1 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
3 * ira.c (ira_init_register_move_cost): Adjust comment.
5 2018-02-05 Martin Liska <mliska@suse.cz>
8 * doc/gcov.texi: Fix typo in documentation.
10 2018-02-05 Martin Liska <mliska@suse.cz>
13 * doc/gcov.texi: Document necessity of --dynamic-list-data when
14 using dlopen functionality.
16 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
18 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
19 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
20 _mm_maskz_range_ss, _mm_mask_range_round_ss,
21 _mm_maskz_range_round_ss): New intrinsics.
22 (__builtin_ia32_rangesd128_round)
23 (__builtin_ia32_rangess128_round): Remove.
24 (__builtin_ia32_rangesd128_mask_round,
25 __builtin_ia32_rangess128_mask_round): New builtins.
26 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
27 __builtin_ia32_rangess128_round): Remove.
28 (__builtin_ia32_rangesd128_mask_round,
29 __builtin_ia32_rangess128_mask_round): New builtins.
30 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
31 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
32 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
33 "<round_saeonly_constraint>")): Changed to ...
34 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
35 "<round_saeonly_scalar_constraint>")): ... this.
36 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
37 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
38 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
39 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
40 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
42 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
44 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
46 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
47 Remove all values except native, 8540 and 8548.
49 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
51 * config/i386/i386.c (ix86_output_function_return): Pass
52 INVALID_REGNUM, instead of -1, as invalid register number to
53 indirect_thunk_name and output_indirect_thunk.
55 2018-02-02 Julia Koval <julia.koval@intel.com>
57 * config.gcc: Add -march=icelake.
58 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
59 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
60 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
61 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
62 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
63 (processor_target_table): Add icelake.
64 (ix86_option_override_internal): Handle new PTAs.
65 (get_builtin_code_for_version): Handle icelake.
66 (M_INTEL_COREI7_ICELAKE): New.
67 (fold_builtin_cpu): Handle icelake.
68 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
69 * doc/invoke.texi: Add -march=icelake.
71 2018-02-02 Julia Koval <julia.koval@intel.com>
73 * config/i386/i386.c (ix86_option_override_internal): Change flags type
75 * wide-int-bitmask.h: New.
77 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
80 * config/i386/i386.md: Replace Pmode with word_mode in
81 builtin_setjmp_setup and builtin_longjmp to support x32.
83 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
87 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
89 (rs6000_supported_cpu_names): New static variable.
90 (linux_cpu_translation_table): Likewise.
91 (elf_platform) <cpu>: Define new static variable and use it.
92 Translate kernel AT_PLATFORM name to canonical name if needed.
93 Error if platform name is unknown.
95 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
98 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
100 2018-02-01 Jeff Law <law@redhat.com>
103 * config/i386/i386.c (release_scratch_register_on_entry): Add new
104 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
105 the scratch if RELEASE_VIA_POP is false.
106 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
107 If we have to save a temporary register, decrement SIZE appropriately.
108 Pass new arguments to release_scratch_register_on_entry.
109 (ix86_adjust_stack_and_probe): Likewise.
110 (ix86_emit_probe_stack_range): Pass new arguments to
111 release_scratch_register_on_entry.
113 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
115 PR rtl-optimization/84157
116 * combine.c (change_zero_ext): Use REG_P predicate in
117 front of HARD_REGISTER_P predicate.
119 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
121 * config/avr/avr.c (avr_option_override): Move disabling of
122 -fdelete-null-pointer-checks to...
123 * common/config/avr/avr-common.c (avr_option_optimization_table):
126 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
128 PR tree-optimization/81635
129 * tree-data-ref.c (split_constant_offset_1): For types that
130 wrap on overflow, try to use range info to prove that wrapping
133 2018-02-01 Renlin Li <renlin.li@arm.com>
136 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
138 (aarch64_register_move_cost): Likewise.
139 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
141 (REG_CLASS_NAMES): Likewise.
142 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
143 TAILCALL_ADDR_REGS. Remove IP registers.
144 * config/aarch64/aarch64.md (Ucs): Update register constraint.
146 2018-02-01 Richard Biener <rguenther@suse.de>
148 * domwalk.h (dom_walker::dom_walker): Add additional constructor
149 for specifying RPO order and allow NULL for that.
150 * domwalk.c (dom_walker::dom_walker): Likewise.
151 (dom_walker::walk): Handle NULL RPO order.
152 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
154 (rewrite_update_dom_walker): Likewise.
155 (mark_def_dom_walker): Likewise.
157 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
159 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
160 (aarch64_maybe_expand_sve_subreg_move): Declare.
161 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
162 * config/aarch64/predicates.md (aarch64_any_register_operand): New
164 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
165 that are semantically a reverse operation.
166 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
167 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
168 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
170 (aarch64_can_change_mode_class): For big-endian, forbid changes
171 between two SVE modes if they have different element sizes.
173 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
175 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
176 the TImode handling for big-endian targets.
178 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
180 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
181 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
183 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
184 Remove BSWAP handing for big-endian targets and use the form of
185 LD1RQ appropariate for the mode.
187 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
189 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
190 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
193 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
196 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
197 check for operands that need to go through aarch64_sve_reload_be.
199 2018-02-01 Jakub Jelinek <jakub@redhat.com>
201 PR tree-optimization/81661
202 PR tree-optimization/84117
203 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
204 * tree-eh.c: Include gimplify.h.
205 (find_trapping_overflow, replace_trapping_overflow,
206 rewrite_to_non_trapping_overflow): New functions.
207 * tree-vect-loop.c: Include tree-eh.h.
208 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
209 * tree-data-ref.c: Include tree-eh.h.
210 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
212 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
214 PR rtl-optimization/84123
215 * combine.c (change_zero_ext): Check if hard register satisfies
216 can_change_dest_mode before calling gen_lowpart_SUBREG.
218 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
221 * ira.c (ira_init_register_move_cost): Remove assert.
223 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
225 PR rtl-optimization/84071
226 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
227 * doc/tm.texi: Regenerate.
229 2018-01-31 Richard Biener <rguenther@suse.de>
231 PR tree-optimization/84132
232 * tree-data-ref.c (analyze_miv_subscript): Properly
233 check whether evolution_function_is_affine_multivariate_p
234 before calling gcd_of_steps_may_divide_p.
236 2018-01-31 Julia Koval <julia.koval@intel.com>
239 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
240 * config/i386/i386.md (rdpid_rex64) New.
241 (rdpid): Make 32bit only.
243 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
246 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
247 an IDENTIFIER_NODE for FUNCTION_TYPE's.
249 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
252 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
254 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
256 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
258 PR rtl-optimization/84071
259 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
260 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
262 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
264 * config/arc/arc.c (arc_handle_aux_attribute): New function.
265 (arc_attribute_table): Add 'aux' attribute.
266 (arc_in_small_data_p): Consider aux like variables.
267 (arc_is_aux_reg_p): New function.
268 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
269 (arc_get_aux_arg): New function.
270 (prepare_move_operands): Handle aux-register access.
271 (arc_handle_aux_attribute): New function.
272 * doc/extend.texi (ARC Variable attributes): Add subsection.
274 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
276 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
277 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
278 (arc_attribute_table): Add 'uncached' attribute.
279 (arc_print_operand): Print '.di' flag for uncached memory
281 (arc_in_small_data_p): Do not consider for small data the uncached
283 (arc_is_uncached_mem_p): New function.
284 * config/arc/predicates.md (compact_store_memory_operand): Check
285 for uncached memory accesses.
286 (nonvol_nonimm_operand): Likewise.
287 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
289 2018-01-31 Jakub Jelinek <jakub@redhat.com>
292 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
293 falign-loops=): Add Optimization flag.
295 2018-01-30 Jeff Law <law@redhat.com>
298 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
299 INT_REGISTERS_SAVED. Check it prior to calling
300 get_scratch_register_on_entry.
301 (ix86_adjust_stack_and_probe): Similarly.
302 (ix86_emit_probe_stack_range): Similarly.
303 (ix86_expand_prologue): Corresponding changes.
305 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
308 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
309 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
311 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
314 * lra-constraints.c (curr_insn_transform): Process AND in the
317 2018-01-30 Jakub Jelinek <jakub@redhat.com>
319 PR rtl-optimization/83986
320 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
321 dependence against last_pending_memory_flush in addition to
324 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
326 PR tree-optimization/81611
327 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
330 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
332 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
335 2018-01-30 Richard Biener <rguenther@suse.de>
336 Jakub Jelinek <jakub@redhat.com>
338 PR tree-optimization/84111
339 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
340 inner loops added during recursion, as they don't have up-to-date
343 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
346 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
347 (can_inline_edge_by_limits_p): ... here.
348 (can_early_inline_edge_p, check_callers,
349 update_caller_keys, update_callee_keys, recursive_inlining,
350 add_new_edges_to_heap, speculation_useful_p,
351 inline_small_functions,
352 inline_small_functions, flatten_function,
353 inline_to_all_callers_1): Update.
355 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
357 * profile-count.c (profile_count::combine_with_ipa_count): Handle
360 2018-01-30 Richard Biener <rguenther@suse.de>
362 PR tree-optimization/83008
363 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
364 invariant and constant vector uses in stmts when they need
367 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
370 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
371 * configure: Regenerate.
373 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
375 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
377 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
378 Use gen_rtx_REG rather than gen_lowpart.
380 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
382 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
383 rather than 0 when creating partial subregs.
385 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
387 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
390 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
393 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
394 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
395 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
396 flags. This restores the settings used before the 2017-07-24.
397 Turning off pre increment/decrement/modify allows IVOPTS to
398 optimize DF/SF loops where the index is an int.
400 2018-01-29 Richard Biener <rguenther@suse.de>
401 Kelvin Nilsen <kelvin@gcc.gnu.org>
404 * tree-vect-stmts.c (vectorizable_call): Don't call
405 targetm.vectorize_builtin_md_vectorized_function if callee is
408 2018-01-22 Carl Love <cel@us.ibm.com>
410 * doc/extend.tex: Fix typo in second arg in
411 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
413 2018-01-29 Richard Biener <rguenther@suse.de>
415 PR tree-optimization/84086
416 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
417 (flush_ssaname_freelist): When SSA names were released reset
420 2018-01-29 Richard Biener <rguenther@suse.de>
422 PR tree-optimization/84057
423 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
424 removed paths when removing edges.
426 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
428 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
429 -mfunction-return=@var{choice}.
431 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
434 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
436 (layout::print_source_line): Likewise.
437 (test_get_line_width_without_trailing_whitespace): Add test cases.
439 2018-01-27 Jakub Jelinek <jakub@redhat.com>
442 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
445 2018-01-26 Jim Wilson <jimw@sifive.com>
447 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
449 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
452 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
454 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
455 and CMP + SUB-immediate -> SUBS.
457 2018-01-26 Martin Sebor <msebor@redhat.com>
459 PR tree-optimization/83896
460 * tree-ssa-strlen.c (get_string_len): Rename...
461 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
462 Avoid assuming length is constant.
463 (handle_char_store): Use HOST_WIDE_INT for string length.
465 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
468 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
469 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
471 2018-01-26 Richard Biener <rguenther@suse.de>
473 PR rtl-optimization/84003
474 * dse.c (record_store): Only record redundant stores when
475 the earlier store aliases at least all accesses the later one does.
477 2018-01-26 Jakub Jelinek <jakub@redhat.com>
479 PR rtl-optimization/83985
480 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
481 REG_CFA_RESTORE insns.
482 (delete_unmarked_insns): Don't ignore separate shrink wrapping
483 REG_CFA_RESTORE insns here.
486 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
487 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
489 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
491 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
492 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
493 (arc_init): Likewise.
494 (arc_override_options): Likewise.
495 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
497 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
499 * config/arc/arc.h (TARGET_DBNZ): Define.
500 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
501 properly set the tune attribute.
502 (dbnz): Use TARGET_DBNZ guard.
503 * config/arc/arc.opt (mtune): Add core3 option.
505 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
507 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
508 recognize new pic like addresses.
509 (arc_delegitimize_address): Clean up.
511 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
513 * config/arc/arc-arches.def: Option mrf16 valid for all
515 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
516 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
517 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
518 * config/arc/arc-tables.opt: Regenerate.
519 * config/arc/arc.c (arc_conditional_register_usage): Handle
520 reduced register file case.
521 (arc_file_start): Set must have build attributes.
522 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
524 * config/arc/arc.opt (mrf16): Add new option.
525 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
526 * config/arc/genmultilib.awk: Handle new mrf16 option.
527 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
528 * config/arc/t-multilib: Regenerate.
529 * doc/invoke.texi (ARC Options): Document mrf16 option.
531 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
533 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
534 * config/arc/arc.c (arc_handle_secure_attribute): New function.
535 (arc_attribute_table): Add 'secure_call' attribute.
536 (arc_print_operand): Print secure call operand.
537 (arc_function_ok_for_sibcall): Don't optimize tail calls when
539 (arc_is_secure_call_p): New function. * config/arc/arc.md
540 (call_i): Add support for sjli instruction.
541 (call_value_i): Likewise.
542 * config/arc/constraints.md (Csc): New constraint.
544 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
545 John Eric Martin <John.Martin@emmicro-us.com>
547 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
548 * config/arc/arc.c (_arc_jli_section): New struct.
549 (arc_jli_section): New type.
550 (rc_jli_sections): New static variable.
551 (arc_handle_jli_attribute): New function.
552 (arc_attribute_table): Add jli_always and jli_fixed attribute.
553 (arc_file_end): New function.
554 (TARGET_ASM_FILE_END): Define.
555 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
556 (arc_add_jli_section): New function.
557 (jli_call_scan): Likewise.
558 (arc_reorg): Call jli_call_scan.
559 (arc_output_addsi): Remove 'S' from printing asm operand.
560 (arc_is_jli_call_p): New function.
561 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
563 (movhi_insn): Likewise.
564 (movsi_insn): Likewise.
565 (movsi_set_cc_insn): Likewise.
566 (loadqi_update): Likewise.
567 (load_zeroextendqisi_update): Likewise.
568 (load_signextendqisi_update): Likewise.
569 (loadhi_update): Likewise.
570 (load_zeroextendhisi_update): Likewise.
571 (load_signextendhisi_update): Likewise.
572 (loadsi_update): Likewise.
573 (loadsf_update): Likewise.
574 (movsicc_insn): Likewise.
575 (bset_insn): Likewise.
576 (bxor_insn): Likewise.
577 (bclr_insn): Likewise.
578 (bmsk_insn): Likewise.
579 (bicsi3_insn): Likewise.
580 (cmpsi_cc_c_insn): Likewise.
581 (movsi_ne): Likewise.
582 (movsi_cond_exec): Likewise.
583 (clrsbsi2): Likewise.
594 (call_i): Remove 'S' asm letter, add jli instruction.
595 (call_value_i): Likewise.
596 * config/arc/arc.op (mjli-always): New option.
597 * config/arc/constraints.md (Cji): New constraint.
598 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
600 (subsf3_fpx): Likewise.
601 (mulsf3_fpx): Likewise.
602 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
604 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
606 * doc/invoke.texi (ARC): Document mjli-always option.
608 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
610 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
611 and use incw and decw where possible
612 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
614 2018-01-26 Richard Biener <rguenther@suse.de>
616 PR tree-optimization/81082
617 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
618 association if it requires casting to unsigned.
619 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
620 from fold_plusminus_mult_expr to catch important cases late when
621 range info is available.
623 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
625 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
626 * configure.ac (hidden_linkonce): New test.
627 * configure: Regenerate.
628 * config.in: Regenerate.
630 2018-01-26 Julia Koval <julia.koval@intel.com>
632 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
633 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
634 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
635 _mm_mask_bitshuffle_epi64_mask): Fix type.
636 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
637 USI_FTYPE_V4DI_V4DI_USI): Remove.
638 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
639 __builtin_ia32_vpshufbitqmb256_mask,
640 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
641 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
642 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
644 2018-01-26 Alan Modra <amodra@gmail.com>
647 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
648 UNSPEC_VBPERMQ. Sort other unspecs.
650 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
652 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
654 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
657 * predict.c (drop_profile): Do not push/pop cfun; update also
659 (handle_missing_profiles): Fix logic looking for zero profiles.
661 2018-01-25 Jakub Jelinek <jakub@redhat.com>
664 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
665 on functions with #pragma omp declare simd or functions with simd
667 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
668 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
669 Remove trailing \n from warning_at calls.
671 2018-01-25 Tom de Vries <tom@codesourcery.com>
674 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
675 for neutered workers.
677 2018-01-24 Joseph Myers <joseph@codesourcery.com>
680 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
681 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
683 2018-01-24 Jeff Law <law@redhat.com>
686 * i386.c (get_probe_interval): Move to earlier point.
687 (ix86_compute_frame_layout): If -fstack-clash-protection and
688 the frame is larger than the probe interval, then use pushes
689 to save registers rather than reg->mem moves.
690 (ix86_expand_prologue): Remove conditional for int_registers_saved
693 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
696 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
697 min/max for never referenced object.
699 2018-01-24 Jakub Jelinek <jakub@redhat.com>
702 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
704 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
705 attributes from DECL_ATTRIBUTES (decl) without affecting
706 DECL_ATTRIBUTES (current_function_decl).
707 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
708 functions with non-NULL DECL_ABSTRACT_ORIGIN.
710 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
712 PR tree-optimization/83979
713 * fold-const.c (fold_comparison): Use constant_boolean_node
714 instead of boolean_{true,false}_node.
716 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
718 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
721 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
723 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
724 Simplify the clause that sets the length attribute.
725 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
726 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
727 clause that sets the length attribute.
728 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
730 2018-01-24 Tom de Vries <tom@codesourcery.com>
733 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
734 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
735 Add strict parameter.
736 (prevent_branch_around_nothing): Insert dummy insn between branch to
737 label and label with no ptx insn inbetween.
738 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
740 2018-01-24 Tom de Vries <tom@codesourcery.com>
743 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
744 for neutered threads in warp.
745 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
747 2018-01-24 Richard Biener <rguenther@suse.de>
749 PR tree-optimization/83176
750 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
753 2018-01-24 Richard Biener <rguenther@suse.de>
755 PR tree-optimization/82819
756 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
757 code generating pluses that are no-ops in the target precision.
759 2018-01-24 Richard Biener <rguenther@suse.de>
762 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
764 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
766 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
767 to merge probabilities.
768 * predict.c (probably_never_executed): Also mark as cold functions
769 with global 0 profile and guessed local profile.
770 * profile-count.c (profile_probability::combine_with_count): New
772 * profile-count.h (profile_probability::operator*,
773 profile_probability::operator*=, profile_probability::operator/,
774 profile_probability::operator/=): Reduce precision to adjusted
775 and set value to guessed on contradictory divisions.
776 (profile_probability::combine_with_freq): Remove.
777 (profile_probability::combine_wiht_count): Declare.
778 (profile_count::force_nonzero):: Set to adjusted.
779 (profile_count::probability_in):: Set quality to adjusted.
780 * tree-ssa-tail-merge.c (replace_block_by): Use
783 2018-01-23 Andrew Waterman <andrew@sifive.com>
784 Jim Wilson <jimw@sifive.com>
786 * config/riscv/riscv.c (riscv_stack_boundary): New.
787 (riscv_option_override): Set riscv_stack_boundary. Handle
788 riscv_preferred_stack_boundary_arg.
789 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
790 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
791 (STACK_BOUNDARY): Set to riscv_stack_boundary.
792 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
793 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
794 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
796 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
799 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
800 of struct ix86_frame.
801 (ix86_expand_epilogue): Likewise. Add a local variable for
802 the reg_save_offset field in struct ix86_frame.
804 2018-01-23 Bin Cheng <bin.cheng@arm.com>
806 PR tree-optimization/82604
807 * tree-loop-distribution.c (enum partition_kind): New enum item
808 PKIND_PARTIAL_MEMSET.
809 (partition_builtin_p): Support above new enum item.
810 (generate_code_for_partition): Ditto.
811 (compute_access_range): Differentiate cases that equality can be
812 proven at all loops, the innermost loops or no loops.
813 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
814 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
815 (finalize_partitions, distribute_loop): Don't fuse partition of
816 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
817 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
820 2018-01-23 Martin Liska <mliska@suse.cz>
822 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
823 order to ignore the predictor.
824 (PRED_POLYMORPHIC_CALL): Likewise.
825 (PRED_RECURSIVE_CALL): Likewise.
827 2018-01-23 Martin Liska <mliska@suse.cz>
829 * tree-profile.c (tree_profiling): Print function header to
830 aware reader which function we are working on.
831 * value-prof.c (gimple_find_values_to_profile): Do not print
832 not interesting value histograms.
834 2018-01-23 Martin Liska <mliska@suse.cz>
836 * profile-count.h (enum profile_quality): Add
837 profile_uninitialized as the first value. Do not number values
838 as they are zero based.
839 (profile_count::verify): Update sanity check.
840 (profile_probability::verify): Likewise.
842 2018-01-23 Nathan Sidwell <nathan@acm.org>
844 * doc/invoke.texi (ffor-scope): Deprecate.
846 2018-01-23 David Malcolm <dmalcolm@redhat.com>
848 PR tree-optimization/83510
849 * domwalk.c (set_all_edges_as_executable): New function.
850 (dom_walker::dom_walker): Convert bool param
851 "skip_unreachable_blocks" to enum reachability. Move setup of
852 edge flags to set_all_edges_as_executable and only do it when
853 reachability is REACHABLE_BLOCKS.
854 * domwalk.h (enum dom_walker::reachability): New enum.
