Likewise.
2405 (vect_get_loop_mask): Likewise.
2406 (vect_transform_loop): Handle the case in which the final loop
2407 iteration might handle a partial vector. Call vect_set_loop_condition
2408 instead of slpeel_make_loop_iterate_ntimes.
2409 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2410 (check_load_store_masking): New function.
2411 (prepare_load_store_mask): Likewise.
2412 (vectorizable_store): Handle fully-masked loops.
2413 (vectorizable_load): Likewise.
2414 (supportable_widening_operation): Use vect_halve_mask_nunits for
2416 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2417 (vect_gen_while): New function.
2418 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2419 (aarch64_uqdec<mode>): New insn.
2421 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2422 Alan Hayward <alan.hayward@arm.com>
2423 David Sherwood <david.sherwood@arm.com>
2425 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2426 (reduc_xor_scal_optab): New optabs.
2427 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2428 (reduc_xor_scal_@var{m}): Document.
2429 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2430 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2432 * fold-const-call.c (fold_const_call): Handle them.
2433 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2434 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2435 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2436 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2437 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2438 (UNSPEC_XORV): New unspecs.
2439 (optab): Add entries for them.
2440 (BITWISEV): New int iterator.
2441 (bit_reduc_op): New int attributes.
2443 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2444 Alan Hayward <alan.hayward@arm.com>
2445 David Sherwood <david.sherwood@arm.com>
2447 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2448 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2449 * optabs.def (vec_shl_insert_optab): New optab.
2450 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2451 (duplicate_and_interleave): Likewise.
2452 * tree-vect-loop.c: Include internal-fn.h.
2453 (neutral_op_for_slp_reduction): New function, split out from
2454 get_initial_defs_for_reduction.
2455 (get_initial_def_for_reduction): Handle option 2 for variable-length
2456 vectors by loading the neutral value into a vector and then shifting
2457 the initial value into element 0.
2458 (get_initial_defs_for_reduction): Replace the code argument with
2459 the neutral value calculated by neutral_op_for_slp_reduction.
2460 Use gimple_build_vector for constant-length vectors.
2461 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2462 but the first group_size elements have a neutral value.
2463 Use duplicate_and_interleave otherwise.
2464 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2465 Update call to get_initial_defs_for_reduction. Handle SLP
2466 reductions for variable-length vectors by creating one vector
2467 result for each scalar result, with the elements associated
2468 with other scalar results stubbed out with the neutral value.
2469 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2470 Require IFN_VEC_SHL_INSERT for double reductions on
2471 variable-length vectors, or SLP reductions that have
2472 a neutral value. Require can_duplicate_and_interleave_p
2473 support for variable-length unchained SLP reductions if there
2474 is no neutral value, such as for MIN/MAX reductions. Also require
2475 the number of vector elements to be a multiple of the number of
2476 SLP statements when doing variable-length unchained SLP reductions.
2477 Update call to vect_create_epilog_for_reduction.
2478 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2479 and remove initial values.
2480 (duplicate_and_interleave): Make public.
2481 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2482 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
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 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2489 (can_duplicate_and_interleave_p): New function.
2490 (vect_get_and_check_slp_defs): Take the vector of statements
2491 rather than just the current one. Remove excess parentheses.
2492 Restriction rejectinon of vect_constant_def and vect_external_def
2493 for variable-length vectors to boolean types, or types for which
2494 can_duplicate_and_interleave_p is false.
2495 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2496 (duplicate_and_interleave): New function.
2497 (vect_get_constant_vectors): Use gimple_build_vector for
2498 constant-length vectors and suitable variable-length constant
2499 vectors. Use duplicate_and_interleave for other variable-length
2500 vectors. Don't defer the update when inserting new statements.
2502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2503 Alan Hayward <alan.hayward@arm.com>
2504 David Sherwood <david.sherwood@arm.com>
2506 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2507 min_profitable_iters doesn't go negative.
2509 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2510 Alan Hayward <alan.hayward@arm.com>
2511 David Sherwood <david.sherwood@arm.com>
2513 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2514 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2515 * optabs.def (vec_mask_load_lanes_optab): New optab.
2516 (vec_mask_store_lanes_optab): Likewise.
2517 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2518 (MASK_STORE_LANES): Likewise.
2519 * internal-fn.c (mask_load_lanes_direct): New macro.
2520 (mask_store_lanes_direct): Likewise.
2521 (expand_mask_load_optab_fn): Handle masked operations.
2522 (expand_mask_load_lanes_optab_fn): New macro.
2523 (expand_mask_store_optab_fn): Handle masked operations.
2524 (expand_mask_store_lanes_optab_fn): New macro.
2525 (direct_mask_load_lanes_optab_supported_p): Likewise.
2526 (direct_mask_store_lanes_optab_supported_p): Likewise.
2527 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2529 (vect_load_lanes_supported): Likewise.
2530 * tree-vect-data-refs.c (strip_conversion): New function.
2531 (can_group_stmts_p): Likewise.
2532 (vect_analyze_data_ref_accesses): Use it instead of checking
2533 for a pair of assignments.
2534 (vect_store_lanes_supported): Take a masked_p parameter.
2535 (vect_load_lanes_supported): Likewise.
2536 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2537 vect_store_lanes_supported and vect_load_lanes_supported.
2538 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2539 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2540 parameter. Don't allow gaps for masked accesses.
2541 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2542 and vect_load_lanes_supported.
2543 (get_load_store_type): Take a masked_p parameter and update
2544 call to get_group_load_store_type.
2545 (vectorizable_store): Update call to get_load_store_type.
2546 Handle IFN_MASK_STORE_LANES.
2547 (vectorizable_load): Update call to get_load_store_type.
2548 Handle IFN_MASK_LOAD_LANES.
2550 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2551 Alan Hayward <alan.hayward@arm.com>
2552 David Sherwood <david.sherwood@arm.com>
2554 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2556 * config/aarch64/aarch64-protos.h
2557 (aarch64_sve_struct_memory_operand_p): Declare.
2558 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2559 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2560 (VPRED, vpred): Handle SVE structure modes.
2561 * config/aarch64/constraints.md (Utx): New constraint.
2562 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2563 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2564 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2565 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2566 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2567 structure modes. Split into pieces after RA.
2568 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2569 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2571 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2572 SVE structure modes.
2573 (aarch64_classify_address): Likewise.
2574 (sizetochar): Move earlier in file.
2575 (aarch64_print_operand): Handle SVE register lists.
2576 (aarch64_array_mode): New function.
2577 (aarch64_sve_struct_memory_operand_p): Likewise.
2578 (TARGET_ARRAY_MODE): Redefine.
2580 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2581 Alan Hayward <alan.hayward@arm.com>
2582 David Sherwood <david.sherwood@arm.com>
2584 * target.def (array_mode): New target hook.
2585 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2586 * doc/tm.texi: Regenerate.
2587 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2588 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2589 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2591 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2594 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2595 Alan Hayward <alan.hayward@arm.com>
2596 David Sherwood <david.sherwood@arm.com>
2598 * fold-const.c (fold_binary_loc): Check the argument types
2599 rather than the result type when testing for a vector operation.
2601 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2603 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2604 * doc/tm.texi: Regenerate.
2606 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2607 Alan Hayward <alan.hayward@arm.com>
2608 David Sherwood <david.sherwood@arm.com>
2610 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2611 (sve): Document new AArch64 extension.
2612 * doc/md.texi (w): Extend the description of the AArch64
2613 constraint to include SVE vectors.
2614 (Upl, Upa): Document new AArch64 predicate constraints.
2615 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2617 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2618 (msve-vector-bits=): New option.
2619 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2620 SVE when these are disabled.
2621 (sve): New extension.
2622 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2623 modes. Adjust their number of units based on aarch64_sve_vg.
2624 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2625 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2626 aarch64_addr_query_type.
2627 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2628 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2629 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2630 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2631 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2632 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2633 (aarch64_simd_imm_zero_p): Delete.
2634 (aarch64_check_zero_based_sve_index_immediate): Declare.
2635 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2636 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2637 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2638 (aarch64_sve_float_mul_immediate_p): Likewise.
2639 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2641 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2642 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2643 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2644 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2645 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2646 (aarch64_regmode_natural_size): Likewise.
2647 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2648 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2650 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2651 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2652 for VG and the SVE predicate registers.
2653 (V_ALIASES): Add a "z"-prefixed alias.
2654 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2655 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2656 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2657 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2658 (REG_CLASS_NAMES): Add entries for them.
2659 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2660 and the predicate registers.
2661 (aarch64_sve_vg): Declare.
2662 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2663 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2664 (REGMODE_NATURAL_SIZE): Define.
2665 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2667 * config/aarch64/aarch64.c: Include cfgrtl.h.
2668 (simd_immediate_info): Add a constructor for series vectors,
2669 and an associated step field.
2670 (aarch64_sve_vg): New variable.
2671 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2672 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2673 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2674 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2675 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2676 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2677 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2678 (aarch64_get_mask_mode): New functions.
2679 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2680 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2681 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2682 predicate modes and predicate registers. Explicitly restrict
2683 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2684 to store a vector mode if it is recognized by
2685 aarch64_classify_vector_mode.
2686 (aarch64_regmode_natural_size): New function.
2687 (aarch64_hard_regno_caller_save_mode): Return the original mode
2689 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2690 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2691 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2692 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2694 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2695 does not overlap dest if the function is frame-related. Handle
2697 (aarch64_split_add_offset): New function.
2698 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2699 them aarch64_add_offset.
2700 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2701 and update call to aarch64_sub_sp.
2702 (aarch64_add_cfa_expression): New function.
2703 (aarch64_expand_prologue): Pass extra temporary registers to the
2704 functions above. Handle the case in which we need to emit new
2705 DW_CFA_expressions for registers that were originally saved
2706 relative to the stack pointer, but now have to be expressed
2707 relative to the frame pointer.
2708 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2710 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2711 IP0 and IP1 values for SVE frames.
2712 (aarch64_expand_vec_series): New function.
2713 (aarch64_expand_sve_widened_duplicate): Likewise.
2714 (aarch64_expand_sve_const_vector): Likewise.
2715 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2716 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2717 into the register, rather than emitting a SET directly.
2718 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2719 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2720 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2721 (offset_9bit_signed_scaled_p): New functions.
2722 (aarch64_replicate_bitmask_imm): New function.
2723 (aarch64_bitmask_imm): Use it.
2724 (aarch64_cannot_force_const_mem): Reject expressions involving
2725 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2726 (aarch64_classify_index): Handle SVE indices, by requiring
2727 a plain register index with a scale that matches the element size.
2728 (aarch64_classify_address): Handle SVE addresses. Assert that
2729 the mode of the address is VOIDmode or an integer mode.
2730 Update call to aarch64_classify_symbol.
2731 (aarch64_classify_symbolic_expression): Update call to
2732 aarch64_classify_symbol.
2733 (aarch64_const_vec_all_in_range_p): New function.
2734 (aarch64_print_vector_float_operand): Likewise.
2735 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2736 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2737 and the FP immediates 1.0 and 0.5.
2738 (aarch64_print_address_internal): Handle SVE addresses.
2739 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2740 (aarch64_regno_regclass): Handle predicate registers.
2741 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2743 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2744 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2745 (aarch64_convert_sve_vector_bits): New function.
2746 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2747 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2749 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2750 Handle SVE vector and predicate modes. Accept VL-based constants
2751 that need only one temporary register, and VL offsets that require
2752 no temporary registers.
2753 (aarch64_conditional_register_usage): Mark the predicate registers
2754 as fixed if SVE isn't available.
2755 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2756 Return true for SVE vector and predicate modes.
2757 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2758 rather than an unsigned int. Handle SVE modes.
2759 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2761 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2763 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2764 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2765 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2766 (aarch64_sve_float_mul_immediate_p): New functions.
2767 (aarch64_sve_valid_immediate): New function.
2768 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2769 Explicitly reject structure modes. Check for INDEX constants.
2770 Handle PTRUE and PFALSE constants.
2771 (aarch64_check_zero_based_sve_index_immediate): New function.
2772 (aarch64_simd_imm_zero_p): Delete.
2773 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2774 vector modes. Accept constants in the range of CNT[BHWD].
2775 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2776 ask for an Advanced SIMD mode.
2777 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2778 (aarch64_simd_vector_alignment): Handle SVE predicates.
2779 (aarch64_vectorize_preferred_vector_alignment): New function.
2780 (aarch64_simd_vector_alignment_reachable): Use it instead of
2782 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2783 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2785 (MAX_VECT_LEN): Delete.
2786 (expand_vec_perm_d): Add a vec_flags field.
2787 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2788 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2789 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2791 (aarch64_evpc_rev): Rename to...
2792 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2793 (aarch64_evpc_rev_global): New function.
2794 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2795 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2797 (aarch64_evpc_sve_tbl): New function.
2798 (aarch64_expand_vec_perm_const_1): Update after rename of
2799 aarch64_evpc_rev. Handle SVE permutes too, trying
2800 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2801 than aarch64_evpc_tbl.
2802 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2803 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2804 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2805 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2806 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2807 (aarch64_expand_sve_vcond): New functions.