855 (dom_walker::dom_walker): Convert bool param
856 "skip_unreachable_blocks" to enum reachability.
857 (set_all_edges_as_executable): New decl.
858 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
859 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
861 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
862 but converting true to REACHABLE_BLOCKS.
863 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
865 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
866 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
867 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
869 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
870 if check_all_array_refs will be called.
872 2018-01-23 David Malcolm <dmalcolm@redhat.com>
874 * tree.c (selftest::test_location_wrappers): Add more test
877 2018-01-23 David Malcolm <dmalcolm@redhat.com>
879 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
880 (selftest::test_bit_in_range): Likewise.
882 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
885 * doc/sourcebuild.texi (vect_float): Say that the selector
886 only describes the situation when -funsafe-math-optimizations is on.
887 (vect_float_strict): Document.
889 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
891 PR tree-optimization/83965
892 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
893 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
894 instead of checking only for a reduction.
895 (vect_recog_widen_sum_pattern): Likewise.
897 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
899 * predict.c (probably_never_executed): Only use precise profile info.
900 (compute_function_frequency): Skip after inlining hack since we now
901 have quality checking.
903 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
905 * profile-count.h (profile_probability::very_unlikely,
906 profile_probability::unlikely, profile_probability::even): Set
907 precision to guessed.
909 2018-01-23 Richard Biener <rguenther@suse.de>
911 PR tree-optimization/83963
912 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
913 Properly terminate dominator walk when crossing the exit edge not
914 when visiting its source block.
916 2018-01-23 Jakub Jelinek <jakub@redhat.com>
919 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
920 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
922 2018-01-22 Jakub Jelinek <jakub@redhat.com>
924 PR tree-optimization/83957
925 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
926 semicolon after for body surrounded by braces.
928 PR tree-optimization/83081
929 * profile-count.h (profile_probability::split): New method.
930 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
931 Use profile_probability::split.
932 (do_compare_rtx_and_jump): Fix adjustment of probabilities
933 when splitting a single conditional jump into 2.
935 2018-01-22 David Malcolm <dmalcolm@redhat.com>
937 PR tree-optimization/69452
938 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
941 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
943 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
944 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
945 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
947 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
949 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
950 * config/rl78/rl78.md: New define_expand "movdi"
951 * config/rl78/rl78.c: New function definition rl78_split_movdi
953 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
956 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
958 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
959 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
960 128-bit to produce an UNSPEC move to get the double word with the
961 signbit and then a shift directly to do signbit.
962 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
963 implementation with a new version that just does either a direct
964 move or a regular move. Move memory interface to separate insns.
965 Move insns so they are next to the expander.
966 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
967 with signbit move. Split big and little endian case.
968 (signbit<mode>2_dm_mem_le): Likewise.
969 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
970 (signbit<mode>2_dm2): Likewise.
972 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
974 * config/rl78/rl78.md: New define_expand "anddi3".
976 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
978 * config/rl78/rl78.md: New define_expand "umindi3".
980 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
982 * config/rl78/rl78.md: New define_expand "smindi3".
984 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
986 * config/rl78/rl78.md: New define_expand "smaxdi3".
988 2018-01-22 Carl Love <cel@us.ibm.com>
990 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
991 LVX_V1TI): Add macro expansion.
992 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
993 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
994 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
995 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
996 Change check to determine if the instruction is a byte reversing
997 entry. Fix typo in comment.
998 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
999 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1000 Add def_builtin calls for new builtins.
1001 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1002 Add define_insn expansion.
1004 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1006 * config/rl78/rl78.md: New define_expand "umaxdi3".
1008 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1010 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1011 for non-QImode registers
1013 2018-01-22 Richard Biener <rguenther@suse.de>
1015 PR tree-optimization/83963
1016 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1017 including the loop exit block.
1018 (scop_detection::merge_sese): Likewise.
1019 (scop_detection::add_scop): Do it here instead.
1021 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1023 * doc/sourcebuild.texi (arm_softfloat): Document.
1025 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1028 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1029 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1030 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1032 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1033 David Edelsohn <dje.gcc@gmail.com>
1036 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1037 Change "crset eq" to "crset 2".
1038 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1039 (*call_indirect_aix<mode>_nospec): Likewise.
1040 (*call_value_indirect_aix<mode>_nospec): Likewise.
1041 (*call_indirect_elfv2<mode>_nospec): Likewise.
1042 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1043 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1044 change assembly output from . to $.
1045 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1046 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1047 (*tablejump<mode>_internal1_nospec): Likewise.
1049 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1052 * config/sh/sh_optimize_sett_clrt.cc:
1053 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1055 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1057 PR tree-optimization/83940
1058 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1059 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1060 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1061 use it to pass back the definition type.
1062 (vect_check_store_rhs): Likewise rhs_dt_out.
1063 (vect_build_gather_load_calls): Add a mask_dt argument and use
1064 it instead of a call to vect_is_simple_use.
1065 (vectorizable_store): Update calls to vect_check_load_store_mask
1066 and vect_check_store_rhs. Use the dt returned by the latter instead
1067 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1068 instead of calls to vect_is_simple_use. Pass the scalar rather
1069 than the vector operand to vect_is_simple_use when handling
1070 second and subsequent copies of an rhs value.
1071 (vectorizable_load): Update calls to vect_check_load_store_mask
1072 and vect_build_gather_load_calls. Use the cached mask_dt and
1073 gs_info.offset_dt instead of calls to vect_is_simple_use.
1075 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1078 * tree-emutls.c: Include gimplify.h.
1079 (lower_emutls_2): New function.
1080 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1081 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1082 it before further processing.
1085 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1086 UINTVAL (trueop1) instead of INTVAL (op1).
1088 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1092 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1093 INCOMING_FRAME_SP_OFFSET if not defined.
1094 (scan_trace): Add ENTRY argument. If true and
1095 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1096 emit a note to adjust the CFA offset.
1097 (create_cfi_notes): Adjust scan_trace callers.
1098 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1099 INCOMING_FRAME_SP_OFFSET in the CIE.
1100 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1101 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1103 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1104 * doc/tm.texi: Regenerated.
1106 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1108 PR rtl-optimization/83147
1109 * lra-constraints.c (remove_inheritance_pseudos): Use
1110 lra_substitute_pseudo_within_insn.
1112 2018-01-19 Tom de Vries <tom@codesourcery.com>
1113 Cesar Philippidis <cesar@codesourcery.com>
1116 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1118 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1121 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1122 spaces for function labels.
1124 2018-01-19 Martin Liska <mliska@suse.cz>
1126 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1127 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1128 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1129 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1130 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1131 (PRED_CONST_RETURN): Change from 69 to 65.
1132 (PRED_NULL_RETURN): Change from 91 to 71.
1133 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1134 (PRED_LOOP_GUARD): Change from 66 to 73.
1136 2018-01-19 Martin Liska <mliska@suse.cz>
1138 * predict.c (predict_insn_def): Add new assert.
1139 (struct branch_predictor): Change type to signed integer.
1140 (test_prediction_value_range): Amend test to cover
1142 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1143 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1144 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1145 (PRED_LOOP_IV_COMPARE): Likewise.
1146 * predict.h (PROB_UNINITIALIZED): Define new constant.
1148 2018-01-19 Martin Liska <mliska@suse.cz>
1150 * predict.c (dump_prediction): Add new format for
1151 analyze_brprob.py script which is enabled with -details
1153 * profile-count.h (precise_p): New function.
1155 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1157 PR tree-optimization/83922
1158 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1159 there are no statements that need masking.
1160 (vect_active_double_reduction_p): New function.
1161 (vect_analyze_loop_operations): Use it when handling phis that
1162 are not in the loop header.
1164 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1166 PR tree-optimization/83914
1167 * tree-vect-loop.c (vectorizable_induction): Don't convert
1168 init_expr or apply the peeling adjustment for inductions
1169 that are nested within the vectorized loop.
1171 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1173 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1176 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1180 * function.h (gimplify_parameters): Add gimple_seq * argument.
1181 * function.c: Include gimple.h and options.h.
1182 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1183 for the added local temporaries if needed.
1184 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1185 if there are any parameter cleanups, wrap whole body into a
1186 try/finally with the cleanups.
1188 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1191 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1192 Use GET_MODE_CLASS for scalar floating point.
1194 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1198 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1199 Fix call of call_cgraph_insertion_hooks.
1201 2018-01-18 Martin Sebor <msebor@redhat.com>
1203 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1205 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1208 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1211 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1214 * common.opt: (-ffile-prefix-map): New option.
1215 * opts.c (common_handle_option): Defer it.
1216 * opts-global.c (handle_common_deferred_options): Handle it.
1217 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1218 * file-prefix-map.h: New file.
1219 (remap_debug_filename, add_debug_prefix_map): ...here.
1220 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1221 * final.c (debug_prefix_map, add_debug_prefix_map
1222 remap_debug_filename): Move to...
1223 * file-prefix-map.c: New file.
1224 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1225 generalize, get rid of alloca(), use strrchr() instead of strchr().
1226 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1227 Implement in terms of add_prefix_map().
1228 (remap_macro_filename, remap_debug_filename): Implement in term of
1230 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1231 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1232 * dbxout.c: Include file-prefix-map.h.
1233 * varasm.c: Likewise.
1234 * vmsdbgout.c: Likewise.
1235 * xcoffout.c: Likewise.
1236 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1237 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1238 * doc/invoke.texi (-ffile-prefix-map): Document.
1239 (-fdebug-prefix-map): Update description.
1241 2018-01-18 Martin Liska <mliska@suse.cz>
1243 * config/i386/i386.c (indirect_thunk_name): Document that also
1245 (output_indirect_thunk): Document why both instructions
1246 (pause and lfence) are generated.
1248 2018-01-18 Richard Biener <rguenther@suse.de>
1250 PR tree-optimization/83887
1251 * graphite-scop-detection.c
1252 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1253 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1254 (scop_detection::merge_sese): Re-implement with a flood-fill
1255 algorithm that properly finds a SESE region if it exists.
1257 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1260 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1261 pointer_diff optimizations use view_convert instead of convert.
1263 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1265 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1266 Generate different code for -mno-speculate-indirect-jumps.
1267 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1268 (*call_indirect_aix<mode>): Disable for
1269 -mno-speculate-indirect-jumps.
1270 (*call_indirect_aix<mode>_nospec): New define_insn.
1271 (*call_value_indirect_aix<mode>): Disable for
1272 -mno-speculate-indirect-jumps.
1273 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1274 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1275 -mno-speculate-indirect-jumps.
1276 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1278 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1280 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1281 long double type, set the flags for noting the default long double
1282 type, even if we don't pass or return a long double type.
1284 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1287 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1290 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1292 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1293 support for merge[hl].
1294 (fold_mergehl_helper): New helper function.
1295 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1296 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1297 (altivec_vmrglw_direct): Add xxmrglw insn.
1299 2018-01-17 Andrew Waterman <andrew@sifive.com>
1301 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1302 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1304 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1307 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1308 call the lto_location_cache before reading the
1309 DECL_SOURCE_LOCATION of the types.
1311 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1312 Richard Sandiford <richard.sandiford@linaro.org>
1314 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1315 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1316 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1317 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1318 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1320 * config/aarch64/constraints.md (aarch64_movti_operand):
1322 * config/aarch64/predicates.md (Uti): Add new constraint.
1324 2018-01-17 Carl Love <cel@us.ibm.com>
1325 * config/rs6000/vsx.md (define_expand xl_len_r,
1326 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1327 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1329 (define_expand, define_insn): Move the shift left from the
1330 define_insn to the define_expand for lxvl and stxvl instructions.
1331 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1332 and XL_LEN_R definitions to PURE.
1334 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1336 * config/i386/i386.c (indirect_thunk_name): Declare regno
1337 as unsigned int. Compare regno with INVALID_REGNUM.
1338 (output_indirect_thunk): Ditto.
1339 (output_indirect_thunk_function): Ditto.
1340 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1341 in the call to output_indirect_thunk_function.
1343 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1346 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1347 rather than the size of inner_type to determine the stack slot size
1348 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1350 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1353 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1356 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1358 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1359 endian Linux systems to optionally enable multilibs for selecting
1360 the long double type if the user configured an explicit type.
1361 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1362 have no long double multilibs if not defined.
1363 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1364 warn if the user used -mabi={ieee,ibm}longdouble and we built
1365 multilibs for long double.
1366 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1367 appropriate multilib option.
1368 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1370 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1371 for building long double multilibs.
1372 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1374 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1376 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1379 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1381 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1384 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1387 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1390 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1392 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1393 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1395 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1397 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1398 different rtl trees depending on TARGET_64BIT.
1399 (rs6000_gen_lvx): Likewise.
1401 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1403 * config/visium/visium.md (nop): Tweak comment.
1404 (hazard_nop): Likewise.
1406 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1408 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1409 -mspeculate-indirect-jumps.
1410 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1411 for -mno-speculate-indirect-jumps.
1412 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1413 (*call_value_indirect_elfv2<mode>): Disable for
1414 -mno-speculate-indirect-jumps.
1415 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1416 (indirect_jump): Emit different RTL for
1417 -mno-speculate-indirect-jumps.
1418 (*indirect_jump<mode>): Disable for
1419 -mno-speculate-indirect-jumps.
1420 (*indirect_jump<mode>_nospec): New define_insn.
1421 (tablejump): Emit different RTL for
1422 -mno-speculate-indirect-jumps.
1423 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1424 (tablejumpsi_nospec): New define_expand.
1425 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1426 (tablejumpdi_nospec): New define_expand.
1427 (*tablejump<mode>_internal1): Disable for
1428 -mno-speculate-indirect-jumps.
1429 (*tablejump<mode>_internal1_nospec): New define_insn.
1430 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1433 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1435 * caller-save.c (insert_save): Drop unnecessary parameter. All
1438 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1439 Richard Biener <rguenth@suse.de>
1442 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1443 return early, inline manually is_gimple_sizepos. Make sure if we
1444 call gimplify_expr we don't end up with a gimple constant.
1445 * tree.c (variably_modified_type_p): Don't return true for
1446 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1447 * gimplify.h (is_gimple_sizepos): Remove.
1449 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1451 PR tree-optimization/83857
1452 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1453 vectorizable_live_operation for pure SLP statements.
1454 (vectorizable_live_operation): Handle PHIs.
1456 2018-01-16 Richard Biener <rguenther@suse.de>
1458 PR tree-optimization/83867
1459 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1460 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1462 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1465 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1466 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1467 If off is not INTEGER_CST, issue a may not be aligned warning
1468 rather than isn't aligned. Use isn%'t rather than isn't.
1469 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1471 <case MULT_EXPR>: Improve the case when bottom and one of the
1472 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1473 operand, in that case check if the other operand is multiple of
1474 bottom divided by the INTEGER_CST operand.
1476 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1479 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1480 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1481 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1482 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1483 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1484 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1485 (pa_function_arg_size): New function.
1487 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1489 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1490 in a separate statement.
1492 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1494 PR tree-optimization/83847
1495 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1496 group gathers and scatters.
1498 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1500 PR rtl-optimization/86620
1501 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1503 PR rtl-optimization/83213
1504 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1505 to last if both are JUMP_INSNs.
1507 PR tree-optimization/83843
1508 * gimple-ssa-store-merging.c
1509 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1510 store_immediate_info for bswap/nop orig_stores.
1512 2018-01-15 Andrew Waterman <andrew@sifive.com>
1514 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1516 <UDIV>: Increase cost if !TARGET_DIV.
1518 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1520 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1521 (define_attr "cr_logical_3op"): New.
1522 (cceq_ior_compare): Adjust.
1523 (cceq_ior_compare_complement): Adjust.
1524 (*cceq_rev_compare): Adjust.
1525 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1526 (is_cracked_insn): Adjust.
1527 (insn_must_be_first_in_group): Adjust.
1528 * config/rs6000/40x.md: Adjust.
1529 * config/rs6000/440.md: Adjust.
1530 * config/rs6000/476.md: Adjust.
1531 * config/rs6000/601.md: Adjust.
1532 * config/rs6000/603.md: Adjust.
1533 * config/rs6000/6xx.md: Adjust.
1534 * config/rs6000/7450.md: Adjust.
1535 * config/rs6000/7xx.md: Adjust.
1536 * config/rs6000/8540.md: Adjust.
1537 * config/rs6000/cell.md: Adjust.
1538 * config/rs6000/e300c2c3.md: Adjust.
1539 * config/rs6000/e500mc.md: Adjust.
1540 * config/rs6000/e500mc64.md: Adjust.
1541 * config/rs6000/e5500.md: Adjust.
1542 * config/rs6000/e6500.md: Adjust.
1543 * config/rs6000/mpc.md: Adjust.
1544 * config/rs6000/power4.md: Adjust.
1545 * config/rs6000/power5.md: Adjust.
1546 * config/rs6000/power6.md: Adjust.
1547 * config/rs6000/power7.md: Adjust.
1548 * config/rs6000/power8.md: Adjust.
1549 * config/rs6000/power9.md: Adjust.
1550 * config/rs6000/rs64.md: Adjust.
1551 * config/rs6000/titan.md: Adjust.
1553 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1555 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1556 ix86_indirect_branch_register logic.
1558 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1560 * config/i386/constraints.md (Bs): Update
1561 ix86_indirect_branch_register check. Don't check
1562 ix86_indirect_branch_register with GOT_memory_operand.
1564 * config/i386/predicates.md (GOT_memory_operand): Don't check
1565 ix86_indirect_branch_register here.
1566 (GOT32_symbol_operand): Likewise.
1568 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1570 * config/i386/predicates.md (constant_call_address_operand):
1571 Rewrite ix86_indirect_branch_register logic.
1572 (sibcall_insn_operand): Likewise.
1574 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1576 * config/i386/constraints.md (Bs): Replace
1577 ix86_indirect_branch_thunk_register with
1578 ix86_indirect_branch_register.
1580 * config/i386/i386.md (indirect_jump): Likewise.
1581 (tablejump): Likewise.
1582 (*sibcall_memory): Likewise.
1583 (*sibcall_value_memory): Likewise.
1584 Peepholes of indirect call and jump via memory: Likewise.
1585 * config/i386/i386.opt: Likewise.
1586 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1587 (GOT_memory_operand): Likewise.
1588 (call_insn_operand): Likewise.
1589 (sibcall_insn_operand): Likewise.
1590 (GOT32_symbol_operand): Likewise.
1592 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1595 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1596 type rather than type addr's type points to.
1597 (expand_omp_atomic_mutex): Likewise.
1598 (expand_omp_atomic): Likewise.
1600 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1603 * config/i386/i386.c (output_indirect_thunk_function): Use
1604 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1605 for __x86_return_thunk.
1607 2018-01-15 Richard Biener <rguenther@suse.de>
1610 * expmed.c (extract_bit_field_1): Fix typo.
1612 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1615 * config/arm/iterators.md (VF): New mode iterator.
1616 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1617 Remove integer-related logic from pattern.
1618 (neon_vabd<mode>_3): Likewise.
1620 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1623 * common.opt (fstrict-overflow): No longer an alias.
1624 (fwrapv-pointer): New option.
1625 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1626 also for pointer types based on flag_wrapv_pointer.
1627 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1628 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1629 opts->x_flag_wrapv got set.
1630 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1631 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1632 POINTER_TYPE_OVERFLOW_UNDEFINED.
1633 * match.pd: Likewise in address comparison pattern.
1634 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1636 2018-01-15 Richard Biener <rguenther@suse.de>
1639 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1640 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1641 Reset type names to their identifier if their TYPE_DECL doesn't
1642 have linkage (and thus is used for ODR and devirt).
1643 (save_debug_info_for_decl): Remove.
1644 (save_debug_info_for_type): Likewise.
1645 (add_tree_to_fld_list): Adjust.
1646 * tree-pretty-print.c (dump_generic_node): Make dumping of
1647 type names more robust.
1649 2018-01-15 Richard Biener <rguenther@suse.de>
1651 * BASE-VER: Bump to 8.0.1.
1653 2018-01-14 Martin Sebor <msebor@redhat.com>
1656 * builtins.c (check_access): Avoid warning when the no-warning bit
1659 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1661 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1662 * ira-color (allocno_hard_regs_compare): Likewise.
1664 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1667 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1668 Use .pushsection/.popsection.
1670 2018-01-14 Martin Sebor <msebor@redhat.com>
1673 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1675 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1677 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1678 entry from extra_headers.
1679 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1680 extra_headers, make the list bitwise identical to the i?86-*-* one.
1682 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1684 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1685 -mcmodel=large with -mindirect-branch=thunk,
1686 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1687 -mfunction-return=thunk-extern.
1688 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1689 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1690 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1692 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1694 * config/i386/i386.c (print_reg): Print the name of the full
1695 integer register without '%'.
1696 (ix86_print_operand): Handle 'V'.
1697 * doc/extend.texi: Document 'V' modifier.
1699 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1701 * config/i386/constraints.md (Bs): Disallow memory operand for
1702 -mindirect-branch-register.
1704 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1705 (GOT_memory_operand): Likewise.
1706 (call_insn_operand): Likewise.
1707 (sibcall_insn_operand): Likewise.
1708 (GOT32_symbol_operand): Likewise.
1709 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1710 for -mindirect-branch-register.
1711 (tablejump): Likewise.
1712 (*sibcall_memory): Likewise.
1713 (*sibcall_value_memory): Likewise.
1714 Disallow peepholes of indirect call and jump via memory for
1715 -mindirect-branch-register.
1716 (*call_pop): Replace m with Bw.