2808 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2809 of aarch64_vector_mode_p.
2810 (aarch64_dwarf_poly_indeterminate_value): New function.
2811 (aarch64_compute_pressure_classes): Likewise.
2812 (aarch64_can_change_mode_class): Likewise.
2813 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2814 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2815 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2816 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2817 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2818 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2819 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2820 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2822 (Dn, Dl, Dr): Accept const as well as const_vector.
2823 (Dz): Likewise. Compare against CONST0_RTX.
2824 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2825 of "vector" where appropriate.
2826 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2827 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2828 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2829 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2830 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2831 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2832 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2833 (v_int_equiv): Extend to SVE modes.
2834 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2836 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2837 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2838 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2839 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2840 (SVE_COND_FP_CMP): New int iterators.
2841 (perm_hilo): Handle the new unpack unspecs.
2842 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2844 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2845 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2846 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2847 (aarch64_equality_operator, aarch64_constant_vector_operand)
2848 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2849 (aarch64_sve_nonimmediate_operand): Likewise.
2850 (aarch64_sve_general_operand): Likewise.
2851 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2852 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2853 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2854 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2855 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2856 (aarch64_sve_float_arith_immediate): Likewise.
2857 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2858 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2859 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2860 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2861 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2862 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2863 (aarch64_sve_float_arith_operand): Likewise.
2864 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2865 (aarch64_sve_float_mul_operand): Likewise.
2866 (aarch64_sve_vec_perm_operand): Likewise.
2867 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2868 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2869 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2870 as well as const_vector.
2871 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2872 in file. Use CONST0_RTX and CONSTM1_RTX.
2873 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2874 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2875 Use aarch64_simd_imm_zero.
2876 * config/aarch64/aarch64-sve.md: New file.
2877 * config/aarch64/aarch64.md: Include it.
2878 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2879 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2880 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2881 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2882 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2883 (sve): New attribute.
2884 (enabled): Disable instructions with the sve attribute unless
2886 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2887 aarch64_expand_mov_immediate.
2888 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2889 CNT[BHSD] immediates.
2890 (movti): Split CONST_POLY_INT moves into two halves.
2891 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2892 Split additions that need a temporary here if the destination
2893 is the stack pointer.
2894 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2895 (*add<mode>3_poly_1): New instruction.
2896 (set_clobber_cc): New expander.
2898 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2900 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2901 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2902 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2903 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2904 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2905 Change innermode from fixed_mode_size to machine_mode.
2906 (simplify_subreg): Update call accordingly. Handle a constant-sized
2907 subreg of a variable-length CONST_VECTOR.
2909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2910 Alan Hayward <alan.hayward@arm.com>
2911 David Sherwood <david.sherwood@arm.com>
2913 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2914 (add_offset_to_base): New function, split out from...
2915 (create_mem_ref): ...here. When handling a scale other than 1,
2916 check first whether the address is valid without the offset.
2917 Add it into the base if so, leaving the index and scale as-is.
2919 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2922 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2923 fold_for_warn before checking if arg2 is INTEGER_CST.
2925 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2927 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2928 (store_multiple_operation): Delete.
2929 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2930 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2931 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2932 guarded by TARGET_STRING.
2933 (rs6000_output_load_multiple): Delete.
2934 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2935 OPTION_MASK_STRING / TARGET_STRING handling.
2936 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2937 (const rs6000_opt_masks) <"string">: Change mask to 0.
2938 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2939 (MASK_STRING): Delete.
2940 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2942 (load_multiple): Delete.
2949 (store_multiple): Delete.
2956 (movmemsi_8reg): Delete.
2957 (corresponding unnamed define_insn): Delete.
2958 (movmemsi_6reg): Delete.
2959 (corresponding unnamed define_insn): Delete.
2960 (movmemsi_4reg): Delete.
2961 (corresponding unnamed define_insn): Delete.
2962 (movmemsi_2reg): Delete.
2963 (corresponding unnamed define_insn): Delete.
2964 (movmemsi_1reg): Delete.
2965 (corresponding unnamed define_insn): Delete.
2966 * config/rs6000/rs6000.opt (mno-string): New.
2967 (mstring): Replace by deprecation warning stub.
2968 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2970 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2972 * regrename.c (regrename_do_replace): If replacing the same
2973 reg multiple times, try to reuse last created gen_raw_REG.
2976 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2977 main to workaround a bug in GDB.
2979 2018-01-12 Tom de Vries <tom@codesourcery.com>
2982 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2984 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2986 PR rtl-optimization/80481
2987 * ira-color.c (get_cap_member): New function.
2988 (allocnos_conflict_by_live_ranges_p): Use it.
2989 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2990 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2992 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2995 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2996 (*saddl_se_1): Ditto.
2998 (*ssubl_se_1): Ditto.
3000 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3002 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3003 rather than wi::to_widest for DR_INITs.
3004 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3005 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3006 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3008 (vect_analyze_group_access_1): Note that here.
3010 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3012 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3013 polynomial type sizes.
3015 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3017 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3018 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3019 (gimple_add_tmp_var): Likewise.
3021 2018-01-12 Martin Liska <mliska@suse.cz>
3023 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3024 (gimple_alloc_sizes): Likewise.
3025 (dump_gimple_statistics): Use PRIu64 in printf format.
3026 * gimple.h: Change uint64_t to int.
3028 2018-01-12 Martin Liska <mliska@suse.cz>
3030 * tree-core.h: Use uint64_t instead of int.
3031 * tree.c (tree_node_counts): Likewise.
3032 (tree_node_sizes): Likewise.
3033 (dump_tree_statistics): Use PRIu64 in printf format.
3035 2018-01-12 Martin Liska <mliska@suse.cz>
3037 * Makefile.in: As qsort_chk is implemented in vec.c, add
3038 vec.o to linkage of gencfn-macros.
3039 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3040 passing the info to record_node_allocation_statistics.
3041 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3043 * ggc-common.c (struct ggc_usage): Add operator== and use
3044 it in operator< and compare function.
3045 * mem-stats.h (struct mem_usage): Likewise.
3046 * vec.c (struct vec_usage): Remove operator< and compare
3047 function. Can be simply inherited.
3049 2018-01-12 Martin Jambor <mjambor@suse.cz>
3052 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3053 * tree-ssa-math-opts.c: Include domwalk.h.
3054 (convert_mult_to_fma_1): New function.
3055 (fma_transformation_info): New type.
3056 (fma_deferring_state): Likewise.
3057 (cancel_fma_deferring): New function.
3058 (result_of_phi): Likewise.
3059 (last_fma_candidate_feeds_initial_phi): Likewise.
3060 (convert_mult_to_fma): Added deferring logic, split actual
3061 transformation to convert_mult_to_fma_1.
3062 (math_opts_dom_walker): New type.
3063 (math_opts_dom_walker::after_dom_children): New method, body moved
3064 here from pass_optimize_widening_mul::execute, added deferring logic
3066 (pass_optimize_widening_mul::execute): Moved most of code to
3067 math_opts_dom_walker::after_dom_children.
3068 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3069 * config/i386/i386.c (ix86_option_override_internal): Added
3070 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3072 2018-01-12 Richard Biener <rguenther@suse.de>
3075 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3076 inline instance vars.
3078 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3081 * config/rx/rx.c (rx_is_restricted_memory_address):
3084 2018-01-12 Richard Biener <rguenther@suse.de>
3086 PR tree-optimization/80846
3087 * target.def (split_reduction): New target hook.
3088 * targhooks.c (default_split_reduction): New function.
3089 * targhooks.h (default_split_reduction): Declare.
3090 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3091 target requests first reduce vectors by combining low and high
3093 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3094 (get_vectype_for_scalar_type_and_size): Export.
3095 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3096 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3097 * doc/tm.texi: Regenerate.
3098 * config/i386/i386.c (ix86_split_reduction): Implement
3099 TARGET_VECTORIZE_SPLIT_REDUCTION.
3101 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3104 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3105 in PIC mode except for TARGET_VXWORKS_RTP.
3106 * config/sparc/sparc.c: Include cfgrtl.h.
3107 (TARGET_INIT_PIC_REG): Define.
3108 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3109 (sparc_pic_register_p): New predicate.
3110 (sparc_legitimate_address_p): Use it.
3111 (sparc_legitimize_pic_address): Likewise.
3112 (sparc_delegitimize_address): Likewise.
3113 (sparc_mode_dependent_address_p): Likewise.
3114 (gen_load_pcrel_sym): Remove 4th parameter.
3115 (load_got_register): Adjust call to above. Remove obsolete stuff.
3116 (sparc_expand_prologue): Do not call load_got_register here.
3117 (sparc_flat_expand_prologue): Likewise.
3118 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3119 (sparc_use_pseudo_pic_reg): New function.
3120 (sparc_init_pic_reg): Likewise.
3121 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3122 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3124 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3126 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3127 Add item for branch_cost.
3129 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3131 PR rtl-optimization/83565
3132 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3133 not extend the result to a larger mode for rotate operations.
3134 (num_sign_bit_copies1): Likewise.
3136 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3139 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3141 Use values-Xc.o for -pedantic.
3142 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3144 2018-01-12 Martin Liska <mliska@suse.cz>
3147 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3149 (possible_polymorphic_call_targets): Use it.
3150 (ipa_devirt): Likewise.
3152 2018-01-12 Martin Liska <mliska@suse.cz>
3154 * profile-count.h (enum profile_quality): Use 0 as invalid
3155 enum value of profile_quality.
3157 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3159 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3160 -mext-string options.
3162 2018-01-12 Richard Biener <rguenther@suse.de>
3164 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3165 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3166 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3168 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3170 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3172 * configure.ac (--with-long-double-format): Add support for the
3173 configuration option to change the default long double format on
3175 * config.gcc (powerpc*-linux*-*): Likewise.
3176 * configure: Regenerate.
3177 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3178 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3179 used without modification.
3181 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3183 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3184 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3185 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3186 MISC_BUILTIN_SPEC_BARRIER.
3187 (rs6000_init_builtins): Likewise.
3188 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3190 (speculation_barrier): New define_insn.
3191 * doc/extend.texi: Document __builtin_speculation_barrier.
3193 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3196 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3197 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3198 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3200 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3201 integral modes instead of "ss" and "sd".
3202 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3203 vectors with 32-bit and 64-bit elements.
3204 (vecdupssescalarmodesuffix): New mode attribute.
3205 (vec_dup<mode>): Use it.
3207 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3210 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3211 frame if argument is passed on stack.
3213 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3216 * ree.c (combine_reaching_defs): Optimize also
3217 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3218 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3220 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3223 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3225 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3228 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3229 after they are computed.
3231 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3233 PR tree-optimization/83695
3234 * gimple-loop-linterchange.cc
3235 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3236 reset cached scev information after interchange.
3237 (pass_linterchange::execute): Remove call to scev_reset_htab.
3239 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3241 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3242 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3243 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3244 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3245 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3246 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3247 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3248 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3249 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3250 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3251 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3252 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3253 (V_lane_reg): Likewise.
3254 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3256 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3257 (vfmal_lane_low<mode>_intrinsic,
3258 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3259 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3260 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3261 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3262 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3263 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3265 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3267 * config/arm/arm-cpus.in (fp16fml): New feature.
3268 (ALL_SIMD): Add fp16fml.
3269 (armv8.2-a): Add fp16fml as an option.
3270 (armv8.3-a): Likewise.
3271 (armv8.4-a): Add fp16fml as part of fp16.
3272 * config/arm/arm.h (TARGET_FP16FML): Define.
3273 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3275 * config/arm/arm-modes.def (V2HF): Define.
3276 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3277 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3278 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3279 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3280 vfmsl_low, vfmsl_high): New set of builtins.
3281 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3282 (vfml_op): New code attribute.
3283 (VFMLHALVES): New int iterator.
3284 (VFML, VFMLSEL): New mode attributes.
3285 (V_reg): Define mapping for V2HF.
3286 (V_hi, V_lo): New mode attributes.
3287 (VF_constraint): Likewise.
3288 (vfml_half, vfml_half_selector): New int attributes.
3289 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3291 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3292 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3294 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3295 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3296 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3297 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3299 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3300 Document new effective target and option set.
3302 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3304 * config/arm/arm-cpus.in (armv8_4): New feature.
3305 (ARMv8_4a): New fgroup.
3306 (armv8.4-a): New arch.
3307 * config/arm/arm-tables.opt: Regenerate.
3308 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3309 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3310 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3311 Add matching rules for -march=armv8.4-a and extensions.
3312 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3314 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3317 * config/rx/rx.md (BW): New mode attribute.
3318 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3320 2018-01-11 Richard Biener <rguenther@suse.de>
3322 PR tree-optimization/83435
3323 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3324 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3325 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3327 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3328 Alan Hayward <alan.hayward@arm.com>
3329 David Sherwood <david.sherwood@arm.com>
3331 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3333 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3334 (aarch64_print_address_internal): Use it to check for a zero offset.
3336 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3337 Alan Hayward <alan.hayward@arm.com>
3338 David Sherwood <david.sherwood@arm.com>
3340 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3341 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3342 Return a poly_int64 rather than a HOST_WIDE_INT.
3343 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3344 rather than a HOST_WIDE_INT.
3345 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3346 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3347 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3348 final_offset from HOST_WIDE_INT to poly_int64.
3349 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3350 to_constant when getting the number of units in an Advanced SIMD
3352 (aarch64_builtin_vectorized_function): Check for a constant number
3354 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3356 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3357 attribute instead of GET_MODE_NUNITS.
3358 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3359 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3360 GET_MODE_SIZE for fixed-size registers.
3361 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3362 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3363 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3364 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3365 (aarch64_print_operand, aarch64_print_address_internal)
3366 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3367 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3368 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3369 Handle polynomial GET_MODE_SIZE.
3370 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3371 wider than SImode without modification.
3372 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3373 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3374 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3375 passing and returning SVE modes.
3376 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3377 rather than GEN_INT.
3378 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3379 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3380 (aarch64_allocate_and_probe_stack_space): Likewise.
3381 (aarch64_layout_frame): Cope with polynomial offsets.
3382 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3383 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3385 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3386 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3387 poly_int64 rather than a HOST_WIDE_INT.
3388 (aarch64_get_separate_components, aarch64_process_components)
3389 (aarch64_expand_prologue, aarch64_expand_epilogue)
3390 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3391 (aarch64_anchor_offset): New function, split out from...
3392 (aarch64_legitimize_address): ...here.
3393 (aarch64_builtin_vectorization_cost): Handle polynomial
3394 TYPE_VECTOR_SUBPARTS.
3395 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3397 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3398 number of elements from the PARALLEL rather than the mode.
3399 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3400 rather than GET_MODE_BITSIZE.
3401 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3402 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3403 (aarch64_expand_vec_perm_const_1): Handle polynomial
3404 d->perm.length () and d->perm elements.
3405 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3406 Apply to_constant to d->perm elements.
3407 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3408 polynomial CONST_VECTOR_NUNITS.
3409 (aarch64_move_pointer): Take amount as a poly_int64 rather
3411 (aarch64_progress_pointer): Avoid temporary variable.
3412 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3413 the mode attribute instead of GET_MODE.
3415 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3416 Alan Hayward <alan.hayward@arm.com>
3417 David Sherwood <david.sherwood@arm.com>
3419 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3420 x exists before using it.
3421 (aarch64_add_constant_internal): Rename to...
3422 (aarch64_add_offset_1): ...this. Replace regnum with separate
3423 src and dest rtxes. Handle the case in which they're different,
3424 including when the offset is zero. Replace scratchreg with an rtx.
3425 Use 2 additions if there is no spare register into which we can
3426 move a 16-bit constant.
3427 (aarch64_add_constant): Delete.
3428 (aarch64_add_offset): Replace reg with separate src and dest
3429 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3430 Use aarch64_add_offset_1.
3431 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3432 an rtx rather than an int. Take the delta as a poly_int64
3433 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3434 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3435 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3436 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3437 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3439 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3440 aarch64_add_constant.
3442 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3444 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3445 Use scalar_float_mode.
3447 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3449 * config/aarch64/aarch64-simd.md
3450 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3451 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3452 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3453 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3454 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3455 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3456 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3457 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3458 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3459 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3461 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3464 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3465 targ_options->x_arm_arch_string is non NULL.
3467 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3469 * config/aarch64/aarch64.h
3470 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3472 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3475 * expmed.c (emit_store_flag_force): Swap if const op0
3476 and change VOIDmode to mode of op0.
3478 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3480 PR rtl-optimization/83761
3481 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3482 than bytes to mode_for_size.
3484 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3487 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3488 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3491 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3494 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3495 when in layout mode.
3496 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3497 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3500 2018-01-10 Michael Collison <michael.collison@arm.com>
3502 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3503 * config/aarch64/aarch64-option-extension.def: Add
3504 AARCH64_OPT_EXTENSION of 'fp16fml'.
3505 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3506 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3507 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3508 * config/aarch64/constraints.md (Ui7): New constraint.
3509 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3510 (VFMLA_SEL_W): Ditto.
3513 (VFMLA16_LOW): New int iterator.
3514 (VFMLA16_HIGH): Ditto.
3515 (UNSPEC_FMLAL): New unspec.
3516 (UNSPEC_FMLSL): Ditto.
3517 (UNSPEC_FMLAL2): Ditto.
3518 (UNSPEC_FMLSL2): Ditto.
3519 (f16mac): New code attribute.
3520 * config/aarch64/aarch64-simd-builtins.def
3521 (aarch64_fmlal_lowv2sf): Ditto.
3522 (aarch64_fmlsl_lowv2sf): Ditto.
3523 (aarch64_fmlalq_lowv4sf): Ditto.
3524 (aarch64_fmlslq_lowv4sf): Ditto.
3525 (aarch64_fmlal_highv2sf): Ditto.
3526 (aarch64_fmlsl_highv2sf): Ditto.
3527 (aarch64_fmlalq_highv4sf): Ditto.
3528 (aarch64_fmlslq_highv4sf): Ditto.
3529 (aarch64_fmlal_lane_lowv2sf): Ditto.
3530 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3531 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3532 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3533 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3534 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3535 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3536 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3537 (aarch64_fmlal_lane_highv2sf): Ditto.
3538 (aarch64_fmlsl_lane_highv2sf): Ditto.
3539 (aarch64_fmlal_laneq_highv2sf): Ditto.
3540 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3541 (aarch64_fmlalq_lane_highv4sf): Ditto.
3542 (aarch64_fmlsl_lane_highv4sf): Ditto.
3543 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3544 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3545 * config/aarch64/aarch64-simd.md:
3546 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3547 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3548 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3549 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3550 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3551 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3552 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3553 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3554 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3555 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3556 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3557 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3558 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3559 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3560 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3561 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3562 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3563 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3564 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3565 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3566 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3567 (vfmlsl_low_u32): Ditto.
3568 (vfmlalq_low_u32): Ditto.
3569 (vfmlslq_low_u32): Ditto.
3570 (vfmlal_high_u32): Ditto.
3571 (vfmlsl_high_u32): Ditto.
3572 (vfmlalq_high_u32): Ditto.
3573 (vfmlslq_high_u32): Ditto.
3574 (vfmlal_lane_low_u32): Ditto.
3575 (vfmlsl_lane_low_u32): Ditto.
3576 (vfmlal_laneq_low_u32): Ditto.
3577 (vfmlsl_laneq_low_u32): Ditto.
3578 (vfmlalq_lane_low_u32): Ditto.
3579 (vfmlslq_lane_low_u32): Ditto.
3580 (vfmlalq_laneq_low_u32): Ditto.
3581 (vfmlslq_laneq_low_u32): Ditto.
3582 (vfmlal_lane_high_u32): Ditto.
3583 (vfmlsl_lane_high_u32): Ditto.
3584 (vfmlal_laneq_high_u32): Ditto.
3585 (vfmlsl_laneq_high_u32): Ditto.
3586 (vfmlalq_lane_high_u32): Ditto.
3587 (vfmlslq_lane_high_u32): Ditto.
3588 (vfmlalq_laneq_high_u32): Ditto.
3589 (vfmlslq_laneq_high_u32): Ditto.
3590 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3591 (AARCH64_FL_FOR_ARCH8_4): New.
3592 (AARCH64_ISA_F16FML): New ISA flag.
3593 (TARGET_F16FML): New feature flag for fp16fml.
3594 (doc/invoke.texi): Document new fp16fml option.
3596 2018-01-10 Michael Collison <michael.collison@arm.com>
3598 * config/aarch64/aarch64-builtins.c:
3599 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3600 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3601 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3602 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3603 (AARCH64_ISA_SHA3): New ISA flag.
3604 (TARGET_SHA3): New feature flag for sha3.
3605 * config/aarch64/iterators.md (sha512_op): New int attribute.
3606 (CRYPTO_SHA512): New int iterator.
3607 (UNSPEC_SHA512H): New unspec.
3608 (UNSPEC_SHA512H2): Ditto.
3609 (UNSPEC_SHA512SU0): Ditto.
3610 (UNSPEC_SHA512SU1): Ditto.
3611 * config/aarch64/aarch64-simd-builtins.def
3612 (aarch64_crypto_sha512hqv2di): New builtin.
3613 (aarch64_crypto_sha512h2qv2di): Ditto.
3614 (aarch64_crypto_sha512su0qv2di): Ditto.
3615 (aarch64_crypto_sha512su1qv2di): Ditto.
3616 (aarch64_eor3qv8hi): Ditto.
3617 (aarch64_rax1qv2di): Ditto.
3618 (aarch64_xarqv2di): Ditto.
3619 (aarch64_bcaxqv8hi): Ditto.
3620 * config/aarch64/aarch64-simd.md:
3621 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3622 (aarch64_crypto_sha512su0qv2di): Ditto.
3623 (aarch64_crypto_sha512su1qv2di): Ditto.
3624 (aarch64_eor3qv8hi): Ditto.
3625 (aarch64_rax1qv2di): Ditto.
3626 (aarch64_xarqv2di): Ditto.
3627 (aarch64_bcaxqv8hi): Ditto.
3628 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3629 (vsha512h2q_u64): Ditto.
3630 (vsha512su0q_u64): Ditto.
3631 (vsha512su1q_u64): Ditto.
3632 (veor3q_u16): Ditto.
3633 (vrax1q_u64): Ditto.
3635 (vbcaxq_u16): Ditto.
3636 * config/arm/types.md (crypto_sha512): New type attribute.
3637 (crypto_sha3): Ditto.
3638 (doc/invoke.texi): Document new sha3 option.
3640 2018-01-10 Michael Collison <michael.collison@arm.com>
3642 * config/aarch64/aarch64-builtins.c:
3643 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3644 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3645 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3646 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3647 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3648 (AARCH64_ISA_SM4): New ISA flag.
3649 (TARGET_SM4): New feature flag for sm4.
3650 * config/aarch64/aarch64-simd-builtins.def
3651 (aarch64_sm3ss1qv4si): Ditto.
3652 (aarch64_sm3tt1aq4si): Ditto.
3653 (aarch64_sm3tt1bq4si): Ditto.
3654 (aarch64_sm3tt2aq4si): Ditto.
3655 (aarch64_sm3tt2bq4si): Ditto.
3656 (aarch64_sm3partw1qv4si): Ditto.
3657 (aarch64_sm3partw2qv4si): Ditto.
3658 (aarch64_sm4eqv4si): Ditto.
3659 (aarch64_sm4ekeyqv4si): Ditto.
3660 * config/aarch64/aarch64-simd.md:
3661 (aarch64_sm3ss1qv4si): Ditto.
3662 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3663 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3664 (aarch64_sm4eqv4si): Ditto.
3665 (aarch64_sm4ekeyqv4si): Ditto.
3666 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3667 (sm3part_op): Ditto.
3668 (CRYPTO_SM3TT): Ditto.
3669 (CRYPTO_SM3PART): Ditto.
3670 (UNSPEC_SM3SS1): New unspec.
3671 (UNSPEC_SM3TT1A): Ditto.
3672 (UNSPEC_SM3TT1B): Ditto.
3673 (UNSPEC_SM3TT2A): Ditto.
3674 (UNSPEC_SM3TT2B): Ditto.
3675 (UNSPEC_SM3PARTW1): Ditto.
3676 (UNSPEC_SM3PARTW2): Ditto.
3677 (UNSPEC_SM4E): Ditto.
3678 (UNSPEC_SM4EKEY): Ditto.
3679 * config/aarch64/constraints.md (Ui2): New constraint.
3680 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3681 * config/arm/types.md (crypto_sm3): New type attribute.
3682 (crypto_sm4): Ditto.
3683 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3684 (vsm3tt1aq_u32): Ditto.
3685 (vsm3tt1bq_u32): Ditto.
3686 (vsm3tt2aq_u32): Ditto.
3687 (vsm3tt2bq_u32): Ditto.
3688 (vsm3partw1q_u32): Ditto.
3689 (vsm3partw2q_u32): Ditto.
3690 (vsm4eq_u32): Ditto.
3691 (vsm4ekeyq_u32): Ditto.
3692 (doc/invoke.texi): Document new sm4 option.
3694 2018-01-10 Michael Collison <michael.collison@arm.com>
3696 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3697 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3698 (AARCH64_FL_FOR_ARCH8_4): New.
3699 (AARCH64_FL_V8_4): New flag.
3700 (doc/invoke.texi): Document new armv8.4-a option.
3702 2018-01-10 Michael Collison <michael.collison@arm.com>
3704 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3705 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3706 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3707 * config/aarch64/aarch64-option-extension.def: Add
3708 AARCH64_OPT_EXTENSION of 'sha2'.
3709 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3710 (crypto): Disable sha2 and aes if crypto disabled.
3711 (crypto): Enable aes and sha2 if enabled.
3712 (simd): Disable sha2 and aes if simd disabled.
3713 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3715 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3716 (TARGET_SHA2): New feature flag for sha2.
3717 (TARGET_AES): New feature flag for aes.