1717 (*call_value_pop): Likewise.
1718 (*sibcall_pop_memory): Replace m with Bs.
1719 * config/i386/i386.opt (mindirect-branch-register): New option.
1720 * doc/invoke.texi: Document -mindirect-branch-register option.
1722 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1724 * config/i386/i386-protos.h (ix86_output_function_return): New.
1725 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1726 set function_return_type.
1727 (indirect_thunk_name): Add ret_p to indicate thunk for function
1729 (output_indirect_thunk_function): Pass false to
1730 indirect_thunk_name.
1731 (ix86_output_indirect_branch_via_reg): Likewise.
1732 (ix86_output_indirect_branch_via_push): Likewise.
1733 (output_indirect_thunk_function): Create alias for function
1734 return thunk if regno < 0.
1735 (ix86_output_function_return): New function.
1736 (ix86_handle_fndecl_attribute): Handle function_return.
1737 (ix86_attribute_table): Add function_return.
1738 * config/i386/i386.h (machine_function): Add
1739 function_return_type.
1740 * config/i386/i386.md (simple_return_internal): Use
1741 ix86_output_function_return.
1742 (simple_return_internal_long): Likewise.
1743 * config/i386/i386.opt (mfunction-return=): New option.
1744 (indirect_branch): Mention -mfunction-return=.
1745 * doc/extend.texi: Document function_return function attribute.
1746 * doc/invoke.texi: Document -mfunction-return= option.
1748 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1750 * config/i386/i386-opts.h (indirect_branch): New.
1751 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1752 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1753 with local indirect jump when converting indirect call and jump.
1754 (ix86_set_indirect_branch_type): New.
1755 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1756 (indirectlabelno): New.
1757 (indirect_thunk_needed): Likewise.
1758 (indirect_thunk_bnd_needed): Likewise.
1759 (indirect_thunks_used): Likewise.
1760 (indirect_thunks_bnd_used): Likewise.
1761 (INDIRECT_LABEL): Likewise.
1762 (indirect_thunk_name): Likewise.
1763 (output_indirect_thunk): Likewise.
1764 (output_indirect_thunk_function): Likewise.
1765 (ix86_output_indirect_branch_via_reg): Likewise.
1766 (ix86_output_indirect_branch_via_push): Likewise.
1767 (ix86_output_indirect_branch): Likewise.
1768 (ix86_output_indirect_jmp): Likewise.
1769 (ix86_code_end): Call output_indirect_thunk_function if needed.
1770 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1772 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1773 (ix86_attribute_table): Add indirect_branch.
1774 * config/i386/i386.h (machine_function): Add indirect_branch_type
1775 and has_local_indirect_jump.
1776 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1778 (tablejump): Likewise.
1779 (*indirect_jump): Use ix86_output_indirect_jmp.
1780 (*tablejump_1): Likewise.
1781 (simple_return_indirect_internal): Likewise.
1782 * config/i386/i386.opt (mindirect-branch=): New option.
1783 (indirect_branch): New.
1786 (thunk-inline): Likewise.
1787 (thunk-extern): Likewise.
1788 * doc/extend.texi: Document indirect_branch function attribute.
1789 * doc/invoke.texi: Document -mindirect-branch= option.
1791 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1794 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1796 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1798 * ipa-inline.c (want_inline_small_function_p): Return false if
1799 inlining has already failed with CIF_FINAL_ERROR.
1800 (update_caller_keys): Call want_inline_small_function_p before
1802 (update_callee_keys): Likewise.
1804 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1806 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1808 (rs6000_quadword_masked_address_p): Likewise.
1809 (quad_aligned_load_p): Likewise.
1810 (quad_aligned_store_p): Likewise.
1811 (const_load_sequence_p): Add comment to describe the outer-most loop.
1812 (mimic_memory_attributes_and_flags): New function.
1813 (rs6000_gen_stvx): Likewise.
1814 (replace_swapped_aligned_store): Likewise.
1815 (rs6000_gen_lvx): Likewise.
1816 (replace_swapped_aligned_load): Likewise.
1817 (replace_swapped_load_constant): Capitalize argument name in
1818 comment describing this function.
1819 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1820 and stores that access quad-word aligned addresses and replace
1821 with stvx or lvx instructions when appropriate.
1822 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1823 New function prototype.
1824 (rs6000_quadword_masked_address_p): Likewise.
1825 (rs6000_gen_lvx): Likewise.
1826 (rs6000_gen_stvx): Likewise.
1827 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1828 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1829 when memory address is aligned.
1830 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1831 this split to select lvx instruction when memory address is aligned.
1832 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1833 instruction when memory address is aligned.
1834 (*vsx_le_perm_load_v16qi): Likewise.
1835 (four unnamed splitters): Modify to select the stvx instruction
1836 when memory is aligned.
1838 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1840 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1841 which appears in the queue multiple times.
1843 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1844 Alan Hayward <alan.hayward@arm.com>
1845 David Sherwood <david.sherwood@arm.com>
1847 * tree-vectorizer.h (vec_lower_bound): New structure.
1848 (_loop_vec_info): Add check_nonzero and lower_bounds.
1849 (LOOP_VINFO_CHECK_NONZERO): New macro.
1850 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1851 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1852 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1853 fields. Make seg_len the distance travelled, not including the
1855 (dr_direction_indicator): Declare.
1856 (dr_zero_step_indicator): Likewise.
1857 (dr_known_forward_stride_p): Likewise.
1858 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1860 (runtime_alias_check_p): Allow runtime alias checks with
1862 (operator ==): Compare access_size and align.
1863 (prune_runtime_alias_test_list): Rework for new distinction between
1864 the access_size and seg_len.
1865 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1867 (get_segment_min_max): New function.
1868 (create_intersect_range_checks): Use it.
1869 (dr_step_indicator): New function.
1870 (dr_direction_indicator): Likewise.
1871 (dr_zero_step_indicator): Likewise.
1872 (dr_known_forward_stride_p): Likewise.
1873 * tree-loop-distribution.c (data_ref_segment_size): Return
1874 DR_STEP * (niters - 1).
1875 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1877 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1878 (vect_preserves_scalar_order_p): New function, split out from...
1879 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1880 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1881 (vect_vfa_access_size): New function.
1882 (vect_vfa_align): Likewise.
1883 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1884 (dump_lower_bound): New function.
1885 (vect_check_lower_bound): Likewise.
1886 (vect_small_gap_p): Likewise.
1887 (vectorizable_with_step_bound_p): Likewise.
1888 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1889 depencies if the vectorization factor is 1. Convert the checks
1890 for nonzero steps into checks on the bounds of DR_STEP. Try using
1891 a bunds check for variable steps if the minimum required step is
1892 relatively small. Update calls to the dr_with_seg_len
1893 constructor and to vect_compile_time_alias.
1894 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1896 (vect_loop_versioning): Call it.
1897 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1899 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1901 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1902 Alan Hayward <alan.hayward@arm.com>
1903 David Sherwood <david.sherwood@arm.com>
1905 * doc/sourcebuild.texi (vect_scatter_store): Document.
1906 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1908 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1910 * genopinit.c (main): Add supports_vec_scatter_store and
1911 supports_vec_scatter_store_cached to target_optabs.
1912 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1913 IFN_MASK_SCATTER_STORE.
1914 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1916 * internal-fn.h (internal_store_fn_p): Declare.
1917 (internal_fn_stored_value_index): Likewise.
1918 * internal-fn.c (scatter_store_direct): New macro.
1919 (expand_scatter_store_optab_fn): New function.
1920 (direct_scatter_store_optab_supported_p): New macro.
1921 (internal_store_fn_p): New function.
1922 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1923 IFN_MASK_SCATTER_STORE.
1924 (internal_fn_mask_index): Likewise.
1925 (internal_fn_stored_value_index): New function.
1926 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1928 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1929 * optabs-query.c (supports_vec_scatter_store_p): New function.
1930 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1931 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1932 true for scatter stores.
1933 (vect_gather_scatter_fn_p): Handle scatter stores too.
1934 (vect_check_gather_scatter): Consider using scatter stores if
1935 supports_vec_scatter_store_p.
1936 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1938 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1939 internal_fn_stored_value_index.
1940 (check_load_store_masking): Handle scatter stores too.
1941 (vect_get_store_rhs): Make public.
1942 (vectorizable_call): Use internal_store_fn_p.
1943 (vectorizable_store): Handle scatter store internal functions.
1944 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1945 when deciding whether the end of the group has been reached.
1946 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1947 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1948 (mask_scatter_store<mode>): New insns.
1950 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1951 Alan Hayward <alan.hayward@arm.com>
1952 David Sherwood <david.sherwood@arm.com>
1954 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1955 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1956 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1958 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1959 Use vect_truncate_gather_scatter_offset if we can't treat the
1960 operation as a normal gather load or scatter store.
1961 (get_group_load_store_type): Take the gather_scatter_info
1962 as argument. Try using a gather load or scatter store for
1963 single-element groups.
1964 (get_load_store_type): Update calls to get_group_load_store_type
1965 and vect_use_strided_gather_scatters_p.
1967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1968 Alan Hayward <alan.hayward@arm.com>
1969 David Sherwood <david.sherwood@arm.com>
1971 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1972 optional tree argument.
1973 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1975 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1976 but continue to use the current value as a fallback.
1977 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1978 to compare the updates.
1979 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1980 (get_load_store_type): Use it when handling a strided access.
1981 (vect_get_strided_load_store_ops): New function.
1982 (vect_get_data_ptr_increment): Likewise.
1983 (vectorizable_load): Handle strided gather loads. Always pass
1984 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1986 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1987 Alan Hayward <alan.hayward@arm.com>
1988 David Sherwood <david.sherwood@arm.com>
1990 * doc/md.texi (gather_load@var{m}): Document.
1991 (mask_gather_load@var{m}): Likewise.
1992 * genopinit.c (main): Add supports_vec_gather_load and
1993 supports_vec_gather_load_cached to target_optabs.
1994 * optabs-tree.c (init_tree_optimization_optabs): Use
1995 ggc_cleared_alloc to allocate target_optabs.
1996 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1997 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1999 * internal-fn.h (internal_load_fn_p): Declare.
2000 (internal_gather_scatter_fn_p): Likewise.
2001 (internal_fn_mask_index): Likewise.
2002 (internal_gather_scatter_fn_supported_p): Likewise.
2003 * internal-fn.c (gather_load_direct): New macro.
2004 (expand_gather_load_optab_fn): New function.
2005 (direct_gather_load_optab_supported_p): New macro.
2006 (direct_internal_fn_optab): New function.
2007 (internal_load_fn_p): Likewise.
2008 (internal_gather_scatter_fn_p): Likewise.
2009 (internal_fn_mask_index): Likewise.
2010 (internal_gather_scatter_fn_supported_p): Likewise.
2011 * optabs-query.c (supports_at_least_one_mode_p): New function.
2012 (supports_vec_gather_load_p): Likewise.
2013 * optabs-query.h (supports_vec_gather_load_p): Declare.
2014 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2015 and memory_type field.
2016 (NUM_PATTERNS): Bump to 15.
2017 * tree-vect-data-refs.c: Include internal-fn.h.
2018 (vect_gather_scatter_fn_p): New function.
2019 (vect_describe_gather_scatter_call): Likewise.
2020 (vect_check_gather_scatter): Try using internal functions for
2021 gather loads. Recognize existing calls to a gather load function.
2022 (vect_analyze_data_refs): Consider using gather loads if
2023 supports_vec_gather_load_p.
2024 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2025 (vect_get_gather_scatter_offset_type): Likewise.
2026 (vect_convert_mask_for_vectype): Likewise.
2027 (vect_add_conversion_to_patterm): Likewise.
2028 (vect_try_gather_scatter_pattern): Likewise.
2029 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2030 (vect_vect_recog_func_ptrs): Add it.
2031 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2032 internal_fn_mask_index and internal_gather_scatter_fn_p.
2033 (check_load_store_masking): Take the gather_scatter_info as an
2034 argument and handle gather loads.
2035 (vect_get_gather_scatter_ops): New function.
2036 (vectorizable_call): Check internal_load_fn_p.
2037 (vectorizable_load): Likewise. Handle gather load internal
2039 (vectorizable_store): Update call to check_load_store_masking.
2040 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2041 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2042 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2043 (aarch64_gather_scale_operand_d): New predicates.
2044 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2045 (mask_gather_load<mode>): New insns.
2047 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2048 Alan Hayward <alan.hayward@arm.com>
2049 David Sherwood <david.sherwood@arm.com>
2051 * optabs.def (fold_left_plus_optab): New optab.
2052 * doc/md.texi (fold_left_plus_@var{m}): Document.
2053 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2054 * internal-fn.c (fold_left_direct): Define.
2055 (expand_fold_left_optab_fn): Likewise.
2056 (direct_fold_left_optab_supported_p): Likewise.
2057 * fold-const-call.c (fold_const_fold_left): New function.
2058 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2059 * tree-parloops.c (valid_reduction_p): New function.
2060 (gather_scalar_reductions): Use it.
2061 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2062 (vect_finish_replace_stmt): Declare.
2063 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2064 (needs_fold_left_reduction_p): New function, split out from...
2065 (vect_is_simple_reduction): ...here. Accept reductions that
2066 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2067 (vect_force_simple_reduction): Also store the reduction type in
2068 the assignment's STMT_VINFO_REDUC_TYPE.
2069 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2070 (merge_with_identity): New function.
2071 (vect_expand_fold_left): Likewise.
2072 (vectorize_fold_left_reduction): Likewise.
2073 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2074 scalar phi in place for it. Check for target support and reject
2075 cases that would reassociate the operation. Defer the transform
2076 phase to vectorize_fold_left_reduction.
2077 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2078 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2079 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2083 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2084 call to ifc_temp_var.
2086 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2087 Alan Hayward <alan.hayward@arm.com>
2088 David Sherwood <david.sherwood@arm.com>
2090 * target.def (legitimize_address_displacement): Take the original
2091 offset as a poly_int.
2092 * targhooks.h (default_legitimize_address_displacement): Update
2094 * targhooks.c (default_legitimize_address_displacement): Likewise.
2095 * doc/tm.texi: Regenerate.
2096 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2097 as an argument, moving assert of ad->disp == ad->disp_term to...
2098 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2099 Try calling targetm.legitimize_address_displacement before expanding
2100 the address rather than afterwards, and adjust for the new interface.
2101 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2102 Match the new hook interface. Handle SVE addresses.
2103 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2108 * Makefile.in (OBJS): Add early-remat.o.
2109 * target.def (select_early_remat_modes): New hook.
2110 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2111 * doc/tm.texi: Regenerate.
2112 * targhooks.h (default_select_early_remat_modes): Declare.
2113 * targhooks.c (default_select_early_remat_modes): New function.
2114 * timevar.def (TV_EARLY_REMAT): New timevar.
2115 * passes.def (pass_early_remat): New pass.
2116 * tree-pass.h (make_pass_early_remat): Declare.
2117 * early-remat.c: New file.
2118 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2120 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2122 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2123 Alan Hayward <alan.hayward@arm.com>
2124 David Sherwood <david.sherwood@arm.com>
2126 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2127 vfm1 with a bound_epilog parameter.
2128 (vect_do_peeling): Update calls accordingly, and move the prologue
2129 call earlier in the function. Treat the base bound_epilog as 0 for
2130 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2131 this base when peeling for gaps.
2132 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2133 with fully-masked loops.
2134 (vect_estimate_min_profitable_iters): Handle the single peeled
2135 iteration in that case.
2137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2138 Alan Hayward <alan.hayward@arm.com>
2139 David Sherwood <david.sherwood@arm.com>
2141 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2142 single-element interleaving even if the size is not a power of 2.
2143 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2144 accesses for single-element interleaving if the group size is
2147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2148 Alan Hayward <alan.hayward@arm.com>
2149 David Sherwood <david.sherwood@arm.com>
2151 * doc/md.texi (fold_extract_last_@var{m}): Document.
2152 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2153 * optabs.def (fold_extract_last_optab): New optab.
2154 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2155 * internal-fn.c (fold_extract_direct): New macro.
2156 (expand_fold_extract_optab_fn): Likewise.
2157 (direct_fold_extract_optab_supported_p): Likewise.
2158 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2159 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2160 EXTRACT_LAST_REDUCTION.
2161 (get_initial_def_for_reduction): Do not create an initial vector
2162 for EXTRACT_LAST_REDUCTION reductions.
2163 (vectorizable_reduction): Leave the scalar phi in place for
2164 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2165 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2166 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2167 transform phase to vectorizable_condition.
2168 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2170 (vect_finish_stmt_generation): ...here.
2171 (vect_finish_replace_stmt): New function.
2172 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2173 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2175 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2177 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2178 Alan Hayward <alan.hayward@arm.com>
2179 David Sherwood <david.sherwood@arm.com>
2181 * doc/md.texi (extract_last_@var{m}): Document.
2182 * optabs.def (extract_last_optab): New optab.
2183 * internal-fn.def (EXTRACT_LAST): New internal function.
2184 * internal-fn.c (cond_unary_direct): New macro.
2185 (expand_cond_unary_optab_fn): Likewise.
2186 (direct_cond_unary_optab_supported_p): Likewise.
2187 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2188 loops using EXTRACT_LAST.
2189 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2190 (extract_last_<mode>): ...this optab.
2191 (vec_extract<mode><Vel>): Update accordingly.
2193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2194 Alan Hayward <alan.hayward@arm.com>
2195 David Sherwood <david.sherwood@arm.com>
2197 * target.def (empty_mask_is_expensive): New hook.
2198 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2199 * doc/tm.texi: Regenerate.
2200 * targhooks.h (default_empty_mask_is_expensive): Declare.
2201 * targhooks.c (default_empty_mask_is_expensive): New function.
2202 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2203 if the target says that empty masks are expensive.
2204 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2206 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2209 Alan Hayward <alan.hayward@arm.com>
2210 David Sherwood <david.sherwood@arm.com>
2212 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2213 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2214 (vect_use_loop_mask_for_alignment_p): New function.
2215 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2216 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2217 niters_skip argument. Make sure that the first niters_skip elements
2218 of the first iteration are inactive.
2219 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2220 Update call to vect_set_loop_masks_directly.
2221 (get_misalign_in_elems): New function, split out from...
2222 (vect_gen_prolog_loop_niters): ...here.
2223 (vect_update_init_of_dr): Take a code argument that specifies whether
2224 the adjustment should be added or subtracted.
2225 (vect_update_init_of_drs): Likewise.
2226 (vect_prepare_for_masked_peels): New function.
2227 (vect_do_peeling): Skip prologue peeling if we're using a mask
2228 instead. Update call to vect_update_inits_of_drs.
2229 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2231 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2232 alignment. Do not include the number of peeled iterations in
2233 the minimum threshold in that case.
2234 (vectorizable_induction): Adjust the start value down by
2235 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2236 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2237 Take the number of skipped iterations into account when calculating
2239 * tree-vect-stmts.c (vect_gen_while_not): New function.
2241 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2242 Alan Hayward <alan.hayward@arm.com>
2243 David Sherwood <david.sherwood@arm.com>
2245 * doc/sourcebuild.texi (vect_fully_masked): Document.
2246 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2248 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2250 (vect_analyze_loop_2): ...here. Don't check the vectorization
2251 factor against the number of loop iterations if the loop is
2254 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2255 Alan Hayward <alan.hayward@arm.com>
2256 David Sherwood <david.sherwood@arm.com>
2258 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2259 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2260 (dump_groups): Update accordingly.
2261 (iv_use::mem_type): New member variable.
2262 (address_p): New function.
2263 (record_use): Add a mem_type argument and initialize the new
2265 (record_group_use): Add a mem_type argument. Use address_p.
2266 Remove obsolete null checks of base_object. Update call to record_use.
2267 (find_interesting_uses_op): Update call to record_group_use.
2268 (find_interesting_uses_cond): Likewise.
2269 (find_interesting_uses_address): Likewise.
2270 (get_mem_type_for_internal_fn): New function.
2271 (find_address_like_use): Likewise.
2272 (find_interesting_uses_stmt): Try find_address_like_use before
2273 calling find_interesting_uses_op.
2274 (addr_offset_valid_p): Use the iv mem_type field as the type
2275 of the addressed memory.
2276 (add_autoinc_candidates): Likewise.
2277 (get_address_cost): Likewise.
2278 (split_small_address_groups_p): Use address_p.
2279 (split_address_groups): Likewise.
2280 (add_iv_candidate_for_use): Likewise.
2281 (autoinc_possible_for_pair): Likewise.
2282 (rewrite_groups): Likewise.
2283 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2284 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2285 (get_alias_ptr_type_for_ptr_address): New function.
2286 (rewrite_use_address): Rewrite address uses in calls that were
2287 identified by find_address_like_use.
2289 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2290 Alan Hayward <alan.hayward@arm.com>
2291 David Sherwood <david.sherwood@arm.com>
2293 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2295 * gimple-expr.h (is_gimple_addressable: Likewise.
2296 * gimple-expr.c (is_gimple_address): Likewise.
2297 * internal-fn.c (expand_call_mem_ref): New function.
2298 (expand_mask_load_optab_fn): Use it.
2299 (expand_mask_store_optab_fn): Likewise.
2301 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2302 Alan Hayward <alan.hayward@arm.com>
2303 David Sherwood <david.sherwood@arm.com>
2305 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2306 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2307 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2308 (cond_umax@var{mode}): Document.
2309 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2310 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2311 (cond_umin_optab, cond_umax_optab): New optabs.