3718 * config/aarch64/aarch64-simd.md:
3719 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3720 conditional on TARGET_AES.
3721 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3722 (aarch64_crypto_sha1hsi): Make pattern conditional
3724 (aarch64_crypto_sha1hv4si): Ditto.
3725 (aarch64_be_crypto_sha1hv4si): Ditto.
3726 (aarch64_crypto_sha1su1v4si): Ditto.
3727 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3728 (aarch64_crypto_sha1su0v4si): Ditto.
3729 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3730 (aarch64_crypto_sha256su0v4si): Ditto.
3731 (aarch64_crypto_sha256su1v4si): Ditto.
3732 (doc/invoke.texi): Document new aes and sha2 options.
3734 2018-01-10 Martin Sebor <msebor@redhat.com>
3736 PR tree-optimization/83781
3737 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3740 2018-01-11 Martin Sebor <msebor@gmail.com>
3741 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3743 PR tree-optimization/83501
3744 PR tree-optimization/81703
3746 * tree-ssa-strlen.c (get_string_cst): Rename...
3747 (get_string_len): ...to this. Handle global constants.
3748 (handle_char_store): Adjust.
3750 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3751 Jim Wilson <jimw@sifive.com>
3753 * config/riscv/riscv-protos.h (riscv_output_return): New.
3754 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3755 (riscv_attribute_table, riscv_output_return),
3756 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3757 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3758 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3760 (riscv_expand_prologue): Add early return for naked function.
3761 (riscv_expand_epilogue): Likewise.
3762 (riscv_function_ok_for_sibcall): Return false for naked function.
3763 (riscv_set_current_function): New.
3764 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3765 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3766 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3767 * doc/extend.texi (RISC-V Function Attributes): New.
3769 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3771 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3772 check for 128-bit long double before checking TCmode.
3773 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3774 128-bit long doubles before checking TFmode or TCmode.
3775 (FLOAT128_IBM_P): Likewise.
3777 2018-01-10 Martin Sebor <msebor@redhat.com>
3779 PR tree-optimization/83671
3780 * builtins.c (c_strlen): Unconditionally return zero for the empty
3782 Use -Warray-bounds for warnings.
3783 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3784 for non-constant array indices with COMPONENT_REF, arrays of
3785 arrays, and pointers to arrays.
3786 (gimple_fold_builtin_strlen): Determine and set length range for
3787 non-constant character arrays.
3789 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3792 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3795 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3797 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3799 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3802 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3803 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3804 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3805 indexed_or_indirect_operand predicate.
3806 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3807 (*vsx_le_perm_load_v8hi): Likewise.
3808 (*vsx_le_perm_load_v16qi): Likewise.
3809 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3810 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3811 (*vsx_le_perm_store_v8hi): Likewise.
3812 (*vsx_le_perm_store_v16qi): Likewise.
3813 (eight unnamed splitters): Likewise.
3815 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3817 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3818 * config/rs6000/emmintrin.h: Likewise.
3819 * config/rs6000/mmintrin.h: Likewise.
3820 * config/rs6000/xmmintrin.h: Likewise.
3822 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3825 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3827 * tree.c (tree_nop_conversion): Return true for location wrapper
3829 (maybe_wrap_with_location): New function.
3830 (selftest::check_strip_nops): New function.
3831 (selftest::test_location_wrappers): New function.
3832 (selftest::tree_c_tests): Call it.
3833 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3834 (maybe_wrap_with_location): New decl.
3835 (EXPR_LOCATION_WRAPPER_P): New macro.
3836 (location_wrapper_p): New inline function.
3837 (tree_strip_any_location_wrapper): New inline function.
3839 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3842 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3843 stack_realign_offset for the largest alignment of stack slot
3845 (ix86_find_max_used_stack_alignment): New function.
3846 (ix86_finalize_stack_frame_flags): Use it. Set
3847 max_used_stack_alignment if we don't realign stack.
3848 * config/i386/i386.h (machine_function): Add
3849 max_used_stack_alignment.
3851 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3853 * config/arm/arm.opt (-mbranch-cost): New option.
3854 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3857 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3860 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3861 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3863 2018-01-10 Richard Biener <rguenther@suse.de>
3866 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3867 early out so it also covers the case where we have a non-NULL
3870 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3872 PR tree-optimization/83753
3873 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3874 for non-strided grouped accesses if the number of elements is 1.
3876 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3879 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3880 * i386.h (TARGET_USE_GATHER): Define.
3881 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3883 2018-01-10 Martin Liska <mliska@suse.cz>
3886 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3887 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3889 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3890 CLEANUP_NO_PARTITIONING is not set.
3892 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3894 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3895 for vectors, as a partial revert of r254296.
3896 * rtl.h (const_vec_p): Delete.
3897 (const_vec_duplicate_p): Don't test for vector CONSTs.
3898 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3899 * expmed.c (make_tree): Likewise.
3902 * common.md (E, F): Use CONSTANT_P instead of checking for
3904 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3905 checking for CONST_VECTOR.
3907 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3910 * predict.c (force_edge_cold): Handle in more sane way edges
3913 2018-01-09 Carl Love <cel@us.ibm.com>
3915 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3917 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3918 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3919 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3920 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3921 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3922 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3923 * config/rs6000/rs6000-protos.h: Add extern defition for
3924 rs6000_generate_float2_double_code.
3925 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3927 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3928 (float2_v2df): Add define_expand.
3930 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3933 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3934 op_mode in the force_to_mode call.
3936 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3938 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3939 instead of checking each element individually.
3940 (aarch64_evpc_uzp): Likewise.
3941 (aarch64_evpc_zip): Likewise.
3942 (aarch64_evpc_ext): Likewise.
3943 (aarch64_evpc_rev): Likewise.
3944 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3945 instead of checking each element individually. Return true without
3947 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3948 whether all selected elements come from the same input, instead of
3949 checking each element individually. Remove calls to gen_rtx_REG,
3950 start_sequence and end_sequence and instead assert that no rtl is
3953 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3955 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3956 order of HIGH and CONST checks.
3958 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3960 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3961 if the destination isn't an SSA_NAME.
3963 2018-01-09 Richard Biener <rguenther@suse.de>
3965 PR tree-optimization/83668
3966 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3968 (canonicalize_loop_form): ... here, renamed from ...
3969 (canonicalize_loop_closed_ssa_form): ... this and amended to
3970 swap successor edges for loop exit blocks to make us use
3971 the RPO order we need for initial schedule generation.
3973 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3975 PR tree-optimization/64811
3976 * match.pd: When optimizing comparisons with Inf, avoid
3977 introducing or losing exceptions from comparisons with NaN.
3979 2018-01-09 Martin Liska <mliska@suse.cz>
3982 * asan.c (shadow_mem_size): Add gcc_assert.
3984 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3986 Don't save registers in main().
3989 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3990 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3991 * config/avr/avr.c (avr_set_current_function): Don't error if
3992 naked, OS_task or OS_main are specified at the same time.
3993 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3995 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3997 * common/config/avr/avr-common.c (avr_option_optimization_table):
3998 Switch on -mmain-is-OS_task for optimizing compilations.
4000 2018-01-09 Richard Biener <rguenther@suse.de>
4002 PR tree-optimization/83572
4003 * graphite.c: Include cfganal.h.
4004 (graphite_transform_loops): Connect infinite loops to exit
4005 and remove fake edges at the end.
4007 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4009 * ipa-inline.c (edge_badness): Revert accidental checkin.
4011 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4014 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4015 symbols; not inline clones.
4017 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4020 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4021 hard registers. Formatting fixes.
4023 PR preprocessor/83722
4024 * gcc.c (try_generate_repro): Pass
4025 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4026 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4029 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4030 Kito Cheng <kito.cheng@gmail.com>
4032 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4033 (riscv_leaf_function_p): Delete.
4034 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4036 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4038 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4040 (do_ifelse): New function.
4041 (do_isel): New function.
4042 (do_sub3): New function.
4043 (do_add3): New function.
4044 (do_load_mask_compare): New function.
4045 (do_overlap_load_compare): New function.
4046 (expand_compare_loop): New function.
4047 (expand_block_compare): Call expand_compare_loop() when appropriate.
4048 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4050 (-mblock-compare-inline-loop-limit): New option.
4052 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4055 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4056 Reverse order of second and third operands in first alternative.
4057 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4058 of first and second elements in UNSPEC_VPERMR vector.
4059 (altivec_expand_vec_perm_le): Likewise.
4061 2018-01-08 Jeff Law <law@redhat.com>
4063 PR rtl-optimizatin/81308
4064 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4065 (process_switch): If group_case_labels makes a change, then set
4067 (pass_convert_switch::execute): If a switch is converted, then
4068 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4070 PR rtl-optimization/81308
4071 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4074 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4076 PR target/83663 - Revert r255946
4077 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4078 generation for cases where splatting a value is not useful.
4079 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4080 across a vec_duplicate and a paradoxical subreg forming a vector
4081 mode to a vec_concat.
4083 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4085 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4086 -march=armv8.3-a variants.
4087 * config/arm/t-multilib: Likewise.
4088 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4090 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4092 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4094 (cceq_ior_compare_complement): Give it a name so I can use it, and
4095 change boolean_or_operator predicate to boolean_operator so it can
4096 be used to generate a crand.
4097 (eqne): New code iterator.
4098 (bd/bd_neg): New code_attrs.
4099 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4100 a single define_insn.
4101 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4102 decrement (bdnzt/bdnzf/bdzt/bdzf).
4103 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4104 with the new names of the branch decrement patterns, and added the
4105 names of the branch decrement conditional patterns.
4107 2018-01-08 Richard Biener <rguenther@suse.de>
4109 PR tree-optimization/83563
4110 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4113 2018-01-08 Richard Biener <rguenther@suse.de>
4116 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4118 2018-01-08 Richard Biener <rguenther@suse.de>
4120 PR tree-optimization/83685
4121 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4122 references to abnormals.
4124 2018-01-08 Richard Biener <rguenther@suse.de>
4127 * dwarf2out.c (output_indirect_strings): Handle empty
4128 skeleton_debug_str_hash.
4129 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4131 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4133 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4134 (emit_store_direct): Likewise.
4135 (arc_trampoline_adjust_address): Likewise.
4136 (arc_asm_trampoline_template): New function.
4137 (arc_initialize_trampoline): Use asm_trampoline_template.
4138 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4139 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4140 * config/arc/arc.md (flush_icache): Delete pattern.
4142 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4144 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4145 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4148 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4151 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4152 by not USED_FOR_TARGET.
4153 (make_pass_resolve_sw_modes): Likewise.
4155 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4157 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4160 2018-01-08 Richard Biener <rguenther@suse.de>
4163 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4165 2018-01-08 Richard Biener <rguenther@suse.de>
4168 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4170 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4173 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4174 basic blocks with a small number of successors.
4175 (convert_control_dep_chain_into_preds): Improve handling of
4177 (dump_predicates): Split apart into...
4178 (dump_pred_chain): ...here...
4179 (dump_pred_info): ...and here.
4180 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4181 (can_chain_union_be_invalidated_p): Improve check for invalidation
4183 (uninit_uses_cannot_happen): Avoid unnecessary if
4184 convert_control_dep_chain_into_preds yielded nothing.
4186 2018-01-06 Martin Sebor <msebor@redhat.com>
4188 PR tree-optimization/83640
4189 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4190 subtracting negative offset from size.
4191 (builtin_access::overlap): Adjust offset bounds of the access to fall
4192 within the size of the object if possible.
4194 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4196 PR rtl-optimization/83699
4197 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4198 extract_bit_field_as_subreg to cases in which the extracted
4199 value is also a vector.
4201 * lra-constraints.c (process_alt_operands): Test for the equivalence
4202 substitutions when detecting a possible reload cycle.
4204 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4207 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4208 by default if flag_selective_schedling{,2}. Formatting fixes.
4210 PR rtl-optimization/83682
4211 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4212 if it has non-VECTOR_MODE element mode.
4213 (vec_duplicate_p): Likewise.
4216 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4217 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4219 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4222 * config/i386/i386-builtin.def
4223 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4224 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4225 Require also OPTION_MASK_ISA_AVX512F in addition to
4226 OPTION_MASK_ISA_GFNI.
4227 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4228 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4229 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4230 to OPTION_MASK_ISA_GFNI.
4231 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4232 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4233 OPTION_MASK_ISA_AVX512BW.
4234 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4235 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4236 addition to OPTION_MASK_ISA_GFNI.
4237 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4238 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4239 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4240 to OPTION_MASK_ISA_GFNI.
4241 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4242 a requirement for all ISAs rather than any of them with a few
4244 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4246 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4247 bitmasks to be enabled with 3 exceptions, instead of requiring any
4248 enabled ISA with lots of exceptions.
4249 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4250 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4251 Change avx512bw in isa attribute to avx512f.
4252 * config/i386/sgxintrin.h: Add license boilerplate.
4253 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4254 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4255 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4256 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4258 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4259 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4260 temporarily sse2 rather than sse if not enabled already.
4263 * config/i386/sse.md (VI248_VLBW): Rename to ...
4264 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4265 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4266 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4267 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4268 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4269 mode iterator instead of VI248_VLBW.
4271 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4273 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4274 (record_modified): Skip clobbers; add debug output.
4275 (param_change_prob): Use sreal frequencies.
4277 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4279 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4280 punt for user-aligned variables.
4282 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4284 * tree-chrec.c (chrec_contains_symbols): Return true for
4287 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4290 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4291 of (x|y) == x for BICS pattern.
4293 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4295 PR tree-optimization/83605
4296 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4297 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4300 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4302 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4303 * config/epiphany/rtems.h: New file.
4305 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4306 Uros Bizjak <ubizjak@gmail.com>
4309 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4310 QIreg_operand instead of register_operand predicate.
4311 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4312 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4313 comments instead of -fmitigate[-_]rop.
4315 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4318 * cgraphunit.c (symbol_table::compile): Switch to text_section
4319 before calling assembly_start debug hook.
4320 * run-rtl-passes.c (run_rtl_passes): Likewise.
4323 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4325 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4326 range_int_cst_p rather than !symbolic_range_p before calling
4327 extract_range_from_multiplicative_op_1.
4329 2018-01-04 Jeff Law <law@redhat.com>
4331 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4332 redundant test in assertion.
4334 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4336 * doc/rtl.texi: Document machine_mode wrapper classes.
4338 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4340 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4343 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4345 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4346 the VEC_PERM_EXPR fold to fail.
4348 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4351 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4352 to switched_sections.
4354 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4357 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4360 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4363 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4364 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4366 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4369 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4370 is BLKmode and bitpos not zero or mode change is needed.
4372 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4375 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4378 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4381 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4382 instead of MULT rtx. Update all corresponding splitters.
4384 (*ssub<modesuffix>): Ditto.
4386 (*cmp_sadd_di): Update split patterns.
4387 (*cmp_sadd_si): Ditto.
4388 (*cmp_sadd_sidi): Ditto.
4389 (*cmp_ssub_di): Ditto.
4390 (*cmp_ssub_si): Ditto.
4391 (*cmp_ssub_sidi): Ditto.
4392 * config/alpha/predicates.md (const23_operand): New predicate.
4393 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4394 Look for ASHIFT, not MULT inner operand.
4395 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4397 2018-01-04 Martin Liska <mliska@suse.cz>
4399 PR gcov-profile/83669
4400 * gcov.c (output_intermediate_file): Add version to intermediate
4402 * doc/gcov.texi: Document new field 'version' in intermediate
4403 file format. Fix location of '-k' option of gcov command.
4405 2018-01-04 Martin Liska <mliska@suse.cz>
4408 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4410 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4412 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4414 2018-01-03 Martin Sebor <msebor@redhat.com>
4416 PR tree-optimization/83655
4417 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4418 checking calls with invalid arguments.
4420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4422 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4423 (vectorizable_mask_load_store): Delete.
4424 (vectorizable_call): Return false for masked loads and stores.
4425 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4426 instead of gimple_assign_rhs1.
4427 (vectorizable_load): Handle IFN_MASK_LOAD.
4428 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4432 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4434 (vectorizable_mask_load_store): ...here.
4435 (vectorizable_load): ...and here.
4437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4439 * tree-vect-stmts.c (vect_build_all_ones_mask)
4440 (vect_build_zero_merge_argument): New functions, split out from...
4441 (vectorizable_load): ...here.
4443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4445 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4447 (vectorizable_mask_load_store): ...here.
4448 (vectorizable_store): ...and here.
4450 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4452 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4454 (vectorizable_mask_load_store): ...here.
4456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4458 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4459 (vect_model_store_cost): Take a vec_load_store_type instead of a
4461 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4462 (vect_model_store_cost): Take a vec_load_store_type instead of a
4464 (vectorizable_mask_load_store): Update accordingly.
4465 (vectorizable_store): Likewise.
4466 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4470 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4471 IFN_MASK_LOAD calls here rather than...
4472 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4474 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4475 Alan Hayward <alan.hayward@arm.com>
4476 David Sherwood <david.sherwood@arm.com>
4478 * expmed.c (extract_bit_field_1): For vector extracts,
4479 fall back to extract_bit_field_as_subreg if vec_extract
4482 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4483 Alan Hayward <alan.hayward@arm.com>
4484 David Sherwood <david.sherwood@arm.com>
4486 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4487 they are variable or constant sized.
4488 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4489 slots for constant-sized data.
4491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4492 Alan Hayward <alan.hayward@arm.com>
4493 David Sherwood <david.sherwood@arm.com>
4495 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4496 handling COND_EXPRs with boolean comparisons, try to find a better
4497 basis for the mask type than the boolean itself.
4499 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4501 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4502 is calculated and how it can be overridden.
4503 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4504 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4506 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4509 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4510 Alan Hayward <alan.hayward@arm.com>
4511 David Sherwood <david.sherwood@arm.com>
4513 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4514 Remove the mode argument.
4515 (aarch64_simd_valid_immediate): Remove the mode and inverse
4517 * config/aarch64/iterators.md (bitsize): New iterator.
4518 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4519 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4520 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4521 aarch64_simd_valid_immediate.
4522 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4523 (aarch64_reg_or_bic_imm): Likewise.
4524 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4525 with an insn_type enum and msl with a modifier_type enum.
4526 Replace element_width with a scalar_mode. Change the shift
4527 to unsigned int. Add constructors for scalar_float_mode and
4528 scalar_int_mode elements.
4529 (aarch64_vect_float_const_representable_p): Delete.
4530 (aarch64_can_const_movi_rtx_p)
4531 (aarch64_simd_scalar_immediate_valid_for_move)
4532 (aarch64_simd_make_constant): Update call to
4533 aarch64_simd_valid_immediate.
4534 (aarch64_advsimd_valid_immediate_hs): New function.
4535 (aarch64_advsimd_valid_immediate): Likewise.
4536 (aarch64_simd_valid_immediate): Remove mode and inverse
4537 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4538 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4539 and aarch64_float_const_representable_p on the result.
4540 (aarch64_output_simd_mov_immediate): Remove mode argument.
4541 Update call to aarch64_simd_valid_immediate and use of
4542 simd_immediate_info.
4543 (aarch64_output_scalar_simd_mov_immediate): Update call
4546 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4547 Alan Hayward <alan.hayward@arm.com>
4548 David Sherwood <david.sherwood@arm.com>
4550 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4551 (mode_nunits): Likewise CONST_MODE_NUNITS.
4552 * machmode.def (ADJUST_NUNITS): Document.
4553 * genmodes.c (mode_data::need_nunits_adj): New field.
4554 (blank_mode): Update accordingly.
4555 (adj_nunits): New variable.
4556 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4558 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4559 listed in adj_nunits.
4560 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4561 listed in adj_nunits. Don't emit case statements for such modes.
4562 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4563 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4564 nothing if adj_nunits is nonnull.
4565 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4566 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4567 (emit_mode_fbit): Update use of print_maybe_const_decl.
4568 (emit_move_size): Likewise. Treat the array as non-const
4570 (emit_mode_adjustments): Handle adj_nunits.
4572 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4574 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4575 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4576 (VECTOR_MODES): Use it.
4577 (make_vector_modes): Take the prefix as an argument.
4579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4580 Alan Hayward <alan.hayward@arm.com>
4581 David Sherwood <david.sherwood@arm.com>
4583 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4584 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4585 for MODE_VECTOR_BOOL.
4586 * machmode.def (VECTOR_BOOL_MODE): Document.
4587 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4588 (make_vector_bool_mode): New function.
4589 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4591 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4592 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4594 * stor-layout.c (int_mode_for_mode): Likewise.
4595 * tree.c (build_vector_type_for_mode): Likewise.
4596 * varasm.c (output_constant_pool_2): Likewise.
4597 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4598 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4599 for MODE_VECTOR_BOOL.
4600 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4601 of mode class checks.
4602 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4603 instead of a list of mode class checks.
4604 (expand_vector_scalar_condition): Likewise.
4605 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4608 Alan Hayward <alan.hayward@arm.com>
4609 David Sherwood <david.sherwood@arm.com>
4611 * machmode.h (mode_size): Change from unsigned short to
4613 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4614 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4615 or if measurement_type is not polynomial.
4616 (fixed_size_mode::includes_p): Check for constant-sized modes.
4617 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4618 return a poly_uint16 rather than an unsigned short.
4619 (emit_mode_size): Change the type of mode_size from unsigned short
4620 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4621 (emit_mode_adjustments): Cope with polynomial vector sizes.
4622 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4624 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4626 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4627 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4628 * caller-save.c (setup_save_areas): Likewise.
4629 (replace_reg_with_saved_mem): Likewise.
4630 * calls.c (emit_library_call_value_1): Likewise.
4631 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4632 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4633 (gen_lowpart_for_combine): Likewise.
4634 * convert.c (convert_to_integer_1): Likewise.
4635 * cse.c (equiv_constant, cse_insn): Likewise.
4636 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4637 (cselib_subst_to_values): Likewise.
4638 * dce.c (word_dce_process_block): Likewise.
4639 * df-problems.c (df_word_lr_mark_ref): Likewise.
4640 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4641 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4642 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4643 (rtl_for_decl_location): Likewise.
4644 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4645 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4646 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4647 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4648 (expand_expr_real_1): Likewise.
4649 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4650 (pad_below): Likewise.
4651 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4652 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4653 * ira.c (get_subreg_tracking_sizes): Likewise.
4654 * ira-build.c (ira_create_allocno_objects): Likewise.
4655 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4656 (ira_sort_regnos_for_alter_reg): Likewise.
4657 * ira-costs.c (record_operand_costs): Likewise.
4658 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4659 (resolve_simple_move): Likewise.
4660 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4661 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4662 (lra_constraints): Likewise.
4663 (CONST_POOL_OK_P): Reject variable-sized modes.
4664 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4665 (add_pseudo_to_slot, lra_spill): Likewise.
4666 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4667 * optabs-query.c (get_best_extraction_insn): Likewise.
4668 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4669 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4670 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4671 * recog.c (offsettable_address_addr_space_p): Likewise.
4672 * regcprop.c (maybe_mode_change): Likewise.
4673 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4674 * regrename.c (build_def_use): Likewise.
4675 * regstat.c (dump_reg_info): Likewise.
4676 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4677 (find_reloads, find_reloads_subreg_address): Likewise.
4678 * reload1.c (eliminate_regs_1): Likewise.
4679 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4680 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4681 (simplify_binary_operation_1, simplify_subreg): Likewise.
4682 * targhooks.c (default_function_arg_padding): Likewise.
4683 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4684 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4685 (verify_gimple_assign_ternary): Likewise.
4686 * tree-inline.c (estimate_move_cost): Likewise.
4687 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4688 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4689 (get_address_cost_ainc): Likewise.
4690 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4691 (vect_supportable_dr_alignment): Likewise.
4692 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4693 (vectorizable_reduction): Likewise.
4694 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4695 (vectorizable_operation, vectorizable_load): Likewise.
4696 * tree.c (build_same_sized_truth_vector_type): Likewise.
4697 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4698 * var-tracking.c (emit_note_insn_var_location): Likewise.
4699 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4700 (ADDR_VEC_ALIGN): Likewise.
4702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4703 Alan Hayward <alan.hayward@arm.com>
4704 David Sherwood <david.sherwood@arm.com>
4706 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4708 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4709 or if measurement_type is polynomial.
4710 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4711 * combine.c (make_extraction): Likewise.
4712 * dse.c (find_shift_sequence): Likewise.
4713 * dwarf2out.c (mem_loc_descriptor): Likewise.
4714 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4715 (extract_bit_field, extract_low_bits): Likewise.
4716 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4717 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4718 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4719 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4720 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4721 * reload.c (find_reloads): Likewise.
4722 * reload1.c (alter_reg): Likewise.
4723 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4724 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4725 * tree-if-conv.c (predicate_mem_writes): Likewise.
4726 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4727 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4728 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4729 * valtrack.c (dead_debug_insert_temp): Likewise.
4730 * varasm.c (mergeable_constant_section): Likewise.
4731 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4733 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4734 Alan Hayward <alan.hayward@arm.com>
4735 David Sherwood <david.sherwood@arm.com>
4737 * expr.c (expand_assignment): Cope with polynomial mode sizes
4738 when assigning to a CONCAT.
4740 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4741 Alan Hayward <alan.hayward@arm.com>
4742 David Sherwood <david.sherwood@arm.com>
4744 * machmode.h (mode_precision): Change from unsigned short to
4746 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4748 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4749 or if measurement_type is not polynomial.
4750 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4751 in which the mode is already known to be a scalar_int_mode.
4752 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4753 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4755 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4756 for GET_MODE_PRECISION.
4757 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4758 for GET_MODE_PRECISION.
4759 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4761 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4762 (expand_field_assignment, make_extraction): Likewise.
4763 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4764 (get_last_value): Likewise.
4765 * convert.c (convert_to_integer_1): Likewise.
4766 * cse.c (cse_insn): Likewise.
4767 * expr.c (expand_expr_real_1): Likewise.
4768 * lra-constraints.c (simplify_operand_subreg): Likewise.