2312 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2313 (COND_IOR, COND_XOR): New internal functions.
2314 * internal-fn.h (get_conditional_internal_fn): Declare.
2315 * internal-fn.c (cond_binary_direct): New macro.
2316 (expand_cond_binary_optab_fn): Likewise.
2317 (direct_cond_binary_optab_supported_p): Likewise.
2318 (get_conditional_internal_fn): New function.
2319 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2320 Cope with reduction statements that are vectorized as calls rather
2322 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2323 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2324 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2325 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2326 (UNSPEC_COND_EOR): New unspecs.
2327 (optab): Add mappings for them.
2328 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2329 (sve_int_op, sve_fp_op): New int attributes.
2331 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2332 Alan Hayward <alan.hayward@arm.com>
2333 David Sherwood <david.sherwood@arm.com>
2335 * optabs.def (while_ult_optab): New optab.
2336 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2337 * internal-fn.def (WHILE_ULT): New internal function.
2338 * internal-fn.h (direct_internal_fn_supported_p): New override
2339 that takes two types as argument.
2340 * internal-fn.c (while_direct): New macro.
2341 (expand_while_optab_fn): New function.
2342 (convert_optab_supported_p): Likewise.
2343 (direct_while_optab_supported_p): New macro.
2344 * wide-int.h (wi::udiv_ceil): New function.
2345 * tree-vectorizer.h (rgroup_masks): New structure.
2346 (vec_loop_masks): New typedef.
2347 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2349 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2350 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2351 (vect_max_vf): New function.
2352 (slpeel_make_loop_iterate_ntimes): Delete.
2353 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2354 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2355 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2356 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2357 internal-fn.h, stor-layout.h and optabs-query.h.
2358 (vect_set_loop_mask): New function.
2359 (add_preheader_seq): Likewise.
2360 (add_header_seq): Likewise.
2361 (interleave_supported_p): Likewise.
2362 (vect_maybe_permute_loop_masks): Likewise.
2363 (vect_set_loop_masks_directly): Likewise.
2364 (vect_set_loop_condition_masked): Likewise.
2365 (vect_set_loop_condition_unmasked): New function, split out from
2366 slpeel_make_loop_iterate_ntimes.
2367 (slpeel_make_loop_iterate_ntimes): Rename to..
2368 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2369 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2370 (vect_do_peeling): Update call accordingly.
2371 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2373 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2374 mask_compare_type, can_fully_mask_p and fully_masked_p.
2375 (release_vec_loop_masks): New function.
2376 (_loop_vec_info): Use it to free the loop masks.
2377 (can_produce_all_loop_masks_p): New function.
2378 (vect_get_max_nscalars_per_iter): Likewise.
2379 (vect_verify_full_masking): Likewise.
2380 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2381 retries, and free the mask rgroups before retrying. Check loop-wide
2382 reasons for disallowing fully-masked loops. Make the final decision
2383 about whether use a fully-masked loop or not.
2384 (vect_estimate_min_profitable_iters): Do not assume that peeling
2385 for the number of iterations will be needed for fully-masked loops.
2386 (vectorizable_reduction): Disable fully-masked loops.
2387 (vectorizable_live_operation): Likewise.
2388 (vect_halve_mask_nunits): New function.
2389 (vect_double_mask_nunits): Likewise.
2390 (vect_record_loop_mask): Likewise.
2391 (vect_get_loop_mask): Likewise.
2392 (vect_transform_loop): Handle the case in which the final loop
2393 iteration might handle a partial vector. Call vect_set_loop_condition
2394 instead of slpeel_make_loop_iterate_ntimes.
2395 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2396 (check_load_store_masking): New function.
2397 (prepare_load_store_mask): Likewise.
2398 (vectorizable_store): Handle fully-masked loops.
2399 (vectorizable_load): Likewise.
2400 (supportable_widening_operation): Use vect_halve_mask_nunits for
2402 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2403 (vect_gen_while): New function.
2404 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2405 (aarch64_uqdec<mode>): New insn.
2407 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2408 Alan Hayward <alan.hayward@arm.com>
2409 David Sherwood <david.sherwood@arm.com>
2411 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2412 (reduc_xor_scal_optab): New optabs.
2413 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2414 (reduc_xor_scal_@var{m}): Document.
2415 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2416 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2418 * fold-const-call.c (fold_const_call): Handle them.
2419 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2420 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2421 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2422 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2423 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2424 (UNSPEC_XORV): New unspecs.
2425 (optab): Add entries for them.
2426 (BITWISEV): New int iterator.
2427 (bit_reduc_op): New int attributes.
2429 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2430 Alan Hayward <alan.hayward@arm.com>
2431 David Sherwood <david.sherwood@arm.com>
2433 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2434 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2435 * optabs.def (vec_shl_insert_optab): New optab.
2436 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2437 (duplicate_and_interleave): Likewise.
2438 * tree-vect-loop.c: Include internal-fn.h.
2439 (neutral_op_for_slp_reduction): New function, split out from
2440 get_initial_defs_for_reduction.
2441 (get_initial_def_for_reduction): Handle option 2 for variable-length
2442 vectors by loading the neutral value into a vector and then shifting
2443 the initial value into element 0.
2444 (get_initial_defs_for_reduction): Replace the code argument with
2445 the neutral value calculated by neutral_op_for_slp_reduction.
2446 Use gimple_build_vector for constant-length vectors.
2447 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2448 but the first group_size elements have a neutral value.
2449 Use duplicate_and_interleave otherwise.
2450 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2451 Update call to get_initial_defs_for_reduction. Handle SLP
2452 reductions for variable-length vectors by creating one vector
2453 result for each scalar result, with the elements associated
2454 with other scalar results stubbed out with the neutral value.
2455 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2456 Require IFN_VEC_SHL_INSERT for double reductions on
2457 variable-length vectors, or SLP reductions that have
2458 a neutral value. Require can_duplicate_and_interleave_p
2459 support for variable-length unchained SLP reductions if there
2460 is no neutral value, such as for MIN/MAX reductions. Also require
2461 the number of vector elements to be a multiple of the number of
2462 SLP statements when doing variable-length unchained SLP reductions.
2463 Update call to vect_create_epilog_for_reduction.
2464 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2465 and remove initial values.
2466 (duplicate_and_interleave): Make public.
2467 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2468 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2470 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2471 Alan Hayward <alan.hayward@arm.com>
2472 David Sherwood <david.sherwood@arm.com>
2474 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2475 (can_duplicate_and_interleave_p): New function.
2476 (vect_get_and_check_slp_defs): Take the vector of statements
2477 rather than just the current one. Remove excess parentheses.
2478 Restriction rejectinon of vect_constant_def and vect_external_def
2479 for variable-length vectors to boolean types, or types for which
2480 can_duplicate_and_interleave_p is false.
2481 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2482 (duplicate_and_interleave): New function.
2483 (vect_get_constant_vectors): Use gimple_build_vector for
2484 constant-length vectors and suitable variable-length constant
2485 vectors. Use duplicate_and_interleave for other variable-length
2486 vectors. Don't defer the update when inserting new statements.
2488 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2489 Alan Hayward <alan.hayward@arm.com>
2490 David Sherwood <david.sherwood@arm.com>
2492 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2493 min_profitable_iters doesn't go negative.
2495 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2496 Alan Hayward <alan.hayward@arm.com>
2497 David Sherwood <david.sherwood@arm.com>
2499 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2500 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2501 * optabs.def (vec_mask_load_lanes_optab): New optab.
2502 (vec_mask_store_lanes_optab): Likewise.
2503 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2504 (MASK_STORE_LANES): Likewise.
2505 * internal-fn.c (mask_load_lanes_direct): New macro.
2506 (mask_store_lanes_direct): Likewise.
2507 (expand_mask_load_optab_fn): Handle masked operations.
2508 (expand_mask_load_lanes_optab_fn): New macro.
2509 (expand_mask_store_optab_fn): Handle masked operations.
2510 (expand_mask_store_lanes_optab_fn): New macro.
2511 (direct_mask_load_lanes_optab_supported_p): Likewise.
2512 (direct_mask_store_lanes_optab_supported_p): Likewise.
2513 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2515 (vect_load_lanes_supported): Likewise.
2516 * tree-vect-data-refs.c (strip_conversion): New function.
2517 (can_group_stmts_p): Likewise.
2518 (vect_analyze_data_ref_accesses): Use it instead of checking
2519 for a pair of assignments.
2520 (vect_store_lanes_supported): Take a masked_p parameter.
2521 (vect_load_lanes_supported): Likewise.
2522 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2523 vect_store_lanes_supported and vect_load_lanes_supported.
2524 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2525 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2526 parameter. Don't allow gaps for masked accesses.
2527 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2528 and vect_load_lanes_supported.
2529 (get_load_store_type): Take a masked_p parameter and update
2530 call to get_group_load_store_type.
2531 (vectorizable_store): Update call to get_load_store_type.
2532 Handle IFN_MASK_STORE_LANES.
2533 (vectorizable_load): Update call to get_load_store_type.
2534 Handle IFN_MASK_LOAD_LANES.
2536 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2537 Alan Hayward <alan.hayward@arm.com>
2538 David Sherwood <david.sherwood@arm.com>
2540 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2542 * config/aarch64/aarch64-protos.h
2543 (aarch64_sve_struct_memory_operand_p): Declare.
2544 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2545 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2546 (VPRED, vpred): Handle SVE structure modes.
2547 * config/aarch64/constraints.md (Utx): New constraint.
2548 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2549 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2550 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2551 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2552 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2553 structure modes. Split into pieces after RA.
2554 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2555 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2557 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2558 SVE structure modes.
2559 (aarch64_classify_address): Likewise.
2560 (sizetochar): Move earlier in file.
2561 (aarch64_print_operand): Handle SVE register lists.
2562 (aarch64_array_mode): New function.
2563 (aarch64_sve_struct_memory_operand_p): Likewise.
2564 (TARGET_ARRAY_MODE): Redefine.
2566 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2567 Alan Hayward <alan.hayward@arm.com>
2568 David Sherwood <david.sherwood@arm.com>
2570 * target.def (array_mode): New target hook.
2571 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2572 * doc/tm.texi: Regenerate.
2573 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2574 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2575 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2577 * stor-layout.c (mode_for_array): Likewise. Support polynomial
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 * fold-const.c (fold_binary_loc): Check the argument types
2585 rather than the result type when testing for a vector operation.
2587 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2589 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2590 * doc/tm.texi: Regenerate.
2592 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2593 Alan Hayward <alan.hayward@arm.com>
2594 David Sherwood <david.sherwood@arm.com>
2596 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2597 (sve): Document new AArch64 extension.
2598 * doc/md.texi (w): Extend the description of the AArch64
2599 constraint to include SVE vectors.
2600 (Upl, Upa): Document new AArch64 predicate constraints.
2601 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2603 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2604 (msve-vector-bits=): New option.
2605 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2606 SVE when these are disabled.
2607 (sve): New extension.
2608 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2609 modes. Adjust their number of units based on aarch64_sve_vg.
2610 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2611 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2612 aarch64_addr_query_type.
2613 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2614 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2615 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2616 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2617 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2618 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2619 (aarch64_simd_imm_zero_p): Delete.
2620 (aarch64_check_zero_based_sve_index_immediate): Declare.
2621 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2622 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2623 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2624 (aarch64_sve_float_mul_immediate_p): Likewise.
2625 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2627 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2628 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2629 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2630 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2631 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2632 (aarch64_regmode_natural_size): Likewise.
2633 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2634 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2636 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2637 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2638 for VG and the SVE predicate registers.
2639 (V_ALIASES): Add a "z"-prefixed alias.
2640 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2641 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2642 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2643 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2644 (REG_CLASS_NAMES): Add entries for them.
2645 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2646 and the predicate registers.
2647 (aarch64_sve_vg): Declare.
2648 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2649 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2650 (REGMODE_NATURAL_SIZE): Define.
2651 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2653 * config/aarch64/aarch64.c: Include cfgrtl.h.
2654 (simd_immediate_info): Add a constructor for series vectors,
2655 and an associated step field.
2656 (aarch64_sve_vg): New variable.
2657 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2658 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2659 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2660 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2661 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2662 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2663 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2664 (aarch64_get_mask_mode): New functions.
2665 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2666 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2667 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2668 predicate modes and predicate registers. Explicitly restrict
2669 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2670 to store a vector mode if it is recognized by
2671 aarch64_classify_vector_mode.
2672 (aarch64_regmode_natural_size): New function.
2673 (aarch64_hard_regno_caller_save_mode): Return the original mode
2675 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2676 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2677 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2678 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2680 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2681 does not overlap dest if the function is frame-related. Handle
2683 (aarch64_split_add_offset): New function.
2684 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2685 them aarch64_add_offset.
2686 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2687 and update call to aarch64_sub_sp.
2688 (aarch64_add_cfa_expression): New function.
2689 (aarch64_expand_prologue): Pass extra temporary registers to the
2690 functions above. Handle the case in which we need to emit new
2691 DW_CFA_expressions for registers that were originally saved
2692 relative to the stack pointer, but now have to be expressed
2693 relative to the frame pointer.
2694 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2696 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2697 IP0 and IP1 values for SVE frames.
2698 (aarch64_expand_vec_series): New function.
2699 (aarch64_expand_sve_widened_duplicate): Likewise.
2700 (aarch64_expand_sve_const_vector): Likewise.
2701 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2702 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2703 into the register, rather than emitting a SET directly.
2704 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2705 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2706 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2707 (offset_9bit_signed_scaled_p): New functions.
2708 (aarch64_replicate_bitmask_imm): New function.
2709 (aarch64_bitmask_imm): Use it.
2710 (aarch64_cannot_force_const_mem): Reject expressions involving
2711 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2712 (aarch64_classify_index): Handle SVE indices, by requiring
2713 a plain register index with a scale that matches the element size.
2714 (aarch64_classify_address): Handle SVE addresses. Assert that
2715 the mode of the address is VOIDmode or an integer mode.
2716 Update call to aarch64_classify_symbol.
2717 (aarch64_classify_symbolic_expression): Update call to
2718 aarch64_classify_symbol.
2719 (aarch64_const_vec_all_in_range_p): New function.
2720 (aarch64_print_vector_float_operand): Likewise.
2721 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2722 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2723 and the FP immediates 1.0 and 0.5.
2724 (aarch64_print_address_internal): Handle SVE addresses.
2725 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2726 (aarch64_regno_regclass): Handle predicate registers.
2727 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2729 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2730 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2731 (aarch64_convert_sve_vector_bits): New function.
2732 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2733 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2735 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2736 Handle SVE vector and predicate modes. Accept VL-based constants
2737 that need only one temporary register, and VL offsets that require
2738 no temporary registers.
2739 (aarch64_conditional_register_usage): Mark the predicate registers
2740 as fixed if SVE isn't available.
2741 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2742 Return true for SVE vector and predicate modes.
2743 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2744 rather than an unsigned int. Handle SVE modes.
2745 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2747 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2749 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2750 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2751 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2752 (aarch64_sve_float_mul_immediate_p): New functions.
2753 (aarch64_sve_valid_immediate): New function.
2754 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2755 Explicitly reject structure modes. Check for INDEX constants.
2756 Handle PTRUE and PFALSE constants.
2757 (aarch64_check_zero_based_sve_index_immediate): New function.
2758 (aarch64_simd_imm_zero_p): Delete.
2759 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2760 vector modes. Accept constants in the range of CNT[BHWD].
2761 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2762 ask for an Advanced SIMD mode.
2763 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2764 (aarch64_simd_vector_alignment): Handle SVE predicates.
2765 (aarch64_vectorize_preferred_vector_alignment): New function.
2766 (aarch64_simd_vector_alignment_reachable): Use it instead of
2768 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2769 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2771 (MAX_VECT_LEN): Delete.
2772 (expand_vec_perm_d): Add a vec_flags field.
2773 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2774 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2775 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2777 (aarch64_evpc_rev): Rename to...
2778 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2779 (aarch64_evpc_rev_global): New function.
2780 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2781 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2783 (aarch64_evpc_sve_tbl): New function.
2784 (aarch64_expand_vec_perm_const_1): Update after rename of
2785 aarch64_evpc_rev. Handle SVE permutes too, trying
2786 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2787 than aarch64_evpc_tbl.
2788 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2789 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2790 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2791 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2792 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2793 (aarch64_expand_sve_vcond): New functions.
2794 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2795 of aarch64_vector_mode_p.
2796 (aarch64_dwarf_poly_indeterminate_value): New function.
2797 (aarch64_compute_pressure_classes): Likewise.
2798 (aarch64_can_change_mode_class): Likewise.
2799 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2800 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2801 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2802 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2803 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2804 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2805 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2806 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2808 (Dn, Dl, Dr): Accept const as well as const_vector.
2809 (Dz): Likewise. Compare against CONST0_RTX.
2810 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2811 of "vector" where appropriate.
2812 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2813 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2814 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2815 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2816 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2817 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2818 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2819 (v_int_equiv): Extend to SVE modes.
2820 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2822 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2823 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2824 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2825 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2826 (SVE_COND_FP_CMP): New int iterators.
2827 (perm_hilo): Handle the new unpack unspecs.
2828 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2830 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2831 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2832 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2833 (aarch64_equality_operator, aarch64_constant_vector_operand)
2834 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2835 (aarch64_sve_nonimmediate_operand): Likewise.
2836 (aarch64_sve_general_operand): Likewise.
2837 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2838 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2839 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2840 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2841 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2842 (aarch64_sve_float_arith_immediate): Likewise.
2843 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2844 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2845 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2846 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2847 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2848 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2849 (aarch64_sve_float_arith_operand): Likewise.
2850 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2851 (aarch64_sve_float_mul_operand): Likewise.
2852 (aarch64_sve_vec_perm_operand): Likewise.
2853 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2854 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2855 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2856 as well as const_vector.
2857 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2858 in file. Use CONST0_RTX and CONSTM1_RTX.
2859 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2860 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2861 Use aarch64_simd_imm_zero.
2862 * config/aarch64/aarch64-sve.md: New file.
2863 * config/aarch64/aarch64.md: Include it.
2864 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2865 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2866 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2867 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2868 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2869 (sve): New attribute.
2870 (enabled): Disable instructions with the sve attribute unless
2872 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2873 aarch64_expand_mov_immediate.
2874 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2875 CNT[BHSD] immediates.
2876 (movti): Split CONST_POLY_INT moves into two halves.
2877 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2878 Split additions that need a temporary here if the destination
2879 is the stack pointer.
2880 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2881 (*add<mode>3_poly_1): New instruction.
2882 (set_clobber_cc): New expander.
2884 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2886 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2887 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2888 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2889 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2890 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2891 Change innermode from fixed_mode_size to machine_mode.
2892 (simplify_subreg): Update call accordingly. Handle a constant-sized
2893 subreg of a variable-length CONST_VECTOR.
2895 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2896 Alan Hayward <alan.hayward@arm.com>
2897 David Sherwood <david.sherwood@arm.com>
2899 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2900 (add_offset_to_base): New function, split out from...
2901 (create_mem_ref): ...here. When handling a scale other than 1,
2902 check first whether the address is valid without the offset.
2903 Add it into the base if so, leaving the index and scale as-is.
2905 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2908 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2909 fold_for_warn before checking if arg2 is INTEGER_CST.
2911 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2913 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2914 (store_multiple_operation): Delete.
2915 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2916 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2917 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2918 guarded by TARGET_STRING.
2919 (rs6000_output_load_multiple): Delete.
2920 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2921 OPTION_MASK_STRING / TARGET_STRING handling.
2922 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2923 (const rs6000_opt_masks) <"string">: Change mask to 0.
2924 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2925 (MASK_STRING): Delete.
2926 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2928 (load_multiple): Delete.
2935 (store_multiple): Delete.
2942 (movmemsi_8reg): Delete.
2943 (corresponding unnamed define_insn): Delete.
2944 (movmemsi_6reg): Delete.
2945 (corresponding unnamed define_insn): Delete.
2946 (movmemsi_4reg): Delete.
2947 (corresponding unnamed define_insn): Delete.
2948 (movmemsi_2reg): Delete.
2949 (corresponding unnamed define_insn): Delete.
2950 (movmemsi_1reg): Delete.
2951 (corresponding unnamed define_insn): Delete.
2952 * config/rs6000/rs6000.opt (mno-string): New.
2953 (mstring): Replace by deprecation warning stub.
2954 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2956 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2958 * regrename.c (regrename_do_replace): If replacing the same
2959 reg multiple times, try to reuse last created gen_raw_REG.
2962 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2963 main to workaround a bug in GDB.
2965 2018-01-12 Tom de Vries <tom@codesourcery.com>
2968 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2970 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2972 PR rtl-optimization/80481
2973 * ira-color.c (get_cap_member): New function.
2974 (allocnos_conflict_by_live_ranges_p): Use it.
2975 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2976 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2978 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2981 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2982 (*saddl_se_1): Ditto.
2984 (*ssubl_se_1): Ditto.
2986 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2988 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2989 rather than wi::to_widest for DR_INITs.
2990 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2991 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2992 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2994 (vect_analyze_group_access_1): Note that here.
2996 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2998 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2999 polynomial type sizes.
3001 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3003 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3004 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3005 (gimple_add_tmp_var): Likewise.
3007 2018-01-12 Martin Liska <mliska@suse.cz>
3009 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3010 (gimple_alloc_sizes): Likewise.
3011 (dump_gimple_statistics): Use PRIu64 in printf format.
3012 * gimple.h: Change uint64_t to int.