4769 * optabs-query.c (can_atomic_load_p): Likewise.
4770 * optabs.c (expand_atomic_load): Likewise.
4771 (expand_atomic_store): Likewise.
4772 * ree.c (combine_reaching_defs): Likewise.
4773 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4774 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4775 * tree.h (type_has_mode_precision_p): Likewise.
4776 * ubsan.c (instrument_si_overflow): Likewise.
4778 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4779 Alan Hayward <alan.hayward@arm.com>
4780 David Sherwood <david.sherwood@arm.com>
4782 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4783 polynomial numbers of units.
4784 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4785 (valid_vector_subparts_p): New function.
4786 (build_vector_type): Remove temporary shim and take the number
4787 of units as a poly_uint64 rather than an int.
4788 (build_opaque_vector_type): Take the number of units as a
4789 poly_uint64 rather than an int.
4790 * tree.c (build_vector_from_ctor): Handle polynomial
4791 TYPE_VECTOR_SUBPARTS.
4792 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4793 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4794 (build_vector_from_val): If the number of units is variable,
4795 use build_vec_duplicate_cst for constant operands and
4796 VEC_DUPLICATE_EXPR otherwise.
4797 (make_vector_type): Remove temporary is_constant ().
4798 (build_vector_type, build_opaque_vector_type): Take the number of
4799 units as a poly_uint64 rather than an int.
4800 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4802 * cfgexpand.c (expand_debug_expr): Likewise.
4803 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4804 (store_constructor, expand_expr_real_1): Likewise.
4805 (const_scalar_mask_from_tree): Likewise.
4806 * fold-const-call.c (fold_const_reduction): Likewise.
4807 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4808 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4809 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4810 (fold_relational_const): Likewise.
4811 (native_interpret_vector): Likewise. Change the size from an
4812 int to an unsigned int.
4813 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4814 TYPE_VECTOR_SUBPARTS.
4815 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4816 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4817 duplicating a non-constant operand into a variable-length vector.
4818 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4819 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4820 * ipa-icf.c (sem_variable::equals): Likewise.
4821 * match.pd: Likewise.
4822 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4823 * print-tree.c (print_node): Likewise.
4824 * stor-layout.c (layout_type): Likewise.
4825 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4826 * tree-cfg.c (verify_gimple_comparison): Likewise.
4827 (verify_gimple_assign_binary): Likewise.
4828 (verify_gimple_assign_ternary): Likewise.
4829 (verify_gimple_assign_single): Likewise.
4830 * tree-pretty-print.c (dump_generic_node): Likewise.
4831 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4832 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4833 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4834 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4835 (vect_shift_permute_load_chain): Likewise.
4836 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4837 (expand_vector_condition, optimize_vector_constructor): Likewise.
4838 (lower_vec_perm, get_compute_type): Likewise.
4839 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4840 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4841 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4842 (vect_recog_mask_conversion_pattern): Likewise.
4843 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4844 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4845 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4846 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4847 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4848 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4849 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4850 (supportable_widening_operation): Likewise.
4851 (supportable_narrowing_operation): Likewise.
4852 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4854 * varasm.c (output_constant): Likewise.
4856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4857 Alan Hayward <alan.hayward@arm.com>
4858 David Sherwood <david.sherwood@arm.com>
4860 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4861 so that both the length == 3 and length != 3 cases set up their
4862 own permute vectors. Add comments explaining why we know the
4863 number of elements is constant.
4864 (vect_permute_load_chain): Likewise.
4866 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4867 Alan Hayward <alan.hayward@arm.com>
4868 David Sherwood <david.sherwood@arm.com>
4870 * machmode.h (mode_nunits): Change from unsigned char to
4872 (ONLY_FIXED_SIZE_MODES): New macro.
4873 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4874 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4875 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4877 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4878 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4879 or if measurement_type is not polynomial.
4880 * genmodes.c (ZERO_COEFFS): New macro.
4881 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4883 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4884 Use ZERO_COEFFS when emitting initializers.
4885 * data-streamer.h (bp_pack_poly_value): New function.
4886 (bp_unpack_poly_value): Likewise.
4887 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4888 for GET_MODE_NUNITS.
4889 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4890 for GET_MODE_NUNITS.
4891 * tree.c (make_vector_type): Remove temporary shim and make
4892 the real function take the number of units as a poly_uint64
4894 (build_vector_type_for_mode): Handle polynomial nunits.
4895 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4896 * emit-rtl.c (const_vec_series_p_1): Likewise.
4897 (gen_rtx_CONST_VECTOR): Likewise.
4898 * fold-const.c (test_vec_duplicate_folding): Likewise.
4899 * genrecog.c (validate_pattern): Likewise.
4900 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4901 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4902 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4903 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4904 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4905 * rtlanal.c (subreg_get_info): Likewise.
4906 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4907 (vect_grouped_load_supported): Likewise.
4908 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4909 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4910 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4911 (simplify_const_unary_operation, simplify_binary_operation_1)
4912 (simplify_const_binary_operation, simplify_ternary_operation)
4913 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4914 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4915 instead of CONST_VECTOR_NUNITS.
4916 * varasm.c (output_constant_pool_2): Likewise.
4917 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4918 explicit-encoded elements in the XVEC for variable-length vectors.
4920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4922 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4925 Alan Hayward <alan.hayward@arm.com>
4926 David Sherwood <david.sherwood@arm.com>
4928 * coretypes.h (fixed_size_mode): Declare.
4929 (fixed_size_mode_pod): New typedef.
4930 * builtins.h (target_builtins::x_apply_args_mode)
4931 (target_builtins::x_apply_result_mode): Change type to
4932 fixed_size_mode_pod.
4933 * builtins.c (apply_args_size, apply_result_size, result_vector)
4934 (expand_builtin_apply_args_1, expand_builtin_apply)
4935 (expand_builtin_return): Update accordingly.
4937 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4939 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4940 * cselib.c (cselib_hash_rtx): Likewise.
4941 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4942 CONST_VECTOR encoding.
4944 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4945 Jeff Law <law@redhat.com>
4948 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4949 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4950 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4951 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4954 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4955 explicitly probe *sp in a noreturn function if there were any callee
4956 register saves or frame pointer is needed.
4958 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4961 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4962 BLKmode for ternary, binary or unary expressions.
4965 * var-tracking.c (delete_vta_debug_insn): New inline function.
4966 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4967 insns from get_insns () to NULL instead of each bb separately.
4968 Use delete_vta_debug_insn. No longer static.
4969 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4970 delete_vta_debug_insns callers.
4971 * rtl.h (delete_vta_debug_insns): Declare.
4972 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4973 instead of variable_tracking_main.
4975 2018-01-03 Martin Sebor <msebor@redhat.com>
4977 PR tree-optimization/83603
4978 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4979 arguments past the endof the argument list in functions declared
4980 without a prototype.
4981 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4982 Avoid checking when arguments are null.
4984 2018-01-03 Martin Sebor <msebor@redhat.com>
4987 * doc/extend.texi (attribute const): Fix a typo.
4988 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4989 issuing -Wsuggest-attribute for void functions.
4991 2018-01-03 Martin Sebor <msebor@redhat.com>
4993 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4994 offset_int::from instead of wide_int::to_shwi.
4995 (maybe_diag_overlap): Remove assertion.
4996 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4997 * gimple-ssa-sprintf.c (format_directive): Same.
4998 (parse_directive): Same.
4999 (sprintf_dom_walker::compute_format_length): Same.
5000 (try_substitute_return_value): Same.
5002 2018-01-03 Jeff Law <law@redhat.com>
5005 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5006 non-constant residual for zero at runtime and avoid probing in
5007 that case. Reorganize code for trailing problem to mirror handling
5010 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5012 PR tree-optimization/83501
5013 * tree-ssa-strlen.c (get_string_cst): New.
5014 (handle_char_store): Call get_string_cst.
5016 2018-01-03 Martin Liska <mliska@suse.cz>
5018 PR tree-optimization/83593
5019 * tree-ssa-strlen.c: Include tree-cfg.h.
5020 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5021 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5022 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5024 (strlen_dom_walker::before_dom_children): Call
5025 gimple_purge_dead_eh_edges. Dump tranformation with details
5027 (strlen_dom_walker::before_dom_children): Update call by adding
5028 new argument cleanup_eh.
5029 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5031 2018-01-03 Martin Liska <mliska@suse.cz>
5034 * cif-code.def (VARIADIC_THUNK): New enum value.
5035 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5038 2018-01-03 Jan Beulich <jbeulich@suse.com>
5040 * sse.md (mov<mode>_internal): Tighten condition for when to use
5041 vmovdqu<ssescalarsize> for TI and OI modes.
5043 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5045 Update copyright years.
5047 2018-01-03 Martin Liska <mliska@suse.cz>
5050 * ipa-visibility.c (function_and_variable_visibility): Skip
5051 functions with noipa attribure.
5053 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5055 * gcc.c (process_command): Update copyright notice dates.
5056 * gcov-dump.c (print_version): Ditto.
5057 * gcov.c (print_version): Ditto.
5058 * gcov-tool.c (print_version): Ditto.
5059 * gengtype.c (create_file): Ditto.
5060 * doc/cpp.texi: Bump @copying's copyright year.
5061 * doc/cppinternals.texi: Ditto.
5062 * doc/gcc.texi: Ditto.
5063 * doc/gccint.texi: Ditto.
5064 * doc/gcov.texi: Ditto.
5065 * doc/install.texi: Ditto.
5066 * doc/invoke.texi: Ditto.
5068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 * vector-builder.h (vector_builder::m_full_nelts): Change from
5071 unsigned int to poly_uint64.
5072 (vector_builder::full_nelts): Update prototype accordingly.
5073 (vector_builder::new_vector): Likewise.
5074 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5075 (vector_builder::operator ==): Likewise.
5076 (vector_builder::finalize): Likewise.
5077 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5078 Take the number of elements as a poly_uint64 rather than an
5080 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5081 from unsigned int to poly_uint64.
5082 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5083 (vec_perm_indices::new_vector): Likewise.
5084 (vec_perm_indices::length): Likewise.
5085 (vec_perm_indices::nelts_per_input): Likewise.
5086 (vec_perm_indices::input_nelts): Likewise.
5087 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5088 number of elements per input as a poly_uint64 rather than an
5089 unsigned int. Use the original encoding for variable-length
5090 vectors, rather than clamping each individual element.
5091 For the second and subsequent elements in each pattern,
5092 clamp the step and base before clamping their sum.
5093 (vec_perm_indices::series_p): Handle polynomial element counts.
5094 (vec_perm_indices::all_in_range_p): Likewise.
5095 (vec_perm_indices_to_tree): Likewise.
5096 (vec_perm_indices_to_rtx): Likewise.
5097 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5098 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5099 (tree_vector_builder::new_binary_operation): Handle polynomial
5100 element counts. Return false if we need to know the number
5101 of elements at compile time.
5102 * fold-const.c (fold_vec_perm): Punt if the number of elements
5103 isn't known at compile time.
5105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5107 * vec-perm-indices.h (vec_perm_builder): Change element type
5108 from HOST_WIDE_INT to poly_int64.
5109 (vec_perm_indices::element_type): Update accordingly.
5110 (vec_perm_indices::clamp): Handle polynomial element_types.
5111 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5112 (vec_perm_indices::all_in_range_p): Likewise.
5113 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5115 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5116 polynomial vec_perm_indices element types.
5117 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5118 * fold-const.c (fold_vec_perm): Likewise.
5119 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5120 * tree-vect-generic.c (lower_vec_perm): Likewise.
5121 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5122 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5123 element type to HOST_WIDE_INT.
5125 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5126 Alan Hayward <alan.hayward@arm.com>
5127 David Sherwood <david.sherwood@arm.com>
5129 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5130 rather than an int. Use plus_constant.
5131 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5132 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5135 Alan Hayward <alan.hayward@arm.com>
5136 David Sherwood <david.sherwood@arm.com>
5138 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5139 a HOST_WIDE_INT to a poly_int64.
5141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5142 Alan Hayward <alan.hayward@arm.com>
5143 David Sherwood <david.sherwood@arm.com>
5145 * calls.c (load_register_parameters): Cope with polynomial
5146 mode sizes. Require a constant size for BLKmode parameters
5147 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5148 forces a parameter to be padded at the lsb end in order to
5149 fill a complete number of words, require the parameter size
5150 to be ordered wrt UNITS_PER_WORD.
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 * reload1.c (spill_stack_slot_width): Change element type
5157 from unsigned int to poly_uint64_pod.
5158 (alter_reg): Treat mode sizes as polynomial.
5160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5161 Alan Hayward <alan.hayward@arm.com>
5162 David Sherwood <david.sherwood@arm.com>
5164 * reload.c (complex_word_subreg_p): New function.
5165 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
5172 operands whose sizes aren't ordered.
5173 (match_reload): Refer to this check here.
5175 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5176 Alan Hayward <alan.hayward@arm.com>
5177 David Sherwood <david.sherwood@arm.com>
5179 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5180 that the mode size is in the set {1, 2, 4, 8, 16}.