3014 2018-01-12 Martin Liska <mliska@suse.cz>
3016 * tree-core.h: Use uint64_t instead of int.
3017 * tree.c (tree_node_counts): Likewise.
3018 (tree_node_sizes): Likewise.
3019 (dump_tree_statistics): Use PRIu64 in printf format.
3021 2018-01-12 Martin Liska <mliska@suse.cz>
3023 * Makefile.in: As qsort_chk is implemented in vec.c, add
3024 vec.o to linkage of gencfn-macros.
3025 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3026 passing the info to record_node_allocation_statistics.
3027 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3029 * ggc-common.c (struct ggc_usage): Add operator== and use
3030 it in operator< and compare function.
3031 * mem-stats.h (struct mem_usage): Likewise.
3032 * vec.c (struct vec_usage): Remove operator< and compare
3033 function. Can be simply inherited.
3035 2018-01-12 Martin Jambor <mjambor@suse.cz>
3038 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3039 * tree-ssa-math-opts.c: Include domwalk.h.
3040 (convert_mult_to_fma_1): New function.
3041 (fma_transformation_info): New type.
3042 (fma_deferring_state): Likewise.
3043 (cancel_fma_deferring): New function.
3044 (result_of_phi): Likewise.
3045 (last_fma_candidate_feeds_initial_phi): Likewise.
3046 (convert_mult_to_fma): Added deferring logic, split actual
3047 transformation to convert_mult_to_fma_1.
3048 (math_opts_dom_walker): New type.
3049 (math_opts_dom_walker::after_dom_children): New method, body moved
3050 here from pass_optimize_widening_mul::execute, added deferring logic
3052 (pass_optimize_widening_mul::execute): Moved most of code to
3053 math_opts_dom_walker::after_dom_children.
3054 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3055 * config/i386/i386.c (ix86_option_override_internal): Added
3056 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3058 2018-01-12 Richard Biener <rguenther@suse.de>
3061 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3062 inline instance vars.
3064 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3067 * config/rx/rx.c (rx_is_restricted_memory_address):
3070 2018-01-12 Richard Biener <rguenther@suse.de>
3072 PR tree-optimization/80846
3073 * target.def (split_reduction): New target hook.
3074 * targhooks.c (default_split_reduction): New function.
3075 * targhooks.h (default_split_reduction): Declare.
3076 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3077 target requests first reduce vectors by combining low and high
3079 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3080 (get_vectype_for_scalar_type_and_size): Export.
3081 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3082 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3083 * doc/tm.texi: Regenerate.
3084 * config/i386/i386.c (ix86_split_reduction): Implement
3085 TARGET_VECTORIZE_SPLIT_REDUCTION.
3087 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3090 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3091 in PIC mode except for TARGET_VXWORKS_RTP.
3092 * config/sparc/sparc.c: Include cfgrtl.h.
3093 (TARGET_INIT_PIC_REG): Define.
3094 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3095 (sparc_pic_register_p): New predicate.
3096 (sparc_legitimate_address_p): Use it.
3097 (sparc_legitimize_pic_address): Likewise.
3098 (sparc_delegitimize_address): Likewise.
3099 (sparc_mode_dependent_address_p): Likewise.
3100 (gen_load_pcrel_sym): Remove 4th parameter.
3101 (load_got_register): Adjust call to above. Remove obsolete stuff.
3102 (sparc_expand_prologue): Do not call load_got_register here.
3103 (sparc_flat_expand_prologue): Likewise.
3104 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3105 (sparc_use_pseudo_pic_reg): New function.
3106 (sparc_init_pic_reg): Likewise.
3107 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3108 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3110 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3112 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3113 Add item for branch_cost.
3115 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3117 PR rtl-optimization/83565
3118 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3119 not extend the result to a larger mode for rotate operations.
3120 (num_sign_bit_copies1): Likewise.
3122 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3125 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3127 Use values-Xc.o for -pedantic.
3128 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3130 2018-01-12 Martin Liska <mliska@suse.cz>
3133 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3135 (possible_polymorphic_call_targets): Use it.
3136 (ipa_devirt): Likewise.
3138 2018-01-12 Martin Liska <mliska@suse.cz>
3140 * profile-count.h (enum profile_quality): Use 0 as invalid
3141 enum value of profile_quality.
3143 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3145 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3146 -mext-string options.
3148 2018-01-12 Richard Biener <rguenther@suse.de>
3150 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3151 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3152 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3154 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3156 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3158 * configure.ac (--with-long-double-format): Add support for the
3159 configuration option to change the default long double format on
3161 * config.gcc (powerpc*-linux*-*): Likewise.
3162 * configure: Regenerate.
3163 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3164 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3165 used without modification.
3167 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3169 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3170 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3171 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3172 MISC_BUILTIN_SPEC_BARRIER.
3173 (rs6000_init_builtins): Likewise.
3174 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3176 (speculation_barrier): New define_insn.
3177 * doc/extend.texi: Document __builtin_speculation_barrier.
3179 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3182 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3183 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3184 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3186 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3187 integral modes instead of "ss" and "sd".
3188 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3189 vectors with 32-bit and 64-bit elements.
3190 (vecdupssescalarmodesuffix): New mode attribute.
3191 (vec_dup<mode>): Use it.
3193 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3196 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3197 frame if argument is passed on stack.
3199 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3202 * ree.c (combine_reaching_defs): Optimize also
3203 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3204 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3206 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3209 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3211 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3214 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3215 after they are computed.
3217 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3219 PR tree-optimization/83695
3220 * gimple-loop-linterchange.cc
3221 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3222 reset cached scev information after interchange.
3223 (pass_linterchange::execute): Remove call to scev_reset_htab.
3225 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3227 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3228 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3229 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3230 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3231 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3232 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3233 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3234 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3235 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3236 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3237 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3238 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3239 (V_lane_reg): Likewise.
3240 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3242 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3243 (vfmal_lane_low<mode>_intrinsic,
3244 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3245 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3246 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3247 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3248 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3249 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3251 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3253 * config/arm/arm-cpus.in (fp16fml): New feature.
3254 (ALL_SIMD): Add fp16fml.
3255 (armv8.2-a): Add fp16fml as an option.
3256 (armv8.3-a): Likewise.
3257 (armv8.4-a): Add fp16fml as part of fp16.
3258 * config/arm/arm.h (TARGET_FP16FML): Define.
3259 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3261 * config/arm/arm-modes.def (V2HF): Define.
3262 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3263 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3264 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3265 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3266 vfmsl_low, vfmsl_high): New set of builtins.
3267 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3268 (vfml_op): New code attribute.
3269 (VFMLHALVES): New int iterator.
3270 (VFML, VFMLSEL): New mode attributes.
3271 (V_reg): Define mapping for V2HF.
3272 (V_hi, V_lo): New mode attributes.
3273 (VF_constraint): Likewise.
3274 (vfml_half, vfml_half_selector): New int attributes.
3275 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3277 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3278 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3280 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3281 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3282 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3283 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3285 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3286 Document new effective target and option set.
3288 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3290 * config/arm/arm-cpus.in (armv8_4): New feature.
3291 (ARMv8_4a): New fgroup.
3292 (armv8.4-a): New arch.
3293 * config/arm/arm-tables.opt: Regenerate.
3294 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3295 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3296 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3297 Add matching rules for -march=armv8.4-a and extensions.
3298 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3300 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3303 * config/rx/rx.md (BW): New mode attribute.
3304 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3306 2018-01-11 Richard Biener <rguenther@suse.de>
3308 PR tree-optimization/83435
3309 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3310 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3311 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3313 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3314 Alan Hayward <alan.hayward@arm.com>
3315 David Sherwood <david.sherwood@arm.com>
3317 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3319 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3320 (aarch64_print_address_internal): Use it to check for a zero offset.
3322 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3323 Alan Hayward <alan.hayward@arm.com>
3324 David Sherwood <david.sherwood@arm.com>
3326 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3327 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3328 Return a poly_int64 rather than a HOST_WIDE_INT.
3329 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3330 rather than a HOST_WIDE_INT.
3331 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3332 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3333 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3334 final_offset from HOST_WIDE_INT to poly_int64.
3335 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3336 to_constant when getting the number of units in an Advanced SIMD
3338 (aarch64_builtin_vectorized_function): Check for a constant number
3340 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3342 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3343 attribute instead of GET_MODE_NUNITS.
3344 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3345 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3346 GET_MODE_SIZE for fixed-size registers.
3347 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3348 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3349 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3350 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3351 (aarch64_print_operand, aarch64_print_address_internal)
3352 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3353 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3354 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3355 Handle polynomial GET_MODE_SIZE.
3356 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3357 wider than SImode without modification.
3358 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3359 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3360 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3361 passing and returning SVE modes.
3362 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3363 rather than GEN_INT.
3364 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3365 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3366 (aarch64_allocate_and_probe_stack_space): Likewise.
3367 (aarch64_layout_frame): Cope with polynomial offsets.
3368 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3369 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3371 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3372 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3373 poly_int64 rather than a HOST_WIDE_INT.
3374 (aarch64_get_separate_components, aarch64_process_components)
3375 (aarch64_expand_prologue, aarch64_expand_epilogue)
3376 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3377 (aarch64_anchor_offset): New function, split out from...
3378 (aarch64_legitimize_address): ...here.
3379 (aarch64_builtin_vectorization_cost): Handle polynomial
3380 TYPE_VECTOR_SUBPARTS.
3381 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3383 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3384 number of elements from the PARALLEL rather than the mode.
3385 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3386 rather than GET_MODE_BITSIZE.
3387 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3388 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3389 (aarch64_expand_vec_perm_const_1): Handle polynomial
3390 d->perm.length () and d->perm elements.
3391 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3392 Apply to_constant to d->perm elements.
3393 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3394 polynomial CONST_VECTOR_NUNITS.
3395 (aarch64_move_pointer): Take amount as a poly_int64 rather
3397 (aarch64_progress_pointer): Avoid temporary variable.
3398 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3399 the mode attribute instead of GET_MODE.
3401 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3402 Alan Hayward <alan.hayward@arm.com>
3403 David Sherwood <david.sherwood@arm.com>
3405 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3406 x exists before using it.
3407 (aarch64_add_constant_internal): Rename to...
3408 (aarch64_add_offset_1): ...this. Replace regnum with separate
3409 src and dest rtxes. Handle the case in which they're different,
3410 including when the offset is zero. Replace scratchreg with an rtx.
3411 Use 2 additions if there is no spare register into which we can
3412 move a 16-bit constant.
3413 (aarch64_add_constant): Delete.
3414 (aarch64_add_offset): Replace reg with separate src and dest
3415 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3416 Use aarch64_add_offset_1.
3417 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3418 an rtx rather than an int. Take the delta as a poly_int64
3419 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3420 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3421 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3422 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3423 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3425 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3426 aarch64_add_constant.
3428 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3430 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3431 Use scalar_float_mode.
3433 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3435 * config/aarch64/aarch64-simd.md
3436 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3437 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3438 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3439 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3440 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3441 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3442 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3443 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3444 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3445 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3447 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3450 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3451 targ_options->x_arm_arch_string is non NULL.
3453 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3455 * config/aarch64/aarch64.h
3456 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3458 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3461 * expmed.c (emit_store_flag_force): Swap if const op0
3462 and change VOIDmode to mode of op0.
3464 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3466 PR rtl-optimization/83761
3467 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3468 than bytes to mode_for_size.
3470 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3473 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3474 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3477 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3480 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3481 when in layout mode.
3482 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3483 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3486 2018-01-10 Michael Collison <michael.collison@arm.com>
3488 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3489 * config/aarch64/aarch64-option-extension.def: Add
3490 AARCH64_OPT_EXTENSION of 'fp16fml'.
3491 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3492 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3493 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3494 * config/aarch64/constraints.md (Ui7): New constraint.
3495 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3496 (VFMLA_SEL_W): Ditto.
3499 (VFMLA16_LOW): New int iterator.
3500 (VFMLA16_HIGH): Ditto.
3501 (UNSPEC_FMLAL): New unspec.
3502 (UNSPEC_FMLSL): Ditto.
3503 (UNSPEC_FMLAL2): Ditto.
3504 (UNSPEC_FMLSL2): Ditto.
3505 (f16mac): New code attribute.
3506 * config/aarch64/aarch64-simd-builtins.def
3507 (aarch64_fmlal_lowv2sf): Ditto.
3508 (aarch64_fmlsl_lowv2sf): Ditto.
3509 (aarch64_fmlalq_lowv4sf): Ditto.
3510 (aarch64_fmlslq_lowv4sf): Ditto.
3511 (aarch64_fmlal_highv2sf): Ditto.
3512 (aarch64_fmlsl_highv2sf): Ditto.
3513 (aarch64_fmlalq_highv4sf): Ditto.
3514 (aarch64_fmlslq_highv4sf): Ditto.
3515 (aarch64_fmlal_lane_lowv2sf): Ditto.
3516 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3517 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3518 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3519 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3520 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3521 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3522 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3523 (aarch64_fmlal_lane_highv2sf): Ditto.
3524 (aarch64_fmlsl_lane_highv2sf): Ditto.
3525 (aarch64_fmlal_laneq_highv2sf): Ditto.
3526 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3527 (aarch64_fmlalq_lane_highv4sf): Ditto.
3528 (aarch64_fmlsl_lane_highv4sf): Ditto.
3529 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3530 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3531 * config/aarch64/aarch64-simd.md:
3532 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3533 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3534 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3535 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3536 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3537 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3538 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3539 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3540 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3541 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3542 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3543 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3544 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3545 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3546 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3547 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3548 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3549 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3550 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3551 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3552 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3553 (vfmlsl_low_u32): Ditto.
3554 (vfmlalq_low_u32): Ditto.
3555 (vfmlslq_low_u32): Ditto.
3556 (vfmlal_high_u32): Ditto.
3557 (vfmlsl_high_u32): Ditto.
3558 (vfmlalq_high_u32): Ditto.
3559 (vfmlslq_high_u32): Ditto.
3560 (vfmlal_lane_low_u32): Ditto.
3561 (vfmlsl_lane_low_u32): Ditto.
3562 (vfmlal_laneq_low_u32): Ditto.
3563 (vfmlsl_laneq_low_u32): Ditto.
3564 (vfmlalq_lane_low_u32): Ditto.
3565 (vfmlslq_lane_low_u32): Ditto.
3566 (vfmlalq_laneq_low_u32): Ditto.
3567 (vfmlslq_laneq_low_u32): Ditto.
3568 (vfmlal_lane_high_u32): Ditto.
3569 (vfmlsl_lane_high_u32): Ditto.
3570 (vfmlal_laneq_high_u32): Ditto.
3571 (vfmlsl_laneq_high_u32): Ditto.
3572 (vfmlalq_lane_high_u32): Ditto.
3573 (vfmlslq_lane_high_u32): Ditto.
3574 (vfmlalq_laneq_high_u32): Ditto.
3575 (vfmlslq_laneq_high_u32): Ditto.
3576 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3577 (AARCH64_FL_FOR_ARCH8_4): New.
3578 (AARCH64_ISA_F16FML): New ISA flag.
3579 (TARGET_F16FML): New feature flag for fp16fml.
3580 (doc/invoke.texi): Document new fp16fml option.
3582 2018-01-10 Michael Collison <michael.collison@arm.com>
3584 * config/aarch64/aarch64-builtins.c:
3585 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3586 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3587 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3588 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3589 (AARCH64_ISA_SHA3): New ISA flag.
3590 (TARGET_SHA3): New feature flag for sha3.
3591 * config/aarch64/iterators.md (sha512_op): New int attribute.
3592 (CRYPTO_SHA512): New int iterator.
3593 (UNSPEC_SHA512H): New unspec.
3594 (UNSPEC_SHA512H2): Ditto.
3595 (UNSPEC_SHA512SU0): Ditto.
3596 (UNSPEC_SHA512SU1): Ditto.
3597 * config/aarch64/aarch64-simd-builtins.def
3598 (aarch64_crypto_sha512hqv2di): New builtin.
3599 (aarch64_crypto_sha512h2qv2di): Ditto.
3600 (aarch64_crypto_sha512su0qv2di): Ditto.
3601 (aarch64_crypto_sha512su1qv2di): Ditto.
3602 (aarch64_eor3qv8hi): Ditto.
3603 (aarch64_rax1qv2di): Ditto.
3604 (aarch64_xarqv2di): Ditto.
3605 (aarch64_bcaxqv8hi): Ditto.
3606 * config/aarch64/aarch64-simd.md:
3607 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3608 (aarch64_crypto_sha512su0qv2di): Ditto.
3609 (aarch64_crypto_sha512su1qv2di): Ditto.
3610 (aarch64_eor3qv8hi): Ditto.
3611 (aarch64_rax1qv2di): Ditto.
3612 (aarch64_xarqv2di): Ditto.
3613 (aarch64_bcaxqv8hi): Ditto.
3614 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3615 (vsha512h2q_u64): Ditto.
3616 (vsha512su0q_u64): Ditto.
3617 (vsha512su1q_u64): Ditto.
3618 (veor3q_u16): Ditto.
3619 (vrax1q_u64): Ditto.
3621 (vbcaxq_u16): Ditto.
3622 * config/arm/types.md (crypto_sha512): New type attribute.
3623 (crypto_sha3): Ditto.
3624 (doc/invoke.texi): Document new sha3 option.
3626 2018-01-10 Michael Collison <michael.collison@arm.com>
3628 * config/aarch64/aarch64-builtins.c:
3629 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3630 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3631 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3632 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3633 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3634 (AARCH64_ISA_SM4): New ISA flag.
3635 (TARGET_SM4): New feature flag for sm4.
3636 * config/aarch64/aarch64-simd-builtins.def
3637 (aarch64_sm3ss1qv4si): Ditto.
3638 (aarch64_sm3tt1aq4si): Ditto.
3639 (aarch64_sm3tt1bq4si): Ditto.
3640 (aarch64_sm3tt2aq4si): Ditto.
3641 (aarch64_sm3tt2bq4si): Ditto.
3642 (aarch64_sm3partw1qv4si): Ditto.
3643 (aarch64_sm3partw2qv4si): Ditto.
3644 (aarch64_sm4eqv4si): Ditto.
3645 (aarch64_sm4ekeyqv4si): Ditto.
3646 * config/aarch64/aarch64-simd.md:
3647 (aarch64_sm3ss1qv4si): Ditto.
3648 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3649 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3650 (aarch64_sm4eqv4si): Ditto.
3651 (aarch64_sm4ekeyqv4si): Ditto.
3652 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3653 (sm3part_op): Ditto.
3654 (CRYPTO_SM3TT): Ditto.
3655 (CRYPTO_SM3PART): Ditto.
3656 (UNSPEC_SM3SS1): New unspec.
3657 (UNSPEC_SM3TT1A): Ditto.
3658 (UNSPEC_SM3TT1B): Ditto.
3659 (UNSPEC_SM3TT2A): Ditto.
3660 (UNSPEC_SM3TT2B): Ditto.
3661 (UNSPEC_SM3PARTW1): Ditto.
3662 (UNSPEC_SM3PARTW2): Ditto.
3663 (UNSPEC_SM4E): Ditto.
3664 (UNSPEC_SM4EKEY): Ditto.
3665 * config/aarch64/constraints.md (Ui2): New constraint.
3666 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3667 * config/arm/types.md (crypto_sm3): New type attribute.
3668 (crypto_sm4): Ditto.
3669 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3670 (vsm3tt1aq_u32): Ditto.
3671 (vsm3tt1bq_u32): Ditto.
3672 (vsm3tt2aq_u32): Ditto.
3673 (vsm3tt2bq_u32): Ditto.
3674 (vsm3partw1q_u32): Ditto.
3675 (vsm3partw2q_u32): Ditto.
3676 (vsm4eq_u32): Ditto.
3677 (vsm4ekeyq_u32): Ditto.
3678 (doc/invoke.texi): Document new sm4 option.
3680 2018-01-10 Michael Collison <michael.collison@arm.com>
3682 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3683 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3684 (AARCH64_FL_FOR_ARCH8_4): New.
3685 (AARCH64_FL_V8_4): New flag.
3686 (doc/invoke.texi): Document new armv8.4-a option.
3688 2018-01-10 Michael Collison <michael.collison@arm.com>
3690 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3691 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3692 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3693 * config/aarch64/aarch64-option-extension.def: Add
3694 AARCH64_OPT_EXTENSION of 'sha2'.
3695 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3696 (crypto): Disable sha2 and aes if crypto disabled.
3697 (crypto): Enable aes and sha2 if enabled.
3698 (simd): Disable sha2 and aes if simd disabled.
3699 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3701 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3702 (TARGET_SHA2): New feature flag for sha2.
3703 (TARGET_AES): New feature flag for aes.
3704 * config/aarch64/aarch64-simd.md:
3705 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3706 conditional on TARGET_AES.
3707 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3708 (aarch64_crypto_sha1hsi): Make pattern conditional
3710 (aarch64_crypto_sha1hv4si): Ditto.
3711 (aarch64_be_crypto_sha1hv4si): Ditto.
3712 (aarch64_crypto_sha1su1v4si): Ditto.
3713 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3714 (aarch64_crypto_sha1su0v4si): Ditto.
3715 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3716 (aarch64_crypto_sha256su0v4si): Ditto.
3717 (aarch64_crypto_sha256su1v4si): Ditto.
3718 (doc/invoke.texi): Document new aes and sha2 options.