5182 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5183 Alan Hayward <alan.hayward@arm.com>
5184 David Sherwood <david.sherwood@arm.com>
5186 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5187 Use plus_constant instead of gen_rtx_PLUS.
5189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5190 Alan Hayward <alan.hayward@arm.com>
5191 David Sherwood <david.sherwood@arm.com>
5193 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5194 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5195 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5196 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5197 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5198 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5199 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5200 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5201 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5202 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5204 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5205 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5206 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5207 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5208 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5209 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5210 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5211 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5212 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5213 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5215 * expr.c (emit_move_resolve_push): Treat the input and result
5216 of PUSH_ROUNDING as a poly_int64.
5217 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5218 (emit_push_insn): Likewise.
5219 * lra-eliminations.c (mark_not_eliminable): Likewise.
5220 * recog.c (push_operand): Likewise.
5221 * reload1.c (elimination_effects): Likewise.
5222 * rtlanal.c (nonzero_bits1): Likewise.
5223 * calls.c (store_one_arg): Likewise. Require the padding to be
5224 known at compile time.
5226 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5227 Alan Hayward <alan.hayward@arm.com>
5228 David Sherwood <david.sherwood@arm.com>
5230 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5231 Use plus_constant instead of gen_rtx_PLUS.
5233 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5234 Alan Hayward <alan.hayward@arm.com>
5235 David Sherwood <david.sherwood@arm.com>
5237 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5245 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5246 via stack temporaries. Treat the mode size as polynomial too.
5248 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5249 Alan Hayward <alan.hayward@arm.com>
5250 David Sherwood <david.sherwood@arm.com>
5252 * expr.c (expand_expr_real_2): When handling conversions involving
5253 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5254 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5255 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5263 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5264 Alan Hayward <alan.hayward@arm.com>
5265 David Sherwood <david.sherwood@arm.com>
5267 * combine.c (can_change_dest_mode): Handle polynomial
5268 REGMODE_NATURAL_SIZE.
5269 * expmed.c (store_bit_field_1): Likewise.
5270 * expr.c (store_constructor): Likewise.
5271 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5272 and polynomial REGMODE_NATURAL_SIZE.
5273 (gen_lowpart_common): Likewise.
5274 * reginfo.c (record_subregs_of_mode): Likewise.
5275 * rtlanal.c (read_modify_subreg_p): Likewise.
5277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5278 Alan Hayward <alan.hayward@arm.com>
5279 David Sherwood <david.sherwood@arm.com>
5281 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5282 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
5290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5291 Alan Hayward <alan.hayward@arm.com>
5292 David Sherwood <david.sherwood@arm.com>
5294 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5295 in a POINTER_PLUS_EXPR.
5297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5298 Alan Hayward <alan.hayward@arm.com>
5299 David Sherwood <david.sherwood@arm.com>
5301 * omp-simd-clone.c (simd_clone_subparts): New function.
5302 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5303 (ipa_simd_modify_function_body): Likewise.
5305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 Alan Hayward <alan.hayward@arm.com>
5307 David Sherwood <david.sherwood@arm.com>
5309 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5310 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5311 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5312 (expand_vector_condition, vector_element): Likewise.
5313 (subparts_gt): New function.
5314 (get_compute_type): Use subparts_gt.
5315 (count_type_subparts): Delete.
5316 (expand_vector_operations_1): Use subparts_gt instead of
5317 count_type_subparts.
5319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5320 Alan Hayward <alan.hayward@arm.com>
5321 David Sherwood <david.sherwood@arm.com>
5323 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5324 (vect_compile_time_alias): ...this new function. Do the calculation
5325 on poly_ints rather than trees.
5326 (vect_prune_runtime_alias_test_list): Update call accordingly.
5328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5329 Alan Hayward <alan.hayward@arm.com>
5330 David Sherwood <david.sherwood@arm.com>
5332 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5334 (vect_schedule_slp_instance): Likewise.
5336 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5337 Alan Hayward <alan.hayward@arm.com>
5338 David Sherwood <david.sherwood@arm.com>
5340 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5341 constant and extern definitions for variable-length vectors.
5342 (vect_get_constant_vectors): Note that the number of units
5343 is known to be constant.
5345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5346 Alan Hayward <alan.hayward@arm.com>
5347 David Sherwood <david.sherwood@arm.com>
5349 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5350 of units as polynomial. Choose between WIDE and NARROW based
5353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5354 Alan Hayward <alan.hayward@arm.com>
5355 David Sherwood <david.sherwood@arm.com>
5357 * tree-vect-stmts.c (simd_clone_subparts): New function.
5358 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5365 vectors as polynomial. Use build_index_vector for
5368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5369 Alan Hayward <alan.hayward@arm.com>
5370 David Sherwood <david.sherwood@arm.com>
5372 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5373 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5374 for variable-length vectors.
5375 (vectorizable_mask_load_store): Treat the number of units as
5376 polynomial, asserting that it is constant if the condition has
5377 already been enforced.
5378 (vectorizable_store, vectorizable_load): Likewise.
5380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5381 Alan Hayward <alan.hayward@arm.com>
5382 David Sherwood <david.sherwood@arm.com>
5384 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5385 of units as polynomial. Punt if we can't tell at compile time
5386 which vector contains the final result.
5388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5389 Alan Hayward <alan.hayward@arm.com>
5390 David Sherwood <david.sherwood@arm.com>
5392 * tree-vect-loop.c (vectorizable_induction): Treat the number
5393 of units as polynomial. Punt on SLP inductions. Use an integer
5394 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5395 cast of such a series for variable-length floating-point
5398 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5399 Alan Hayward <alan.hayward@arm.com>
5400 David Sherwood <david.sherwood@arm.com>
5402 * tree.h (build_index_vector): Declare.
5403 * tree.c (build_index_vector): New function.
5404 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5405 of units as polynomial, forcibly converting it to a constant if
5406 vectorizable_reduction has already enforced the condition.
5407 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5408 to create a {1,2,3,...} vector.
5409 (vectorizable_reduction): Treat the number of units as polynomial.
5410 Choose vectype_in based on the largest scalar element size rather
5411 than the smallest number of units. Enforce the restrictions
5414 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5415 Alan Hayward <alan.hayward@arm.com>
5416 David Sherwood <david.sherwood@arm.com>
5418 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5419 number of units as polynomial.
5421 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5422 Alan Hayward <alan.hayward@arm.com>
5423 David Sherwood <david.sherwood@arm.com>
5425 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5426 * target.def (autovectorize_vector_sizes): Return the vector sizes
5427 by pointer, using vector_sizes rather than a bitmask.
5428 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5429 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5430 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5432 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5433 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5434 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5435 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5436 * omp-general.c (omp_max_vf): Likewise.
5437 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5438 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5439 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5440 * tree-vect-slp.c (vect_slp_bb): Likewise.
5441 * doc/tm.texi: Regenerate.
5442 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5444 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5445 the vector size as a poly_uint64 rather than an unsigned int.
5446 (current_vector_size): Change from an unsigned int to a poly_uint64.
5447 (get_vectype_for_scalar_type): Update accordingly.
5448 * tree.h (build_truth_vector_type): Take the size and number of
5449 units as a poly_uint64 rather than an unsigned int.
5450 (build_vector_type): Add a temporary overload that takes
5451 the number of units as a poly_uint64 rather than an unsigned int.
5452 * tree.c (make_vector_type): Likewise.
5453 (build_truth_vector_type): Take the number of units as a poly_uint64
5454 rather than an unsigned int.
5456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5457 Alan Hayward <alan.hayward@arm.com>
5458 David Sherwood <david.sherwood@arm.com>
5460 * target.def (get_mask_mode): Take the number of units and length
5461 as poly_uint64s rather than unsigned ints.
5462 * targhooks.h (default_get_mask_mode): Update accordingly.
5463 * targhooks.c (default_get_mask_mode): Likewise.
5464 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5465 * doc/tm.texi: Regenerate.
5467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5468 Alan Hayward <alan.hayward@arm.com>
5469 David Sherwood <david.sherwood@arm.com>
5471 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5472 * omp-general.c (omp_max_vf): Likewise.
5473 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5474 (expand_omp_simd): Handle polynomial safelen.
5475 * omp-low.c (omplow_simd_context): Add a default constructor.
5476 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5477 (lower_rec_simd_input_clauses): Update accordingly.
5478 (lower_rec_input_clauses): Likewise.
5480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5481 Alan Hayward <alan.hayward@arm.com>
5482 David Sherwood <david.sherwood@arm.com>
5484 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5485 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5486 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5487 (vect_analyze_slp_cost): Likewise.
5488 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5489 (vect_model_load_cost): Likewise.
5491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5492 Alan Hayward <alan.hayward@arm.com>
5493 David Sherwood <david.sherwood@arm.com>
5495 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5496 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5497 from an unsigned int * to a poly_uint64_pod *.
5498 (calculate_unrolling_factor): New function.
5499 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5501 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5502 Alan Hayward <alan.hayward@arm.com>
5503 David Sherwood <david.sherwood@arm.com>
5505 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5506 from an unsigned int to a poly_uint64.
5507 (_loop_vec_info::slp_unrolling_factor): Likewise.
5508 (_loop_vec_info::vectorization_factor): Change from an int
5510 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5511 (vect_get_num_vectors): New function.
5512 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5513 (vect_get_num_copies): Use vect_get_num_vectors.
5514 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5515 to an unsigned int *.
5516 (vect_analyze_data_refs): Change min_vf from an int * to a
5518 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5519 than an unsigned HOST_WIDE_INT.
5520 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5521 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5522 to an unsigned int *.
5523 (vect_analyze_data_ref_dependences): Likewise.
5524 (vect_compute_data_ref_alignment): Handle polynomial vf.
5525 (vect_enhance_data_refs_alignment): Likewise.
5526 (vect_prune_runtime_alias_test_list): Likewise.
5527 (vect_shift_permute_load_chain): Likewise.
5528 (vect_supportable_dr_alignment): Likewise.
5529 (dependence_distance_ge_vf): Take the vectorization factor as a
5530 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5531 (vect_analyze_data_refs): Change min_vf from an int * to a
5533 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5534 vfm1 as a poly_uint64 rather than an int. Make the same change
5535 for the returned bound_scalar.
5536 (vect_gen_vector_loop_niters): Handle polynomial vf.
5537 (vect_do_peeling): Likewise. Update call to
5538 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5539 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5541 * tree-vect-loop.c (vect_determine_vectorization_factor)
5542 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5543 (vect_get_known_peeling_cost): Likewise.
5544 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5545 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5546 (vect_transform_loop): Likewise. Use the lowest possible VF when
5547 updating the upper bounds of the loop.
5548 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5550 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5551 polynomial unroll factors.
5552 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5553 (vect_make_slp_decision): Likewise.
5554 (vect_supported_load_permutation_p): Likewise, and polynomial
5556 (vect_analyze_slp_cost): Handle polynomial vf.
5557 (vect_slp_analyze_node_operations): Likewise.
5558 (vect_slp_analyze_bb_1): Likewise.
5559 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5560 than an unsigned HOST_WIDE_INT.
5561 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5562 (vectorizable_load): Handle polynomial vf.
5563 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5565 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5568 Alan Hayward <alan.hayward@arm.com>
5569 David Sherwood <david.sherwood@arm.com>
5571 * match.pd: Handle bit operations involving three constants
5572 and try to fold one pair.
5574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5576 * tree-vect-loop-manip.c: Include gimple-fold.h.
5577 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5578 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5579 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5580 Add a path that uses a step of VF instead of 1, but disable it
5582 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5583 and niters_no_overflow parameters. Update calls to
5584 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5585 Create a new SSA name if the latter choses to use a ste other
5586 than zero, and return it via niters_vector_mult_vf_var.
5587 * tree-vect-loop.c (vect_transform_loop): Update calls to
5588 vect_do_peeling, vect_gen_vector_loop_niters and
5589 slpeel_make_loop_iterate_ntimes.
5590 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5591 (vect_gen_vector_loop_niters): Update declarations after above changes.
5593 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5595 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5596 128-bit round to integer instructions.
5597 (ceil<mode>2): Likewise.
5598 (btrunc<mode>2): Likewise.
5599 (round<mode>2): Likewise.
5601 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5603 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5604 unaligned VSX load/store on P8/P9.
5605 (expand_block_clear): Allow the use of unaligned VSX
5606 load/store on P8/P9.
5608 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5610 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5612 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5613 swap associated with both a load and a store.
5615 2018-01-02 Andrew Waterman <andrew@sifive.com>
5617 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5618 * config/riscv/riscv.md (clear_cache): Use it.
5620 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5622 * web.c: Remove out-of-date comment.
5624 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5626 * expr.c (fixup_args_size_notes): Check that any existing
5627 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5628 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5629 (emit_single_push_insn): ...here.
5631 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5633 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5634 (const_vector_encoded_nelts): New function.
5635 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5636 (const_vector_int_elt, const_vector_elt): Declare.
5637 * emit-rtl.c (const_vector_int_elt_1): New function.
5638 (const_vector_elt): Likewise.
5639 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5640 of CONST_VECTOR_ELT.