3720 2018-01-10 Martin Sebor <msebor@redhat.com>
3722 PR tree-optimization/83781
3723 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3726 2018-01-11 Martin Sebor <msebor@gmail.com>
3727 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3729 PR tree-optimization/83501
3730 PR tree-optimization/81703
3732 * tree-ssa-strlen.c (get_string_cst): Rename...
3733 (get_string_len): ...to this. Handle global constants.
3734 (handle_char_store): Adjust.
3736 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3737 Jim Wilson <jimw@sifive.com>
3739 * config/riscv/riscv-protos.h (riscv_output_return): New.
3740 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3741 (riscv_attribute_table, riscv_output_return),
3742 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3743 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3744 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3746 (riscv_expand_prologue): Add early return for naked function.
3747 (riscv_expand_epilogue): Likewise.
3748 (riscv_function_ok_for_sibcall): Return false for naked function.
3749 (riscv_set_current_function): New.
3750 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3751 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3752 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3753 * doc/extend.texi (RISC-V Function Attributes): New.
3755 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3757 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3758 check for 128-bit long double before checking TCmode.
3759 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3760 128-bit long doubles before checking TFmode or TCmode.
3761 (FLOAT128_IBM_P): Likewise.
3763 2018-01-10 Martin Sebor <msebor@redhat.com>
3765 PR tree-optimization/83671
3766 * builtins.c (c_strlen): Unconditionally return zero for the empty
3768 Use -Warray-bounds for warnings.
3769 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3770 for non-constant array indices with COMPONENT_REF, arrays of
3771 arrays, and pointers to arrays.
3772 (gimple_fold_builtin_strlen): Determine and set length range for
3773 non-constant character arrays.
3775 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3778 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3781 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3783 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3785 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3788 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3789 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3790 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3791 indexed_or_indirect_operand predicate.
3792 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3793 (*vsx_le_perm_load_v8hi): Likewise.
3794 (*vsx_le_perm_load_v16qi): Likewise.
3795 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3796 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3797 (*vsx_le_perm_store_v8hi): Likewise.
3798 (*vsx_le_perm_store_v16qi): Likewise.
3799 (eight unnamed splitters): Likewise.
3801 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3803 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3804 * config/rs6000/emmintrin.h: Likewise.
3805 * config/rs6000/mmintrin.h: Likewise.
3806 * config/rs6000/xmmintrin.h: Likewise.
3808 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3811 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3813 * tree.c (tree_nop_conversion): Return true for location wrapper
3815 (maybe_wrap_with_location): New function.
3816 (selftest::check_strip_nops): New function.
3817 (selftest::test_location_wrappers): New function.
3818 (selftest::tree_c_tests): Call it.
3819 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3820 (maybe_wrap_with_location): New decl.
3821 (EXPR_LOCATION_WRAPPER_P): New macro.
3822 (location_wrapper_p): New inline function.
3823 (tree_strip_any_location_wrapper): New inline function.
3825 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3828 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3829 stack_realign_offset for the largest alignment of stack slot
3831 (ix86_find_max_used_stack_alignment): New function.
3832 (ix86_finalize_stack_frame_flags): Use it. Set
3833 max_used_stack_alignment if we don't realign stack.
3834 * config/i386/i386.h (machine_function): Add
3835 max_used_stack_alignment.
3837 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3839 * config/arm/arm.opt (-mbranch-cost): New option.
3840 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3843 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3846 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3847 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3849 2018-01-10 Richard Biener <rguenther@suse.de>
3852 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3853 early out so it also covers the case where we have a non-NULL
3856 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3858 PR tree-optimization/83753
3859 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3860 for non-strided grouped accesses if the number of elements is 1.
3862 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3865 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3866 * i386.h (TARGET_USE_GATHER): Define.
3867 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3869 2018-01-10 Martin Liska <mliska@suse.cz>
3872 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3873 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3875 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3876 CLEANUP_NO_PARTITIONING is not set.
3878 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3880 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3881 for vectors, as a partial revert of r254296.
3882 * rtl.h (const_vec_p): Delete.
3883 (const_vec_duplicate_p): Don't test for vector CONSTs.
3884 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3885 * expmed.c (make_tree): Likewise.
3888 * common.md (E, F): Use CONSTANT_P instead of checking for
3890 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3891 checking for CONST_VECTOR.
3893 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3896 * predict.c (force_edge_cold): Handle in more sane way edges
3899 2018-01-09 Carl Love <cel@us.ibm.com>
3901 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3903 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3904 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3905 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3906 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3907 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3908 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3909 * config/rs6000/rs6000-protos.h: Add extern defition for
3910 rs6000_generate_float2_double_code.
3911 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3913 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3914 (float2_v2df): Add define_expand.
3916 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3919 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3920 op_mode in the force_to_mode call.
3922 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3924 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3925 instead of checking each element individually.
3926 (aarch64_evpc_uzp): Likewise.
3927 (aarch64_evpc_zip): Likewise.
3928 (aarch64_evpc_ext): Likewise.
3929 (aarch64_evpc_rev): Likewise.
3930 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3931 instead of checking each element individually. Return true without
3933 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3934 whether all selected elements come from the same input, instead of
3935 checking each element individually. Remove calls to gen_rtx_REG,
3936 start_sequence and end_sequence and instead assert that no rtl is
3939 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3941 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3942 order of HIGH and CONST checks.
3944 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3946 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3947 if the destination isn't an SSA_NAME.
3949 2018-01-09 Richard Biener <rguenther@suse.de>
3951 PR tree-optimization/83668
3952 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3954 (canonicalize_loop_form): ... here, renamed from ...
3955 (canonicalize_loop_closed_ssa_form): ... this and amended to
3956 swap successor edges for loop exit blocks to make us use
3957 the RPO order we need for initial schedule generation.
3959 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3961 PR tree-optimization/64811
3962 * match.pd: When optimizing comparisons with Inf, avoid
3963 introducing or losing exceptions from comparisons with NaN.
3965 2018-01-09 Martin Liska <mliska@suse.cz>
3968 * asan.c (shadow_mem_size): Add gcc_assert.
3970 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3972 Don't save registers in main().
3975 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3976 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3977 * config/avr/avr.c (avr_set_current_function): Don't error if
3978 naked, OS_task or OS_main are specified at the same time.
3979 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3981 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3983 * common/config/avr/avr-common.c (avr_option_optimization_table):
3984 Switch on -mmain-is-OS_task for optimizing compilations.
3986 2018-01-09 Richard Biener <rguenther@suse.de>
3988 PR tree-optimization/83572
3989 * graphite.c: Include cfganal.h.
3990 (graphite_transform_loops): Connect infinite loops to exit
3991 and remove fake edges at the end.
3993 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3995 * ipa-inline.c (edge_badness): Revert accidental checkin.
3997 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4000 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4001 symbols; not inline clones.
4003 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4006 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4007 hard registers. Formatting fixes.
4009 PR preprocessor/83722
4010 * gcc.c (try_generate_repro): Pass
4011 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4012 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4015 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4016 Kito Cheng <kito.cheng@gmail.com>
4018 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4019 (riscv_leaf_function_p): Delete.
4020 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4022 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4024 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4026 (do_ifelse): New function.
4027 (do_isel): New function.
4028 (do_sub3): New function.
4029 (do_add3): New function.
4030 (do_load_mask_compare): New function.
4031 (do_overlap_load_compare): New function.
4032 (expand_compare_loop): New function.
4033 (expand_block_compare): Call expand_compare_loop() when appropriate.
4034 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4036 (-mblock-compare-inline-loop-limit): New option.
4038 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4041 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4042 Reverse order of second and third operands in first alternative.
4043 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4044 of first and second elements in UNSPEC_VPERMR vector.
4045 (altivec_expand_vec_perm_le): Likewise.
4047 2018-01-08 Jeff Law <law@redhat.com>
4049 PR rtl-optimizatin/81308
4050 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4051 (process_switch): If group_case_labels makes a change, then set
4053 (pass_convert_switch::execute): If a switch is converted, then
4054 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4056 PR rtl-optimization/81308
4057 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4060 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4062 PR target/83663 - Revert r255946
4063 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4064 generation for cases where splatting a value is not useful.
4065 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4066 across a vec_duplicate and a paradoxical subreg forming a vector
4067 mode to a vec_concat.
4069 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4071 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4072 -march=armv8.3-a variants.
4073 * config/arm/t-multilib: Likewise.
4074 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4076 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4078 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4080 (cceq_ior_compare_complement): Give it a name so I can use it, and
4081 change boolean_or_operator predicate to boolean_operator so it can
4082 be used to generate a crand.
4083 (eqne): New code iterator.
4084 (bd/bd_neg): New code_attrs.
4085 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4086 a single define_insn.
4087 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4088 decrement (bdnzt/bdnzf/bdzt/bdzf).
4089 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4090 with the new names of the branch decrement patterns, and added the
4091 names of the branch decrement conditional patterns.
4093 2018-01-08 Richard Biener <rguenther@suse.de>
4095 PR tree-optimization/83563
4096 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4099 2018-01-08 Richard Biener <rguenther@suse.de>
4102 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4104 2018-01-08 Richard Biener <rguenther@suse.de>
4106 PR tree-optimization/83685
4107 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4108 references to abnormals.
4110 2018-01-08 Richard Biener <rguenther@suse.de>
4113 * dwarf2out.c (output_indirect_strings): Handle empty
4114 skeleton_debug_str_hash.
4115 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4117 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4119 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4120 (emit_store_direct): Likewise.
4121 (arc_trampoline_adjust_address): Likewise.
4122 (arc_asm_trampoline_template): New function.
4123 (arc_initialize_trampoline): Use asm_trampoline_template.
4124 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4125 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4126 * config/arc/arc.md (flush_icache): Delete pattern.
4128 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4130 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4131 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4134 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4137 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4138 by not USED_FOR_TARGET.
4139 (make_pass_resolve_sw_modes): Likewise.
4141 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4143 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4146 2018-01-08 Richard Biener <rguenther@suse.de>
4149 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4151 2018-01-08 Richard Biener <rguenther@suse.de>
4154 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4156 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4159 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4160 basic blocks with a small number of successors.
4161 (convert_control_dep_chain_into_preds): Improve handling of
4163 (dump_predicates): Split apart into...
4164 (dump_pred_chain): ...here...
4165 (dump_pred_info): ...and here.
4166 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4167 (can_chain_union_be_invalidated_p): Improve check for invalidation
4169 (uninit_uses_cannot_happen): Avoid unnecessary if
4170 convert_control_dep_chain_into_preds yielded nothing.
4172 2018-01-06 Martin Sebor <msebor@redhat.com>
4174 PR tree-optimization/83640
4175 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4176 subtracting negative offset from size.
4177 (builtin_access::overlap): Adjust offset bounds of the access to fall
4178 within the size of the object if possible.
4180 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4182 PR rtl-optimization/83699
4183 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4184 extract_bit_field_as_subreg to cases in which the extracted
4185 value is also a vector.
4187 * lra-constraints.c (process_alt_operands): Test for the equivalence
4188 substitutions when detecting a possible reload cycle.
4190 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4193 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4194 by default if flag_selective_schedling{,2}. Formatting fixes.
4196 PR rtl-optimization/83682
4197 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4198 if it has non-VECTOR_MODE element mode.
4199 (vec_duplicate_p): Likewise.
4202 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4203 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4205 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4208 * config/i386/i386-builtin.def
4209 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4210 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4211 Require also OPTION_MASK_ISA_AVX512F in addition to
4212 OPTION_MASK_ISA_GFNI.
4213 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4214 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4215 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4216 to OPTION_MASK_ISA_GFNI.
4217 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4218 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4219 OPTION_MASK_ISA_AVX512BW.
4220 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4221 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4222 addition to OPTION_MASK_ISA_GFNI.
4223 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4224 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4225 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4226 to OPTION_MASK_ISA_GFNI.
4227 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4228 a requirement for all ISAs rather than any of them with a few
4230 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4232 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4233 bitmasks to be enabled with 3 exceptions, instead of requiring any
4234 enabled ISA with lots of exceptions.
4235 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4236 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4237 Change avx512bw in isa attribute to avx512f.
4238 * config/i386/sgxintrin.h: Add license boilerplate.
4239 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4240 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4241 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4242 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4244 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4245 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4246 temporarily sse2 rather than sse if not enabled already.
4249 * config/i386/sse.md (VI248_VLBW): Rename to ...
4250 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4251 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4252 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4253 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4254 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4255 mode iterator instead of VI248_VLBW.
4257 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4259 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4260 (record_modified): Skip clobbers; add debug output.
4261 (param_change_prob): Use sreal frequencies.
4263 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4265 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4266 punt for user-aligned variables.
4268 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4270 * tree-chrec.c (chrec_contains_symbols): Return true for
4273 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4276 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4277 of (x|y) == x for BICS pattern.
4279 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4281 PR tree-optimization/83605
4282 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4283 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4286 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4288 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4289 * config/epiphany/rtems.h: New file.
4291 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4292 Uros Bizjak <ubizjak@gmail.com>
4295 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4296 QIreg_operand instead of register_operand predicate.
4297 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4298 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4299 comments instead of -fmitigate[-_]rop.
4301 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4304 * cgraphunit.c (symbol_table::compile): Switch to text_section
4305 before calling assembly_start debug hook.
4306 * run-rtl-passes.c (run_rtl_passes): Likewise.
4309 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4311 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4312 range_int_cst_p rather than !symbolic_range_p before calling
4313 extract_range_from_multiplicative_op_1.
4315 2018-01-04 Jeff Law <law@redhat.com>
4317 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4318 redundant test in assertion.
4320 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4322 * doc/rtl.texi: Document machine_mode wrapper classes.
4324 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4326 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4329 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4331 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4332 the VEC_PERM_EXPR fold to fail.
4334 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4337 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4338 to switched_sections.
4340 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4343 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4346 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4349 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4350 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4352 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4355 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4356 is BLKmode and bitpos not zero or mode change is needed.
4358 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4361 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4364 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4367 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4368 instead of MULT rtx. Update all corresponding splitters.
4370 (*ssub<modesuffix>): Ditto.
4372 (*cmp_sadd_di): Update split patterns.
4373 (*cmp_sadd_si): Ditto.
4374 (*cmp_sadd_sidi): Ditto.
4375 (*cmp_ssub_di): Ditto.
4376 (*cmp_ssub_si): Ditto.
4377 (*cmp_ssub_sidi): Ditto.
4378 * config/alpha/predicates.md (const23_operand): New predicate.
4379 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4380 Look for ASHIFT, not MULT inner operand.
4381 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4383 2018-01-04 Martin Liska <mliska@suse.cz>
4385 PR gcov-profile/83669
4386 * gcov.c (output_intermediate_file): Add version to intermediate
4388 * doc/gcov.texi: Document new field 'version' in intermediate
4389 file format. Fix location of '-k' option of gcov command.
4391 2018-01-04 Martin Liska <mliska@suse.cz>
4394 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4396 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4398 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4400 2018-01-03 Martin Sebor <msebor@redhat.com>
4402 PR tree-optimization/83655
4403 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4404 checking calls with invalid arguments.
4406 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4408 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4409 (vectorizable_mask_load_store): Delete.
4410 (vectorizable_call): Return false for masked loads and stores.
4411 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4412 instead of gimple_assign_rhs1.
4413 (vectorizable_load): Handle IFN_MASK_LOAD.
4414 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4418 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4420 (vectorizable_mask_load_store): ...here.
4421 (vectorizable_load): ...and here.
4423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4425 * tree-vect-stmts.c (vect_build_all_ones_mask)
4426 (vect_build_zero_merge_argument): New functions, split out from...
4427 (vectorizable_load): ...here.
4429 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4431 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4433 (vectorizable_mask_load_store): ...here.
4434 (vectorizable_store): ...and here.
4436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4438 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4440 (vectorizable_mask_load_store): ...here.
4442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4444 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4445 (vect_model_store_cost): Take a vec_load_store_type instead of a
4447 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4448 (vect_model_store_cost): Take a vec_load_store_type instead of a
4450 (vectorizable_mask_load_store): Update accordingly.
4451 (vectorizable_store): Likewise.
4452 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4456 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4457 IFN_MASK_LOAD calls here rather than...
4458 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4461 Alan Hayward <alan.hayward@arm.com>
4462 David Sherwood <david.sherwood@arm.com>
4464 * expmed.c (extract_bit_field_1): For vector extracts,
4465 fall back to extract_bit_field_as_subreg if vec_extract
4468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4469 Alan Hayward <alan.hayward@arm.com>
4470 David Sherwood <david.sherwood@arm.com>
4472 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4473 they are variable or constant sized.
4474 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4475 slots for constant-sized data.
4477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4478 Alan Hayward <alan.hayward@arm.com>
4479 David Sherwood <david.sherwood@arm.com>
4481 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4482 handling COND_EXPRs with boolean comparisons, try to find a better
4483 basis for the mask type than the boolean itself.
4485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4487 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4488 is calculated and how it can be overridden.
4489 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4490 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4492 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4495 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4496 Alan Hayward <alan.hayward@arm.com>
4497 David Sherwood <david.sherwood@arm.com>
4499 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4500 Remove the mode argument.
4501 (aarch64_simd_valid_immediate): Remove the mode and inverse
4503 * config/aarch64/iterators.md (bitsize): New iterator.
4504 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4505 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4506 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4507 aarch64_simd_valid_immediate.
4508 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4509 (aarch64_reg_or_bic_imm): Likewise.
4510 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4511 with an insn_type enum and msl with a modifier_type enum.
4512 Replace element_width with a scalar_mode. Change the shift
4513 to unsigned int. Add constructors for scalar_float_mode and
4514 scalar_int_mode elements.
4515 (aarch64_vect_float_const_representable_p): Delete.
4516 (aarch64_can_const_movi_rtx_p)
4517 (aarch64_simd_scalar_immediate_valid_for_move)
4518 (aarch64_simd_make_constant): Update call to
4519 aarch64_simd_valid_immediate.
4520 (aarch64_advsimd_valid_immediate_hs): New function.
4521 (aarch64_advsimd_valid_immediate): Likewise.
4522 (aarch64_simd_valid_immediate): Remove mode and inverse
4523 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4524 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4525 and aarch64_float_const_representable_p on the result.
4526 (aarch64_output_simd_mov_immediate): Remove mode argument.
4527 Update call to aarch64_simd_valid_immediate and use of
4528 simd_immediate_info.
4529 (aarch64_output_scalar_simd_mov_immediate): Update call
4532 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4533 Alan Hayward <alan.hayward@arm.com>
4534 David Sherwood <david.sherwood@arm.com>
4536 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4537 (mode_nunits): Likewise CONST_MODE_NUNITS.
4538 * machmode.def (ADJUST_NUNITS): Document.
4539 * genmodes.c (mode_data::need_nunits_adj): New field.
4540 (blank_mode): Update accordingly.
4541 (adj_nunits): New variable.
4542 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4544 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4545 listed in adj_nunits.
4546 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4547 listed in adj_nunits. Don't emit case statements for such modes.
4548 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4549 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4550 nothing if adj_nunits is nonnull.
4551 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4552 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4553 (emit_mode_fbit): Update use of print_maybe_const_decl.
4554 (emit_move_size): Likewise. Treat the array as non-const
4556 (emit_mode_adjustments): Handle adj_nunits.
4558 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4560 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4561 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4562 (VECTOR_MODES): Use it.
4563 (make_vector_modes): Take the prefix as an argument.
4565 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4566 Alan Hayward <alan.hayward@arm.com>
4567 David Sherwood <david.sherwood@arm.com>
4569 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4570 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4571 for MODE_VECTOR_BOOL.
4572 * machmode.def (VECTOR_BOOL_MODE): Document.
4573 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4574 (make_vector_bool_mode): New function.
4575 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4577 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4578 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4580 * stor-layout.c (int_mode_for_mode): Likewise.
4581 * tree.c (build_vector_type_for_mode): Likewise.
4582 * varasm.c (output_constant_pool_2): Likewise.
4583 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4584 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4585 for MODE_VECTOR_BOOL.
4586 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4587 of mode class checks.
4588 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4589 instead of a list of mode class checks.
4590 (expand_vector_scalar_condition): Likewise.
4591 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4593 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4594 Alan Hayward <alan.hayward@arm.com>
4595 David Sherwood <david.sherwood@arm.com>
4597 * machmode.h (mode_size): Change from unsigned short to
4599 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4600 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4601 or if measurement_type is not polynomial.
4602 (fixed_size_mode::includes_p): Check for constant-sized modes.
4603 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4604 return a poly_uint16 rather than an unsigned short.
4605 (emit_mode_size): Change the type of mode_size from unsigned short
4606 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4607 (emit_mode_adjustments): Cope with polynomial vector sizes.
4608 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4610 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4612 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4613 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4614 * caller-save.c (setup_save_areas): Likewise.
4615 (replace_reg_with_saved_mem): Likewise.
4616 * calls.c (emit_library_call_value_1): Likewise.
4617 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4618 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4619 (gen_lowpart_for_combine): Likewise.
4620 * convert.c (convert_to_integer_1): Likewise.
4621 * cse.c (equiv_constant, cse_insn): Likewise.
4622 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4623 (cselib_subst_to_values): Likewise.
4624 * dce.c (word_dce_process_block): Likewise.
4625 * df-problems.c (df_word_lr_mark_ref): Likewise.
4626 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4627 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4628 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4629 (rtl_for_decl_location): Likewise.
4630 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4631 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4632 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4633 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4634 (expand_expr_real_1): Likewise.
4635 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4636 (pad_below): Likewise.
4637 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4638 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4639 * ira.c (get_subreg_tracking_sizes): Likewise.
4640 * ira-build.c (ira_create_allocno_objects): Likewise.
4641 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4642 (ira_sort_regnos_for_alter_reg): Likewise.
4643 * ira-costs.c (record_operand_costs): Likewise.
4644 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4645 (resolve_simple_move): Likewise.
4646 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4647 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4648 (lra_constraints): Likewise.
4649 (CONST_POOL_OK_P): Reject variable-sized modes.
4650 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4651 (add_pseudo_to_slot, lra_spill): Likewise.
4652 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4653 * optabs-query.c (get_best_extraction_insn): Likewise.
4654 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4655 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4656 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4657 * recog.c (offsettable_address_addr_space_p): Likewise.
4658 * regcprop.c (maybe_mode_change): Likewise.
4659 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4660 * regrename.c (build_def_use): Likewise.
4661 * regstat.c (dump_reg_info): Likewise.
4662 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4663 (find_reloads, find_reloads_subreg_address): Likewise.
4664 * reload1.c (eliminate_regs_1): Likewise.
4665 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4666 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4667 (simplify_binary_operation_1, simplify_subreg): Likewise.
4668 * targhooks.c (default_function_arg_padding): Likewise.
4669 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4670 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4671 (verify_gimple_assign_ternary): Likewise.
4672 * tree-inline.c (estimate_move_cost): Likewise.
4673 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4674 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4675 (get_address_cost_ainc): Likewise.
4676 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4677 (vect_supportable_dr_alignment): Likewise.
4678 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4679 (vectorizable_reduction): Likewise.
4680 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4681 (vectorizable_operation, vectorizable_load): Likewise.
4682 * tree.c (build_same_sized_truth_vector_type): Likewise.
4683 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4684 * var-tracking.c (emit_note_insn_var_location): Likewise.
4685 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4686 (ADDR_VEC_ALIGN): Likewise.
4688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4689 Alan Hayward <alan.hayward@arm.com>
4690 David Sherwood <david.sherwood@arm.com>
4692 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4694 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4695 or if measurement_type is polynomial.
4696 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4697 * combine.c (make_extraction): Likewise.
4698 * dse.c (find_shift_sequence): Likewise.
4699 * dwarf2out.c (mem_loc_descriptor): Likewise.
4700 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4701 (extract_bit_field, extract_low_bits): Likewise.
4702 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4703 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4704 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4705 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4706 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4707 * reload.c (find_reloads): Likewise.
4708 * reload1.c (alter_reg): Likewise.
4709 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4710 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4711 * tree-if-conv.c (predicate_mem_writes): Likewise.
4712 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4713 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4714 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4715 * valtrack.c (dead_debug_insert_temp): Likewise.
4716 * varasm.c (mergeable_constant_section): Likewise.
4717 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4720 Alan Hayward <alan.hayward@arm.com>
4721 David Sherwood <david.sherwood@arm.com>
4723 * expr.c (expand_assignment): Cope with polynomial mode sizes
4724 when assigning to a CONCAT.
4726 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4727 Alan Hayward <alan.hayward@arm.com>
4728 David Sherwood <david.sherwood@arm.com>
4730 * machmode.h (mode_precision): Change from unsigned short to
4732 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4734 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4735 or if measurement_type is not polynomial.
4736 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4737 in which the mode is already known to be a scalar_int_mode.
4738 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4739 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4741 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4742 for GET_MODE_PRECISION.
4743 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4744 for GET_MODE_PRECISION.
4745 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4747 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4748 (expand_field_assignment, make_extraction): Likewise.
4749 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4750 (get_last_value): Likewise.
4751 * convert.c (convert_to_integer_1): Likewise.
4752 * cse.c (cse_insn): Likewise.
4753 * expr.c (expand_expr_real_1): Likewise.
4754 * lra-constraints.c (simplify_operand_subreg): Likewise.
4755 * optabs-query.c (can_atomic_load_p): Likewise.
4756 * optabs.c (expand_atomic_load): Likewise.
4757 (expand_atomic_store): Likewise.
4758 * ree.c (combine_reaching_defs): Likewise.
4759 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4760 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4761 * tree.h (type_has_mode_precision_p): Likewise.
4762 * ubsan.c (instrument_si_overflow): Likewise.
4764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4765 Alan Hayward <alan.hayward@arm.com>
4766 David Sherwood <david.sherwood@arm.com>
4768 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4769 polynomial numbers of units.
4770 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4771 (valid_vector_subparts_p): New function.
4772 (build_vector_type): Remove temporary shim and take the number
4773 of units as a poly_uint64 rather than an int.
4774 (build_opaque_vector_type): Take the number of units as a
4775 poly_uint64 rather than an int.
4776 * tree.c (build_vector_from_ctor): Handle polynomial
4777 TYPE_VECTOR_SUBPARTS.
4778 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4779 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4780 (build_vector_from_val): If the number of units is variable,
4781 use build_vec_duplicate_cst for constant operands and
4782 VEC_DUPLICATE_EXPR otherwise.
4783 (make_vector_type): Remove temporary is_constant ().
4784 (build_vector_type, build_opaque_vector_type): Take the number of
4785 units as a poly_uint64 rather than an int.
4786 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4788 * cfgexpand.c (expand_debug_expr): Likewise.
4789 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4790 (store_constructor, expand_expr_real_1): Likewise.
4791 (const_scalar_mask_from_tree): Likewise.
4792 * fold-const-call.c (fold_const_reduction): Likewise.
4793 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4794 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4795 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4796 (fold_relational_const): Likewise.
4797 (native_interpret_vector): Likewise. Change the size from an
4798 int to an unsigned int.
4799 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4800 TYPE_VECTOR_SUBPARTS.
4801 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4802 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4803 duplicating a non-constant operand into a variable-length vector.
4804 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4805 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4806 * ipa-icf.c (sem_variable::equals): Likewise.
4807 * match.pd: Likewise.
4808 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4809 * print-tree.c (print_node): Likewise.
4810 * stor-layout.c (layout_type): Likewise.
4811 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4812 * tree-cfg.c (verify_gimple_comparison): Likewise.
4813 (verify_gimple_assign_binary): Likewise.
4814 (verify_gimple_assign_ternary): Likewise.
4815 (verify_gimple_assign_single): Likewise.
4816 * tree-pretty-print.c (dump_generic_node): Likewise.
4817 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4818 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4819 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4820 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4821 (vect_shift_permute_load_chain): Likewise.
4822 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4823 (expand_vector_condition, optimize_vector_constructor): Likewise.
4824 (lower_vec_perm, get_compute_type): Likewise.
4825 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4826 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4827 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4828 (vect_recog_mask_conversion_pattern): Likewise.
4829 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4830 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4831 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4832 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4833 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4834 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4835 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4836 (supportable_widening_operation): Likewise.
4837 (supportable_narrowing_operation): Likewise.
4838 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4840 * varasm.c (output_constant): Likewise.
4842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4843 Alan Hayward <alan.hayward@arm.com>
4844 David Sherwood <david.sherwood@arm.com>
4846 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4847 so that both the length == 3 and length != 3 cases set up their
4848 own permute vectors. Add comments explaining why we know the
4849 number of elements is constant.
4850 (vect_permute_load_chain): Likewise.
4852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4853 Alan Hayward <alan.hayward@arm.com>
4854 David Sherwood <david.sherwood@arm.com>
4856 * machmode.h (mode_nunits): Change from unsigned char to
4858 (ONLY_FIXED_SIZE_MODES): New macro.
4859 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4860 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4861 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4863 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4864 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4865 or if measurement_type is not polynomial.
4866 * genmodes.c (ZERO_COEFFS): New macro.
4867 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4869 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4870 Use ZERO_COEFFS when emitting initializers.
4871 * data-streamer.h (bp_pack_poly_value): New function.
4872 (bp_unpack_poly_value): Likewise.
4873 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4874 for GET_MODE_NUNITS.
4875 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4876 for GET_MODE_NUNITS.
4877 * tree.c (make_vector_type): Remove temporary shim and make
4878 the real function take the number of units as a poly_uint64
4880 (build_vector_type_for_mode): Handle polynomial nunits.
4881 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4882 * emit-rtl.c (const_vec_series_p_1): Likewise.
4883 (gen_rtx_CONST_VECTOR): Likewise.
4884 * fold-const.c (test_vec_duplicate_folding): Likewise.
4885 * genrecog.c (validate_pattern): Likewise.
4886 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4887 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4888 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4889 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4890 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4891 * rtlanal.c (subreg_get_info): Likewise.
4892 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4893 (vect_grouped_load_supported): Likewise.
4894 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4895 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4896 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4897 (simplify_const_unary_operation, simplify_binary_operation_1)
4898 (simplify_const_binary_operation, simplify_ternary_operation)
4899 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4900 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4901 instead of CONST_VECTOR_NUNITS.
4902 * varasm.c (output_constant_pool_2): Likewise.
4903 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4904 explicit-encoded elements in the XVEC for variable-length vectors.
4906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4908 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4911 Alan Hayward <alan.hayward@arm.com>
4912 David Sherwood <david.sherwood@arm.com>
4914 * coretypes.h (fixed_size_mode): Declare.
4915 (fixed_size_mode_pod): New typedef.
4916 * builtins.h (target_builtins::x_apply_args_mode)
4917 (target_builtins::x_apply_result_mode): Change type to
4918 fixed_size_mode_pod.
4919 * builtins.c (apply_args_size, apply_result_size, result_vector)
4920 (expand_builtin_apply_args_1, expand_builtin_apply)
4921 (expand_builtin_return): Update accordingly.
4923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4925 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4926 * cselib.c (cselib_hash_rtx): Likewise.
4927 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4928 CONST_VECTOR encoding.
4930 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4931 Jeff Law <law@redhat.com>
4934 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4935 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4936 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4937 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4940 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4941 explicitly probe *sp in a noreturn function if there were any callee
4942 register saves or frame pointer is needed.
4944 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4947 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4948 BLKmode for ternary, binary or unary expressions.
4951 * var-tracking.c (delete_vta_debug_insn): New inline function.
4952 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4953 insns from get_insns () to NULL instead of each bb separately.
4954 Use delete_vta_debug_insn. No longer static.
4955 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4956 delete_vta_debug_insns callers.
4957 * rtl.h (delete_vta_debug_insns): Declare.
4958 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4959 instead of variable_tracking_main.
4961 2018-01-03 Martin Sebor <msebor@redhat.com>
4963 PR tree-optimization/83603
4964 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4965 arguments past the endof the argument list in functions declared
4966 without a prototype.
4967 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4968 Avoid checking when arguments are null.
4970 2018-01-03 Martin Sebor <msebor@redhat.com>
4973 * doc/extend.texi (attribute const): Fix a typo.
4974 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4975 issuing -Wsuggest-attribute for void functions.
4977 2018-01-03 Martin Sebor <msebor@redhat.com>
4979 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4980 offset_int::from instead of wide_int::to_shwi.
4981 (maybe_diag_overlap): Remove assertion.
4982 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4983 * gimple-ssa-sprintf.c (format_directive): Same.
4984 (parse_directive): Same.
4985 (sprintf_dom_walker::compute_format_length): Same.
4986 (try_substitute_return_value): Same.
4988 2018-01-03 Jeff Law <law@redhat.com>
4991 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4992 non-constant residual for zero at runtime and avoid probing in
4993 that case. Reorganize code for trailing problem to mirror handling
4996 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4998 PR tree-optimization/83501
4999 * tree-ssa-strlen.c (get_string_cst): New.
5000 (handle_char_store): Call get_string_cst.
5002 2018-01-03 Martin Liska <mliska@suse.cz>
5004 PR tree-optimization/83593
5005 * tree-ssa-strlen.c: Include tree-cfg.h.
5006 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5007 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5008 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5010 (strlen_dom_walker::before_dom_children): Call
5011 gimple_purge_dead_eh_edges. Dump tranformation with details
5013 (strlen_dom_walker::before_dom_children): Update call by adding
5014 new argument cleanup_eh.
5015 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5017 2018-01-03 Martin Liska <mliska@suse.cz>
5020 * cif-code.def (VARIADIC_THUNK): New enum value.
5021 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5024 2018-01-03 Jan Beulich <jbeulich@suse.com>
5026 * sse.md (mov<mode>_internal): Tighten condition for when to use
5027 vmovdqu<ssescalarsize> for TI and OI modes.
5029 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5031 Update copyright years.
5033 2018-01-03 Martin Liska <mliska@suse.cz>
5036 * ipa-visibility.c (function_and_variable_visibility): Skip
5037 functions with noipa attribure.
5039 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5041 * gcc.c (process_command): Update copyright notice dates.
5042 * gcov-dump.c (print_version): Ditto.
5043 * gcov.c (print_version): Ditto.
5044 * gcov-tool.c (print_version): Ditto.
5045 * gengtype.c (create_file): Ditto.
5046 * doc/cpp.texi: Bump @copying's copyright year.
5047 * doc/cppinternals.texi: Ditto.
5048 * doc/gcc.texi: Ditto.
5049 * doc/gccint.texi: Ditto.
5050 * doc/gcov.texi: Ditto.
5051 * doc/install.texi: Ditto.
5052 * doc/invoke.texi: Ditto.
5054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5056 * vector-builder.h (vector_builder::m_full_nelts): Change from
5057 unsigned int to poly_uint64.
5058 (vector_builder::full_nelts): Update prototype accordingly.
5059 (vector_builder::new_vector): Likewise.
5060 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5061 (vector_builder::operator ==): Likewise.
5062 (vector_builder::finalize): Likewise.
5063 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5064 Take the number of elements as a poly_uint64 rather than an
5066 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5067 from unsigned int to poly_uint64.
5068 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5069 (vec_perm_indices::new_vector): Likewise.
5070 (vec_perm_indices::length): Likewise.
5071 (vec_perm_indices::nelts_per_input): Likewise.
5072 (vec_perm_indices::input_nelts): Likewise.
5073 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5074 number of elements per input as a poly_uint64 rather than an
5075 unsigned int. Use the original encoding for variable-length
5076 vectors, rather than clamping each individual element.
5077 For the second and subsequent elements in each pattern,
5078 clamp the step and base before clamping their sum.
5079 (vec_perm_indices::series_p): Handle polynomial element counts.
5080 (vec_perm_indices::all_in_range_p): Likewise.
5081 (vec_perm_indices_to_tree): Likewise.
5082 (vec_perm_indices_to_rtx): Likewise.
5083 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5084 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5085 (tree_vector_builder::new_binary_operation): Handle polynomial
5086 element counts. Return false if we need to know the number
5087 of elements at compile time.
5088 * fold-const.c (fold_vec_perm): Punt if the number of elements
5089 isn't known at compile time.
5091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5093 * vec-perm-indices.h (vec_perm_builder): Change element type
5094 from HOST_WIDE_INT to poly_int64.
5095 (vec_perm_indices::element_type): Update accordingly.
5096 (vec_perm_indices::clamp): Handle polynomial element_types.
5097 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5098 (vec_perm_indices::all_in_range_p): Likewise.
5099 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5101 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5102 polynomial vec_perm_indices element types.
5103 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5104 * fold-const.c (fold_vec_perm): Likewise.
5105 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5106 * tree-vect-generic.c (lower_vec_perm): Likewise.
5107 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5108 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5109 element type to HOST_WIDE_INT.
5111 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5112 Alan Hayward <alan.hayward@arm.com>
5113 David Sherwood <david.sherwood@arm.com>
5115 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5116 rather than an int. Use plus_constant.
5117 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5118 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5121 Alan Hayward <alan.hayward@arm.com>
5122 David Sherwood <david.sherwood@arm.com>
5124 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5125 a HOST_WIDE_INT to a poly_int64.
5127 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5128 Alan Hayward <alan.hayward@arm.com>
5129 David Sherwood <david.sherwood@arm.com>
5131 * calls.c (load_register_parameters): Cope with polynomial
5132 mode sizes. Require a constant size for BLKmode parameters
5133 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5134 forces a parameter to be padded at the lsb end in order to
5135 fill a complete number of words, require the parameter size
5136 to be ordered wrt UNITS_PER_WORD.
5138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5139 Alan Hayward <alan.hayward@arm.com>
5140 David Sherwood <david.sherwood@arm.com>
5142 * reload1.c (spill_stack_slot_width): Change element type
5143 from unsigned int to poly_uint64_pod.
5144 (alter_reg): Treat mode sizes as polynomial.
5146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5147 Alan Hayward <alan.hayward@arm.com>
5148 David Sherwood <david.sherwood@arm.com>
5150 * reload.c (complex_word_subreg_p): New function.
5151 (reload_inner_reg_of_subreg, push_reload): Use it.
5153 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5154 Alan Hayward <alan.hayward@arm.com>
5155 David Sherwood <david.sherwood@arm.com>
5157 * lra-constraints.c (process_alt_operands): Reject matched
5158 operands whose sizes aren't ordered.
5159 (match_reload): Refer to this check here.
5161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5162 Alan Hayward <alan.hayward@arm.com>
5163 David Sherwood <david.sherwood@arm.com>
5165 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5166 that the mode size is in the set {1, 2, 4, 8, 16}.
5168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5169 Alan Hayward <alan.hayward@arm.com>
5170 David Sherwood <david.sherwood@arm.com>
5172 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5173 Use plus_constant instead of gen_rtx_PLUS.
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 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5180 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5181 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5182 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5183 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5184 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5185 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5186 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5187 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5188 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5190 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5191 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5192 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5193 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5194 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5195 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5196 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5197 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5198 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5199 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5201 * expr.c (emit_move_resolve_push): Treat the input and result
5202 of PUSH_ROUNDING as a poly_int64.
5203 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5204 (emit_push_insn): Likewise.
5205 * lra-eliminations.c (mark_not_eliminable): Likewise.
5206 * recog.c (push_operand): Likewise.
5207 * reload1.c (elimination_effects): Likewise.
5208 * rtlanal.c (nonzero_bits1): Likewise.
5209 * calls.c (store_one_arg): Likewise. Require the padding to be
5210 known at compile time.
5212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5213 Alan Hayward <alan.hayward@arm.com>
5214 David Sherwood <david.sherwood@arm.com>
5216 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5217 Use plus_constant instead of gen_rtx_PLUS.
5219 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5220 Alan Hayward <alan.hayward@arm.com>
5221 David Sherwood <david.sherwood@arm.com>
5223 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 (expand_expr_real_1): Use tree_to_poly_uint64
5231 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5232 via stack temporaries. Treat the mode size as polynomial too.
5234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5235 Alan Hayward <alan.hayward@arm.com>
5236 David Sherwood <david.sherwood@arm.com>
5238 * expr.c (expand_expr_real_2): When handling conversions involving
5239 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5240 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5241 as a poly_uint64 too.
5243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5244 Alan Hayward <alan.hayward@arm.com>
5245 David Sherwood <david.sherwood@arm.com>
5247 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5250 Alan Hayward <alan.hayward@arm.com>
5251 David Sherwood <david.sherwood@arm.com>
5253 * combine.c (can_change_dest_mode): Handle polynomial
5254 REGMODE_NATURAL_SIZE.
5255 * expmed.c (store_bit_field_1): Likewise.
5256 * expr.c (store_constructor): Likewise.
5257 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5258 and polynomial REGMODE_NATURAL_SIZE.
5259 (gen_lowpart_common): Likewise.
5260 * reginfo.c (record_subregs_of_mode): Likewise.
5261 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5268 numbers of elements.
5270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5271 Alan Hayward <alan.hayward@arm.com>
5272 David Sherwood <david.sherwood@arm.com>
5274 * match.pd: Cope with polynomial numbers of vector elements.
5276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5277 Alan Hayward <alan.hayward@arm.com>
5278 David Sherwood <david.sherwood@arm.com>
5280 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5281 in a POINTER_PLUS_EXPR.
5283 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5284 Alan Hayward <alan.hayward@arm.com>
5285 David Sherwood <david.sherwood@arm.com>
5287 * omp-simd-clone.c (simd_clone_subparts): New function.
5288 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5289 (ipa_simd_modify_function_body): Likewise.
5291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5292 Alan Hayward <alan.hayward@arm.com>
5293 David Sherwood <david.sherwood@arm.com>
5295 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5296 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5297 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5298 (expand_vector_condition, vector_element): Likewise.
5299 (subparts_gt): New function.
5300 (get_compute_type): Use subparts_gt.
5301 (count_type_subparts): Delete.
5302 (expand_vector_operations_1): Use subparts_gt instead of
5303 count_type_subparts.
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-data-refs.c (vect_no_alias_p): Replace with...
5310 (vect_compile_time_alias): ...this new function. Do the calculation
5311 on poly_ints rather than trees.
5312 (vect_prune_runtime_alias_test_list): Update call accordingly.
5314 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5315 Alan Hayward <alan.hayward@arm.com>
5316 David Sherwood <david.sherwood@arm.com>
5318 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5320 (vect_schedule_slp_instance): Likewise.
5322 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5323 Alan Hayward <alan.hayward@arm.com>
5324 David Sherwood <david.sherwood@arm.com>
5326 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5327 constant and extern definitions for variable-length vectors.
5328 (vect_get_constant_vectors): Note that the number of units
5329 is known to be constant.
5331 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5332 Alan Hayward <alan.hayward@arm.com>
5333 David Sherwood <david.sherwood@arm.com>
5335 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5336 of units as polynomial. Choose between WIDE and NARROW based
5339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5340 Alan Hayward <alan.hayward@arm.com>
5341 David Sherwood <david.sherwood@arm.com>
5343 * tree-vect-stmts.c (simd_clone_subparts): New function.