5642 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5644 * expr.c: Include rtx-vector-builder.h.
5645 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5646 directly on the tree encoding.
5647 (const_vector_from_tree): Likewise.
5648 * optabs.c: Include rtx-vector-builder.h.
5649 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5650 sequence of "u" values.
5651 * vec-perm-indices.c: Include rtx-vector-builder.h.
5652 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5653 directly on the vec_perm_indices encoding.
5655 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5657 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5658 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5659 * rtx-vector-builder.h: New file.
5660 * rtx-vector-builder.c: Likewise.
5661 * rtl.h (rtx_def::u2): Add a const_vector field.
5662 (CONST_VECTOR_NPATTERNS): New macro.
5663 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5664 (CONST_VECTOR_DUPLICATE_P): Likewise.
5665 (CONST_VECTOR_STEPPED_P): Likewise.
5666 (CONST_VECTOR_ENCODED_ELT): Likewise.
5667 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5668 (unwrap_const_vec_duplicate): Likewise.
5669 (const_vec_series_p): Check for a non-duplicated vector encoding.
5670 Say that the function only returns true for integer vectors.
5671 * emit-rtl.c: Include rtx-vector-builder.h.
5672 (gen_const_vec_duplicate_1): Delete.
5673 (gen_const_vector): Call gen_const_vec_duplicate instead of
5674 gen_const_vec_duplicate_1.
5675 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5676 (gen_const_vec_duplicate): Use rtx_vector_builder.
5677 (gen_const_vec_series): Likewise.
5678 (gen_rtx_CONST_VECTOR): Likewise.
5679 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5680 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5681 Build a new vector rather than modifying a CONST_VECTOR in-place.
5682 (handle_special_swappables): Update call accordingly.
5683 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5684 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5685 Build a new vector rather than modifying a CONST_VECTOR in-place.
5686 (handle_special_swappables): Update call accordingly.
5688 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5690 * simplify-rtx.c (simplify_const_binary_operation): Use
5691 CONST_VECTOR_ELT instead of XVECEXP.
5693 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5695 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5696 the selector elements to be different from the data elements
5697 if the selector is a VECTOR_CST.
5698 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5699 ssizetype for the selector.
5701 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5703 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5704 before testing each element individually.
5705 * tree-vect-generic.c (lower_vec_perm): Likewise.
5707 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5709 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5710 * selftest-run-tests.c (selftest::run_tests): Call it.
5711 * vector-builder.h (vector_builder::operator ==): New function.
5712 (vector_builder::operator !=): Likewise.
5713 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5714 (vec_perm_indices::all_from_input_p): New function.
5715 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5716 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5717 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5718 instead of reading the VECTOR_CST directly. Detect whether both
5719 vector inputs are the same before constructing the vec_perm_indices,
5720 and update the number of inputs argument accordingly. Use the
5721 utility functions added above. Only construct sel2 if we need to.
5723 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5725 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5726 the broadcast of the low byte.
5727 (expand_mult_highpart): Use an explicit encoding for the permutes.
5728 * optabs-query.c (can_mult_highpart_p): Likewise.
5729 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5730 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5731 (vectorizable_bswap): Likewise.
5732 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5733 explicit encoding for the power-of-2 permutes.
5734 (vect_permute_store_chain): Likewise.
5735 (vect_grouped_load_supported): Likewise.
5736 (vect_permute_load_chain): Likewise.
5738 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5740 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5741 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5742 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5743 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5744 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5745 (vect_gen_perm_mask_any): Likewise.
5747 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5749 * int-vector-builder.h: New file.
5750 * vec-perm-indices.h: Include int-vector-builder.h.
5751 (vec_perm_indices): Redefine as an int_vector_builder.
5752 (auto_vec_perm_indices): Delete.
5753 (vec_perm_builder): Redefine as a stand-alone class.
5754 (vec_perm_indices::vec_perm_indices): New function.
5755 (vec_perm_indices::clamp): Likewise.
5756 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5757 (vec_perm_indices::new_vector): New function.
5758 (vec_perm_indices::new_expanded_vector): Update for new
5759 vec_perm_indices class.
5760 (vec_perm_indices::rotate_inputs): New function.
5761 (vec_perm_indices::all_in_range_p): Operate directly on the
5762 encoded form, without computing elided elements.
5763 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5764 encoding. Update for new vec_perm_indices class.
5765 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5766 the given vec_perm_builder.
5767 (expand_vec_perm_var): Update vec_perm_builder constructor.
5768 (expand_mult_highpart): Use vec_perm_builder instead of
5769 auto_vec_perm_indices.
5770 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5771 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5772 or double series encoding as appropriate.
5773 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5774 vec_perm_indices instead of auto_vec_perm_indices.
5775 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5776 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5777 (vect_permute_store_chain): Likewise.
5778 (vect_grouped_load_supported): Likewise.
5779 (vect_permute_load_chain): Likewise.
5780 (vect_shift_permute_load_chain): Likewise.
5781 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5782 (vect_transform_slp_perm_load): Likewise.
5783 (vect_schedule_slp_instance): Likewise.
5784 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5785 (vectorizable_mask_load_store): Likewise.
5786 (vectorizable_bswap): Likewise.
5787 (vectorizable_store): Likewise.
5788 (vectorizable_load): Likewise.
5789 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5790 vec_perm_indices instead of auto_vec_perm_indices. Use
5791 tree_to_vec_perm_builder to read the vector from a tree.
5792 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5793 vec_perm_builder instead of a vec_perm_indices.
5794 (have_whole_vector_shift): Use vec_perm_builder and
5795 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5796 truncation to calc_vec_perm_mask_for_shift.
5797 (vect_create_epilog_for_reduction): Likewise.
5798 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5799 from auto_vec_perm_indices to vec_perm_indices.
5800 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5801 instead of changing individual elements.
5802 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5803 the vector in d.perm.
5804 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5805 from auto_vec_perm_indices to vec_perm_indices.
5806 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5807 instead of changing individual elements.
5808 (arm_vectorize_vec_perm_const): Use new_vector to install
5809 the vector in d.perm.
5810 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5811 Update vec_perm_builder constructor.
5812 (rs6000_expand_interleave): Likewise.
5813 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5814 (rs6000_expand_interleave): Likewise.
5816 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5818 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5819 to qimode could truncate the indices.
5820 * optabs.c (expand_vec_perm_var): Likewise.
5822 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5824 * Makefile.in (OBJS): Add vec-perm-indices.o.
5825 * vec-perm-indices.h: New file.
5826 * vec-perm-indices.c: Likewise.
5827 * target.h (vec_perm_indices): Replace with a forward class
5829 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5830 * optabs.h: Include vec-perm-indices.h.
5831 (expand_vec_perm): Delete.
5832 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5833 (expand_vec_perm_const): Declare.
5834 * target.def (vec_perm_const_ok): Replace with...
5835 (vec_perm_const): ...this new hook.
5836 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5837 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5838 * doc/tm.texi: Regenerate.
5839 * optabs.def (vec_perm_const): Delete.
5840 * doc/md.texi (vec_perm_const): Likewise.
5841 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5842 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5843 expand_vec_perm for constant permutation vectors. Assert that
5844 the mode of variable permutation vectors is the integer equivalent
5845 of the mode that is being permuted.
5846 * optabs-query.h (selector_fits_mode_p): Declare.
5847 * optabs-query.c: Include vec-perm-indices.h.
5848 (selector_fits_mode_p): New function.
5849 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5850 is defined, instead of checking whether the vec_perm_const_optab
5851 exists. Use targetm.vectorize.vec_perm_const instead of
5852 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5853 fit in the vector mode before using a variable permute.
5854 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5855 vec_perm_indices instead of an rtx.
5856 (expand_vec_perm): Replace with...
5857 (expand_vec_perm_const): ...this new function. Take the selector
5858 as a vec_perm_indices rather than an rtx. Also take the mode of
5859 the selector. Update call to shift_amt_for_vec_perm_mask.
5860 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5861 Use vec_perm_indices::new_expanded_vector to expand the original
5862 selector into bytes. Check whether the indices fit in the vector
5863 mode before using a variable permute.
5864 (expand_vec_perm_var): Make global.
5865 (expand_mult_highpart): Use expand_vec_perm_const.
5866 * fold-const.c: Includes vec-perm-indices.h.
5867 * tree-ssa-forwprop.c: Likewise.
5868 * tree-vect-data-refs.c: Likewise.
5869 * tree-vect-generic.c: Likewise.
5870 * tree-vect-loop.c: Likewise.
5871 * tree-vect-slp.c: Likewise.
5872 * tree-vect-stmts.c: Likewise.
5873 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5875 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5876 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5877 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5878 (aarch64_vectorize_vec_perm_const): ...this new function.
5879 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5880 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5881 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5882 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5883 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5884 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5885 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5887 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5888 check for NEON modes.
5889 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5890 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5891 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5892 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5894 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5895 the old VEC_PERM_CONST conditions.
5896 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5897 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5898 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5899 (ia64_vectorize_vec_perm_const_ok): Merge into...
5900 (ia64_vectorize_vec_perm_const): ...this new function.
5901 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5902 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5903 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5904 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5905 * config/mips/mips.c (mips_expand_vec_perm_const)
5906 (mips_vectorize_vec_perm_const_ok): Merge into...
5907 (mips_vectorize_vec_perm_const): ...this new function.
5908 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5909 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5910 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5911 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5912 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5913 (rs6000_expand_vec_perm_const): Delete.
5914 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5916 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5917 (altivec_expand_vec_perm_const_le): Take each operand individually.
5918 Operate on constant selectors rather than rtxes.
5919 (altivec_expand_vec_perm_const): Likewise. Update call to
5920 altivec_expand_vec_perm_const_le.
5921 (rs6000_expand_vec_perm_const): Delete.
5922 (rs6000_vectorize_vec_perm_const_ok): Delete.
5923 (rs6000_vectorize_vec_perm_const): New function.
5924 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5925 an element count and rtx array.
5926 (rs6000_expand_extract_even): Update call accordingly.
5927 (rs6000_expand_interleave): Likewise.
5928 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5929 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5930 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5931 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5932 (rs6000_expand_vec_perm_const): Delete.
5933 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5934 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5935 (altivec_expand_vec_perm_const_le): Take each operand individually.
5936 Operate on constant selectors rather than rtxes.
5937 (altivec_expand_vec_perm_const): Likewise. Update call to
5938 altivec_expand_vec_perm_const_le.
5939 (rs6000_expand_vec_perm_const): Delete.
5940 (rs6000_vectorize_vec_perm_const_ok): Delete.
5941 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5942 reference to the SPE evmerge intructions.
5943 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5944 an element count and rtx array.
5945 (rs6000_expand_extract_even): Update call accordingly.
5946 (rs6000_expand_interleave): Likewise.
5947 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5948 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5950 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5952 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5954 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5955 vector mode and that that mode matches the mode of the data
5957 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5958 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5959 directly using expand_vec_perm_1 when forcing selectors into
5961 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5963 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5965 * optabs-query.h (can_vec_perm_p): Delete.
5966 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5967 * optabs-query.c (can_vec_perm_p): Split into...
5968 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5969 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5970 particular selector is valid.
5971 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5972 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5973 (vect_grouped_load_supported): Likewise.
5974 (vect_shift_permute_load_chain): Likewise.
5975 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5976 (vect_transform_slp_perm_load): Likewise.
5977 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5978 (vectorizable_bswap): Likewise.
5979 (vect_gen_perm_mask_checked): Likewise.
5980 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5981 implementations of variable permutation vectors into account
5982 when deciding which selector to use.
5983 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5984 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5985 with a false third argument.
5986 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5987 to test whether the constant selector is valid and can_vec_perm_var_p
5988 to test whether a variable selector is valid.
5990 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5992 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5993 * optabs-query.c (can_vec_perm_p): Likewise.
5994 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5995 instead of vec_perm_indices.
5996 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5997 (vect_gen_perm_mask_checked): Likewise,
5998 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5999 (vect_gen_perm_mask_checked): Likewise,
6001 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6003 * optabs-query.h (qimode_for_vec_perm): Declare.
6004 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6005 (qimode_for_vec_perm): ...this new function.
6006 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6008 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6010 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6011 does not have a conditional at the top.
6013 2018-01-02 Richard Biener <rguenther@suse.de>
6015 * ipa-inline.c (big_speedup_p): Fix expression.
6017 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6020 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6023 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6027 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6028 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6029 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6030 cond_taken_branch_cost 3->4.
6032 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6034 PR tree-optimization/83581
6035 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6036 TODO_cleanup_cfg if any changes have been made.
6039 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6040 convert_modes if target mode has the right side, but different mode
6044 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6045 last argument when extracting from CONCAT. If either from_real or
6046 from_imag is NULL, use expansion through memory. If result is not
6047 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6048 the parts directly to inner mode, if even that fails, use expansion
6052 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6053 check for bswap in mode rather than HImode and use that in expand_unop
6056 Copyright (C) 2018 Free Software Foundation, Inc.
6058 Copying and distribution of this file, with or without modification,
6059 are permitted in any medium without royalty provided the copyright
6060 notice and this notice are preserved.