5344 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5347 Alan Hayward <alan.hayward@arm.com>
5348 David Sherwood <david.sherwood@arm.com>
5350 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5351 vectors as polynomial. Use build_index_vector for
5354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5355 Alan Hayward <alan.hayward@arm.com>
5356 David Sherwood <david.sherwood@arm.com>
5358 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5359 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5360 for variable-length vectors.
5361 (vectorizable_mask_load_store): Treat the number of units as
5362 polynomial, asserting that it is constant if the condition has
5363 already been enforced.
5364 (vectorizable_store, vectorizable_load): Likewise.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5371 of units as polynomial. Punt if we can't tell at compile time
5372 which vector contains the final result.
5374 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5375 Alan Hayward <alan.hayward@arm.com>
5376 David Sherwood <david.sherwood@arm.com>
5378 * tree-vect-loop.c (vectorizable_induction): Treat the number
5379 of units as polynomial. Punt on SLP inductions. Use an integer
5380 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5381 cast of such a series for variable-length floating-point
5384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5385 Alan Hayward <alan.hayward@arm.com>
5386 David Sherwood <david.sherwood@arm.com>
5388 * tree.h (build_index_vector): Declare.
5389 * tree.c (build_index_vector): New function.
5390 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5391 of units as polynomial, forcibly converting it to a constant if
5392 vectorizable_reduction has already enforced the condition.
5393 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5394 to create a {1,2,3,...} vector.
5395 (vectorizable_reduction): Treat the number of units as polynomial.
5396 Choose vectype_in based on the largest scalar element size rather
5397 than the smallest number of units. Enforce the restrictions
5400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5401 Alan Hayward <alan.hayward@arm.com>
5402 David Sherwood <david.sherwood@arm.com>
5404 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5405 number of units as polynomial.
5407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5408 Alan Hayward <alan.hayward@arm.com>
5409 David Sherwood <david.sherwood@arm.com>
5411 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5412 * target.def (autovectorize_vector_sizes): Return the vector sizes
5413 by pointer, using vector_sizes rather than a bitmask.
5414 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5415 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5416 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5418 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5419 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5420 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5421 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5422 * omp-general.c (omp_max_vf): Likewise.
5423 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5424 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5425 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5426 * tree-vect-slp.c (vect_slp_bb): Likewise.
5427 * doc/tm.texi: Regenerate.
5428 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5430 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5431 the vector size as a poly_uint64 rather than an unsigned int.
5432 (current_vector_size): Change from an unsigned int to a poly_uint64.
5433 (get_vectype_for_scalar_type): Update accordingly.
5434 * tree.h (build_truth_vector_type): Take the size and number of
5435 units as a poly_uint64 rather than an unsigned int.
5436 (build_vector_type): Add a temporary overload that takes
5437 the number of units as a poly_uint64 rather than an unsigned int.
5438 * tree.c (make_vector_type): Likewise.
5439 (build_truth_vector_type): Take the number of units as a poly_uint64
5440 rather than an unsigned int.
5442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5443 Alan Hayward <alan.hayward@arm.com>
5444 David Sherwood <david.sherwood@arm.com>
5446 * target.def (get_mask_mode): Take the number of units and length
5447 as poly_uint64s rather than unsigned ints.
5448 * targhooks.h (default_get_mask_mode): Update accordingly.
5449 * targhooks.c (default_get_mask_mode): Likewise.
5450 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5451 * doc/tm.texi: Regenerate.
5453 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5454 Alan Hayward <alan.hayward@arm.com>
5455 David Sherwood <david.sherwood@arm.com>
5457 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5458 * omp-general.c (omp_max_vf): Likewise.
5459 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5460 (expand_omp_simd): Handle polynomial safelen.
5461 * omp-low.c (omplow_simd_context): Add a default constructor.
5462 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5463 (lower_rec_simd_input_clauses): Update accordingly.
5464 (lower_rec_input_clauses): Likewise.
5466 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5467 Alan Hayward <alan.hayward@arm.com>
5468 David Sherwood <david.sherwood@arm.com>
5470 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5471 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5472 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5473 (vect_analyze_slp_cost): Likewise.
5474 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5475 (vect_model_load_cost): Likewise.
5477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5478 Alan Hayward <alan.hayward@arm.com>
5479 David Sherwood <david.sherwood@arm.com>
5481 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5482 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5483 from an unsigned int * to a poly_uint64_pod *.
5484 (calculate_unrolling_factor): New function.
5485 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5488 Alan Hayward <alan.hayward@arm.com>
5489 David Sherwood <david.sherwood@arm.com>
5491 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5492 from an unsigned int to a poly_uint64.
5493 (_loop_vec_info::slp_unrolling_factor): Likewise.
5494 (_loop_vec_info::vectorization_factor): Change from an int
5496 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5497 (vect_get_num_vectors): New function.
5498 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5499 (vect_get_num_copies): Use vect_get_num_vectors.
5500 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5501 to an unsigned int *.
5502 (vect_analyze_data_refs): Change min_vf from an int * to a
5504 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5505 than an unsigned HOST_WIDE_INT.
5506 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5507 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5508 to an unsigned int *.
5509 (vect_analyze_data_ref_dependences): Likewise.
5510 (vect_compute_data_ref_alignment): Handle polynomial vf.
5511 (vect_enhance_data_refs_alignment): Likewise.
5512 (vect_prune_runtime_alias_test_list): Likewise.
5513 (vect_shift_permute_load_chain): Likewise.
5514 (vect_supportable_dr_alignment): Likewise.
5515 (dependence_distance_ge_vf): Take the vectorization factor as a
5516 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5517 (vect_analyze_data_refs): Change min_vf from an int * to a
5519 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5520 vfm1 as a poly_uint64 rather than an int. Make the same change
5521 for the returned bound_scalar.
5522 (vect_gen_vector_loop_niters): Handle polynomial vf.
5523 (vect_do_peeling): Likewise. Update call to
5524 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5525 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5527 * tree-vect-loop.c (vect_determine_vectorization_factor)
5528 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5529 (vect_get_known_peeling_cost): Likewise.
5530 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5531 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5532 (vect_transform_loop): Likewise. Use the lowest possible VF when
5533 updating the upper bounds of the loop.
5534 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5536 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5537 polynomial unroll factors.
5538 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5539 (vect_make_slp_decision): Likewise.
5540 (vect_supported_load_permutation_p): Likewise, and polynomial
5542 (vect_analyze_slp_cost): Handle polynomial vf.
5543 (vect_slp_analyze_node_operations): Likewise.
5544 (vect_slp_analyze_bb_1): Likewise.
5545 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5546 than an unsigned HOST_WIDE_INT.
5547 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5548 (vectorizable_load): Handle polynomial vf.
5549 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5551 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5554 Alan Hayward <alan.hayward@arm.com>
5555 David Sherwood <david.sherwood@arm.com>
5557 * match.pd: Handle bit operations involving three constants
5558 and try to fold one pair.
5560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5562 * tree-vect-loop-manip.c: Include gimple-fold.h.
5563 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5564 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5565 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5566 Add a path that uses a step of VF instead of 1, but disable it
5568 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5569 and niters_no_overflow parameters. Update calls to
5570 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5571 Create a new SSA name if the latter choses to use a ste other
5572 than zero, and return it via niters_vector_mult_vf_var.
5573 * tree-vect-loop.c (vect_transform_loop): Update calls to
5574 vect_do_peeling, vect_gen_vector_loop_niters and
5575 slpeel_make_loop_iterate_ntimes.
5576 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5577 (vect_gen_vector_loop_niters): Update declarations after above changes.
5579 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5581 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5582 128-bit round to integer instructions.
5583 (ceil<mode>2): Likewise.
5584 (btrunc<mode>2): Likewise.
5585 (round<mode>2): Likewise.
5587 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5589 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5590 unaligned VSX load/store on P8/P9.
5591 (expand_block_clear): Allow the use of unaligned VSX
5592 load/store on P8/P9.
5594 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5596 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5598 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5599 swap associated with both a load and a store.
5601 2018-01-02 Andrew Waterman <andrew@sifive.com>
5603 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5604 * config/riscv/riscv.md (clear_cache): Use it.
5606 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5608 * web.c: Remove out-of-date comment.
5610 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5612 * expr.c (fixup_args_size_notes): Check that any existing
5613 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5614 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5615 (emit_single_push_insn): ...here.
5617 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5619 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5620 (const_vector_encoded_nelts): New function.
5621 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5622 (const_vector_int_elt, const_vector_elt): Declare.
5623 * emit-rtl.c (const_vector_int_elt_1): New function.
5624 (const_vector_elt): Likewise.
5625 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5626 of CONST_VECTOR_ELT.
5628 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5630 * expr.c: Include rtx-vector-builder.h.
5631 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5632 directly on the tree encoding.
5633 (const_vector_from_tree): Likewise.
5634 * optabs.c: Include rtx-vector-builder.h.
5635 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5636 sequence of "u" values.
5637 * vec-perm-indices.c: Include rtx-vector-builder.h.
5638 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5639 directly on the vec_perm_indices encoding.
5641 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5643 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5644 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5645 * rtx-vector-builder.h: New file.
5646 * rtx-vector-builder.c: Likewise.
5647 * rtl.h (rtx_def::u2): Add a const_vector field.
5648 (CONST_VECTOR_NPATTERNS): New macro.
5649 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5650 (CONST_VECTOR_DUPLICATE_P): Likewise.
5651 (CONST_VECTOR_STEPPED_P): Likewise.
5652 (CONST_VECTOR_ENCODED_ELT): Likewise.
5653 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5654 (unwrap_const_vec_duplicate): Likewise.
5655 (const_vec_series_p): Check for a non-duplicated vector encoding.
5656 Say that the function only returns true for integer vectors.
5657 * emit-rtl.c: Include rtx-vector-builder.h.
5658 (gen_const_vec_duplicate_1): Delete.
5659 (gen_const_vector): Call gen_const_vec_duplicate instead of
5660 gen_const_vec_duplicate_1.
5661 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5662 (gen_const_vec_duplicate): Use rtx_vector_builder.
5663 (gen_const_vec_series): Likewise.
5664 (gen_rtx_CONST_VECTOR): Likewise.
5665 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5666 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5667 Build a new vector rather than modifying a CONST_VECTOR in-place.
5668 (handle_special_swappables): Update call accordingly.
5669 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5670 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5671 Build a new vector rather than modifying a CONST_VECTOR in-place.
5672 (handle_special_swappables): Update call accordingly.
5674 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5676 * simplify-rtx.c (simplify_const_binary_operation): Use
5677 CONST_VECTOR_ELT instead of XVECEXP.
5679 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5681 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5682 the selector elements to be different from the data elements
5683 if the selector is a VECTOR_CST.
5684 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5685 ssizetype for the selector.
5687 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5689 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5690 before testing each element individually.
5691 * tree-vect-generic.c (lower_vec_perm): Likewise.
5693 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5695 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5696 * selftest-run-tests.c (selftest::run_tests): Call it.
5697 * vector-builder.h (vector_builder::operator ==): New function.
5698 (vector_builder::operator !=): Likewise.
5699 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5700 (vec_perm_indices::all_from_input_p): New function.
5701 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5702 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5703 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5704 instead of reading the VECTOR_CST directly. Detect whether both
5705 vector inputs are the same before constructing the vec_perm_indices,
5706 and update the number of inputs argument accordingly. Use the
5707 utility functions added above. Only construct sel2 if we need to.
5709 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5711 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5712 the broadcast of the low byte.
5713 (expand_mult_highpart): Use an explicit encoding for the permutes.
5714 * optabs-query.c (can_mult_highpart_p): Likewise.
5715 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5716 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5717 (vectorizable_bswap): Likewise.
5718 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5719 explicit encoding for the power-of-2 permutes.
5720 (vect_permute_store_chain): Likewise.
5721 (vect_grouped_load_supported): Likewise.
5722 (vect_permute_load_chain): Likewise.
5724 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5726 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5727 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5728 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5729 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5730 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5731 (vect_gen_perm_mask_any): Likewise.
5733 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5735 * int-vector-builder.h: New file.
5736 * vec-perm-indices.h: Include int-vector-builder.h.
5737 (vec_perm_indices): Redefine as an int_vector_builder.
5738 (auto_vec_perm_indices): Delete.
5739 (vec_perm_builder): Redefine as a stand-alone class.
5740 (vec_perm_indices::vec_perm_indices): New function.
5741 (vec_perm_indices::clamp): Likewise.
5742 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5743 (vec_perm_indices::new_vector): New function.
5744 (vec_perm_indices::new_expanded_vector): Update for new
5745 vec_perm_indices class.
5746 (vec_perm_indices::rotate_inputs): New function.
5747 (vec_perm_indices::all_in_range_p): Operate directly on the
5748 encoded form, without computing elided elements.
5749 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5750 encoding. Update for new vec_perm_indices class.
5751 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5752 the given vec_perm_builder.
5753 (expand_vec_perm_var): Update vec_perm_builder constructor.
5754 (expand_mult_highpart): Use vec_perm_builder instead of
5755 auto_vec_perm_indices.
5756 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5757 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5758 or double series encoding as appropriate.
5759 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5760 vec_perm_indices instead of auto_vec_perm_indices.
5761 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5762 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5763 (vect_permute_store_chain): Likewise.
5764 (vect_grouped_load_supported): Likewise.
5765 (vect_permute_load_chain): Likewise.
5766 (vect_shift_permute_load_chain): Likewise.
5767 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5768 (vect_transform_slp_perm_load): Likewise.
5769 (vect_schedule_slp_instance): Likewise.
5770 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5771 (vectorizable_mask_load_store): Likewise.
5772 (vectorizable_bswap): Likewise.
5773 (vectorizable_store): Likewise.
5774 (vectorizable_load): Likewise.
5775 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5776 vec_perm_indices instead of auto_vec_perm_indices. Use
5777 tree_to_vec_perm_builder to read the vector from a tree.
5778 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5779 vec_perm_builder instead of a vec_perm_indices.
5780 (have_whole_vector_shift): Use vec_perm_builder and
5781 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5782 truncation to calc_vec_perm_mask_for_shift.
5783 (vect_create_epilog_for_reduction): Likewise.
5784 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5785 from auto_vec_perm_indices to vec_perm_indices.
5786 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5787 instead of changing individual elements.
5788 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5789 the vector in d.perm.
5790 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5791 from auto_vec_perm_indices to vec_perm_indices.
5792 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5793 instead of changing individual elements.
5794 (arm_vectorize_vec_perm_const): Use new_vector to install
5795 the vector in d.perm.
5796 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5797 Update vec_perm_builder constructor.
5798 (rs6000_expand_interleave): Likewise.
5799 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5800 (rs6000_expand_interleave): Likewise.
5802 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5804 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5805 to qimode could truncate the indices.
5806 * optabs.c (expand_vec_perm_var): Likewise.
5808 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5810 * Makefile.in (OBJS): Add vec-perm-indices.o.
5811 * vec-perm-indices.h: New file.
5812 * vec-perm-indices.c: Likewise.
5813 * target.h (vec_perm_indices): Replace with a forward class
5815 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5816 * optabs.h: Include vec-perm-indices.h.
5817 (expand_vec_perm): Delete.
5818 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5819 (expand_vec_perm_const): Declare.
5820 * target.def (vec_perm_const_ok): Replace with...
5821 (vec_perm_const): ...this new hook.
5822 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5823 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5824 * doc/tm.texi: Regenerate.
5825 * optabs.def (vec_perm_const): Delete.
5826 * doc/md.texi (vec_perm_const): Likewise.
5827 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5828 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5829 expand_vec_perm for constant permutation vectors. Assert that
5830 the mode of variable permutation vectors is the integer equivalent
5831 of the mode that is being permuted.
5832 * optabs-query.h (selector_fits_mode_p): Declare.
5833 * optabs-query.c: Include vec-perm-indices.h.
5834 (selector_fits_mode_p): New function.
5835 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5836 is defined, instead of checking whether the vec_perm_const_optab
5837 exists. Use targetm.vectorize.vec_perm_const instead of
5838 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5839 fit in the vector mode before using a variable permute.
5840 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5841 vec_perm_indices instead of an rtx.
5842 (expand_vec_perm): Replace with...
5843 (expand_vec_perm_const): ...this new function. Take the selector
5844 as a vec_perm_indices rather than an rtx. Also take the mode of
5845 the selector. Update call to shift_amt_for_vec_perm_mask.
5846 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5847 Use vec_perm_indices::new_expanded_vector to expand the original
5848 selector into bytes. Check whether the indices fit in the vector
5849 mode before using a variable permute.
5850 (expand_vec_perm_var): Make global.
5851 (expand_mult_highpart): Use expand_vec_perm_const.
5852 * fold-const.c: Includes vec-perm-indices.h.
5853 * tree-ssa-forwprop.c: Likewise.
5854 * tree-vect-data-refs.c: Likewise.
5855 * tree-vect-generic.c: Likewise.
5856 * tree-vect-loop.c: Likewise.
5857 * tree-vect-slp.c: Likewise.
5858 * tree-vect-stmts.c: Likewise.
5859 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5861 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5862 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5863 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5864 (aarch64_vectorize_vec_perm_const): ...this new function.
5865 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5866 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5867 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5868 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5869 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5870 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5871 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5873 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5874 check for NEON modes.
5875 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5876 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5877 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5878 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5880 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5881 the old VEC_PERM_CONST conditions.
5882 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5883 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5884 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5885 (ia64_vectorize_vec_perm_const_ok): Merge into...
5886 (ia64_vectorize_vec_perm_const): ...this new function.
5887 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5888 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5889 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5890 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5891 * config/mips/mips.c (mips_expand_vec_perm_const)
5892 (mips_vectorize_vec_perm_const_ok): Merge into...
5893 (mips_vectorize_vec_perm_const): ...this new function.
5894 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5895 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5896 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5897 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5898 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5899 (rs6000_expand_vec_perm_const): Delete.
5900 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5902 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5903 (altivec_expand_vec_perm_const_le): Take each operand individually.
5904 Operate on constant selectors rather than rtxes.
5905 (altivec_expand_vec_perm_const): Likewise. Update call to
5906 altivec_expand_vec_perm_const_le.
5907 (rs6000_expand_vec_perm_const): Delete.
5908 (rs6000_vectorize_vec_perm_const_ok): Delete.
5909 (rs6000_vectorize_vec_perm_const): New function.
5910 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5911 an element count and rtx array.
5912 (rs6000_expand_extract_even): Update call accordingly.
5913 (rs6000_expand_interleave): Likewise.
5914 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5915 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5916 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5917 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5918 (rs6000_expand_vec_perm_const): Delete.
5919 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5920 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5921 (altivec_expand_vec_perm_const_le): Take each operand individually.
5922 Operate on constant selectors rather than rtxes.
5923 (altivec_expand_vec_perm_const): Likewise. Update call to
5924 altivec_expand_vec_perm_const_le.
5925 (rs6000_expand_vec_perm_const): Delete.
5926 (rs6000_vectorize_vec_perm_const_ok): Delete.
5927 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5928 reference to the SPE evmerge intructions.
5929 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5930 an element count and rtx array.
5931 (rs6000_expand_extract_even): Update call accordingly.
5932 (rs6000_expand_interleave): Likewise.
5933 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5934 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5936 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5938 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5940 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5941 vector mode and that that mode matches the mode of the data
5943 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5944 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5945 directly using expand_vec_perm_1 when forcing selectors into
5947 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5949 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5951 * optabs-query.h (can_vec_perm_p): Delete.
5952 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5953 * optabs-query.c (can_vec_perm_p): Split into...
5954 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5955 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5956 particular selector is valid.
5957 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5958 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5959 (vect_grouped_load_supported): Likewise.
5960 (vect_shift_permute_load_chain): Likewise.
5961 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5962 (vect_transform_slp_perm_load): Likewise.
5963 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5964 (vectorizable_bswap): Likewise.
5965 (vect_gen_perm_mask_checked): Likewise.
5966 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5967 implementations of variable permutation vectors into account
5968 when deciding which selector to use.
5969 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5970 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5971 with a false third argument.
5972 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5973 to test whether the constant selector is valid and can_vec_perm_var_p
5974 to test whether a variable selector is valid.
5976 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5978 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5979 * optabs-query.c (can_vec_perm_p): Likewise.
5980 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5981 instead of vec_perm_indices.
5982 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5983 (vect_gen_perm_mask_checked): Likewise,
5984 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5985 (vect_gen_perm_mask_checked): Likewise,
5987 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5989 * optabs-query.h (qimode_for_vec_perm): Declare.
5990 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5991 (qimode_for_vec_perm): ...this new function.
5992 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5994 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5996 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5997 does not have a conditional at the top.
5999 2018-01-02 Richard Biener <rguenther@suse.de>
6001 * ipa-inline.c (big_speedup_p): Fix expression.
6003 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6006 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6009 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6013 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6014 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6015 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6016 cond_taken_branch_cost 3->4.
6018 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6020 PR tree-optimization/83581
6021 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6022 TODO_cleanup_cfg if any changes have been made.
6025 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6026 convert_modes if target mode has the right side, but different mode
6030 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6031 last argument when extracting from CONCAT. If either from_real or
6032 from_imag is NULL, use expansion through memory. If result is not
6033 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6034 the parts directly to inner mode, if even that fails, use expansion
6038 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6039 check for bswap in mode rather than HImode and use that in expand_unop
6042 Copyright (C) 2018 Free Software Foundation, Inc.
6044 Copying and distribution of this file, with or without modification,
6045 are permitted in any medium without royalty provided the copyright
6046 notice and this notice are preserved.